A new tool for assessing psychosocial factors at work: The

A new tool for assessing psychosocial factors at work: The

Volume 38 Supplement 3 February 2010 The Copenhagen Psychosocial Questionnaire This supplement was published in honour of Professor Tage Søndergaar...

3MB Sizes 0 Downloads 0 Views

Recommend Documents

Lean manufacturing and psychosocial factors at work
May 6, 2010 - Lean manufacturing was invented for the car industry in the early 90ties. • Based on Japanese production

Interaction between physical and psychosocial work risk factors for low
My sincere thanks go also to Associate Professor Mark Stevenson, my ... Lis Ellison-Loschmann, Dr Dave McLean, Professor

Psychosocial work factors and social inequalities - BMC Public Health
population active” (EPA) and “L'Enquête sur la santé dans les collectivités canadiennes” (ESCC) [41]. Socioecon

Aerial photographs as a tool for assessing the regional - CiteSeerX
aerial photographs documenting the invasion from the outset facilitate the quantifica- ... species is easily detectable

A novel tool for assessing and summarizing the built environment
Oct 17, 2012 - Our collaboration with community partners at all stages of the tool development, data collection, and dis

Work Interrupted: A Questionnaire Assessing the Relationship
Psychological Adjustment to Cancer. Betsy A. Bates Freed. Antioch University - Santa Barbara. Follow this and additional

Nursery production: A tool for assessing - BIO Web of Conferences
promise of afterlife in the 6th century B.C.; symbolism that .... 37th World Congress of Vine and Wine and 12th General

Nursery production: A tool for assessing - BIO Web of Conferences
promise of afterlife in the 6th century B.C.; symbolism that .... 37th World Congress of Vine and Wine and 12th General

Psychosocial Risk Factors Associated With Cyberbullying Among
victims. Cybervictim-only status was associated with living in a family with other than 2 biological parents, per- ceive

Community Safety, Perceptions, and Psychosocial Factors: A Selective
Abstract. Community safety, regardless of setting and specific defining features is now a global concern of significant

Volume 38

Supplement 3

February 2010

The Copenhagen Psychosocial Questionnaire This supplement was published in honour of Professor Tage Søndergaard Kristensen, Denmark Guest Editors: Jakob B. Bjorner and Reiner Rugulies

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 3

EDITORIAL

Introduction by the Chief Editor

FINN KAMPER-JØRGENSEN An important field within Public Health is working life. Scandinavian Journal of Public Health often publishes scientific articles on working life, working conditions and working environment. The journal has agreed to honour Professor Tage Sønderga˚rd Kristensen by devoting this supplement to a number of articles related to a genuine questionnaire, ‘‘The Copenhagen Psychosocial Questionanire’’ developed by Tage Sønderga˚rd Kristensen. Professor Kristensen belongs to a fairly small international scientific elite researching psychosocial factors related to work, and he has a strong basis in methodological and theoretical work. Much of his work originates from The National Research Centre for the Working Environment in Denmark. ‘‘My main message’’ - he explains in his article – ‘‘is that a questionnaire is not just a questionnaire. A questionnaire, such as COPSOQ, is a tool for

creating theoretical insight, an eye opener for employees and employers, a way to create a new language, a bridge for building long-lasting ties between researchers and workplaces, a way to give legitimacy to the field of psychosocial factors at work, an instrument for creating new personal and professional friendships, and – last but not least – a tool for improvement of the working conditions of thousands of employees and for increasing the productivity of companies.’’ As Chief Editor of this journal I am pleased to honour a fine scientist who has been very original and productive throughout his scientific life. All articles have been exposed to the usual critical external scientific review. I thank Jakob Bue Bjørner and Reiner Rugulies for their valuable contribution as Guest Editors of this supplement.

Finn Kamper-Jørgensen Chief Editor, Scandinavian Journal of Public Health

Correspondence: Finn Kamper-Jørgensen, National Institute of Public Health, University of Southern Denmark, Øster Farimagsgade 5A, 2nd Floor, DK-1353 Copenhagen K. Denmark. E-mail: [email protected]

ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809354434

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 4–7

INTRODUCTION

Introduction to the supplement on the Copenhagen Psychosocial Questionnaire – in honour of Tage Søndergaard Kristensen

JAKOB BUE BJORNER1,2, KAREN ALBERTSEN1,2 & REINER RUGULIES1,2 1

National Research Centre for the Working Environment, Copenhagen, Denmark, and 2University of Copenhagen, Denmark

This special issue was conceived as a tribute to Tage Søndergaard Kristensen when he retired from the position as professor at the National Research Centre for the Working Environment to pursue a career as independent researcher and consultant. We chose the Copenhagen Psychosocial Questionnaire (COPSOQ) as a theme, since this questionnaire was developed by Tage, has been the focus of his research activities for the past 10 years, and synthesizes much of his thinking about the psychosocial working environment. The COPSOQ aims to provide occupational health practitioners and researchers with a tool for assessing the psychosocial working environment that is generic in the sense that it is applicable across job groups and provides a broad and detailed description of the working environment, rather than being linked to one particular theory. The questionnaire has generated broad international interest and has so far been translated into 12 languages.

Tage Søndergaard Kristensen and the occupational health field Psychology and health psychology have a long tradition of developing, translating and cross-culturally adapting standardized questionnaires. However, this approach has come comparatively late to the occupational health psychology field and has, in fact, also come late in Tage’s research career. His initial work in the occupational health field focused on empirical description of the working environment for women in different occupations [1]. While this early work was not noted internationally (being published in Danish) this work brought him fame in Denmark

and was instrumental in demonstrating to the Danish occupational health community that the psychosocial working environment was important for workers’ health and could be assessed in a valid way by questionnaires. The work was initiated before the publication of the prominent theoretical models of the field, such as the demand-control model. However, when the first papers on the demand-control model appeared, Tage immediately realized the relevance of the model and incorporated it into his thinking and empirical studies, as evidenced e.g. in his work on slaughterhouse workers [2]. Always aware of potential methodological pitfalls, Tage avoided the tautology problem by basing the evaluation of the working environment on a careful analysis by job type rather than analyzing only individual self-report exposure data. In addition, Tage also developed new theoretical perspectives, e.g. regarding disease, absence and use of medicine as coping strategies [3,4]. Tage’s next project on stress as a risk factor for cardiovascular disease used structured reviews at a time where the movement for meta-analyses hadn’t really taken off yet [5–8]. This work can be seen as contributing to improved research methodology by focusing on potential design problems. Tage’s subsequent work focused on broadening our understanding of the psychosocial work environment, by stressing the importance of factors such as meaning of work, predictability and social relations at work [9,10]. This work eventually led to the construction of the COPSOQ [11]. In addition to the COPSOQ, Tage and his colleagues also developed the Copenhagen Burnout Inventory (CBI) [12], which gives a new perspective on how to

Correspondence: Jakob Bue Bjorner, National Research Centre for the Working Environment, Lersø Parkalle´ 105, 2100 Copenhagen, Denmark. Tel: þ45 3916 5476. Fax: þ45 3916 5202. E-mail: [email protected]

ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809354842

Introduction to the special issue on COPSOQ conceptualize and measure burnout. Moreover, Tage has been a major contributor to workplace intervention research, both by designing intervention studies in Denmark [9,12–14] and by reflecting on how to conduct and evaluate intervention studies [15]. Most recently, Tage has championed the concept of social capital at work, which he sees as key to promote healthy work as well as work quality, and work productivity.

The Copenhagen Psychosocial Questionnaire In many ways, the COPSOQ reflects many of the qualities of Tage’s research. Firstly, COPSOQ reflects much of the current theoretical thinking in the field but takes a critical and independent perspective. Thus, while some items and scales were directly adapted from other questionnaires, Tage did not hesitate in revising items or measurement approaches if he felt that they could be improved. Secondly, the selection of domains in COPSOQ is informed by a direct contact with work sites, occupational health professionals, and labour organizations. Such contacts have been an integral part of Tage’s life as a researcher. Thirdly, item wording in COPSOQ shows good craftsmanship, always a hallmark of Tage’s work.

About this special issue The first three papers in this special issue focus on methodology. Pejtersen et al. describe the considerations and results that informed the development of the second version of the COPSOQ. In addition to being the prime documentation of COPSOQ II, the paper also provides a short description of all COPSOQ domains and psychometric results, in particular with regard to the newly developed scales. Psychometrics, in particular reliability analysis, is the focus of the paper by Thorsen and Bjorner. They argue that internal consistency reliability coefficients (i.e. Cronbach’s alpha) may not be appropriate for all COPSOQ scales, since many scale items must be regarded as causal rather than effect indicators. For this reason, reliability was evaluated by a test–retest approach. Another aspect of validity is interpretability. The second paper by Pejtersen et al. focuses on one aspect of interpretability, namely the determination of when an observed score change or score difference is of practical or clinical significance. Large scale studies have the statistical power to detect even very small

5

differences that may not be of practical significance. Thus, determining the minimally important difference (MID) is crucial to the planning of studies and interpretation of results. With inspiration from outcomes research, Pejtersen et al. use respondents’ self-evaluation of change as a benchmark for establishing the MID. For evaluation of the psychosocial working environment, the evaluation of validity is tricky. For most concepts, no external gold standard exists and it may be argued that the important issue is the perceived psychosocial working environment rather than the objective psychosocial working environment, if such can even be meaningfully defined. In the absence of a gold standard, validity is usually inferred from the associations that questionnaire responses have with other measures. Such results may on one hand be interpreted as substantive empirical findings and on the other hand as evidence of the validity of the questionnaire. This dualism is evidenced in the next section of five papers, all of which are longitudinal studies concerning the association between COPSOQ scales and various outcomes. The paper by Rugulies et al. concerns COPSOQ scales as predictors of register-based long-term sickness absence in a sample of the general population. The paper by Clausen et al. also examines predictors of register-based sickness absence, but focuses on employees in the eldercare sector. The paper focuses on two COPSOQ scales that are considered indicators of positive work-related states: meaning of work and commitment to the workplace. The paper by Burr et al. uses assessment of vitality and mental health as outcomes in a longitudinal study with particular focus on the potential extra benefit of examining more domains than the traditional domains of demand-control and effort-reward when studying potential psychosocial work-related health risk factors. Li et al. study nurses in China and use intention-to-quit the nursing profession as the outcome variable in both cross-sectional and longitudinal analyses. The paper by Albertsen et al. focuses on knowledge workers and uses cognitive stress symptoms as the endpoint variable in longitudinal analyses. The final paper in this section, by Bjorner and Pejtersen, differs from the rest of the papers in that it examines each scale item, relative to the rest of the scale, as predictor or outcome variable in analyses with important workplace or outcomes variables. Bjorner and Pejtersen argue that such tests of differential item functioning and differential item effect constitute the most appropriate approach to the evaluation of the construct validity of the COPSOQ.

6

J. B. Bjorner et al.

The final section of papers deals with the application of COPSOQ in research and policy analysis. In these papers, the work environment scales are generally considered the dependent variable. Aust et al. use COPSOQ to test the success of a workplace intervention study. Their study demonstrates how quantitative and qualitative methods can supplement each other to improve the understanding of a research problem – in this case, why the intervention had a negative impact on the working environment. The short communication by Nu¨bling and Hasselhorn uses the German COPSOQ experience to discuss the construction of a large database from workplace surveillance studies, and how such a database can enhance the interpretation of workplace surveillance data – even in the absence of a representative sample. The paper by Llorenz et al. evaluates the associations between COPSOQ scales and indicators of labour management practices. The paper by Moncada et al. compares the social class gradient in COPSOQ scales between Denmark and Spain. This paper is the first to use COPSOQ for cross-national comparative studies. Further applications of this kind will hopefully be made possible by the international collaboration on COPSOQ studies. The special issue concludes with a commentary to all papers by Tage Søndergaard Kristensen. We asked Tage to comment on the papers, but basically gave him freedom to write whatever he wanted. Tage discusses the past, present, and future of the COPSOQ – the latter with particular attention to the concept of social capital. Making COPSOQ the unifying theme of this special issue may at first seem like a boring technical approach. However, we hope to show that the focus on one particular methodology, such as the COPSOQ, provides a useful framework for interpretation of empirical studies, for conceptual discussions and for practical applications. However, from the perspective of a tribute to Tage, the choice of the COPSOQ theme had one important disadvantage: it severely limited the potential group of authors, excluding many friends, colleagues, and former students who would otherwise have been natural contributors. We apologize for this choice, but note that many old friends, colleagues, and students have generously acted as reviewers for the papers. Since we asked the reviewers to be rigorous and critical – and they obeyed – they need to remain anonymous, but we are deeply thankful for their help and for all the other help we have received in putting the special issue together. The willingness of so many researchers in the field to contribute

and help is perhaps a testimony to the respect and affection that Tage enjoys on both a professional and personal level. In particular, we would like to thank Palle Ørbæk, director of the National Research Centre for the Working Environment, for funding this special issue and supporting the idea from the start, and Finn Kamper-Jørgensen, editor of the Scandinavian Journal of Public Health, who immediately committed the journal to the project. Finally, we would like to thank Vibeke Rosendal, who has dealt with all the technical issues of the review process. A final pitfall of a tribute issue like this is to be too much in awe of the person who has inspired all the work. We believe that uncritical praise of Tage’s ideas would be very much against the spirit of his own work. Thus, we have encouraged authors to present original research and critical assessment. We leave it to the reader to decide to which degree we have succeeded.

References [1] Kristensen TS. Kvinders helbred og arbejde [Women’s health status and occupation]. København: FADL’s forlag; 1978. [2] Kristensen TS. Arbejdsmiljø, stress og helbred i den danske slagteribranche [Working environment, stress and health status among Danish slaughterhouse workers]. København: FADL’s forlag; 1994. [3] Kristensen TS. Use of medicine as a coping strategy among Danish slaughterhouse workers. J Soc Admin Pharmacy 1991;8(2):53–65. [4] Kristensen TS. Sickness absence and work strain among Danish slaughterhouse workers: an analysis of absence from work regarded as coping behaviour. Soc Sci Med 1991; 32(1):15–27. [5] Kristensen TS. Cardiovascular diseases and the work enviroment. A critical review of the epidemiologic literature on nonchemical factors. Scand J Work Environ Health 1989; 15:165–79. [6] Kristensen TS. Cardiovascular diseases and the work environment. A critical review of the epidemiologic literature on chemical factors. Scand J Work Environ Health 1989; 15(4):245–64. [7] Kristensen TS. The demand-control-support model: methodological challenges for future research. Stress Medicine 1995;11(1):17–26. [8] Kristensen TS. Job stress and cardiovascular disease: a theoretic critical review. J Occup Health Psychol 1996;1(3):246–60. [9] Nielsen ML, Rugulies R, Christensen KB, Smith-Hansen L, Bjorner JB, Kristensen TS. Impact of the psychosocial work environment on registered absence from work: a two-year longitudinal study using the IPAW cohort. Work & Stress 2002;18(4):323–35. [10] Borritz M, Bultmann U, Rugulies R, Christensen KB, Villadsen E, Kristensen TS. Psychosocial work

Introduction to the special issue on COPSOQ characteristics as predictors for burnout: findings from 3-year follow up of the PUMA Study. J Occup Environ Med 2005;47(10):1015–25. [11] Kristensen TS, Hannerz H, Hogh A, Borg V. The Copenhagen Psychosocial Questionnaire – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31(6):438–49. [12] Borritz M, Rugulies R, Bjorner JB, Villadsen E, Mikkelsen OA, Kristensen TS. Burnout among employees

7

in human service work: design and baseline findings of the PUMA study. Scand J Public Health 2006;34(1):49–58. [13] Nielsen ML, Kristensen TS, Smith-Hansen L. The intervention project on absence and well-being (IPAW): design and results from the baseline of a 5-year study. Work & Stress 2002;16(3):191–206. [14] Kristensen TS. Workplace intervention studies. Occup Med 2000;15(2):293–306. [15] Kristensen TS. Intervention studies in occupational epidemiology. Occup Environ Med 2005;62(3):205–10.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 8–24

ORIGINAL ARTICLE

The second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II)

JAN HYLD PEJTERSEN, TAGE SØNDERGAARD KRISTENSEN, VILHELM BORG & JAKOB BUE BJORNER National Research Centre for the Working Environment, Copenhagen, Denmark

Abstract Aims: The aim of the present paper is to present the development of the second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II). Methods: The development of COPSOQ II took place in five main steps: (1) We considered practical experience from the use of COPSOQ I, in particular feedback from workplace studies where the questionnaire had been used; (2) All scales concerning workplace factors in COPSOQ I were analyzed for differential item functioning (DIF) with regard to gender, age and occupational status; (3) A test version of COPSOQ II including new scales and items was developed and tested in a representative sample of working Danes between 20 and 59 years of age. In all, 3,517 Danish employees participated in the study. The overall response rate was 60.4%; (4) Based on psychometric analyses, the final questionnaire was developed; and (5) Criteria-related validity of the new scales was tested. Results: The development of COPSOQ II resulted in a questionnaire with 41 scales and 127 items. New scales on values at the workplace were introduced including scales on Trust, Justice and Social inclusiveness. Scales on Variation, Work pace, Recognition, Work-family conflicts and items on offensive behaviour were also added. New scales regarding health symptoms included: Burnout, Stress, Sleeping troubles and Depressive symptoms. In general, the new scales showed good criteria validity. All in all, 57% of the items of COPSOQ I were retained in COPSOQ II. Conclusions: The COPSOQ I concept has been further developed and new validated scales have been included.

Key Words: Psychosocial factors, psychosocial work environment, questionnaire, stress, survey

Background The Copenhagen Psychosocial Questionnaire (COPSOQ I) was developed in 1997 to satisfy the need of Danish work environment professionals and researchers for a standardized and validated questionnaire that covered a broad range of psychosocial factors [1,2]. It was developed in three versions of different lengths: a long version for research use, a medium-length version for work environment professionals, and a short version for the workplace. This questionnaire concept has now become the national Danish standard for assessing psychosocial work environment, and both the short and the middle version questionnaire are widely used by workplaces and work environment professionals. For example, COPSOQ I is a standard choice when Danish

companies perform their mandatory workplace risk assessment, which is required every third year and needs to include the psychosocial work environment [3]. The workplaces benchmark themselves against the national average for the different COPSOQ I scales based on the population study from 1997 [4]. The COPSOQ I questionnaire was developed based on the following principles and theoretical considerations [1]: (i) the questionnaire should cover all important aspects of the psychosocial work environment stressors as well as resources, (ii) the questionnaire should be theory-based, but not attached to one single theory, (iii) the dimensions of the questionnaire should be related to different analytical levels (company, department, job, person-work interface, and individual), (iv) the questionnaire should be generic.

Correspondence: Jan Hyld Pejtersen, National Research Centre for the Working Environment, Lersø Parkalle´ 105, DK 2100 Copenhagen, Denmark. Tel: þ45 39 16 52 99. Fax: þ45 39 16 52 01. E-mail: [email protected] (Accepted 31 August 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809349858

Second version of the Copenhagen Psychosocial Questionnaire So far no single theory or model covers all important aspects of the psychosocial work environment but the seven major theories on psychosocial factors at work show considerable overlap of dimensions [5]. The COPSOQ I included a majority of the main dimensions of the seven theories in occupational health psychology but also lacked some important factors related to work: reward, justice, trust and discrimination [1]. The COPSOQ I questionnaires have now been translated into several languages: Chinese, English, Flemish, German, Japanese, Malaysian, Norwegian, Persian, Portuguese, Spanish, Swedish and Turkish. Especially in Spain and Germany the questionnaire has been adopted as a standard for measuring the psychosocial work environment [6,7]. COPSOQ I scales have been used in several large Danish and international studies since 2000. These studies concern the work environment for: women [8], human service workers [9], computer users [10], dentists [11], correctional officers [12], hospital workers [13], pig-farm workers [14], office workers [15] and also workplace interventions [16]. Also, the Danish Work Environment Cohort Study (DWECS) [17], which has been performed every fifth year since 1990, has used COPSOQ I scales. Therefore COPSOQ I scales have been used in analyses of shift work [18], violence and stress [19], depressive symptoms [20], sickness absence [21] and early retirement [22]. The extensive use of the questionnaire, both in research and as a practical tool for assessing the psychosocial work environment at workplace level, has convinced us that the concept has to be maintained and further developed. The purpose of the present paper is to describe the COPSOQ II questionnaire and to clarify and explain the changes made from the previous version, the COPSOQ I.

Methods The development of COPSOQ II took place in five main steps: (1) We considered practical experience from the use of COPSOQ I, in particular feedback from workplace studies that had used the COPSOQ I. (2) All scales concerning workplace factors in COPSOQ I [1] were analyzed for differential item functioning (DIF) [23] with regard to gender, age and occupational status. An example of such an analysis has been published previously [2]. (3) A test version of COPSOQ II was developed and tested in a representative sample of working Danes between 20 and 59 years of age.

9

(4) Based on psychometric analyses, the final COPSOQ II was developed. (5) Criteria-related validity of the new scales was tested.

The sample, data collection procedures, item development, and psychometric and statistical analysis are described below. COPSOQ II study sample The total sample included 8,000 adult respondents randomly selected from the Danish Centralized Civil Register (in Danish CPR). On their change of address form, Danish citizens have the possibility of indicating whether they would like to have survey exemption [17], hence, when the sample was drawn, approximately 10% of the population had survey exemption – in particular the age group 20–29 years [24]. Survey exemption and a lower response rate in the youngest age group have led to some underrepresentation of the age group 20–29 in our sample. COPSOQ II study procedure The respondents received a questionnaire and a stamped response envelope by mail. Non-respondents received two mailed reminders, the second one with a new questionnaire. Non-respondents were contacted a third time by telephone and asked to fill in the questionnaire and if necessary a new questionnaire was mailed to the respondents. The respondents could also choose to fill in the questionnaire electronically on a website. This option was used by 10.4% of the respondents. The study took place in the autumn and winter 2004/2005. Of the 8,000 selected participants 166 were excluded from the study: 12 had emigrated, 50 had unknown addresses, 62 were mentally handicapped, 37 were abroad for a longer period, two were dead and three persons were also in the COPSOQ I cohort. Furthermore, 53 persons filled in the questionnaire but had too many missing values or inconsistent data for both gender and day of birth compared to the Civil Register and were regarded as having invalid responses. We received a total of 4,732 valid responses corresponding to a response rate of 60.4% – 1,215 respondents indicated that they were not in the work force or that they were self employed, leaving us with a final sample of 3,517 wage earners. In the statistical analyses of the scales we used all 4,732 respondents for the scales on health and self-efficacy whereas the sample of 3,517 wage earners was used for analyzing the work environment scales. The Civil Register provided data on the respondents’ age and gender whereas the questionnaire

10

J. H. Pejtersen et al.

contained a number of background questions including: working hours, industry, occupation, education, and socioeconomic status. Job groups were classified on the basis of selfreported information on occupation, industry, education and socioeconomic status using the 1986 Danish extended version of the International Standard Classification of Occupations (ISCO) [25]. In all, 56 major job groups with more than 20 employees in each group were identified based on the three first digits of the modified ISCO code supplied with information on education, socioeconomic status and industry. Social class was classified according to the European Classification of Social Class based on the three digits ISCO88 code [26] (For details, see Moncada in this issue [27]). Characteristics of the study population are given in Table I. Study samples for testing criteria related validity Criteria-related validity for the scales with regard to the company level was analyzed by looking at the relation between self-reported sickness absence and Table I. Characteristics of the study population. % Number of respondents Women Average age, years Age distribution 20–29 years 30–39 years 40–49 years 50–59 years Social class 1 Higher professionals and managers 2 Lower professionals and managers 3 Higher clerical, services and sales workers 4 Small employers and self-employed 5 Farmers 6 Lower supervisors and technicians 7 Lower clerical, services and sales workers 8 Skilled workers 9 Semi- and unskilled workers Not classified Sectors Private Public Semi-public Not classified Working hours per week <30 30–34 35–39 40–44 444 Not stated

3,517 52.6 42.3 13.5 26.4 30.3 29.8 12.8 17.9 21.3 0 0 0.4 16.5 9.8 17.1 4.1 53.8 38.9 3.8 3.5 7.1 8.5 39.3 21.1 21.1 2.9

scale values at the company level. The data was taken from a study of 10,600 care workers in the long-term care sector. For further details about the study, see Winsløw and Borg [28]. The care workers were employed in 310 different organizational units within 40 different municipalities. The scales on individual factors were analyzed by relating scale values for individuals to long-term sickness absence derived from the national register on social transfer payment DREAM [29]. The DREAM register contains information on the compensation that employers receive when their employees are sick. The employers are entitled to get compensation if their employees are sick for more than 14 days. Structure of the questionnaires Like COPSOQ I, COPSOQ II is available in three versions of different length but the structure was changed. The number of items in the scales for the long and medium questionnaire was kept the same in both questionnaires, but in the medium questionnaire the number of scales was reduced from 41 to 28, (see Table II). In the short questionnaire, the scales were generally based on two questions and the number of scales was further reduced to 23. In COPSOQ II, we aimed at a scale length of three to four items. In our experience, this scale length represents a reasonable trade-off between precision and response burden. Because our test questionnaire contained more than four items for many scales, we used the psychometric analysis to select the best items for each scale. Psychometric and statistical analysis The items in COPSOQ II were analyzed using explorative factor analyses, separately within each major domain. The number of factors was decided based on Eigen value analysis and interpretable factor loadings. DIF analyses were performed on selected scales using the logistic regression approach [2]. Internal consistency reliability was analyzed using Cronbach’s alpha. Floor and ceiling effects, defined as the proportion of respondents selecting the lowest (floor) and highest (ceiling) response options for all items in a scale, were determined for all scales. Each item was scored 0–100 (i.e. 0, 25, 50, 75, and 100 for a five response category item). The scale score was computed as the mean item score. If respondents had answered less than half of the questions in the particular scale, the scale score was set to missing. Each scale was scored in the direction indicated by the scale name.

Work-individual Interface

Interpersonal relations and leadership

Work organization and job contents

Demands at work

Domain

4 – 8

3 2 5 10

Emotional demands

Demands for hiding emotions Sensory demands Influence

4 4 8 4

Quality of leadership 9 > > > =

Role clarity Role conflicts

2 2 3 4 7

6



Recognition (Reward)

Social support from supervisor Social support > > > from colleagues ; Feedback Social relations Social community at work Job insecurity Job satisfaction

– 2

4 2

2 – 3 4 6

> > > > ;

9 4> > > > =

8

3 4

5

2 3

– 3 4

5

7

Commitment to the workplace Degrees of freedom at work Predictability

Possibilities for development Variation Meaning of work

4 6

– 8

Work pace Cognitive demands 4

5

II-test

7

I

Quantitative demands

Dimension

– – 3/0 0/0 4/1

3/0

3 – – 3 4 4

3/2

4/2

3/2 4/0

3/2

– 2/2

4/2

0/0 3/2

4/2

– 4/2

0/0

4/2

3/2 0/0

4/2

II middle/ short

3

4

3 4

3

– 2

4

2 3

4

– 4

3

4

3 4

4

II-long

COPSOQ version

– – JDC W W

JD

JD

D

JD JD

DC

– D

IJC

J IJ

J

– J

JD

J

J J

J

Level of dimension

RE1 Recognised by management*; RE2 Respected by management; RE3 Treated fairly* CL1 Clear objectives*; CL2 Responsibility; CL3 Expectation* CO1 Mixed acceptance; CO2 Contradictory demands; CO3 Do things wrongly; CO4 Unnecessary tasks QL1 Development opportunities; QL2 Prioritise job satisfaction*; QL3 Work planning*; QL4 Solving conflicts SC1 Support supervisor*; SC2 Supervisor listens to problems*; SC3 Supervisor talks about performance SS1 Support colleagues*; SS2 Colleagues listen to problems*; SS3 Colleagues talk about performance Included in support scales Not included SW1 Atmosphere; SW2 Cooperation; SW3 Community JI1 Unemployed; JI2 Redundant; JI3 Finding new job; JI4 Transferred JS1 Work prospects; JS2 Work conditions; JS3 Work abilities; JS4 Job in general*

Not included IN1 Influence work*; IN2 Say in choosing colleges; IN3 Amount of work*; IN4 Influence work task PD1 Take initiative*; PD2 Learning new things*; PD3 Use skills; PD4 Develop skills VA1 Work varied; VA2 Do things over and over again MW1 Work meaningful*; MW2 Work important*; MW3 Motivated and involved CW1 Enjoy telling others; CW2 Workplace great importance*; CW3 Recommend a friend*; CW4 Looking for work elsewhere Not included PR1 Informed about changes*; PR2 Information to work well*

QD1 Work piles up; QD2 Complete task; QD3 Get behind*; QD4 Enough time* WP1 Work fast; WP2 High pace*; WP3 High pace necessary* CD1 Eyes on lots of things; CD2 Remember a lot; CD 3 New ideas; CD4 Difficult decisions ED1 Emotional disturbing*; ED2 Relate to other people’s problems*; ED3 Emotionally demanding; ED4 Emotionally involved HE1 Treat equally; HE2 Hide feelings; HE 3 Emotionally involved

Items, abbreviation and short label

Table II. Dimension and items in the COPSOQ questionnaire. The scales and items in italics are new. Scales and items in grey are dropped from COPSOQ II.

Second version of the Copenhagen Psychosocial Questionnaire 11

2 9

9 7

– –

– –

Family-work conflict 9 Trust regarding > > = management Mutual trust between > > ; employees Justice

Social inclusiveness

7 4 8

– – – 7 4 5 4 8 –

9 2 2 2 1 1 1 – 1 1 1 36 147

Stress Sleeping troubles

Depressive symptoms

Somatic stress symptoms

Cognitive stress symptoms

Mental health Vitality Behavioural stress Self-efficacy

Sense of coherence Problem focused coping Selective coping Resigning coping Sexual harassment Threats of violence Physical violence Bullying Unpleasant teasing Conflicts and quarrels Gossip and slander Dimensions Items

– – – – 1 1 1 1 1 1 1 40 169

– – – 7

4

6

1 6

5 –

General health perception Burnout



4

II-test



I

Work-family conflict

Dimension

:

– – – – 1 1 1 1 1 1 1 41 127

– – – 6

4

4

4

4 4

1 4

4

4

3

2 8 < 4

4

II-long

COPSOQ version

1/1 1/1 1/1 1/1 0/0 0/0 0/0 28/23 87/40

– – –

– – – 0/0

0/0

0/0

0/0

4/2 4/0

1/1 4/2

0/0

4/2

3/0

0/0 4/2

4/2

II middle/ short

– – – – JW JW JW W W W W

– – – I

I

I

I

I I

I I

C

C

C

W C

W

Level of dimension Items, abbreviation and short label

WF1 Being in both places; WF2 Energy conflict*; WF3 Time conflict*;WF4 Family think you work too much FW1 Energy conflict; FW2 Time conflict TM1 Management trust employees*; TM2 Employees trust information*; TM3 Management withhold information; TM4 Employees express views TE1 Colleagues withhold information; TE2 Withhold information management; TE3 Trust colleagues JU1 Conflicts resolved fairly*; JU2 Employees appreciated; JU3 Suggestions treated seriously; JU4 Work distributed fairly* SI1 Gender discrimination; SI2 Race/religion discrimination; SI3 Age discrimination; SI4 Health discrimination GH1 Health all in all* BO1 Worn out*; BO2 Physically exhausted; BO3 Emotionally exhausted*; BO4 Tired ST1 Problems relaxing; ST2 Irritable*; ST3 Tense; ST4 Stressed* SL1 Slept badly; SL2 Hard to sleep; SL3 Woken up early; SL4 Woken up several times DS1 Sadness; DS2 Lack of self-confidence; DS3 Feel guilty; DS4 Lack of interest in daily activity SO1 Stomach ache; SO2 Headache; SO3 Palpitations; SO4 Muscle tension CS1 Problems concentrating; CS2 Difficult thinking clearly; CS3 Difficult taking decisions; CS4 Difficult remembering Not included Not included Not included SE1 Solve problems; SE2 Achieving what I want; SE3 Reach objectives; SE4 Handle unexpected events; SE5 Several ways solving problems; SE6 Usually manage Not included Not included Not included Not included SH single item* TV single item* PV single item* BU single item* UT single item CQ single item GS single item

I, Individual level; J, Job level; D, Department level; C, Company level; W, Work-individual interphase. *Items used in the short version.

Total

Offensive behaviours

Personality

Health and well-being

Values at workplace level

Domain

Table II. Continued.

12 J. H. Pejtersen et al.

Second version of the Copenhagen Psychosocial Questionnaire Selection of scales and items The selection of scales and items for the test questionnaire was based on COPSOQ I, but a number of new scales and items were also constructed. The COPSOQ I had 20 scales covering workplace factors, six scales covering individual factors of health and well-being and four scales covering individual factors of personality. In the new questionnaire, only one scale for personality was included and a major revision was made for scales on health and wellbeing. New scales concerning values at the workplace were introduced. To avoid increasing response burden a number of scales and questions from COPSOQ I were discarded. All in all, 57% of the items of COPSOQ I were retained. The scales and items of the COPSOQ II questionnaire are given in Table II and in more details in the appendix. Unchanged scales The following COPSOQ I scales were incorporated in COPSOQ II without any changes: Meaning of work, Predictability, Role conflicts, Social community at work and Cognitive stress symptoms. Deleted scales from COPSOQ I The scales Sensory demands, Degree of freedom and Social relations were discarded mainly because they often revealed conditions that could be interpreted from the job title and therefore would be impossible to change (for example, bus drivers and teachers have few degrees of freedom; nurses have many relations to colleagues at work and truck drivers have few). Feedback at work was deleted as a separate scale and the items were included in two new scales on social support. The scales on Sense of coherence and Coping were abandoned because they had not been widely used in research projects. The revision of the health scales meant that we excluded the scale for Behavioural stress and the two SF-36 scales Mental health and Vitality. The Mental health scale correlated highly with the scale for Vitality and it contained two aspects of mental health, namely anxiety and depressive symptoms. New scales A new scale of Work pace was included in the questionnaire. The purpose of the scale was to measure the intensity aspect of the quantitative demands at work [2]. We had four items in the test questionnaire but chose to discard one of the items that had a more individual character than the others.

13

Our DIF analyses of the scale Possibilities for development showed that two items had DIF in relation to job category. These two items were originally intended to form a specific scale on Variation in COPSOQ I and we therefore formed this scale. We wanted to include a new scale on rewards at work, as we consider rewards to be a very important factor in the psychosocial work environment. However, the three components recognition, salary, and career prospects that are included under the label of rewards in the Siegrist Effort-Reward Imbalance model [30] do not necessarily reflect the same underlying quality of the work of the individual. This expectation was confirmed by the statistical analyses. We had to discard items both due to poor correlation with other items and due to content considerations. Our final scale consists of three homogeneous items but covers only one of the three sub-components of Siegrist’s reward concept, namely Recognition. Two scales were constructed on work-family conflict that reflects the direction of the conflict, work interfering with private life and private life interfering with work. The items cover two aspects – time and energy. However, very few employees felt that their work was influenced by their private life. Scales on values at the workplace are new in COPSOQ II. We included items intended to cover scales of Trust, Justice and Social inclusiveness. The purpose of these scales was to get a picture of the whole workplace (company) and not just the person’s own job or department. Trust and justice, also referred to as Social Capital [31], are important human values in the workplace [32,33] and it is our hypothesis that living up to these values has a great impact not only on the recruitment and the wellbeing of the employees but also on the social processes in the workplace. The items chosen were inspired by a number of researchers in the fields of ‘‘trust’’ (Cook and Wall [34]) and ‘‘justice’’ (Carless [35], Elovainio and Vahtera [36]). The factor analyses of the trust items showed three items loaded on a common factor about the employees’ trust in each other and their behaviour in relation to the management. The other items covered trust between management and employees. Therefore we chose to form an independent scale for Trust regarding management with four items and a scale on Mutual trust between employees with three items as in accordance with Cook and Wall [34]. The scale on Justice was formed on the basis of nine test questions. The final scale had four items and the two items that used the words ‘‘justice’’ and ‘‘respect’’, respectively, had the highest correlation with the total scale. This suggests that the scale measures what it is intended to measure.

14

J. H. Pejtersen et al.

Table III. Mean score, standard deviation, ceiling, floor, and missing values for the COPSOQ II questionnaire (n ¼ 3,517).

Quantitative demands Work pace Cognitive demands Emotional demands Demands for hiding emotions Influence Possibilities for development Variation Meaning of work Commitment to the workplace Predictability Recognition (Reward) Role clarity Role conflicts Quality of leadershipb Social support from supervisorb Social support from colleaguesc Social community at workd Job insecurity Job satisfactione Work–family conflict Family–work conflict Trust regarding management Mutual trust between employees Justice Social inclusiveness Self-rated health Burnout Stress Sleeping troubles Depressive symptoms Somatic stress symptoms Cognitive stress symptoms Self-efficacy Sexual harassment Threats of violence Physical violence Bullying Unpleasant teasing Conflicts and quarrels Gossip and slander

Cronbach alpha

Meana

SD

% Floor

% Ceiling

Range of item missing (%)

Scale missing (%)

0.82 0.84 0.74 0.87 0.57 0.73 0.77 0.50 0.74 0.76 0.74 0.83 0.78 0.67 0.89 0.79 0.70 0.85 0.77 0.82 0.80 0.79 0.80 0.77 0.83 0.63 – 0.83 0.81 0.86 0.76 0.68 0.83 0.80 – – – – – – –

40.2 59.5 63.9 40.7 50.6 49.8 65.9 60.4 73.8 60.9 57.7 66.2 73.5 42.0 55.3 61.6 57.3 78.7 23.7 65.3 33.5 7.6 67.0 68.6 59.2 67.5 66.0 34.1 26.7 21.3 21.0 17.8 17.8 67.5 2.9% 7.8% 3.9% 8.3% 8.3% 51.2% 38.9%

20.5 19.1 18.7 24.3 20.8 21.2 17.6 21.4 15.8 20.4 20.9 19.9 16.4 16.6 21.1 22.4 19.7 18.9 20.8 18.2 24.3 15.3 17.7 16.9 17.7 16.3 20.9 18.2 17.7 19.0 16.5 16.0 15.7 16.0 – – – – – – –

2.9 0.5 0.3 5.7 1.5 1.6 0.4 2.0 0.1 0.7 1.5 0.9 0.0 1.3 1.2 0.9 1.1 0.2 19.0 0.7 9.7 74.6 0.2 0.0 0.4 0.1 0.8 1.7 5.2 17.4 10.3 16.6 18.6 0.0 97.0 92.2 96.1 91.7 91.7 48.8 61.1

0.3 3.4 1.1 0.4 0.9 0.5 2.3 4.2 7.3 2.2 4.2 5.8 7.5 0.2 1.9 4.4 1.9 24.4 0.5 5.1 1.2 0.2 3.9 5.6 1.6 3.8 14.8 0.2 0.1 0.0 0.0 0.0 0.0 1.8 0.1 0.3 0.0 0.5 0.3 1.3 3.5

2.3–2.6 2.4–2.9 2.3–2.5 2.4–2.9 2.6–3.0 2.3–2.7 2.8–3.1 2.4–2.6 2.7–2.8 2.3–3.0 2.6–2.8 2.8–3.0 2.7–3.0 2.8–3.5 2.1–2.7 1.9–2.1 2.7–3.0 2.7–2.8 2.8–2.9 2.9–3.1 3.1–3.8 2.9–3.1 2.6–4.2 3.0–3.8 2.8–3.6 3.4–5.2 – 0.7–1.0 0.7–1.0 0.7 0.7–0.9 0.7–0.8 0.8–0.9 1.3–1.5 – – – – – – –

2.2 2.2 2.2 2.2 2.3 2.2 2.6 2.2 2.8 2.2 2.3 2.8 2.7 2.6 2.0 2.0 2.7 2.6 2.3 2.8 2.9 2.9 2.5 3.2 2.6 2.8 1.2 0.6 0.6 0.6 0.7 0.6 0.7 1.3 3.3 3.2 3.3 2.5 3.2 2.5 2.6

a Prevalence proportions for the single items. bn ¼ 2,719 did have a leader. cn ¼ 3,422, 95 answered not relevant. dn ¼ 3,481, 36 answered not relevant. en ¼ 3,494 23 answered not relevant.

In Danish society there has been increasing interest in the issues of social inclusiveness and social responsibility of the workplace. Therefore we included seven items on this aspect of the psychosocial work environment in our test questionnaire. Although the two items on gender and race/religion had the lowest correlation with the others, we chose to disregard statistics for the final scale and gave priority to four key domains regarding inclusiveness: Gender, ethnicity, age and health. The scale on sleeping quality from the Karolinska Sleep Questionnaire was included in the questionnaire [37]. The four items loaded clearly on the same

factor in the analyses and the scale had high internal reliability (see Table III). Furthermore, the scale has worked well in Swedish research [38] and in our own study on burnout among human service workers [9]. We included the scale for personal burnout from the Copenhagen Burnout Inventory [39], which was developed in connection with the study on burnout [9]. The questions were changed so that they fitted the COPSOQ I format and the time window of four weeks. The items Vulnerable and Cannot take it anymore were discarded as they had a very skewed response distribution and also showed weak correlations with the other items. We found that

Second version of the Copenhagen Psychosocial Questionnaire the item Emotionally exhausted loaded on the scale about Depressive symptoms but we chose to keep it in the scale since we wanted to cover emotional as well as physical fatigue. In COPSOQ II we have chosen to separate the two phenomena stress and depressive symptoms. We define stress as an individual state characterized by a combination of high arousal and displeasure. In the choice of symptoms we were inspired by Peter Warr’s circle model for psychological states from which we have chosen examples characterized by the combination of arousal and displeasure [40]. In the scale for stress, we have chosen not to combine positive and negative questions since it was our experience that the positive and negative symptoms tended to form separate scales. We included seven items in our test questionnaire and had a number of considerations regarding content as well as statistics. We retained the four items that were the most appropriate for a conceptualization of stress as an intra-individual state (see Table II). After a thorough review of the internationally acknowledged questionnaires on depression and depressive symptoms we chose to include eight items slightly modified from Bech et al. in our test questionnaire [41]. The purpose of this scale was not to try to diagnose clinical depression but to develop a simple scale measuring the degree of depressive symptoms in persons belonging to the working population. After our analyses and content considerations we excluded four items: Lacked energy because it was close to the dimension Burnout, Lacked appetite since 76% of the respondents did not have a problem with appetite, and both Bad mood and Upset since the items were too similar to the item on Sadness. We ended up with the four items covering Sadness, Lack of self-confidence, Feel guilty and Lack of interest in daily activities. In order to assess the respondents’ level of selfconfidence or faith in their own abilities to solve unexpected or difficult problems in life, we chose seven items on self-efficacy from Bandura [42]. The scale worked well statistically but we excluded the item I keep calm as it has a hidden assumption, namely that the person is always calm. In this scale, we did not give high priority to reaching a scale with four items since the scale was only to be included in the long questionnaire. A number of single items measuring offensive behaviour were included in the questionnaire, (see Table II), except for bullying they were all taken from COPSOQ I [1]. Offensive behaviour seems to be an important factor in the psychosocial work environment and is now included also in the middle and short version of the questionnaire [43].

15

Shortening of scales from COPSOQ I In order to keep the general layout of a maximum four items per scale, a number of COPSOQ I scales were reduced based on the statistical analysis of DIF and by looking at the distributions of the items. For the scales Cognitive demands, Influence and Job satisfaction the analyses showed that the old scales from the medium size COPSOQ I worked quite well and these are now used in the long version of COPSOQ II. Role clarity was reduced to three items due to DIF for one of the items. We have previously found that traditional scales for quantitative demands contains two dimensions of intensity (tempo, pace) and extensity (amount of work, deadlines, workload) that ought to be separated in specific scales [2]. Thus, the quantitative demand scale was reduced from seven to four items and a new Work pace scale has been formed (see New scales). This change also reflects our experiences from using the COPSOQ I scale on quantitative demands in practical workplace surveys. For General health perception we selected only one global item, which has been used in the SF-36 [44] and in numerous other questionnaires, and has been shown to predict many different endpoints including mortality, cardiovascular diseases, hospitalizations, use of medicine, absence, and early retirement [45]. Change of items and new items on COPSOQ I scales We included new items (Relate to other people’s problems and Treat equally) in the scales for Emotional demands and Demands for hiding emotions to make the scales broader. These items performed well in the psychometric analyses. The scale Demands for hiding emotions aims at catching the essence of ‘‘emotional labour’’ where the employee is expected to keep a neutral fac¸ade regardless of the behaviour of the clients or customers [46]. We have included two new items (Recommend a friend, Looking for work elsewhere) in the scale for Commitment to the workplace. The items have been used in other studies and the last of the items can be seen as a measure of the concept ‘‘intention to quit’’ [47]. In COPSOQ I, the Social support scale included items on support from supervisors and colleagues. In discussions with workplaces, the respondents felt that support from supervisors and support from colleagues were two different things. Also, our statistical analyses of scale showed that the items on colleagues correlated poorly with the items on supervisors. Finally, items on support from colleagues had strong correlations with items on Feedback at work

16

J. H. Pejtersen et al.

from colleagues, while items on support from supervisors had strong correlations with items on Feedback at work from supervisors. We ended up with two scales on social support at work including items on feedback, namely one for Social support from colleagues and one for Social support from supervisors. The items on the Somatic stress symptoms have been changed slightly compared with COPSOQ I in order to fit the general layout. We included two new symptoms, Nauseous and Headache, and removed Chest pain, Short of breath and Tendency to sweat in the test questionnaire. The final four-item scale on Somatic stress included the items with the least skew and highest interim correlation: Stomach ache; Headache; Palpitations; Muscle tension. Also, this scale did not show DIF in relation to gender.

Change of response categories in COPSOQ I scales We changed the response options for Insecurity at work from yes-no to five response options as with most of the other items in the questionnaire. The categories for the scales on Somatic and Cognitive stress were changed to: All the time; A large part of the time; Part of the time; A small part of the time; Not at all. These categories were used for most of the health scales (see appendix).

Criteria-related validity of the new scales We looked at criteria validity only for the new scales that have not been used in other studies before. The scales were constructed to cover different analytical levels: Scales mainly related to job factors (Work pace, Variation), scales mainly related to the company level (Justice, Trust and Social inclusiveness), scales mainly related to the department level (Recognition), individually based scales (Stress and Depression) and scales related to work–individual interface (Workfamily conflict), see Table II. The job-related scales Work pace and Variation were analyzed by looking at their ability to discriminate among job groups. The scale on Work-family conflict was also analyzed this way because it has a strong element related to the job content. Analysis of variance was performed on the scales with job group as the independent variable. We hypothesised that: (1) Work pace is high for industrial workers, slaughterhouse workers etc and is low for drivers, family childcare providers, childcare workers and janitors; (2) Variation in work is high for academic groups and is low for the industrial job groups, postal workers etc; (3) Work family conflict is high for academic groups and teachers and is low for industrial groups.

Criteria-related validity for the scales on Justice was analyzed by looking at the relation between selfreported sickness absence and scale values at company level. We also analyzed Recognition this way even though the scale is more related to department level than company level. It was expected that low scale values for Justice and Recognition, respectively, were related to high rates of sickness absence for the organizational units. The mean scale scores for Justice and Recognition were calculated for each organizational unit and related to the mean number of sickness absence days for the unit within the last year. The mean number of sickness absence days was categorized into low level (0–5 days), medium level (6–20 days) and high level (more than 20 days). Logistic regression was used with the categorized variable on sickness absence as the dependent variable and the scale score as the independent variable. We calculated the odds ratio for a score difference of 10 points. We were not able to validate the other company-related scales on Trust and Social inclusiveness since we did not have a workplace study where these scales were included. Because both the scales for Burnout and Sleeping problems have been used in other studies [37–39] we decided only to look at criteria-related validity for the new scales on Stress and Depression. The hazard ratios (HR) for long-term sickness absence were calculated using the Cox regression model. We calculated the hazard ratios for scale value differences of 10 points. The analyses were adjusted for age, gender and social class. Separate analyses were performed including interaction between gender and scale value for the independent variable.

Results Scale characteristics for the dimensions in COPSOQ II are shown in Table III. The internal consistency reliability measured by Cronbach alpha was high and above 0.7 for most of the scales. However, low values were seen for the scales Demands for hiding emotions (0.57) and Variation (0.50). The proportion of missing values for the scales was between 0.6% and 3.3%. The items on offensive behaviour had the highest number of missing values. Most of the scales had low floor or ceiling effect but problems were seen for the scales: Family-work conflicts, Job-insecurity and Social community at work. The scale Family-work conflict had high floor effect (74.6%) and a very low mean value (7.6) showing that private life is not interfering with work in general. Floor effect was also seen for the scale Job insecurity (19.0%) and ceiling effect was found for the scale Social community at work (24.4%).

Second version of the Copenhagen Psychosocial Questionnaire Also the scales on health had some floor or ceiling effect indicating a high proportion of respondents with no health symptoms. The analyses of variance on the scales Work pace, Variation and Work-family conflict showed that job group was significantly (p < 0.0001) related to the scale score for all three dimensions. Table IV shows the scale scores on the dimensions for the 10 job groups with the lowest and highest scores, respectively. The table shows that the scales, in general, are able to discriminate between job groups as we had expected. As hypothesised Work pace was high for the industrial groups, slaughterhouse workers, packing and bottling plant employees and also high for mailworkers, managers in the private sector and doctors and dentists. Work pace was low for childminders, drivers and janitors as expected and also low for pre-school teachers and teachers. However, surprisingly, unskilled metal workers also had low work pace, which we do not have an explanation for. The degree of variation in the work was, as expected, low for mailworkers, bus drivers and industrial workers and high for academics, engineers and managers. Work-family conflict was low for industrial workers and high for academics, engineers, school teachers and managers. The logistic regressions showed that mean sickness absence at the organizational units was significantly related to mean score values of Justice and Recognition. For Justice, the risk of being absent due to sickness was 1.66 (confidence interval (CI) 1.12–2.40) for a mean scale score difference of 10 points. For Recognition, the odds ratio was 1.58 (CI 1.09–2.29). The Cox regression on the Stress scale showed increased risk of long-term sickness absence with increasing scale value (step of 10 points) (hazard ratio (HR) ¼ 1.16 (CI 1.11–1.21)) when adjusting for age, gender and social class. A similar result was found for the Depression scale where the risk of long-term sickness absence increased with increasing scale value (HR ¼ 1.16 (CI 1.11–1.22)) when adjusting for age, gender and social class. For both scales, the results are in the expected direction. There was no significant interaction between gender and the scale value for any of the tested models.

Discussion and conclusions Standardized generic questionnaires face an inherent conflict between conservatism and innovation. If the instrument is revised frequently, comparisons between studies are hampered since the questions will not be identical. On the other hand, if problems have been identified, revisions in order to improve

17

validity and reliability seem logical. After developing COPSOQ I version I, we decided to use the following guidelines for revisions: (1) In order to have a standardized measuring tool we would not like to make changes too often and at the most every fifth year; (2) We would only make changes in scales if tests had shown problems with scales or items or if the practical use of the scales had shown problems, as for instance with the scale of quantitative demands [2]; (3) We would delete scales which had not been used for research or practical purposes (e.g sensory demand, freedoms at work); and (4) We would include new scales that reflected the development of new theories and new perspectives (e.g Recognition, Trust, Justice, Work-family conflicts and Depressive symptoms). The development of COPSOQ II was based on theoretical considerations and on feedback from the users of the questionnaire. Following standard approaches in the field of occupational psychology and sociology, our psychometric testing used classical psychometric techniques evaluating dimensionality and internal consistency (e.g. explorative factor analyses, Cronbach’s analysis of internal reliability) as well as evaluation of criteria validity. In addition, we used modern psychometric methods such as DIF analysis. We found high Cronbach alphas for most but not all scales. Interestingly, a reliability study of COPSOQ I scales using a test–retest design have found higher reliabilities for most of the scales where Cronbach’s alphas were low in this study (Thorsen et al in this issue [48]). A possible reason for this discrepancy is the implicit logic behind Cronbach’s alpha and classical psychometrics in general. Generally, these criteria are most appropriate for so-called effect-indicator scales (i.e. scales where the items can be seen as effects of a single latent cause such as an intra-individual trait or a state like depression or anxiety) [49]. As discussed in other papers in this issue (Thorsen et al. [48] and Bjorner et al. [50]) these criteria may set too narrow limits on scales where items are not all effects of a single intraindividual state. This suggests that internal consistency may underestimate reliability for scales that are not comprised solely of effect indicators. Our analysis showed low floor and ceiling effects for most scales. From a technical perspective, this is important, since floor or ceiling effects limit the ability to show changes over time and reduce the scales’ power to predict other outcomes. However, in scales for phenomena such as violence or harassment floor (or ceiling) is unavoidable since such outcomes are rare. Generally, our tests of criteria validity supported the validity of the new scales. For the job-related scales Work pace, Variation, and Work-family conflict our overall hypotheses about which job groups had

1 2 3 4 5 6 7 8 9 10 47 48 49 50 51 52 53 54 55 56

Order

Child-minders High school teachers Bus drivers Pre-school teacher, unskilled Metal workers, unskilled Teachers, other Janitors Truck drivers Social education worker Caregivers, hospital Store managers Administrator, public health Media employees Construction workers, skilled Medical doctors and dentists Manager, private sector Packing and bottling plant employees Slaughterhouse workers Medical secretary Mailperson

Work pace 52 25 22 41 24 67 62 48 70 22 27 26 20 28 35 107 21 22 30 21

n 38.5 47.0 47.2 47.8 50.0 50.7 51.4 52.0 52.8 53.8 67.0 67.0 67.9 68.2 68.6 68.8 70.8 71.2 71.4 73.4

Score Mailperson Bus drivers Food, drink and tobacco workers Cleaners Slaughterhouse workers Packing and bottling plant employees Warehouse workers and dockworkers Metal workers, unskilled Electronic workers, unskilled Medical secretary Administrator, public health High school teachers Social worker Physical and occupational therapist Teachers, other Academics, social sciences and humanities Graduate engineers and architects Academics, natural sciences Manager, private sector Manager, public sector

Variation 21 22 29 48 22 21 41 24 22 30 26 25 28 22 67 50 62 32 107 21

n 27.4 35.1 36.2 36.5 41.5 43.8 45.0 46.7 49.4 61.4 69.2 70.0 70.1 71.0 71.1 71.8 72.0 72.7 73.5 76.8

Score

Machinists Metal workers, unskilled Cleaners Warehouse workers and dockworkers Electronic workers, unskilled Pre-school teacher, unskilled Clerk, public sector Mechanics Construction workers, skilled Aides for handicapped children or adults Social education worker Medical doctors and dentists Physical and occupational therapist Supervisors Media employees Graduate engineers and architects Academics, natural sciences Academics, social sciences and humanities Primary and secondary school teachers Manager, private sector

Work family conflict

Table IV. The 10 job groups with the lowest and highest, respectively scale score for work pace, variation and work–family conflict. The order is from low to high.

41 24 48 41 22 41 104 37 28 61 70 35 22 68 20 62 32 50 120 107

n

18.4 20.5 21.9 22.8 22.9 23.1 23.5 25.5 26.5 27.3 39.4 40.7 41.3 41.5 42.1 42.2 42.4 42.8 43.1 43.3

Score

18 J. H. Pejtersen et al.

Second version of the Copenhagen Psychosocial Questionnaire the highest and lowest scale scores, respectively, were fulfilled for the three scales in general. Unskilled (and to some degree skilled) metal workers reported lower Work pace than we expected. One explanation could be that these jobs may involve automatic processing and therefore process monitoring rather than manual work. The company level analyses of the scales on Justice showed that higher mean company score on justice was associated with low company frequency of sickness absence. These analyses ignored score variation within companies and are therefore rather conservative (for a more sophisticated multilevel approach see [51]). In our register-based individual-level analyses, the scales on Stress and Depression also showed good criteria validity by being able to predict long term sickness absence. We did not have a workplace study where the scales on Trust and Social inclusiveness had been included, so we were not able to evaluate these scales. However, other studies have shown that the dimensions Trust and Justice combined into the concept of ‘‘Social capital’’ [31] are related to human health, since decreasing social capital at work was related to low self rated health [52]. We believe that social capital at the company level is an important factor for the psychosocial work environment. An advantage of the present study is that the development of the COPSOQ I questionnaire is based on a representative national sample of wage earners in Denmark. A weakness of the current study is the somewhat low response rate (60.4%) and the disproportionately higher exclusion of young people due to general Danish survey exception policy. A previous analysis of response rates in this sample using logistic regression evaluated gender, age group and degree of urbanization [53]. No effect was found for urbanization, but the response rates were higher for women and increased with age [53]. To evaluate the possibility that the response rate differences had an impact on the reported mean values we calculated standardized regression coefficients for the regression of scale values on age group and gender. For one scale, Emotional demands, we found a standardized regression coefficient of 0.25 for gender. All other coefficients were below 0.2, suggesting that response rate differences across age and gender are unlikely to have major impact on the mean values. With regard to reliability the study by Thorsen et al. in this issue [48] showed that there were no major age and gender differences for reliability. All in all, while the low response rate is still a limitation of the study we have no indication that it had any major impact on the results. We conclude that these initial results support the validity of the COPSOQ II questionnaire. Further information on reliability, construct validity and

19

criteria validity are presented in other papers in this issue. (Construct validity, Bjorner et al. [50]; Test– retest reliability, Thorsen et al. [48]; and Predictive validity, Rugulies et al. [54]).

References [1] Kristensen TS, Hannerz H, Høgh A, Borg V. The Copenhagen Psychosocial Questionnaire – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31:438–49. [2] Kristensen TS, Bjorner JB, Christensen KB, Borg V. The distinction between work pace and working hours in the measurement of quantitative demands at work. Work & Stress 2004;18:305–22. [3] Jensen PL. Assessing assessment: the Danish experience of worker participation in risk assessment. Econ Ind Democracy 2002;23:201–28. [4] Kristensen TS, Borg V, Hannerz H. Socioeconomic status and psychosocial work environment: results from a Danish national study. Scand J Public Health Supplement 2002;59:41–8. [5] Kompier M. Job design and well-being. In: Schabracq MJ, Winnubst JAM, Cooper CL, editors. The handbook of work and health psychology, Chap. 20. Chichester: John Wiley & Sons; 2003. pp.429–54. [6] Nu¨bling M, Sto¨ssel U, Hasselhorn HM, Michaelis M, Hofmann F. Measuring psychological stress and strain at work: evaluation of the COPSOQ I Questionnaire in Germany. GMS Psycho-Social-Medicine 2006;3(Doc05):1–14. [7] Navarro A, Llorens C, Kristensen TS, Moncada i Luis S. ISTAS21: The Spanish version of the Copenhagen Psyhosocial Questionnaire (COPSOQ I) (in Spanish). Archivos de prevencio´n de riesgos laborales 2005;8:18–29. [8] Larsson A, Karlqvist L, Gard G. Effects of work ability and health promoting interventions for women with musculoskeletal symptoms: a 9-month prospective study. BMC Musculoskeletal Disorders 2008;9:105. [9] Borritz M, Rugulies R, Bjorner JB, Villadsen E, Mikkelsen OA, Kristensen TS. Burnout among employees in human service work: design and baseline findings of the PUMA study. Scand J Public Health 2006;34:49–58. [10] Arvidsson I, Axmon A, Skerfving S. Follow-up study of musculoskeletal disorders 20 months after the introduction of a mouse-based computer system. Scand J Work Environ Health 2008;34:374–80. [11] Tsutsumi A, Umehara K, Ono H, Kawakami N. Types of psychosocial job demands and adverse events due to dental mismanagement: a cross sectional study. BMC Oral Health 2007;7:3. [12] Ghaddar A, Mateo I, Sanchez P. Occupational stress and mental health among correctional officers: a cross-sectional study. J Occup Health 2008;50:92–8. [13] Aust B, Rugulies R, Skakon J, Scherzer T, Jensen C. Psychosocial work environment of hospital workers: validation of a comprehensive assessment scale. Int J Nurs Studies 2007;44:814–25. [14] Kolstrup C, Lundqvist P, Pinzke S. Psychosocial work environment among employed Swedish dairy and pig farmworkers. J Agromedicine 2008;13:23–36. [15] Pejtersen J, Allermann L, Kristensen TS, Poulsen OM. Indoor climate, psychosocial work environment and symptoms in open-plan offices. Indoor Air 2006;16:392–401.

20

J. H. Pejtersen et al.

[16] Nielsen K, Fredslund H, Christensen K, Albertsen K. Success or failure? Interpreting and understanding the impact of interventions in four similar worksites. Work & Stress 2006;20:272–87. [17] Feveile H, Olsen O, Burr H, Bach E. Danish work environment cohort study 2005: from idea to sampling design. Statistics in Transition (new series) 2007;8:441–58. [18] Bøggild H, Burr H, Tu¨chsen F, Jeppesen HJ. Work environment of Danish shift and day workers. Scand J Work Environ Health 2001;27:97–105. [19] Wieclaw J, Agerbo E, Mortensen PB, Burr H, Tuchsen F, Bonde JP. Work related violence and threats and the risk of depression and stress disorders. J Epidemiol Comm Health 2006;60(9):771–5. [20] Rugulies R, Bu¨ltmann U, Aust B, Burr H. Psychosocial work environment and incidence of severe depressive symptoms: prospective findings from a 5-year follow-up of the Danish Work Environment Cohort Study. Am J Epidemiol 2006;163:877–87. [21] Lund T, Labriola M, Christensen KB, Bu¨ltmann U, Villadsen E, Burr H. Psychosocial work environment exposures as risk factors for long-term sickness absence among Danish employees: results from DWECS/DREAM. J Occup Environ Med 2005;47:1141–7. [22] Lund T, Villadsen E. Who retires early and why? Determinants of early retirement pension among Danish employees 57–62 years. Eur J Ageing 2005;2:275–80. [23] Holland PW, Wainer H. Differential Item Functioning. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.; 1993. [24] Ortega A, Høgh A, Pejtersen JH, Olsen O. Prevalence of workplace bullying and risk groups: a representative population study. Int Arch Occup Environ Health 2009;82: 417–26. [25] The Directorate of Labor. The Danish job classification (in Danish). Copenhagen: The Directorate of Labor; 1986. [26] Rose D, Pevalin DJ, Elias P, Martin J. Towards a European socio-economic classification: final report to Eurostat of the Expert Group. London and Colchester: ONS and ISER, University of Essex; 2001. [27] Moncada S, Pejtersen JH, Navarro A, Llorens C, Burr H, Hasle P, Bjorner JB. Psychosocial work environment and its association with socioeconomic status. A comparison of Spain and Denmark. Scand J Public Health 2010;38(Suppl 3): 137–148. [28] Winsløw JH, Borg V. Resources and quality of care in services for the elderly. Scand J Public Health 2008;36:272–8. [29] Hjollund NH, Larsen FB, Andersen JH. Register-based follow-up of social benefits and other transfer payments: accuracy and degree of completeness in a Danish interdepartmental administrative database compared with a population-based survey. Scand J Public Health 2007;35:497–502. [30] Siegrist J. Adverse health effects of high-effort/low-reward conditions. J Occup Health Psychol 1996;1:27–41. [31] Coleman JS. Social capital in the creation of human capital. Am J Sociol 1988;94:95–120. [32] Handbook of organizational justice. In: Greenberg J, Colquitt JA, editors. Mahwah, New Jersey: Lawrence Erlbaum Associates, Publishers; 2005. [33] The trust process in organizations. In: Nooteboom E, Six F, editors. Cheltenham: Edward Elgar Publishing Ltd; 2003. [34] Cook J, Wall T. New work attitude measures of trust, organizational commitment and personal need non-fulfilment. J Occup Psychol 1980;53:39–52. [35] Carless SA, De Paola C. The measurement of cohesion in work teams. Small Group Res 2000;31:71–88.

[36] Elovainio M, Kivima¨ki M, Steen N, Vahtera J. Job decision latitude, organizational justice and health: multilevel covariance structure analysis. Soc Sci Med 2004;58:1659–69. [37] Harma M, Tenkanen L, Sjoblom T, Alikoski T, Heinsalmi P. Combined effects of shift work and life-style on the prevalence of insomnia, sleep deprivation and daytime sleepiness. Scand J Work Environ Health 1998;24:300–7. [38] Akerstedt T, Knutsson A, Westerholm P, Theorell T, Alfredsson L, Kecklund G. Sleep disturbances, work stress and work hours: a cross-sectional study. J Psychosom Res 2002;53:741–8. [39] Kristensen TS, Borritz M, Villadsen E, Christensen KB. The Copenhagen Burnout Inventory: a new tool for the assessment of burnout. Work & Stress 2005;19:192–207. [40] Warr P. The measurement of well-being and other aspects of mental health. J Occup Psychol 1990;63:193–210. [41] Bech P, Rasmussen NA, Olsen LR, Noerholm V, Abildgaard W. The sensitivity and specificity of the Major Depression Inventory, using the Present State Examination as the index of diagnostic validity. J Affect Disorders 2001;66(2–3):159–64. [42] Bandura A. Self-efficacy: the exercise of control. New York: W. H. Freeman & Company; 1997. [43] European Foundation for the Improvement of Living and Working Conditions. Violence, bullying and harassment in the workplace. Dublin: European Foundation for the Improvement of Living and Working Conditions; 2007. [44] Ware JE, Kosinski M, Bjorner JB, Turner-Bowker DM, Gandek B, Maruish ME. User’s manual for the SF-36v2 Health Survey, 2nd edn. Lincoln, RI: QualityMetric Incorporated; 2007. [45] Idler EL, Benyamini Y. Self-rated health and mortality: a review of twenty-seven community studies. J Health Soc Behavior 1997;38:21–37. [46] Hochschild AR. The managed heart – commercialization of human ffeling, University of Califonia Press; 1983. [47] Meyer JP, Stanley DJ, Herscovitch L, Topolnytsky L. Affective, continuance, and normative commitment to the organization: a meta-analysis of antecedents, correlates, and consequences. Journal of Vocational Behavior 2002;61:20–52. [48] Thorsen SV, Bjorner JB. Reliability of the Copenhagen Psychosocial Questionnaire. Scand J Public Health 2010; 38(Suppl 3):8–24. [49] Bollen K, Lennox R. Conventional wisdom on measurement – a structural equation perspective. Psychological Bull 1991;110:305–14. [50] Bjorner JB, Pejtersen JH. Evaluating construct validity of the Copenhagen Psychosocial Questionnaire through analysis of differential item functioning and differential item effect. Scand J Public Health 2010;38(Suppl 3):90–105. [51] Christensen KB, Nielsen ML, Rugulies R, Smith-Hansen L, Kristensen TS. Workplace levels of psychosocial factors as prospective predictors of registered sickness absence. J Occup Environ Med 2005;47:933–40. [52] Oksanen T, Kouvonen A, Kivimaki M, Pentti J, Virtanen M, Linnaa A, et al. Social capital at work as a predictor of employee health: multilevel evidence from work units in Finland. Soc Sci Med 2008;66:637–49. [53] Pejtersen JH, Kristensen TS. The development of the psychosocial work environment in Denmark from 1997 to 2005. Scand J Work Environ Health 2009;35(4):284–93. [54] Rugulies R, Aust B, Pejtersen JH. Do psychosocial work environment factors measured with scales from the COPSOQ predict register-based sickness absence of 3 weeks or more in Denmark? Scand J Public Health 2010;38(Suppl 3):42–50.

Second version of the Copenhagen Psychosocial Questionnaire Appendix Appendix: Scales of the COPSOQ II questionnaire The response option for questions in COPSOQ II is given below: (a) Always; Often; Sometimes; Seldom; Never/hardly ever (b) Always; Often; Sometimes; Seldom; Never/hardly ever (reversed scoring) (c) To a very large extent; To a large extent; Somewhat; To a small extent; To a very small extent (d) Always; Often; Sometimes; Seldom; Never/hardly ever; Not relevant (e) Very satisfied; Satisfied; Unsatisfied; Very unsatisfied; Not relevant (f) Yes, often; Yes, sometimes; Rarely; No, never (g) Yes, certainly; Yes, to a certain degree; Yes, but only very little; No, not at all

21

(h) To a very large extent; To a large extent; Somewhat; To a small extent; To a very small extent (reversed scoring) (i) Excellent; Very good; Good; Fair; Poor (j) All the time; A large part of the time; Part of the time; A small part of the time; Not at all (k) Fits perfectly; Fits quite well; Fits a little bit; Does not fit (l) Yes, daily; Yes, weekly; Yes, monthly; Yes, a few times; No



If yes, with whom? (You may tick off more than one); Colleagues; Manager/superior; Subordinates; Clients/customers/patients (m) Yes, daily; Yes, weekly; Yes, monthly; Yes, a few times; No – If yes, from whom? (You may tick off more than one); Colleagues; Manager/superior; Subordinates; Clients/customers/patients

Demands at work

Scale

Item #

Quantitative demands

QD1 QD2 QD3 QD4 WP1 WP2 WP3 CD1 CD2 CD3 CD4 ED1 ED2

Work pace

Cognitive demands

Emotional demands

Demands for hiding emotions

ED3 ED4 HE1 HE2 HE3

Item Is your workload unevenly distributed so it piles up? How often do you not have time to complete all your work tasks? Do you get behind with your work? Do you have enough time for your work tasks? Do you have to work very fast? Do you work at a high pace throughout the day? Is it necessary to keep working at a high pace? Do you have to keep your eyes on lots of things while you work? Does your work require that you remember a lot of things? Does your work demand that you are good at coming up with new ideas? Does your work require you to make difficult decisions? Does your work put you in emotionally disturbing situations? Do you have to relate to other people’s personal problems as part of your work? Is your work emotionally demanding? Do you get emotionally involved in your work? Are you required to treat everyone equally, even if you do not feel like it? Does your work require that you hide your feelings? Are you required to be kind and open towards everyone – regardless of how they behave towards you?

Response option a a a b a c c a a a a a a c c a c c

22

J. H. Pejtersen et al.

Work organization and job contents

Item#

Influence

IN1 IN2 IN3 IN4 PD1

Do you have a large degree of influence concerning your work? Do you have a say in choosing who you work with? Can you influence the amount of work assigned to you? Do you have any influence on what you do at work? Does your work require you to take the initiative?

a a a a c

PD2 PD3 PD4 VA1 VA2 MW1 MW2 MW3 CW1 CW2 CW3

Do you have the possibility of learning new things through your work? Can you use your skills or expertise in your work? Does your work give you the opportunity to develop your skills? Is your work varied? Do you have to do the same thing over and over again? Is your work meaningful? Do you feel that the work you do is important? Do you feel motivated and involved in your work? Do you enjoy telling others about your place of work? Do you feel that your place of work is of great importance to you? Would you recommend a good friend to apply for a position at your workplace? How often do you consider looking for work elsewhere?

c c c a b c c c c c c

Possibilities for development (skill discretion)

Variation Meaning of work

Commitment to the workplace

CW4

Item

Response option

Scale

b

Interpersonal relations and leadership

Scale

Item #

Predictability

PR1

Recognition

Role clarity

Role conflicts

PR2 RE1 RE2 RE3 CL1 CL2 CL3 CO1 CO2 CO3 CO4

Quality of leadership QL1 QL2 QL3 QL4 Social support from colleagues

SC1 SC2 SC3

Social support from supervisors*

SS1 SS2 SS3

Social community at work

SW1 SW2 SW3

Item At your place of work, are you informed well in advance concerning for example important decisions, changes, or plans for the future? Do you receive all the information you need in order to do your work well? Is your work recognised and appreciated by the management? Does the management at your workplace respect you? Are you treated fairly at your workplace? Does your work have clear objectives? Do you know exactly which areas are your responsibility? Do you know exactly what is expected of you at work? Do you do things at work, which are accepted by some people but not by others? Are contradictory demands placed on you at work? Do you sometimes have to do things which ought to have been done in a different way? Do you sometimes have to do things which seem to be unnecessary? To what extent would you say that your immediate superior: – makes sure that the individual member of staff has good development opportunities? – gives high priority to job satisfaction? – is good at work planning? – is good at solving conflicts? How often do you get help and support from your colleagues? How often are your colleagues willing to listen to your problems at work? How often do your colleagues talk with you about how well you carry out your work? How often is your nearest superior willing to listen to your problems at work? How often do you get help and support from your nearest superior? How often does your nearest superior talk with you about how well you carry out your work? Is there a good atmosphere between you and your colleagues? Is there good co-operation between the colleagues at work? Do you feel part of a community at your place of work?

*These questions were only addressed to respondents who were not supervisors themselves and who had a supervisor.

Response option c c c c c c c c c c c c c c c c c d d d a a a a a a

Second version of the Copenhagen Psychosocial Questionnaire

23

Work-individual interface

Scale

Item #

Job insecurity

JI1 JI2 JI3 JI4

Job satisfaction

Work–family conflict

JS1 JS2 JS3 JS4 WF1

WF2 WF3 WF4 Family–work conflict FW1 FW2

Response option

Item Are you worried about becoming unemployed? Are you worried about new technology making you redundant? Are you worried about it being difficult for you to find another job if you became unemployed? Are you worried about being transferred to another job against your will? Regarding your work in general. How pleased are you with: – your work prospects? – the physical working conditions? – the way your abilities are used? – your job as a whole, everything taken into consideration? Do you often feel a conflict between your work and your private life, making you want to be in both places at the same time? The next three questions concern the ways in which your work affects your private life: Do you feel that your work drains so much of your energy that it has a negative effect on your private life? Do you feel that your work takes so much of your time that it has a negative effect on your private life? Do your friends or family tell you that you work too much? The next two questions concern the ways in which your private life affects your work: Do you feel that your private life takes so much of your energy that it has a negative effect on your work? Do you feel that your private life takes so much of your time that it has a negative effect on your work?

c c c c e e e e f

g g g g g

Values at the workplace The next questions are not about your own job but about the workplace as a whole

Scale

Item #

Mutual trust between employees

TE1 TE2 TE3 TM1 TM2 TM3 TM4 JU1 JU2 JU3 JU4 SI1 SI2 SI3 SI4

Trust regarging management

Justice

Social inclusiveness

Item Do the employees withhold information from each other? Do the employees withhold information from the management? Do the employees in general trust each other? Does the management trust the employees to do their work well? Can you trust the information that comes from the management? Does the management withhold important information from the employees? Are the employees able to express their views and feelings? Are conflicts resolved in a fair way? Are employees appreciated when they have done a good job? Are all suggestions from employees treated seriously by the management? Is the work distributed fairly? Are men and women treated equally at your workplace? Is there space for employees of a different race and religion? Is there space for elderly employees? Is there space for employees with various illnesses or disabilities?

Response option h h c c c h c c c c c c c c c

24

J. H. Pejtersen et al.

Health and well-being

Scale

Item #

General health perception

GH1

Sleeping troubles

SL1 SL2 SL3 SL4 BO1 BO2 BO3 BO4 ST1 ST2 ST3 ST4 DS1 DS2 DS3 DS4 SO1 SO2 SO3 SO4 CS1 CS2 CS3 CS4

Burnout

Stress

Depressive symptoms

Somatic stress

Cognitive stress

Self-efficacy SE1 SE2 SE3 SE4 SE5 SE6

Item In general, would you say your health is: (Excellent, Very good, Good, Fair, Poor) These questions are about how you have been during the last 4 weeks. How often have you slept badly and restlessly? How often have you found it hard to go to sleep? How often have you woken up too early and not been able to get back to sleep? How often have you woken up several times and found it difficult to get back to sleep? How often have you felt worn out? How often have you been physically exhausted? How often have you been emotionally exhausted? How often have you felt tired? How often have you had problems relaxing? How often have you been irritable? How often have you been tense? How often have you been stressed? How often have you felt sad? How often have you lacked self-confidence? How often have you had a bad conscience or felt guilty? How often have you lacked interest in everyday things? How often have you had stomach ache? How often have you had a headache? How often have you had palpitations? How often have you had tension in various muscles? How often have you had problems concentrating? How often have you found it difficult to think clearly? How often have you had difficulty in taking decisions? How often have you had difficulty with remembering? How well do these descriptions fit you as a person? I am always able to solve difficult problems, if I try hard enough. If people work against me, I find a way of achieving what I want. It is easy for me to stick to my plans and reach my objectives. I feel confident that I can handle unexpected events. When I have a problem, I can usually find several ways of solving it. Regardless of what happens, I usually manage.

Response option i j j j j j j j j j j j j j j j j j j j j j j j j k k k k k k

Offensive behaviour

Scale

Item #

Sexual harassment

SH

Threats of violence

TV

Physical violence

PV

Bullying

BU

Unpleasant teasing

UT

Conflicts and quarrels

CQ

Gossip and slander

GS

Item Have you been exposed to undesired sexual attention at your workplace during the last 12 months? Have you been exposed to threats of violence at your workplace during the last 12 months? Have you been exposed to physical violence at your workplace during the last 12 months? Bullying means that a person repeatedly is exposed to unpleasant or degrading treatment, and that the person finds it difficult to defend himself or herself against it. Have you been exposed to bullying at your workplace during the last 12 months? Have you been exposed to unpleasant teasing at your workplace during the last 12 months? Have you been involved in quarrels or conflicts at your workplace during the last 12 months? Have you been exposed to gossip and slander at your workplace during the last 12 months?

Response option l l l

l l l m

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 25–32

ORIGINAL ARTICLE

Reliability of the Copenhagen Psychosocial Questionnaire

SANNIE V. THORSEN & JAKOB B. BJORNER National Research Centre for the Working Environment, Copenhagen, Denmark

Abstract Aims: Reliabilities of the work environment questionnaire Copenhagen Psychosocial Questionnaire (COPSOQ) have previously been estimated by Cronbach’s alpha, but since the internal consistency assumption may not apply to all COPSOQ scales, Cronbach’s alpha may underestimate true reliability. This study aims to evaluate reliability in a test–retest design. Methods: We analyzed postal questionnaire data from 349 persons (of whom 283 were employees) who completed two forms with a median interval of 22 (range 6–65) days between baseline and follow-up. Test–retest reliabilities were estimated by the intraclass correlation (ICC). For scales where the internal consistency assumption was theoretically plausible, reliabilities were also estimated by Cronbach’s alpha and by Green’s test–retest alpha. Results: With one exception, the ICC estimated reliabilities of the COPSOQ scales were adequate or good (range 0.70–0.89). A scale concerning mutual trust between employees had a low reliability of 0.64. Among the scales where the internal consistency assumption was plausible, Cronbach’s alpha was adequate or good (0.75–0.85) for seven out of eight scales. Green’s retest alpha was adequate or good for six out of eight scales (0.72–0.81). Conclusions: Standard criteria for acceptable intraclass correlation reliability were achieved for all COPSOQ scales but one. The test–retest design and intraclass correlation appears to be more appropriate than Cronbach’s alpha for assessing the reliability of psychosocial work environment scales.

Key Words: Copenhagen Psychosocial Questionnaire, Cronbach’s alpha, health psychology, occupational, questionnaires, reliability (epidemiology), work environment

Background The reliability of a questionnaire scale reflects the amount of variance in the scale that is explained by the construct that scale is intended to measure as opposed to random error (page 28 in [1]]. If a scale that is valid but unreliable is used as an endpoint in statistical analyses, the statistical power will be low [2]. If such a scale is used as an independent variable, results will be biased [3]. Therefore, reliability is an important measurement property of any scale used in research. For assessment of individuals, even higher precision is required than necessary for the group level analyses typically used in research [4]. The most frequently used reliability estimator is Cronbach’s alpha [5]. This estimator is easy to achieve since it only requires cross-sectional data. However, all reliability estimators are calculated under a number of assumptions, and the assumptions

of a chosen estimator should be carefully evaluated in the particular situation it is used. Cronbach’s alpha assumes ‘‘internal consistency’’ [6] and that item specific errors are uncorrelated [7]. These assumptions are crucial to the use of Cronbach’s alpha [6,8–11]. A third assumption is that items in a scale have the same true score (tau equivalency) [1]. Due to violation of this less crucial assumption, alpha is sometimes considered a lower bound of reliability. The ‘‘internal consistency assumption’’ is the assumption that items belonging to the same scale have to have a positive correlation because they are effect indicators of a common unidimensional cause [6,8]. As discussed by Bjorner et al in this issue, this assumption seems reasonable for some COPSOQ constructs (e.g. depression, burnout and stress) but not for other constructs (e.g. possibilities for development, recognition, social inclusiveness). For such scales, where items are combined because of

Correspondence: Sannie V. Thorsen, National Research Centre for the Working Environment, Lersø Parkalle´ 105, DK2100 Copenhagen, Denmark. Tel: þ45 3916 5200. Fax: þ45 3916 5201. E-mail: [email protected] (Accepted 31 August 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809349859

26

S. V. Thorsen & J. B. Bjorner

their hypothesized common effect rather than their common cause, high inter-item correlation is not a necessary criterion of construct validity [6,8]. For such scales with ‘‘causal indicators’’, Cronbach’s alpha might not be a good measure of reliability [8], because it might underestimate true reliability. Cronbach’s alpha is sometimes interpreted as a measure of ‘‘internal consistency’’, i.e. that the items measure one unidimensional construct. However, this use of Cronbach’s alpha is also doubtful since simulation studies show that Cronbach’s alpha is not a good indicator of unidimensionality [12]. Another potential problem of Cronbach’s alpha is the assumption that item specific errors are uncorrelated, i.e. that all correlation between items is due only to the latent construct measure by the scale. Such violation may be caused by local item dependence [13] (caused, for example, by similarities in wording between two items) or by transient errors (i.e. errors due to specific conditions at the time the test is taken, e.g. the mood of the test-person at the particular time). In both cases, Cronbach’s alpha may overestimate true reliability. The aim of the current study is to estimate reliability of the Copenhagen Psychosocial Questionnaire version II (COPSOQ II) scales, using a test–retest design to avoid the problems of Cronbach’s alpha. For the scales that fulfil the internal consistency assumption, we report three reliability estimates; the intraclass coefficient (ICC), Cronbach’s alpha and Green’s test–retest alpha. Although ICC reliabilities avoid some of the assumptions of Cronbach’s alpha, they have assumptions of their own – in particular that the 1) concept measured is stable in the interval between test and retest, and 2) that the error terms are not correlated over time. If the concept measures changes over time, the test–retest design might underestimate the reliability and a cross-sectional assessment of reliability (e.g. Cronbach’s alpha) may be preferable. If the assumption of uncorrelated error terms between baseline and follow-up is not fulfilled, the test–retest design might overestimate the reliability. An example of correlated error terms in test– retest design is when the respondents’ answers at time two are influenced by their recollection of their responses to the same items at time one, the ‘‘carryover effect’’ [14]. Finally, the answers in the second test may also be influenced by ‘‘practice effect’’, i.e. that the respondent learns how to answer the questionnaire the first time [14]. For scales where the internal consistency assumption applies, Green’s retest alpha is an alternative to the ICC and Cronbach’s alpha reliability estimators. The Green’s test–retest alpha estimator is similar to Cronbach’s

alpha but removes transient error from the estimate by using two time points in the calculation [10]. It also takes problems due to ‘‘carry-over effect’’ into account because it compares different items at the two time points. For further comparisons, we report Green’s reliabilities for the scales where we find the internal consistency assumption to be plausible.

Methods The study was designed as a test–retest study using postal questionnaires. A random sample of 1,000 Danes between 18-59 years, selected from the CPR registry, received the baseline questionnaire and an introductory letter explaining that respondents would receive another questionnaire with similar questions shortly after answering the first questionnaire. Two weeks after receiving the first questionnaire, we mailed a second questionnaire to all responders. A letter included with the retest questionnaire instructed the respondent to answer not from memory of the answers to the first test but to choose the answer that best described his or her current situation. Non-responders to the first or the second questionnaire received two reminders by post, unless they had indicated that they did not want to take part in the study. The study aimed to achieve test and retest responses from at least 200 employees, which would result in a confidence interval of 0.68–0.81 for a scale with a reliability of 0.75. The study exceeded these minimum requirements, achieving responses from 457 persons for the baseline questionnaire and from 349 persons for the follow-up questionnaire, 283 of whom were currently employed (see Table I).

Questionnaire The study used the medium length version of the COPSOQ II (Pejtersen et al in this issue) to avoid excessive response burden. Abbreviated content of the items can be seen in Table II. The total number of items was 112 in the first questionnaire and 97 in the second questionnaire. The questionnaire included the standard 23 scales of the medium length COPSOQ plus two extra scales from the full length questionnaire. We added the scales on Demands for hiding emotions and Social inclusiveness based on the hypothesis that they did not fulfil the internal consistency assumption and thus that previous reliability estimates might be biased. Also, based on the same logic we included the question ‘‘Is your salary fair in relation to your effort at work?’’ that was part of

Reliability of the Copenhagen Psychosocial Questionnaire

27

Table I. Sample characteristics.

Age of respondents Median (range) Days between assessments. Median (range) Percent males Percent blue collarb

First questionnaire (n ¼ 457)

Both questionnaires (n ¼ 349)

Employeesa (n ¼ 283)

42 (18–58) – 42% 34%b

43 (18–58) 22 (6–65) 41% 32%b

44 (19–58) 22 (6–65) 42% 32%b

a Not including self employed. bPercentage of blue collar workers to blue and white collar. Unemployed and self employed are not included. Data for blue/white collar status is available for 330 employees who answered the first questionnaire and for 258 employees who answered both questionnaires.

the ‘‘Recognition’’ scale in the COPSOQ II test version but was subsequently removed due to low correlations with other items. The internal consistency assumption was found to be reasonable for eight scales (see Table II). The assessment was based on considerations regarding the items’ status as causes or effects of the latent construct. The construct lacks internal consistency if at least one item is considered a cause [6,8]. Statistical analysis All COPSOQ scales were calculated using the standard recommendations of simple sum scoring and transformation to a 0–100 metric (see Pejtersen et al in this issue). An average value of each scale’s score is given in Table III. Test–retest reliabilities were calculated using an intraclass correlation [15] for all 25 scales. We identified eight scales where the internal consistency seemed plausible (Table II). For these scales Cronbach’s alpha and Green’s retest alpha were calculated. To enable comparisons, all reliability coefficients were calculated only for those respondents who had completed the questionnaire both at time one and two. Cronbach’s alpha was calculated separately from baseline data and from retest data. We used empirical bootstrap [16] to estimate confidence intervals for all reliability estimates (1,000 bootstrap samples) and to test whether reliabilities were significantly different across age, gender, white/blue collar, and time between assessments. For these comparisons we dichotomized age and time between assessments at their median value. We used SAS version 9.1 for all analyses and wrote an SAS macro to calculate Green’s test–retest coefficient alpha (available from the authors).

Results The respondent rate was 46% for the first questionnaire and 35% for the second questionnaire.

Of these, 283 (81%) were employees at a workplace (as opposed to being unemployed or self-employed). Sample characteristics are shown in Table I. The time interval between test and retest responses of the questionnaire was approximately three weeks. The range was from 6 days to 65 days. Time intervals shorter than 14 days were possible because a few respondents answered the first questionnaire late and then answered a questionnaire (actually intended to be the first) sent out with a reminder. Intraclass correlation provided acceptable to high reliability estimates for 24 out of 25 scales (ICC range 0.70–0.89, see Table III). The scale concerning mutual trust between employees had an ICC reliability of 0.64. For the eight scales where we found the assumption of internal consistency reasonable, Cronbach’s alpha was generally high, although one scale (Meaning of work) had a reliability of 0.68. Among these eight scales, ICC reliability estimates were higher than Cronbach’s alpha for the scales concerning the psychosocial working environment or the work-individual interface (Meaning of work, Commitment to the workplace, Role clarity, and Work– family conflict), while Cronbach’s alpha was higher for scales measuring health outcomes (Burnout, Stress and Sleep troubles). Green’s retest alpha was lowest for all scales. On average, the transient ‘‘error’’ (the difference between Cronbach’s alpha and Green’s retest alpha) was 0.07 (range 0.02–0.18). The three scales with the highest transient dependency were the scales regarding health (Stress, Burnout and Sleep problems), while the scales concerning the working environment had lower transient ‘‘error’’. Cronbach’s alpha reliabilities of all eight scales increased from test data to retest data (results not shown). The average increase was 0.03 (range 0.02–0.06). The ICC reliabilities calculated from respondents with short time intervals between responses was compared to the reliability from respondents with larger time intervals between responses.

28

S. V. Thorsen & J. B. Bjorner

Table II. Internal consistency of scales. Scale (number of items)

Items in scale: causal indicators and effect indicatorsa

Consistency

Demands at work Quantitative demands (4)

No

Work pace (2) Emotional demands (4)

Yes No

Demands for hiding emotions (3) Work organization and job contents Influence (4)

No No

Possibilities for development (4)

No

Meaning of work (3)

Yes

Commitment to the workplace (4)

Yes

Interpersonal relations and leadership Predictability (2) Recognition (rewards) (4)

No No

Role clarity (3) Role conflicts (4)

Yes No

Quality of leadership (4)

No

Social support from supervisor (3)

No

Social support from colleagues (3)

No

IN1 Influence work; IN2 Say in choosing colleges; IN3 Amount of work; IN4 Influence work task PD1 Take initiative; PD2 Learning new things; PD3 Use skills; PD4 Develop skills MW1 Work meaningful; MW2 Work important; MW3 Motivated and involved CW1 Enjoy telling others; CW2 Workplace great importance; CW3 Recommend a friend; CW4 Looking for work elsewhere PR1 Informed about changes; PR2 Information to work well RE1 Recognised by management; RE2 Respected by management; RE3 Treated fairly; RE4 Fair salary CL1 Clear objectives; CL2 Responsibility; CL3 Expectation CO1 Mixed acceptance; CO2 Contradictory demands; CO3 Do things wrongly; CO4 Unnecessary tasks QL1 Development opportunities; QL2 Prioritise job satisfaction; QL3 Work planning; QL4 Solving conflicts SC1 Support supervisor; SC2 Supervisor listens to problems; SC3 Supervisor talks about performance SS1 Support colleagues; SS2 Colleagues listen to problems; SS3 Colleagues talk about performance SW1 Atmosphere; SW2 Cooperation; SW3 Community

Social community at work (3) Work–individual interface Job satisfaction (4) Work–family conflict (4)

No Yes

JS1 Work prospects; JS2 Work conditions; JS3 Work abilities; JS4 Job in general WF1 Being both places; WF2 Energy conflict; WF3 Time conflict; WF4 Family think you work too much

Values at the workplace Trust regarding management (4)

No

TM1 Management trust employees; TM2 Employees trust information; TM3 Management withhold information; TM4 Employees express views TE1 Colleagues withhold information; TE2 Withhold information management; TE3 Trust colleagues JU1 Conflicts resolved fairly; JU2 Employees appreciated; JU3 Suggestions treated seriously; JU4 Work distributed fairly SI1 Gender discrimination; SI2 Race/religion discrimination; SI3 Age discrimination; SI4 Health discrimination

Mutual trust between employees (3)

No

Justice (4)

No

Social inclusiveness (3)

No

Health and well-being Self-rated health (1) Burnout (3) Stress (4) Sleeping troubles (4)

a

No

QD1 Work piles up; QD2 Complete task; QD3 Get behind; QD4 Enough time WP1 Work fast; WP2 High pace ED1 Emotional disturbing; ED2 Other problems; ED3 Emotionally demanding; ED4 Emotionally involved HE1 Treat equally; HE2 Hide feelings; HE 3 Emotionally involved

No Yes Yes Yes

GH1 Health all in all BO1 Worn out; BO2 Physically exhausted; BO3 Emotionally exhausted; ST1 Problems relaxing; ST2 Irritable; ST3 Tense; ST4 Stressed SL1 Slept badly; SL2 Hard to sleep; SL3 Woken up early; SL4 Woken up several times

Each item/question in the scale is given in abbreviated form. Assumed causal items are written in italics.

Only one difference had a p value below 0.05 and it was insignificant after Bonferroni adjustment (see Table IV). Similar tests were performed for age and for blue/ white collar workers and for gender (see Table IV). We found four tests to be below p value 0.05 and one of the p values could pass a Bonferroni adjustment test. The scale Burnout had significant higher

reliability for older respondents compared to younger (p ¼ 0.0006). Discussion The design of the current study was prompted by the hypothesis that the internal consistency approach to estimation of reliability of the COPSOQ scales may

Reliability of the Copenhagen Psychosocial Questionnaire

29

Table III. Reliability estimates for COPSOQ scales.

Internal consistency

Score time-one (std)

Score time-two (std)

271 275 275 268

No Yes No No

40 (20) 54 (20) 38 (23) 50 (21)

38 (19) 53 (20) 38 (24) 53 (22)

0.87 0.85 0.89 0.75

(0.83–0.89) (0.81–0.88) (0.86–0.91) (0.70–0.80)

270 272 275 271

No No Yes Yes

50 (22) 33 (17) 25 (14) 39 (19)

51 (21) 34 (17) 27 (14) 39 (19)

0.83 0.80 0.74 0.87

(0.78–0.87) (0.76–0.84) (0.67–0.79) (0.83–0.90)

274 274 271 266 222 226 256 268

No No Yes No No No No No

44 (21) 36 (16) 29 (16) 36 (17) 43 (20) 38 (22) 40 (19) 20 (17)

45 (20) 36 (16) 31 (17) 38 (18) 43 (20) 37 (22) 42 (19) 24 (17)

0.70 0.80 0.80 0.74 0.83 0.73 0.70 0.73

(0.63–0.77) (0.76–0.84) (0.76–0.84) (0.68–0.80) (0.77–0.87) (0.64–0.80) (0.62–0.76) (0.66–0.79)

247 273

No Yes

25 (13) 24 (17)

26 (12) 24 (18)

0.73 (0.64–0.80) 0.86 (0.82–0.89)

263 265 265 260

No No No No

32 (16) 29 (16) 39 (16) 28 (15)

33 (16) 30 (16) 39 (15) 29 (15)

0.80 0.64 0.80 0.75

(0.74–0.84) (0.55–0.73) (0.74–0.85) (0.69–0.80)

347 344 345 343

Noa Yes Yes Yes

40 (23) 22 (19) 28 (18) 22 (19)

41 (21) 23 (19) 28 (18) 23 (19)

0.78 0.79 0.72 0.81

(0.72–0.82) (0.75–0.83) (0.65–0.78) (0.76–0.85)

Scale Demands at work Quantitative demands Work pace Emotional demands Demands for hiding emotions Work organization and job contents Influence Possibilities for development Meaning of work Commitment to the workplace Interpersonal relations and leadership Predictability Recognition (reward) Role clarity Role conflicts Quality of leadership Social support from supervisor Social support from colleagues Social community at work Work–individual interface Job satisfaction Work-family conflict Values at the workplace Trust regarding management Mutual trust between employees Justice Social inclusiveness Health and well-being Self-rated health Burnout Stress Sleeping troubles

n

ICC (95% CI)

Cronbach’s alpha (95% CI)

Green’s retest alpha (95% CI)

0.85 (0.82–0.88)

0.81 (0.77–0.85)

0.68 (0.59–0.74) 0.75 (0.67–0.80)

0.61 (0.52–0.68) 0.72 (0.66–0.78)

0.77 (0.71–0.82)

0.72 (0.65–0.77)

0.80 (0.75–0.84)

0.76 (0.70–0.81)

0.81 (0.77–0.84) 0.85 (0.81–0.88) 0.84 (0.80–0.87)

0.72 (0.66–0.76) 0.67 (0.60–0.73) 0.74 (0.68–0.79)

a

Self-rated health only includes one item.

underestimate true reliability, because many of the indicators of the psychosocial working environment must be considered causal indicators rather than effect indicators [6,8]. Previous studies of the COPSOQ using Cronbach’s alpha (Pejtersen et al in this issue) have found adequate reliability for most scales, but low alphas for the following scales from the medium length COPSOQ: Demands for hiding emotions (p ¼ 0.57), Role conflicts (q ¼ 0.67), and Social inclusiveness (q ¼ 0.63). We found higher ICC retest reliabilities (0.75, 0.74, and 0.75, respectively) for these scales, which could suggest that Cronbach’s alpha coefficient provided biased estimates of reliabilities for these scales. The average ICC test–retest reliability across all the ‘‘not internally consistent’’ scales was only slightly higher than the average Cronbach’s alpha coefficient previously reported for the same scales (and also only slightly higher than Cronbach’s alpha for the ‘‘not internally consistent’’ scales when estimated in this study – data not shown). This may appear to be in

contrast with our argument about the dangers of using Cronbach’s alpha on ‘‘not internally consistent’’ scales. However, since item selection for the COPSOQ scales was partly determined by the ability of the items to increase alpha (Pejtersen et al in this issue), Cronbach’s alpha may be high for the COPSOQ scales, even in cases where the internal consistency assumption is not theoretically plausible. There is nothing inherently wrong in choosing items to maximize Cronbach’s alpha in a scale without internal consistency. However, such an approach may put an unnecessary constraint on the scale by rejecting good causal items because they fail to fulfil an unnecessary constraint. In turn, this may hamper content validity and predictive validity. Although lack of internal consistency may cause Cronbach’s alpha to be lower than the true reliability, transient error and local dependence may inflate alpha. Green’s retest alpha removes the transient errors from the estimate, but is similar to Cronbach’s alpha in other respects. However, both Green’s retest

30

S. V. Thorsen & J. B. Bjorner Table IV. Reliability (ICC) estimates for subgroups differing in response time-interval, age, blue/white collar or gender. Interval time

Scale Demands at work Quantitative demands Work pace Emotional demands Demands for hiding emotions Work organization and job contents Influence Possibilities for development Meaning of work Commitment to the workplace Interpersonal relations and leadership Predictability Recognition (rewards) Role clarity Role conflicts Quality of leadership Social support from supervisor Social support from colleagues Social community at work Work–individual interface Job satisfaction Work-family conflict Values at the workplace Trust regarding management Mutual trust between employees Justice Social inclusiveness Health and well-being Self-rated health Burnout Stress Sleeping troubles

Age

Gender

6–22 days

23–65 days

18–42 years

43–58 years

Blue collar

White collar

Male

Female

0.88 0.88 0.90 0.75

0.85 0.83 0.88 0.76

0.87 0.85 0.90 0.78

0.86 0.85 0.89 0.74

0.81 0.86 0.88 0.84*

0.88 0.82 0.88 0.68*

0.86 0.87 0.87 0.75

0.88 0.84 0.89 0.76

0.82 0.84* 0.77 0.88

0.84 0.75* 0.71 0.84

0.79 0.81 0.79 0.85

0.85 0.80 0.69 0.88

0.82 0.78 0.72 0.87

0.83 0.78 0.76 0.87

0.86 0.79 0.74 0.86

0.80 0.81 0.73 0.87

0.66 0.84 0.81 0.81 0.81 0.75 0.74 0.75

0.75 0.77 0.80 0.72 0.86 0.75 0.65 0.72

0.71 0.79 0.81 0.74 0.83 0.72 0.66 0.77

0.70 0.82 0.79 0.75 0.82 0.73 0.72 0.71

0.70 0.86 0.73 0.72 0.86 0.77 0.76 0.67

0.69 0.79 0.83 0.77 0.82 0.70 0.66 0.77

0.72 0.81 0.81 0.72 0.80 0.70 0.66 0.68

0.69 0.80 0.80 0.74 0.83 0.75 0.71 0.78

0.73 0.87

0.72 0.82

0.78 0.86

0.71 0.87

0.70 0.78*

0.77 0.88*

0.77 0.87

0.70 0.85

0.83 0.53 0.78 0.74

0.77 0.72 0.82 0.76

0.73 0.61 0.81 0.73

0.84 0.67 0.80 0.76

0.85 0.63 0.87 0.77

0.77 0.66 0.79 0.74

0.84 0.60 0.85 0.79

0.76 0.67 0.76 0.72

0.78 0.82 0.73 0.84

0.77 0.76 0.72 0.79

0.70 0.70* 0.74 0.71*

0.81 0.85* 0.71 0.85*

0.76 0.77 0.62 0.79

0.75 0.78 0.77 0.82

0.75 0.78 0.73 0.78

0.79 0.80 0.72 0.82

*Pairs of reliability estimates with an asterisk are significantly different (p < 0.05, no Bonferroni adjustment).

alpha and the ICC test–retest reliability coefficient assume that there is no true change in the domains measured. For many COPSOQ domains, such assumption is reasonable within the retest lag used in this study, but change may well occur within some domains, e.g. stress and sleeping troubles. We suspect this is the reason that test–retest reliability is lower than Cronbach’s alpha for the three healthrelated domains (i.e. the ‘‘transient error’’ is a true change and a one time point assessment is the optimal). In our opinion, Green’s retest alpha represents a lower bound estimate of reliability, since the coefficient is unaffected by most of the factors that may cause upwards bias in reliability estimates, but still prone to factors that may cause downwards bias. Thus, it is unsurprising that the retest alpha provides the lowest reliability estimate for all scales where it is used. The test–retest design may introduce errors due to a ‘‘practice effect’’ or a ‘‘carry-over effect’’.

In this context, a practice effect means that respondents achieve a better understanding of the items by answering them in round one and that this makes their answers in round two more reliable. This would again lead to increased Cronbach’s alpha reliability in round two and to some (although not as large) increase in the ICC and Green’s retest alpha reliability. We did find slightly higher Cronbach’s alpha values at time two compared to time one. The differences were fairly consistent and on average 0.03. This would lead to a small (app. 0.015) upward bias in the ICC test–retest coefficient as a measure of scale reliability for a person who has never answered the questionnaire before. A carry-over effect means that a person remembers his or her last answers and copies these answers in the retest. The round two introduction letter instructed the respondent not to take time one answers into account, but nevertheless a carry-over effect cannot be ruled out. A carry-over effect may influence the

Reliability of the Copenhagen Psychosocial Questionnaire ICC but not Green’s retest alpha, because covariances in this estimate are between different times and different items. For the five work environment scales where Green’s retest alpha was calculated, we found that Green’s retest alpha was 0.04 to 0.13 lower than the ICC estimate. This might be due to a carry-over effect or occur because our assumption of internal consistency and unidimensionality does not hold. The standard criterion for reliability is 0.70. Green’s retest alpha exceeded 0.70 for four of the five work environment scales, but not for the scale Meaning of work (Green’s retest alpha ¼ 0.61). The ICC is 0.74 for this scale, while Cronbach’s alpha was 0.68 at time one and 0.74 at time two. A previous study (Pejdersen et al in this issue) found a Cronbach’s alpha of 0.74. The low Green’s retest alpha for the scale Meaning of work could be a true low reliability of the scale, but since Cronbach’s alpha is lower at baseline than in other studies and data sets, it is likely that the low retest alpha is an aberrant result. Another explanation is that the internal consistency/ unidimensionality assumption does not hold for the scale, in which case the ICC retest reliability estimate is the most appropriate estimate for this scale. While a reliability of 0.70 is traditionally regarded a threshold for adequate reliability for group analyses [4], we would like to emphasize that this threshold is arbitrary and that adequate reliability depends on the study and the purpose. For scales that are used as endpoint variables, low reliability will not bias the estimates, but merely weaken the power of the study [2], an effect that can be countered by increasing the sample size. For scales that are used as independent variables, low reliabilities may bias results, but this effect may be countered by applying analytic methods, e.g. structural equation models, that can take measurement error into account [17]. Thus, while high reliability is preferable, sensible statistical analyses can still be made with scales that have less than perfect reliability. Even higher reliability is required for assessment of individual respondents, but COPSOQ has rarely been used for that purpose.

Conclusion Using a test–retest design, we found ICC test–retest reliabilities that according to standard guidelines were adequate to good in 24 out of 25 scales in the medium length COPSOQ questionnaire. In particular, we found adequate reliabilities for scales that had low reliability in previous studies based on Cronbach’s alpha. For eight scales considered to be internally consistent Cronbach’s alpha and Green’s retest alpha were calculated and one scale is below

31

standard recommendations for both ‘‘internal consistency’’ estimates. We found indications of practice effects and transient differences between scale estimates of test and retest. It appears that older respondents have higher ICC reliability than younger respondents of the scale Burnout. Based on conceptual considerations and the results of the current study, we recommend the ICC coefficient as a reliability estimate and that the internal consistency assumptions are carefully considered before Cronbach’s alpha is used to estimate reliabilities for scales concerning the psychosocial working environment.

Acknowledgements This study was conducted under funding from the Danish Ministry of Employment. The authors wish to thank Pia Gøtterup, Christian Trolle Strandfelt, Christian Roepstorff, and Helle Soll-Johanning for help with data collection and thank Reiner Rugulies and two anonymous reviewers for helpful comments on a previous version of the paper.

References [1] Thissen D, Wainer H. True score theory: the traditional method. In: Thissen D, Wainer H, editors. Test scoring, Lawrence Erlbaum Associates, Publishers; 2001. pp. 23–71. [2] Kraemer HC. To increase power in randomized clinical trials without increasing sample size. Psychopharmacol Bull 1991;27(3):217–24. [3] Carroll RJ. Covariance analysis in generalized linear measurement error models. Stat in Med 1989;8(9):1075–93. [4] Nunnally JC, Bernstein IH. Psychometric theory. New York: McGraw-Hill, Inc.; 1994. [5] Cronbach LJ. Coefficient alpha and the internal structure of tests. Psychometrika 1951;16(3):297–334. [6] Bollen KA. Multiple indicators – internal consistency or no necessary relationship. Quality & Quantity 1984;18(4):377–85. [7] Green SB, Lissitz RW, Mulaik SA. Limitations of coefficient alpha as an index of test unidimensionality. Educ Psychol Measure 1977;37(4):827–38 . [8] Bollen K, Lennox R. Conventional wisdom on measurement – a structural equation perspective. Psychol Bull 1991; 110(2):305–14. [9] Curtis RF, Jackson EF. Multiple indicators in surveyresearch. Am J Sociol 1962;68(2):195–204. [10] Green SB. A coefficient alpha for test-retest data. Psychol Meth 2003;8(1):88–101. [11] Osburn HG. Coefficient alpha and related internal consistency reliability coefficients. Psychol Meth 2000; 5(3):343–55. [12] Hattie JA. Methodology review: assessing unidimensionality of tests and items. Appl Psychol Measure 1985;9(2):139–64. [13] Chen W-H, Thissen D. Local dependence indexes for item pairs using item response theory. Educ Behav Stat 1997; 22(3):265–89.

32

S. V. Thorsen & J. B. Bjorner

[14] Mckelvie SJ. Does memory contaminate test-retest reliability. J Gen Psychol 1992;119(1):59–72. [15] Shrout PE, Fleiss JL. Intraclass correlations – uses in assessing rater reliability. Psychol Bull 1979;86(2):420–8. [16] Henderson AR. The bootstrap: a technique for datadriven statistics. Using computer-intensive analyses

to explore experimental data. Clin Chim Acta 2005; 359(1–2):1–26. [17] Budtz-Jorgensen E. Estimation of the benchmark dose by structural equation models. Biostatistics 2007; 8(4):675–88.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 33–41

ORIGINAL ARTICLE

Determining minimally important score differences in scales of the Copenhagen Psychosocial Questionnaire

JAN HYLD PEJTERSEN, JAKOB BUE BJORNER & PETER HASLE National Research Centre for the Working Environment, Copenhagen, Denmark

Abstract Aim: To determine minimally important differences (MIDs) for scales in the first version of the Copenhagen Psychosocial Questionnaire (COPSOQ). Methods: Data were taken from two separate studies: a national population survey (N ¼ 1062), and an intervention study at 14 workplaces (N ¼ 1505). On the basis of the population survey, the MID for each COPSOQ scale was calculated as one-half of a standard deviation (0.5 SD). For the core COPSOQ scales on ‘‘Quantitative demands’’, ‘‘Influence at work’’, ‘‘Predictability’’, ‘‘Social support (from colleagues and supervisors, respectively)’’, and ‘‘Job satisfaction’’, the MIDs were evaluated in the intervention study, where score differences for the scales were linked to the respondents’ global self-evaluation of the impact of the interventions. The scales were scored from 0 to 100 in both studies. Results: The MIDs calculated as 0.5 SD were, on average, 9.2 (range 6.8–14.9) for the long version scales, and 10.8 (range 7.6–14.9) for the medium-length version scales. The analysis of the self-evaluated changes on the scale scores for the core COPSOQ scales showed that the anchor-based estimates of MID were generally lower than 0.5 SD. Conclusions: We recommend the following MID values for the COPSOQ scales: ‘‘Quantitative demands’’, 0.3 SD; ‘‘Influence’’, 0.2 SD; ‘‘Predictability’’, 0.3 SD; ‘‘Social support from colleagues’’, 0.3 SD; ‘‘Social support from supervisor’’, 0.7 SD; and ‘‘Job satisfaction’’, 0.4 SD. For all other COPSOQ scales, where we do not have anchor-based results, we recommend the conventional MID value of 0.5 SD.

Key Words: Meaningful change, minimally important difference, psychosocial factors, psychosocial work environment, questionnaire

Background The Copenhagen Psychosocial Questionnaire (COPSOQ) is a standardized measuring instrument designed to monitor different aspects of the psychosocial work environment [1]. It is widely used at workplaces in Denmark when enterprises survey the psychosocial work environment as part of their mandatory workplace risk assessment (in Denmark, called workplace assessment), which has to be carried out every third year [2,3]. It has been used to study the following: (a) differences in the psychosocial work environment for various population groups, e.g. looking at differences among various professions [4]; (b) whether differences in the psychosocial work environment are associated with differences in subsequent health outcomes, sick days, or early retirement [5–7];

and (c) changes in the psychosocial work environment in intervention studies [8–10] (see also Pejtersen et al. [11] and other articles in this issue). When interpreting results from work environment surveillance or follow-up studies of workplace interventions, it is crucial to know whether a difference in the questionnaire scores is of a magnitude that has practical consequences at the workplace. The research results could be expected to serve as guidelines for practitioners who design and implement real-life interventions in workplaces, and it is important for them to obtain an understanding of whether an intervention – even if it is successful – would lead to a magnitude of improvement that is of practical importance. In many research fields, and especially in epidemiology, differences are often exclusively interpreted

Correspondence: Jan Hyld Pejtersen, National Research Centre for the Working Environment, Lersø Parkalle´ 105, DK 2100 Copenhagen, Denmark. Tel: +45 39 16 52 99. Fax: +45 39 16 52 01. E-mail: [email protected] (Accepted 5 August 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809347024

34

J. H. Pejtersen et al.

on the basis of statistical methods. This means that small differences may be statistically significant in large studies but insignificant in studies based on smaller samples. In such large studies, it is difficult to know whether a statistically significant result is of practical importance. The identification of minimally important differences (MIDs) has been the focus of much recent medical research, e.g. in the health outcomes field [12–15]. Researchers in the health outcomes field have focused on MIDs for individual scores, sometimes called responder criteria, or group-level comparison [16]. Criteria for individual scores have to take reliability and measurement precision into account, and therefore MIDs are larger for individual scores than for group-level comparison [17]. The COPSOQ was developed for assessment of the psychosocial work environment at workplace level, and therefore the present article will focus on MIDs for group-level comparison. Two different approaches can be used for estimating the MID (sometimes referred to as minimal clinically important difference): distribution-based methods and anchor-based methods [16]. In the distribution-based methods, the MID is linked to a statistical parameter such as the standard deviation (SD) or effect size (average change divided by SD at baseline). However, the distribution-based methods should be validated against anchor-based methods to obtain benchmarks for interpretation [14]. Such benchmarks should make sense to practitioners in the field. Examples of benchmarks could be employees’ evaluations of improvements, reduction in sickness absence days [18], or increased productivity. The more an instrument is used, the more benchmarks will be available for interpretation. In the anchor-based methods, the changes in scores are related to an external event, rating, condition, clinical changes, etc. [16]. One approach is to link score changes in longitudinal studies to the respondents’ self-evaluation of the magnitude of change. Several studies in health-related quality-of-life research have looked into MIDs for various questionnaires [13,14]. The studies are surprisingly consistent in their findings of MIDs. Across many different instruments, the general finding is that the MID is close to 0.5 SD [13]. In their review of different methods and instruments, Norman et al. [13] argued that the consistent finding of 0.5 SD is not a coincidence, but it is related to humans’ limitation in discrimination. Over a wide range of tasks (tastes, points on a line, pitch and loudness of sounds), Miller [19] has shown that people’s ability to achieve absolute discrimination is accurate until the number of categories reaches approximately seven (range five to nine). On a uniform rectangular

distribution, a step of one out of seven corresponds to a standard deviation of 0.5. However, some recent studies, using a variety of anchors, have found MIDs that are smaller than 0.5 SD [15,20]. These studies suggest that MIDs based on 0.5 SD are conservative, and that even smaller score differences may be of practical importance in many situations. The purpose of the present study was to determine MIDs for scales in COPSOQ in two ways. On the basis of the COPSOQ population survey [1], the MID for each COPSOQ scale was calculated as 0.5 SD, as suggested by Norman et al. [13]. For selected scales, the MID was evaluated in an intervention study where score differences for the scales were linked to the respondents’ global self-evaluation of the impact of the interventions.

Material and methods Data were taken from two studies: the COPSOQ survey [1] and the BEST study [9,21]. The study population, the measurements and the statistical analyzes for the two studies are described below. COPSOQ questionnaire This paper concerns the first version of COPSOQ which exists in three lengths: long (30 scales, 141 items), medium (26 scales, 95 items), and short (eight scales, 44 items). All scales from the long and the medium-length version of COPSOQ (except for coping scales) were included in the analyzes. The medium-length version is the most frequently used. For details about the questionnaire, see Kristensen et al. [1]. Furthermore, in order to compare the respondents’ self-evaluations in the BEST study, two scales on social support were constructed. These scales included the items on support from colleagues and support from supervisors that are used in the second version of the questionnaire (COPSOQ II) [11]. For all scales, the items were scored as 0–100 (i.e. 0, 25, 50, 75 and 100 for a five-category item). The scale score was computed as the mean item score. If respondents had answered less than half of the questions on the particular scale, the scale score was set to missing. Each scale was scored in the direction indicated by the scale name. Study 1: COPSOQ survey This study was conducted in 1997 and was a representative national questionnaire survey of working Danes between 20 and 60 years of age.

Determining minimally important score differences in scales of the COPSOQ

35

COPSOQ population. In all, 1857 wage earners answered the questionnaire (response rate of 60%). The respondents were randomly assigned a mailed questionnaire (67%) or a telephone interview (33%). The present study used the mailed questionnaire data only. Thus, the population in the present study consisted of 1062 respondents, of whom 51.2% were women, and 33% were high-school graduates; the average age was 39.9 years (SD 10.7). For a detailed description of the study, see Kristensen et al. [1,4].

BEST population. In total, 3116 employees filled in the baseline questionnaire in 2004–2005, and 2351 filled in the follow-up questionnaire in 2006–2007, providing response rates of 88% and 78%, respectively. Only respondents who participated in both the baseline and follow-up study were included in the present analysis. The study population consisted of 1505 employees. The mean age of the participants was 44.1 years (SD 9.3 years), and 56% of the sample was female. Further details about the sample and the participating companies are shown in Table I.

COPSOQ analysis. Mean scores and SDs for the population were calculated for all scales. The MIDs were determined as 0.5 SD for the scales corresponding to the findings by Norman et al. [13].

BEST measurements. The surveys at baseline and follow-up were performed by using the mediumlength COPSOQ questionnaire. The questionnaire was either distributed in a paper version or available as a web-based version. In the follow-up survey, the employees were also asked directly whether they perceived any difference in six of the central psychosocial factors at work after the interventions. The factors included changes in quantitative demands, influence at work, social support from colleagues, social support from supervisor, predictability, and an overall assessment of the changes at the workplace. The global items used in the follow-up and the corresponding COPSOQ scales are shown in Table II. This article reports data on six COPSOQ scales that correspond to the six change items. The original COPSOQ scales ‘‘Social support’’ and ‘‘Feedback’’ were transformed into two new scales: ‘‘Support from colleagues’’ and ‘‘Support from supervisor’’. This was done in order to allow comparison of the scales with the global questions that were directed towards help and support from colleagues and from supervisor respectively. The two new support scales were identical to the support scales of COPSOQ II [11].

Study 2: The BEST study The BEST study (Better Psychosocial Work Environment: a Study of Workplace Interventions) is an intervention study in 14 Danish enterprises and public institutions. The companies were evenly distributed within the industrial sector, the healthcare sector, and a group of companies characterized by information and knowledge work. The study included qualitative methods such as field observations and interviews with selected employees and managements, as well as quantitative data collection using the COPSOQ questionnaire. The main objective was to study the possibilities for a group of companies to organize and implement improvements on the psychosocial work environment on the basis of feedback from COPSOQ. The interventions comprised the following elements: (1) a contract with each of the participating companies, outlining the intentions of the company to (a) improve the psychosocial work environment, (b) form a steering group, and (c) work out an action plan for the improvement in the psychosocial work environment; (2) implementation of a questionnaire survey using COPSOQ, with feedback to each company concerning the results; (3) the companies being subsequently responsible for the preparation of the action plan and for the implementation of organizational changes aimed at improving the psychosocial work environment; and (4) the initial COPSOQ survey serving also as baseline result and a follow-up questionnaire survey being conducted 2 years after the baseline. The baseline survey took place between October 2004 and April 2005, and the follow-up survey was carried out between October 2006 and April 2007. For a detailed description of the study, see Hasle et al. [21].

BESTanalysis. For comparison with the results of the COPSOQ survey, the total sample SDs for the six scales were calculated, as well as the average withincompany SD. The self-evaluations of change were then compared with the changes in scale scores from repeated measurements based on COPSOQ, thus providing an unadjusted assessment of the MID for core COPSOQ scales. A T-test was used to evaluate the mean change. The MIDs were further analyzed by adjusting for the mean change score for the group that reported no change on the global change item. The statistical analyzes were based on the changes in the scale scores, using the linear regression model GLM in SAS. The analyzes evaluated the magnitude of change in scale scores. The independent variable was self-reported change (comparing the group indicating deterioration with the stable group and the group indicating improvement with the stable group). On the basis of the analysis of variance, the within-company variance, which can be regarded as

36

J. H. Pejtersen et al.

Table I. Characteristics of the study sample and the participating enterprises.

Enterprise

N

Information and knowledge work Bank 106 IT company 389

Gender, % women

Mean age, years (SD)

High-school graduate (%)

44.3 49.5

41.9 (8.8) 44.3 (8.4)

60.0 56.7

Consulting house

46

57.8

42.0 (10.2)

57.8

Saving bank

104

57.7

44.7 (9.8)

40.4

Industrial sector Catering

58

44.8

43.8 (7.6)

20.4

Machine factory

94

12.9

38.4 (9.1)

17.0

Transport

92

18.5

44.2 (8.9)

31.9

Print shop

82

16.1

48.2 (8.3)

9.9

218

86.1

43.1 (9.8)

35.4

Residential care for the disabled

45

77.8

46.2 (9.3)

20.0

Home care for the elderly Social welfare centre, family counselling Social welfare centre, visiting nurses Total

248 11

83.8 100

45.8 (10.0) 47.5 (7.9)

25.3 63.6

12

100

47.9 (7.6)

66.7

1505

56.3

44.1 (9.3)

38.8

Fish factorya Healthcare sector Daycare centres

a

Main activities Information and training in coping with stress Detailed analysis of COPSOQ results and department-specific plans Leader training, employee assessment interviews, and regular job satisfaction surveys Value-based leadership and employee assessment interviews Management–employee collaboration on staff reductions Participatory first-line management and value-based leadership Team organization and management–employee collaboration on staff reductions Conflicts between management and employees, no real activities implemented No activities, owing to closure Specific action plans for the 42 independent units. Special focus on absenteeism Specific action plans for the eight independent units. Focus on integration of care policy for residents and psychosocial work environment Leadership training Supervision and cross-sectional collaboration Introduction of newly appointed, collegiate supervision and cross-sectional collaboration

The fish factory closed before the follow-up.

Table II. Self-evaluations of the change in work conditions and the corresponding COPSOQ scale. Scale

Items

Self-evaluation of the change

Response options

Quantitative demands

4

Has your work pressure increased?

No, work pressure has decreased The same as before Yes, work pressure has increased

Influence at work

4

Do you have more influence on your own work?

Yes, more influence The same as before No, less influence

Predictability

2

Are you better informed about changes and plans for the future?

Yes, better informed than before The same as before No, worse informed than before

Social support from colleagues

3

Do you get better help and support from your colleagues?

Yes, better help and support than before The same as before No, worse help and support than before

Social support from supervisor

3

Do you get better help and support from your immediate superior?

Yes, better help and support than before The same as before No, worse help and support than before

Job satisfaction

4

In general, would you say that your workplace is better than before?

Yes, better than before The same as before No, worse than before

37

Determining minimally important score differences in scales of the COPSOQ the average of the variances within each company, was estimated. Finally, analyzes were conducted to evaluate the validity of the global change scores. Some studies of health outcomes have found that selfratings of change at follow-up are strongly associated with the state at follow-up but not with the baseline state [22]. Such results could suggest that self-rating of change is a measure of health state rather than change. To evaluate validity, we hypothesized that a valid self-rating of change should have a positive association with the follow-up score, but a negative association with the baseline score for given levels of the follow-up score. This was tested in a logistic

regression model with self-rated change as the dependent variable and baseline and follow-up scores as independent variables. Note that these criteria should apply no matter what the overall change is in sample mean over time. Other validity criteria have been proposed [22], but they only apply when the overall sample mean does not change over time.

Results Mean values, SDs and MIDs for the COPSOQ scales in the COPSOQ study are given in Table III.

Table III. Average scores and standard deviations (SDs) of COPSOQ scales for mailed responses; included is minimally important difference (MID) calculated as 0.5 SD. COPSOQ version Scale Quantitative demands Quantitative demands Cognitive demands Cognitive demands Emotional demands Demands for hiding emotions Sensory demands Sensory demands Influence Influence Possibilities for development Possibilities for development Degrees of freedom at work Meaning of work Commitment to the workplace Predictability Role clarity Role conflicts Quality of leadership Quality of leadership Social support Feedback Social support from colleaguesa Social support from supervisora Social relations Sense of community Job insecurity Job satisfaction Job satisfaction General health Mental health Vitality Behavioural stress Behavioural stress Somatic stress Somatic stress Cognitive stress Sense of coherence Average MID for long version scales Average MID for medium version scales a

Long

Medium

x x x x x x

x x x x

x x x x x x x x x x x x – – x x x x x x x x

x x x x x x x x x x – – x x x x x x x x

x x x

x x

Items

N

Mean

SD

MID 0.5 SD

7 4 8 4 3 2 5 4 10 4 7 4 4 3 4 2 4 4 8 4 4 2 3 3 2 3 4 7 4 5 5 4 8 4 7 4 4 9

1010 1010 1008 1008 1022 1030 1006 1008 1014 1014 1024 1024 1015 1022 1023 1035 1004 1004 978 976 1012 997 988 996 1012 982 980 1001 1005 1045 1038 1048 1042 1042 1041 1047 1039 1035

44.8 46.7 62.3 63.1 38.9 31.3 63.7 62.4 60.6 53.3 67.9 71.3 63.5 77.6 54.6 57.5 74.4 36.7 53.9 52.6 65.8 38.5 58.7 55.1 66.6 80.0 18.1 66.6 65.5 78.8 77.7 63.3 15.2 16.8 17.6 20.1 20.7 81.4

17.1 17.4 18.4 20.1 25.7 22.7 20.5 22.8 18.0 22.5 18.9 19.8 25.0 17.0 21.2 23.1 15.2 18.3 21.5 22.2 20.3 23.2 18.9 22.9 29.8 17.9 26.2 16.0 17.0 17.1 15.6 19.7 15.2 17.1 15.0 16.3 18.9 13.7

8.6 8.7 9.2 10.0 12.9 11.4 10.2 11.4 9.0 11.2 9.4 9.9 12.5 8.5 10.6 11.5 7.6 9.2 10.7 11.1 10.2 11.6 9.5 11.5 14.9 9.0 13.1 8.0 8.5 8.6 7.8 9.9 7.6 8.5 7.5 8.1 9.5 6.8 9.2 10.8

Support scales including feedback at work. Similar to the support scales in COPSOQ II [11].

38

J. H. Pejtersen et al.

The MID calculated as 0.5 SD was, on average, 9.2 (range 6.8–14.9) for the long version scales, and 10.8 (range 7.6–14.9) for the medium-length version scales. As expected, the longer scales had lower SDs and therefore a lower MID. In the BEST study, the SDs of the six scales evaluated were, generally, slightly smaller than in the COPSOQ study. The following overall sample and within-company SDs, respectively, were found: ‘‘Quantitative demands’’, 16.0 and 15.1; ‘‘Influence’’, 20.0 and 17.7; ‘‘Predictability’’, 19.8 and 17.5; ‘‘Social support from colleagues’’, 16.6 and 16.2; ‘‘Social support from supervisor’’, 20.0 and 18.9; and ‘‘Job satisfaction’’, 16.6 and 15.2. Analyzes supported the validity of the global change items. As hypothesized, all items had a positive association with the follow-up score and negative associations with the baseline scores (data not shown). The unadjusted estimates of MIDs based on the mean changes on the scales according to the response on the change items are shown in Table IV. The average change in scale score ranged from 2.5 to 10.0 for self-reported improvement and from 4.9 to 20.9 for self-reported deterioration. The group that reported no change had a change in mean scale score from 0.8 to 2.6.

The results of the linear regressions of selfevaluated changes on the scale scores from the BEST study are shown in Table IV. All regression coefficients are in the expected direction and, except for ‘‘Quantitative demands’’, there were significant relationships between self-evaluated changes and changes in scale scores. The non-significant result for ‘‘Quantitative demands’’ may be a matter of statistical power, since only 44 people evaluated the work pressure as having decreased after the interventions. In the fifth column in Table IV, the results in the fourth column have been normalized by dividing by the general population SD for the particular scale (based on results from the COPSOQ study – Table III). The fifth column shows that the anchorbased estimates of MID are generally lower than 0.5 SD. In only two of 12 cases were the MIDs larger than the 0.5 SD, and this applied for the scale ‘‘Social support from supervisors’’. On average, the MID was 0.3 SD for self-reported improvement and 0.4 SD for self-reported deterioration. This corresponds to an average change in scale score of 6.2 points for improvement and an average change of 8.9 points for deterioration. In five of the six scales, the MIDs were higher for deterioration than for improvement.

Table IV. Self-evaluated changes and estimated changes in scale scores for selected COPSOQ scales.

Self-evaluated change

N

Estimated change in scale value (95% CI)

Adjusted change in scale value (95% CI)

Normalized adjusted changea, SD units (95% CI)

Work pressure Better Same Worse

44 511 913

Quantitative demands  4.5 (9.1–0.2)  0.9 ( 1.9–0.1) 4.9 (4.0–5.8)***

 3.5 ( 7.6–0.5) 0.0 5.9 (4.4–7.3)***

 0.20 ( 0.44–0.03) 0.0 0.34 (0.26–0.42)***

Influence Better Same Worse

390 911 169

Influence at work 2.7 (1.1–4.3)**  0.8 ( 1.9–0.3)  6.1 ( 8.6 to  3.6)***

3.5 (1.6–5.5)** 0.0  5.3 ( 8.0 to  2.6)***

0.16 (0.07–0.24)** 0.0  0.24 ( 0.36 to  0.12)***

Information Better Same Worse

285 929 243

Predictability 4.6 (2.6–6.7)***  2.6 ( 3.7 to 1.4)***  11.1 ( 13.6 to  8.6)***

7.2 (4.8–9.7)*** 0.0  8.6 ( 11.2 to  5.9)***

0.31 (0.21–0.42)*** 0.0  0.37 ( 0.48 to  0.26)***

Support from colleagues Better Same Worse

214 1157 90

Social support from colleagues 2.5 (0.1–4.9)*  1.8 ( 2.8 to  0.9)**  9.0 ( 12.9 to  5.1)***

4.3 (1.8–6.7)** 0.0  7.2 ( 10.8 to  3.5)**

0.23 (0.10–0.36)** 0.0  0.38 ( 0.57 to  0.19)**

Support from supervisor Better Same Worse

275 1018 162

Social support from supervisor 10.0 (7.7–12.4)***  1.6 ( 2.8 to  0.5)*  20.9 ( 24.4 to  17.4)***

11.7 (9.1–14.2)*** 0.0  19.3 ( 22.5 to  16.1)***

0.51 (0.40–0.62)*** 0.0  0.84 ( 0.98 to  0.70)***

Workplace in general Better Same Worse

295 835 329

Job satisfaction 5.8 (4.0–7.6)***  1.6 ( 2.6 to  0.6)*  8.0 ( 9.8 to  6.2)***

7.4 (5.3–9.4)*** 0.0  6.4 ( 8.4 to  4.4)***

0.43 (0.31–0.55)*** 0.0 0.37 (0.49 to 0.26)***

a

Normalized by the SD derived from the population study in 1997 (Table III). *p < 0.05, **p < 0.001, ***p < 0.0001.

Determining minimally important score differences in scales of the COPSOQ Discussion The MIDs have been established for all scales in the COPSOQ, on the basis of the 0.5 SD rule. This yielded MIDs from 6.6 points (‘‘Sense of coherence’’ scale) to 14.9 points (‘‘Social relations’’ scale). These MIDs are based on a representative national population study in which the respondents were selected randomly. Analyzes of the BEST data showed that the within-company SDs were generally slightly lower, although the differences were not dramatic. This difference between the two studies may be explained by differences in the compositions of the two groups. In the national representative sample, all kinds of job groups and workplaces were represented in the sample, whereas the BEST study only included a subset of job groups. In this study, use of within-company SDs would have reduced the MIDs by from 0 to 2 points, depending on the scale. Whether distribution-based MIDs should be based on general population data or distributions in the relevant subpopulation is not clear. In the second part of our study, we estimated MIDs for six of the scales in COPSOQ, using an anchor-based method. The analyzes of the unadjusted mean values showed that, for the group that reported no change, the estimated changes in scale values were from 0.8 to 2.6 (Table IV). Except for ‘‘Quantitative demands’’, the group that reported no change showed slightly worse scores at follow-up – although these changes were much smaller than the estimated MIDs. The MIDs based on the unadjusted estimates for these scales may therefore be underestimated for improvement and overestimated for deterioration. For ‘‘Quantitative demands’’, this trend was in the opposite direction. From Table IV, it can be seen that the MID estimates were generally smaller than 0.5 SD. The average MID was 0.3 SD for improved conditions and 0.4 SD for worsened conditions. These results suggest that the MIDs established on the basis of 0.5 SD are fairly conservative estimates. Only for ‘‘Social support from supervisor’’ was the MID estimate larger than 0.5 SD. This estimate was very high as compared with the MIDs for the other scales. The scale concerns support from supervisor, and is the only tested scale directed towards the management. However, since the MID was large for both the positive and the negative changes, it is difficult to argue that the employees were biased in the reporting of their management. We do not have an explanation for the high MID estimate for this scale. Our study included three scales from the SF-36 Health Survey: ‘‘General health’’, ‘‘Vitality’’, and ‘‘Mental health’’ [17]. MIDs for these scales have

39

been estimated by using anchor-based methods relating score differences to risk of hospitalization, inability to work, and mortality [17]. In line with our results, these MID estimates were also smaller than 0.5 SD. For ‘‘General health’’ and ‘‘Vitality’’, the recommended MID is 0.2 SD for scores one or more SDs below the mean, and 0.3 SD for scores above this threshold. For ‘‘Mental health’’, the corresponding MID is 0.3 SD. The results from SF-36 thus support the idea that the MIDs in this study represent conservative estimates. It seems that the MID established on the basis of the intervention study was larger for worsening effects than for improvements (Table IV). Even though several studies showed that the MIDs for positive and negative changes were approximately the same [13], at least one study showed that the change score for global worsening was larger than the change score for improvement [23]. Bjorner argues that this could be because the MID may depend on the scale value [17]. Thus, when an MID for worsening is larger than a corresponding MID for improvement, this may just be because scores at different ends of the measurement scale are being compared. However, our data are too modest to allow us to draw any conclusion on this matter. The basic weakness of anchor-based methods is mainly related to the validity of the respondents’ response to the global change item [22]. One problem may be related to the validity of the change items, as it may be difficult to capture the content of a multiitem scale in a single change item. When carefully evaluating the content of the items for the single scales, we found that the change items seemed to be valid as global measures for the dimensions of ‘‘Influence at work’’ (Influence work; Say in choosing colleagues; Influence amount of work; Influence work task), ‘‘Predictability’’ (Informed about changes; Information provided to work well), ‘‘Social support from colleagues’’ (Support colleagues; Colleagues listen to problems; Colleagues talk about performance), and ‘‘Social support from supervisor’’ (Support supervisor; Supervisor listens to problems; Supervisor talks about performance). We also think that the global item of work pressure was central for the dimension of ‘‘Quantitative demands’’ (Work piles up; Complete task; Overtime; Work fast), although we, in the second version of the COPSOQ, have made separate measures for the intensive (work fast) and extensive (overtime) aspects of quantitative demands [11]. For ‘‘Job satisfaction’’, the change item was phrased ‘‘In general, would you say that your workplace is better than before?’’, whereas the job satisfaction items were about work prospects, work conditions,

40

J. H. Pejtersen et al.

work abilities, and the job in general. Although two of the items are especially related to the workplace, one may argue that ‘‘Job satisfaction’’ is more generally related to the job than the change item that is specifically related to the workplace. Evaluating MID within an intervention study (the BEST study) instead of an observational study (as is typically done in health outcomes research) has both disadvantages and advantages. On the one hand, the change item may be influenced by the respondents’ attitude towards the intervention, in addition to true changes. On the other hand, the intervention increases the likelihood that a true change has occurred and that changes in the scales and the global items are not just reflections of random ‘‘noise’’. Another limitation of the present study is that the global change items had only three response options and therefore had only one category for improvement and one for deterioration. This means that it was not possible to separate small changes from large changes in the present study. Therefore, the estimated MIDs based on the anchor-based method in the present study may be overestimated. Since the estimated MIDs still were smaller than the conventional 0.5 SD, the present results suggest that the 0.5 SD is a conservative estimate, as found by others [15,20]. Our study had a recall period of 2 years, which is long in comparison with the typical recall periods used in MID research in the health outcome field. This may increase the risk of recall bias in assessing change, and thus weaken the study. However, we think that the choice of recall period must reflect the time needed to achieve a true change within the studied field. While changes in health may occur over a fairly short time span, changes in organizations usually take longer, and a time span of 2 years is not uncommon in organizational studies [24,25]. Thus, we think that a recall period of 2 years is the best compromise for our study. While some of the potential biases discussed above might lead to underestimation of the MIDs, we believe that, all in all, the potential biases in the anchor-based methods used in this study are more likely to lead to a minor overestimation of the MIDs – mostly due to the use of only three categories in the change items. Thus, we believe that the MID estimates are conservative.

Conclusions In conclusion, we recommend the following MID values for the COPSOQ scales: ‘‘Quantitative demands’’, 0.3 SD; ‘‘Influence’’, 0.2 SD; ‘‘Predictability’’, 0.3 SD; ‘‘Social support from colleagues’’, 0.3 SD;

‘‘Social support from supervisor’’, 0.7 SD; and ‘‘Job satisfaction’’, 0.4 SD. For all other COPSOQ scales, where we do not have anchor-based results, we recommend the conventional MID value of 0.5 SD. This recommendation may be revised as more data become available.

Acknowledgements The authors would like to thank two anonymous reviewers for valuable comments and suggestions that improved the manuscript.

References [1] Kristensen TS, Hannerz H, Høgh A, Borg V. The Copenhagen Psychosocial Questionnaire – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environment Health 2005; 31:438–49. [2] Jensen PL. Assessing assessment: the Danish experience of worker participation in risk assessment. Econ Indust Democracy 2002;23:201–28. [3] Riis AH, Jensen PL. Denmark: transforming risk assessment to workplace assessment. In: Walters D, ed. Regulating health and safety management in the European Union – a study of the dynamics of change. Brussels: P. I. E. – Peter Lang; 2002. pp. 59–80. [4] Kristensen TS, Borg V, Hannerz H. Socioeconomic status and psychosocial work environment: results from a Danish national study. Scand J Public Health Suppl 2002;59:41–8. [5] Lund T, Labriola M, Christensen KB, Bu¨ltmann U, Villadsen E, Burr H. Psychosocial work environment exposures as risk factors for long-term sickness absence among Danish employees: results from DWECS/DREAM. J Occup Environ Med 2005;47:1141–7. [6] Labriola M, Lund T, Burr H. Prospective study of physical and psychosocial risk factors for sickness absence. Occup Med 2006;56:469–74. [7] Lund T, Villadsen E. Who retires early and why? Determinants of early retirement pension among Danish employees 57–62 years. Eur J Ageing 2005;2:275–80. [8] Borritz M, Rugulies R, Bjorner JB, Villadsen E, Mikkelsen OA, Kristensen TS. Burnout among employees in human service work: design and baseline findings of the PUMA study. Scand J Public Health 2006;34:49–58. [9] Hvid H, Lund H, Pejtersen J. Control, flexibility and rhythms. SJWEH Suppl 2008;6:83–90. [10] Nielsen K, Fredslund H, Christensen K, Albertsen K. Success or failure? Interpreting and understanding the impact of interventions in four similar worksites. Work Stress 2006;20:272–87. [11] Pejtersen JH, Kristensen TS, Borg V, Bjorner JB. The second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II). Scand J Public Health 2010; 38(Suppl 3):8–24. [12] Kosinski M, Zhao SZ, Dedhiya S, Osterhaus JT, Ware JE. Determining minimally important changes in generic and disease-specific health-related quality of life questionnaires in clinical trials of rheumatoid arthritis. Arthritis Rheum 2000;43:1478–87.

Determining minimally important score differences in scales of the COPSOQ [13] Norman GR, Sloan JA, Wyrwich KW. Interpretation of changes in health-related quality of life – the remarkable universality of half a standard deviation. Med Care 2003; 41:582–92. [14] Wyrwich KW, Bullinger M, Aaronson N, Hays RD, Patrick DL, Symonds T. Estimating clinically significant differences in quality of life outcomes. Qual Life Res 2005;14:285–95. [15] Bjorner JB, Wallenstein GV, Martin MC, Lin P, BlaisdellGross B, Piech CT, Mody SH. Interpreting score differences in the SF-36 Vitality scale: using clinical conditions and functional outcomes to define the minimally important difference. Curr Med Res Opin 2007;23:731–9. [16] Lydick E, Epstein RS. Interpretation of quality-of-life changes. Qual Life Res 1993;2:221–6. [17] Ware JE, Kosinski M, Bjorner JB, Turner-Bowker DM, Gandek B, Maruish ME. User’s manual for the SF-36v2 Health Survey, 2nd edn. Lincoln, RI: QualityMetric Incorporated; 2007. [18] Rugulies R, Aust B, Pejtersen JH. Do psychosocial work environment factors measured with scales from the COPSOQ predict register-based sickness absence of 3 weeks or more in Denmark? Scand J Public Health 2010;38(Suppl 3):42–50. [19] Miller GA. The magical number 7, plus or minus 2 – some limits on our capacity for processing information. Psychol Rev 1956;63:81–97.

41

[20] Spiegel BM, Younossi ZM, Hays RD, Revicki D, Robbins S, Kanwal F. The impact of hepatitis C on health related quality of life: a systematic review and quantitative assessment. Hepatology 2005;41:790–800. [21] Hasle P, Hvid H, Kristensen TS, Limborg HJ, Møller N, Pejtersen J, Hvenegaard H. Better psychosocial work environment: a study of workplace interventions – report from the BEST study. København: Det Nationale Forskningscenter for Arbejdsmiljø; 2008. [22] Norman GR, Stratford P, Regehr G. Methodological problems in the retrospective computation of responsiveness to change: the lesson of Cronbach. J Clin Epidemiol 1997;50:869–79. [23] Cella D, Hahn EA, Dineen K. Meaningful change in cancer-specific quality of life scores: differences between improvement and worsening. Qual Life Res 2002; 11:207–21. [24] Bambra C, Egan M, Thomas S, Petticrew M, Whitehead M. The psychosocial and health effects of workplace reorganisation. 2. A systematic review of task restructuring interventions. J Epidemiol Community Health 2007;61:1028–37. [25] Egan M, Bambra C, Thomas S, Petticrew M, Whitehead M, Thomson H. The psychosocial and health effects of workplace reorganisation. 1. A systematic review of organisational-level interventions that aim to increase employee control. J Epidemiol Community Health 2007;61:945–54.

Scandinavian Journal of Public Health, 2010; 38: 42–50

ORIGINAL ARTICLE

Do psychosocial work environment factors measured with scales from the Copenhagen Psychosocial Questionnaire predict register-based sickness absence of 3 weeks or more in Denmark?

REINER RUGULIES1,2,3, BIRGIT AUST1 & JAN HYLD PEJTERSEN1 1

National Research Centre for the Working Environment, Copenhagen, Denmark, 2Institute of Public Health, University of Copenhagen, Denmark, and 3Department of Psychology, University of Copenhagen, Denmark

Abstract Aims: To analyse the predictive validity of 18 psychosocial work environment scales from the Copenhagen Psychosocial Questionnaire version II (COPSOQ II) with regard to risk of sickness absence. Methods: The study population consisted of 3188 wage earners (52% women) from a representative sample of Danish residents. Participants received the long version of the COPSOQ II in autumn and winter 2004–2005, including 18 psychosocial work environment scales from the domains ‘‘Demands at work’’, ‘‘Work organization and job contents’’, and ‘‘Interpersonal relations and leadership’’. The study endpoint was register-based sickness absence of 3 weeks or more in the 1-year period following completion of the COPSOQ II. Associations between COPSOQ scales at baseline and sickness absence at follow-up were analysed with Cox proportional hazards models, adjusted for age, gender, prevalence of a health problem at baseline, and occupational grade. Results: Sickness absence during follow-up was predicted by a one standard deviation increase on the scales of cognitive demands (hazard ratio (HR) 1.17, 95% confidence interval (CI) 1.00–1.37), emotional demands (HR 1.28, 95% CI 1.10–1.50), and role conflicts (HR 1.32, 95% CI 1.15–1.52). After applying adjustment for multiple testing, the effect of emotional demands and of role conflict remained statistically significant, but not the effect of cognitive demands. Conclusions: Selected psychosocial work environment factors from the COPSOQ predict register-based sickness absence in the Danish workforce.

Key Words: Absenteeism, emotional demands, longitudinal studies, occupational health, prospective studies, psychological, questionnaires, role conflicts, sickness absence, stress, workplace

Background The Copenhagen Psychosocial Questionnaire (COPSOQ) is a relatively new instrument for measurement of the psychosocial work environment, and is used by a rapidly increasing number of researchers and work environment professionals, both in Denmark and in other countries [1]. According to Kristensen et al., the COPSOQ is ‘‘theory-based, but not attached to one specific theory, . . . should include dimensions related to worktasks, the organization of work, interpersonal relations at work, cooperation, and leadership, . . . [and] should cover potential work stressors, as well as resources’’ [2]. The COPSOQ is a comprehensive instrument that not only measures

specifically defined potentially health-hazardous constellations at work (as other questionnaires do, for example with regard to ‘‘job strain’’ or ‘‘effort–reward imbalance’’) [3,4], but has the objective of assessing all relevant aspects of the psychosocial work environment. In the words of Kristensen et al: ‘‘there should not be any significant ‘white spots’ in the picture painted’’ [2]. Social epidemiological evidence for the predictive validity of the COPSOQ scales is scarce. Neither the COPSOQ I sample (from 1997) nor the COPSOQ II sample (from 2004–2005) have ever been used to analyse prospective associations between work environment exposures and health endpoints. In this

Correspondence: Reiner Rugulies, National Research Centre for the Working Environment, Lersø Parkalle´ 105, DK-2100 Copenhagen, Denmark. Tel: þ45 39 16 52 18. Fax: þ45 39 16 52 01. E-mail: [email protected] (Accepted 21 July 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809346873

Do COPSOQ factors predict sickness absence?

43

study, we investigated whether risk of sickness absence is predicted by the COPSOQ II scales from the domains ‘‘Demands at work’’, ‘‘Work organization and job contents’’, and ‘‘Interpersonal relations and leadership’’. We chose sickness absence as the endpoint because it is regarded as a good indicator for ill-health, and it strongly predicts both disability and mortality [5–7]. Several prospective studies on work environment and sickness absence have been carried out in recent years [8]. However, most of these studies have focused only on psychosocial work environment factors that are related to specific theories, such as the demand-control model [9], the effort–reward imbalance model [10], or the organizational justice model [11]. Although conducting occupational health studies based on a specific theoretical exposure model undoubtedly has its merits [12], the aim here was to take a different approach, and investigate which psychosocial work environment factors predict sickness absence when the psychosocial work environment is measured as comprehensively as possible.

the 3-month period preceding the completion of the COPSOQ II were excluded, yielding a final study sample of 3188 participants.

Aims

Measurement of the psychosocial work environment

The aim of this study was to investigate the predictive validity of the COPSOQ. This was done by analysing the effect of 18 psychosocial work environment scales from the COPSOQ II on risk of sickness absence of 3 weeks or more during a 1-year follow-up period.

The COPSOQ II includes 18 scales in the domains ‘‘Demands at work’’ (five scales), ‘‘Work organization and job contents’’ (five scales), and ‘‘Interpersonal relations and leadership’’ (eight scales) [1]. The scales are based on the long version of the COPSOQ II, and consist of two to four items. Scales were scored from 0 to 100. The direction of the score follows the label of the scale; that is, a high score indicates high demands, high influence, and so on. The scales on quality of leadership and on social support from supervisors had substantially more missing values than the other scales, because they could only be answered by employees with superiors. A detailed description of the COPSOQ I and COPSOQ II scales and their psychometric properties have been published elsewhere [1,2,14,15], including the article by Pejtersen et al. in this special issue [1].

Material and methods Study design and sample This was a prospective analysis, linking survey data from the COPSOQ II study with register data on sickness absence. The follow-up period was 1 year. The COPSOQ II study is described in detail by Pejtersen et al. in another article in this special issue [1]. Briefly, 8000 Danish residents were randomly selected from the Danish Centralized Civil Register, of which 7834 were eligible and received a questionnaire in autumn and winter 2004–2005. Of these, 4732 provided valid responses (60.4% response rate). Among the responders, 3517 were wage earners and were selected for the data analyses in the present study. Of these, 72 participants were excluded because they had missing values on all 18 psychosocial work environment scales used in the analyses. Furthermore, 57 participants whose occupational position could not be determined were excluded. Finally, 200 participants who had an entry of a sickness absence spell in the registry for

Measurement of sickness absence The incidence of an episode of sickness absence during the 1-year follow-up period was obtained from the Danish National Register of Social Transfer Payments (DREAM). A more detailed description of DREAM and its use in epidemiological studies has been published elsewhere [13]. Briefly, DREAM contains, among other things, weekly updated information on sickness absence compensation. Sickness absence spells of 3 weeks or more were investigated because, at this point, the municipalities become responsible for managing the sickness absence case. Hence, for the purpose of this study, a study participant was recorded as having sickness absence when he or she was registered with three consecutive weeks of sickness absence within the 1-year follow-up period after completion of the COPSOQ II baseline questionnaire.

Measurement of covariates As covariates, gender, age, prevalence of a health problem at baseline and occupational grade were included. Prevalence of a health problem was measured with a 17-item checklist that included both severe chronic physical diseases, such as cancer, diabetes, or cardiovascular disease, and more unspecific disorders, such as musculoskeletal disorders, psychological disorders, or stomach-ache. Participants who responded that they currently had at least

44

R. Rugulies et al.

one of these diseases or disorders were scored as having a health problem at baseline. Occupational grade was measured by asking participants to categorize themselves as executive, non-manual worker, skilled manual worker, or unskilled manual worker. Non-manual workers were by far the largest group (n ¼ 1694 (53%)). Consequently, this group was further differentiated by adding information on post-high school education, resulting into an occupational grade variable with five categories: I, executive; II, non-manual worker with more advanced post-high school education (i.e. academic education or vocational training of more than 3 years); III, non-manual worker with less advanced post-high school education (i.e. no vocational training or vocational training of 3 years or less); IV, skilled manual worker; and V, unskilled manual worker. Data analysis Correlations between variables were calculated with Pearson’s correlation coefficients. The prospective associations between predictor variables at baseline and sickness absence during follow-up were analysed with Cox proportional hazards models, adjusted for gender, age, health problems at baseline, and occupational grade. Participants were censored at first onset of sickness absence of 3 weeks or more, emigration, death, or end of follow-up, whichever came first. Statistical significance was determined before and after adjusting for multiple testing, in accordance with the method suggested by Holm, which is also sometimes referred to as the Holm– Bonferroni method [16]. For psychosocial work environment factors that showed statistically significant associations with sickness absence (before adjusting for multiple testing), it was further investigated whether effects differed across the five occupational grades.

Results Characteristics of study participants Of the 3188 study participants, 199 (6.2%) were registered as being absent sick at one point in time during the 1 year of follow-up. Table I shows the basic characteristics of the study participants and the univariate associations of these characteristics with incidence of sickness absence. Men and women had similar hazard ratios (HRs) for sickness absence. HR increased with age. Participants with a health problem at baseline showed a statistically significantly increased HR for sickness absence during follow-up.

There was a strong inverse dose–response association between occupational grade and risk of sickness absence, with the lowest occupational grade showing the highest risk. When the variables in Table I were adjusted for each other, the HRs of all variables remained virtually the same, indicating that the effects of age, baseline health problems and occupational grade were largely independent of each other (data not shown). Correlations of the 18 psychosocial work environment scales with each other and with occupational grade Table II shows how the 18 psychosocial work environment scales correlated with each other and with occupational grade. Many of the psychosocial scales were moderately correlated with each other. Meaning of work, for example, showed a correlation coefficient 30 with 10 scales and a coefficient 60 with another two scales. The highest correlation coefficient was found for quality of leadership and social support from supervisors, with a coefficient of 0.70. Occupational grade showed correlation coefficients of 30 with the scales on cognitive demands (0.35) and possibilities for development (0.31), indicating that people of higher occupational grade were more likely to be exposed to high cognitive demands and high possibilities for development than people of low occupational grade. Psychosocial work environment at baseline as predictor for sickness absence during follow-up Table III shows the prospective association between the 18 psychosocial work environment factors at baseline and risk of sickness absence during followup. A one standard deviation increase on the scales of cognitive demands (HR 1.17), emotional demands (HR 1.28) and role conflicts (HR 1.32) predicted risk of sickness absence, after adjustment for gender, age, prevalence of a health problem at baseline, and occupational grade. After adjustment for multiple testing, emotional demands and role conflicts, but not cognitive demands, remained as statistically significant predictors of sickness absence. Effect modification by occupational grade For the three psychosocial work environment factors that had shown a statistically significant association with sickness absence before adjustment for multiple testing, further analysis was performed to determine whether the effects were different within the five occupational grades. Figure 1 shows that, for cognitive demands, the HRs were highest in occupational

Do COPSOQ factors predict sickness absence?

45

Table I. Characteristics of study participants and their association with incident sickness absence of 3 weeks or more during 1 year of follow-up. Participants

Gender Women Men Age (years) 20–29 30–39 40–49 50–60 Health problems at baseline No Yes Occupational grade I: Executive II: Non-manual worker, more advanced vocational training III: Non-manual worker, less advanced vocational training IV: Skilled manual worker V: Unskilled manual worker

Incident sickness absence

n

%

n

%

HR

95% CI

1652 1536

51.8 48.2

108 91

6.5 5.9

1 0.92

Reference 0.69–1.22

415 857 980 936

13.0 26.9 30.7 29.4

11 50 68 70

2.7 5.8 6.9 7.5

1 2.36 2.81 3.11

Reference 1.19–4.66 1.44–5.46 1.60–6.03

1904 1284

59.7 40.3

104 95

5.5 7.4

1 1.39

Reference 1.05–1.85

411 843

12.9 26.4

7 30

1.7 3.6

1 1.98

Reference 0.86–4.53

851

26.7

58

6.8

4.00

1.82–8.77

558 525

17.5 16.5

52 52

9.3 9.9

5.27 6.12

2.39–11.65 2.78–13.46

CI, confidence interval; HR, hazard ratio. HRs are based on univariate analyses.

grade I (HR 2.25). Emotional demands had the strongest effects in occupational grades II and IV (HR 1.78 and HR 1.55, respectively), whereas the effect of role conflicts was strongest in occupational grades I (HR 1.51) and II (HR 1.49). Because of the small sample sizes in the different strata, confidence intervals were wide (data not shown), and none of the interaction effects was statistically significant (data not shown).

Discussion This is the first study to use the COPSOQ II sample to investigate the effect of all factors from the domains of ‘‘Demands at work’’, ‘‘Work organization and job contents’’, and ‘‘Interpersonal relations and leadership’’ on sickness absence. We found statistically significant effects for high cognitive demands, high emotional demands, and high role conflict, although the effect of high cognitive demands was no longer statistically significant after adjustment for multiple testing. Effect estimates differed somewhat by occupational grade; however, these differences were not statistically significant. Previous studies used selected COPSOQ I scales to investigate predictors for sickness absence in Denmark. Lund et al. analysed the effect on incidence of sickness absence of 8 weeks or more in the Danish Work Environment Cohort Study [17], whereas

Rugulies et al. studied the effect on number of self-reported sickness absence days in the PUMA study [18]. In line with the results from the present study, both previous studies showed that emotional demands and role conflicts were associated with sickness absence, although strengths of association and statistical significance differed by gender and by the type of covariates included in the statistical models. Taking into consideration that the two previous studies and the present study all used different definitions of sickness absence, the consistency of the findings strongly suggests that high emotional demands and high role conflicts are important risk factors for sickness absence in Denmark. To our knowledge, the effect of cognitive demands on sickness absence have not been studied before in Denmark. The results are especially interesting, because it is thought that higher values of cognitive demands indicate a ‘‘better’’ work environment [19]. It is assumed that high cognitive demands in general are positive challenges, but with the reservation that the demands should fit with the abilities of the person [19]. One might speculate that, in the present study, the positive association between cognitive demands and sickness absence was caused by a mismatch between demands on the one hand and abilities on the other. It is also possible that high cognitive demands only have a negative effect if people lack the resources needed to cope with cognitive demands (e.g. not having enough time to think things through,

Quantitative demands Work pace Cognitive demands Emotional demands Demands for hiding emotions Influence Possibilities for development Variation Meaning of work Commitment to the workplace Predictability Reward Role clarity Role conflicts Quality of leadership Social support from supervisors Social support from colleagues Social community at work Occupational grade

1.00 0.45 0.42 0.29 0.15 0.01 0.13 0.15 0.02 0.15 0.20 0.20 0.23 0.32 0.22 0.18 0.15 0.20 0.27

3

1.00 0.35 1.00 0.18 0.48 0.18 0.25 0.09 0.37 0.10 0.56 0.03 0.36 0.11 0.38 0.07 0.17 0.08 0.07 0.07 0.10 0.03 0.07 0.23 0.23 0.08 0.06 0.08 0.05 0.07 0.10 0.07 0.03 0.03 0.35

2

1.00 0.40 0.14 0.30 0.12 0.22 0.03 0.01 0.07 0.06 0.33 0.05 0.11 0.03 0.13 0.26

4

6

7

8

9

10

11

12

13

14

15

1.00 0.07 1.00 0.04 0.49 1.00 0.05 0.37 0.46 1.00 0.04 0.37 0.66 0.34 1.00 0.08 0.39 0.56 0.33 0.62 1.00 0.08 0.36 0.42 0.18 0.46 0.58 1.00 0.14 0.42 0.43 0.23 0.49 0.63 0.65 1.00 0.04 0.23 0.31 0.06 0.49 0.46 0.54 0.53 1.00 0.28 0.09 0.04 0.08 0.15 0.29 0.29 0.32 0.26 1.00 0.11 0.33 0.42 0.19 0.42 0.60 0.63 0.68 0.46 0.31 1.00 0.11 0.29 0.33 0.15 0.33 0.51 0.50 0.63 0.39 0.22 0.70 0.09 0.28 0.34 0.16 0.33 0.42 0.36 0.44 0.28 0.15 0.44 0.13 0.24 0.27 0.19 0.37 0.49 0.43 0.52 0.39 0.32 0.48 0.04 0.19 0.31 0.26 0.13 0.09 0.06 0.04 0.06 0.00 0.06

5

Correlation coefficients 30 are in bold, and correlation coefficients 60 are in bold italics.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

1

Table II. Correlation coefficients for the 18 psychosocial work environment scales and occupational grade. 17

18

19

1.00 0.49 1.00 0.44 0.54 1.00 0.01 0.02 0.02 1.00

16

46 R. Rugulies et al.

Do COPSOQ factors predict sickness absence?

47

Table III. Prospective associations between Copenhagen Psychosocial Questionnaire version II (COPSOQ II) psychosocial work environment scales at baseline and incident sickness absence of 3 weeks or more during 1 year of follow-up. Adjusted HR Demands at work Quantitative demands (n ¼ 3164) Work pace (n ¼ 3164) Cognitive demands (n ¼ 3164) Emotional demands (n ¼ 3166) Demands for hiding emotions (n ¼ 3163) Work organization and job contents Influence (n ¼ 3164) Possibilities for development (n ¼ 3154) Variation (n ¼ 3164) Meaning of work (n ¼ 3148) Commitment to the workplace (n ¼ 3164) Interpersonal relations and leadership Predictability (n ¼ 3162) Reward (n ¼ 3147) Role clarity (n ¼ 3148) Role conflicts (n ¼ 3153) Quality of leadership (n ¼ 2443) Social support from supervisors (n ¼ 2442) Social support from colleagues (n ¼ 3063) Social community at work (n ¼ 3117)

95% CI

p

0.96 1.08 1.17 1.28 1.08

0.82–1.11 0.94–1.24 1.00–1.37 1.10–1.50 0.93–1.24

0.560 0.298 0.046a 0.001b 0.306

0.96 1.16 0.96 1.06 0.93

0.83–1.11 0.99–1.35 0.83–1.11 0.92–1.23 0.81–1.07

0.604 0.065 0.592 0.428 0.328

0.97 0.93 0.96 1.32 0.96 1.00 1.14 1.06

0.84–1.11 0.81–1.06 0.83–1.10 1.15–1.52 0.82–1.11 0.86–1.16 0.99–1.32 0.92–1.23

0.636 0.285 0.536 <0.001b 0.565 0.992 0.078 0.419

3

Cognitive demands 2.25

2 1.19

1.27

1.07

1.13

1 0 I (highest)

II

III

IV

Hazard ratio (HR) for sickness absence

Hazard ratio (HR) for sickness absence

CI, confidence interval; HR, hazard ratio. HRs are adjusted for gender, age (continuous), prevalence of a health problem at baseline, and occupational grade. Psychosocial work environment scales are not adjusted for each other. aStatistically significant before but not after adjustment for multiple testing. bStatistically significant before and after adjustment for multiple testing.

3

Emotional demands

1.78

2

0

Hazard ratio (HR) for sickness absence

2

III

II

Occupational grade

3

1.04

1

I (highest)

V (lowest)

1.55 1.15

1.09

IV

V (lowest)

Occupational grade

Role conflicts

1.51

1.49

I

II

1.27

1.35

III

IV

1.25

1 0 (highest)

V (lowest)

Occupational grade

Hazard ratios are adjusted for gender, age (continuous) and prevalence of a health problem at baseline.

Figure 1. Hazard ratios of incident sickness absence for a one standard deviation increase on three psychosocial work environment scales stratified by occupational grade.

48

R. Rugulies et al.

or facing role conflicts that distract them from concentrating on problems). The role of occupational grade Occupational grade was a very strong predictor for sickness absence in this study. We first treated occupational grade as a confounder; that is, we adjusted the effect of the psychosocial work environment scales on sickness absence by occupational grade. Our reasoning for this was that occupational grade might be a proxy measure for physical workload (e.g. heavy lifting or awkward working postures) that could be associated with both psychosocial working conditions and sickness absence. In the analyses by Lund et al., adjustment for physical working conditions substantially decreased the effect of several psychosocial work environment variables [17]. Moreover, occupational grade is also an indicator of socioeconomic position and therefore might also capture non-work-related stressors, such as daily problems, financial strains, or conflicts in private life, that might influence both self-report of working conditions and sickness absence. In the next step, we treated occupational grade not as a confounder, but as an effect modifier. The results have to be viewed with caution, because confidence intervals were wide, owing to the relatively low number of participants in the different strata. However, if one just focuses on the HRs, one can see that the harmful effect of high cognitive demands on sickness absence was strongest in the highest occupational grade. This means that high cognitive demands had the strongest effect in the occupational grade with the lowest prevalence of sickness absence. It would have been interesting to explore this further by stratifying the analyses not just by the rather crude measure of occupational grade, but by job groups, such as engineers, nurses, police force, and cleaners. However, a meaningful analysis on this basis was not possible, because of the lack of statistical power resulting from the limited number of study participants. Methodological and theoretical considerations Adverse psychosocial working conditions might affect health directly via psychoneuroendocrinological and psychoneuroimmunological processes, or more indirectly via poor health behaviours (such as smoking, excessive alcohol consumption, and lack of exercise) [20]. There is good evidence that both socioeconomic position [21] and psychosocial working conditions influence health behaviours [22], and there is excellent evidence that health behaviours

greatly impact on health [23], but there is little evidence that health behaviours influence reporting of the psychosocial working environment. Hence, health behaviours are likely to be an intermediate step in the causal pathway that links the psychosocial work environment to health. Consequently, we decided not to adjust our analyses for indicators of health behaviours, such as smoking, alcohol consumption, or body mass index. However, we adjusted the analyses for prevalence of health problems at baseline. This adjustment might partly be an appropriate confounder control, if we assume that people with health problems will be more likely to report adverse working conditions (because of reduced ability to work) and also will be at higher risk for sickness absence. However, it also might partly be an inappropriate adjustment for a causal step in the pathway, if we assume that psychosocial working conditions have contributed to the onset of the health problems. Hence, the reader should be aware that, with regard to health problems at baseline, our analyses might, to some extent, be overadjusted. From a theoretical point of view, we believe that it will be important for the further development of the COPSOQ to obtain more insights into how the different psychosocial constructs are related to each other. To what extent are the constructs independent from each other and to what extent do they overlap? How do the exposures influence each other, and which exposures might cluster under which conditions? The correlation coefficients in Table II showed that several of the scales were indeed moderately correlated with each other, and some scales even showed correlations greater than 0.60. However, it remains unclear whether these correlations indicate conceptual overlap between scales or whether they indicate that scores on some scales had caused scores on other scales. For example, exposure to poor leadership quality might cause both a high level of role conflicts and a low level of predictability at work. Because of the unclear nature of the associations between the 18 different psychosocial work environment scales, we decided not to adjust the scales for each other in the analyses on sickness absence. If scores on some scales had indeed influenced scores on other scales, mutual adjustment would not be an appropriate confounder control, but would be inappropriate adjustment for a step in the causal pathway. Moreover, it is also possible that the effects of some scales on sickness absence were modified by other scales. For example, high emotional demands might have a less adverse effect on sickness absence if a person experiences his or her work as very meaningful. In this case of effect modification, mutual

Do COPSOQ factors predict sickness absence? adjustment of the psychosocial work environment scales would also be inappropriate. Limitations Several limitations of the study have to be considered. First, because of sample size restrictions, we could not analyse whether the psychosocial work environment scales had differential effects in different job groups. It is possible that some of the scales affect health only in certain job groups and not in others. This needs to be further investigated. Second, it would have been interesting to analyse whether the effect of the psychosocial work environment was stronger on sickness absence because of mental or because of physical health problems. However, this was not possible, because data on cause-specific sickness absence were not available for this study. Third, we did not have information about physical demands, and we had only very limited information about non-work-related determinants of sickness absence. Adjusting for occupational grade might have captured some of these potential confounders, as discussed above, but certainly not all. Fourth, our analyses were restricted to the COPSOQ domains of ‘‘Demands at work’’, ‘‘Work organization and job contents’’, and ‘‘Interpersonal relations at leadership’’. We also considered analysing the new domain ‘‘Values at the workplace level’’, including the construct of ‘‘Social capital at work’’, which has become an emerging topic in psychosocial occupational health research [24,25]. However, we felt that we needed more time to think about the theoretical implications of this new domain, before testing it in an epidemiological analysis.

Conclusion Selected COPSOQ scales predicted register-based sickness absence of 3 weeks or more in this representative sample of the Danish workforce. With regard to emotional demands and role conflicts, we conclude that there now is evidence from three Danish prospective studies for the contribution of these two factors to sickness absence. With regard to cognitive demands, the findings presented here are new and need to be replicated in future studies. The effect of the psychosocial work environment on sickness absence was, to some degree, moderated by occupational grade. Further studies should investigate whether the strength and direction of the associations between psychosocial work environment scales from the COPSOQ and health endpoints differ by job group.

49

Acknowledgements We thank Tage Søndergaard Kristensen for many years of enlightening, critical and thought-provoking discussions. It was a great pleasure and privilege to be your colleague at NRCWE, and we are very much looking forward to our future collaborations. The analyses of this study were partly funded by a grant from the Danish Working Environment Research Fund (Grant Number: 5-2006-04). References [1] Pejtersen JH, Kristensen TS, Borg V, Bjorner JB. The second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II). Scand J Public Health 2010;38(Suppl 3): 8–24. [2] Kristensen T, Hannerz H, Høgh A, Borg V. The Copenhagen Psychosocial Questionnaire. A tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31:438–49. [3] Karasek R, Brisson C, Kawakami N, Houtman I, Bongers P, Amick B. The Job Content Questionnaire (JCQ): an instrument for internationally comparative assessments of psychosocial job characteristics. J Occup Health Psychol 1998;3:322–55. [4] Siegrist J, Starke D, Chandola T, Godin I, Marmot M, Niedhammer I, et al. The measurement of effort–reward imbalance at work: European comparisons. Soc Sci Med 2004;58:1483–99. [5] Kivima¨ki M, Forma P, Wikstrom J, Halmeenmaki T, Pentti J, Elovainio M, et al. Sickness absence as a risk marker of future disability pension: the 10-town study. J Epidemiol Community Health 2004;58:710–11. [6] Kivima¨ki M, Head J, Ferrie JE, Shipley MJ, Vahtera J, Marmot MG. Sickness absence as a global measure of health: evidence from mortality in the Whitehall II prospective cohort study. BMJ 2003;327:364. [7] Vahtera J, Pentti J, Kivima¨ki M. Sickness absence as a predictor of mortality among male and female employees. J Epidemiol Community Health 2004;58:321–6. [8] Duijts SF, Kant I, Swaen GM, van den Brandt PA, Zeegers MP. A meta-analysis of observational studies identifies predictors of sickness absence. J Clin Epidemiol 2007;60:1105–15. [9] Theorell T, Karasek R. Current issues relating to psychological job strain and cardiovascular disease research. J Occup Health Psychol 1996;1:9–26. [10] Siegrist J. Adverse health effects of high-effort/low-reward conditions. J Occup Health Psychol 1996;1:27–41. [11] Elovainio M, Kivima¨ki M, Vahtera J. Organizational justice: evidence of a new psychosocial predictor of health. Am J Public Health 2002;92:105–8. [12] Levi L, Bartley M, Marmot M, Karasek R, Theorell T, Siegrist J, et al. Stressors at the workplace: theoretical models. Occup Med 2000;15:69–106. [13] Hjollund NH, Larsen FB, Andersen JH. Register-based follow-up of social benefits and other transfer payments: accuracy and degree of completeness in a Danish interdepartmental administrative database compared with a population-based survey. Scand J Public Health 2007;35:497–502. [14] Kristensen TS, Borg V, Hannerz H. Socioeconomic status and psychosocial work environment: results from a Danish national study. Scand J Public Health 2002;30:41–8.

50

R. Rugulies et al.

[15] Aust B, Rugulies R, Skakon J, Scherzer T, Jensen C. Psychosocial work environment of hospital workers: validation of a comprehensive assessment scale. Int J Nurs Studies 2007;44:814–25. [16] Holm S. A simple sequentially rejective multiple test procedure. Scand J Statistics 1979;6:65–70. [17] Lund T, Labriola M, Christensen KB, Bu¨ltmann U, Villadsen E, Burr H. Psychosocial work environment exposures as risk factors for long-term sickness absence among Danish employees: results from DWECS/DREAM. J Occup Environ Med 2005;47:1141–7. [18] Rugulies R, Christensen KB, Borritz M, Villadsen E, Bu¨ltmann U, Kristensen TS. The contribution of the psychosocial work environment to sickness absence in human service workers. Results from a 3-year follow-up. Work Stress 2007;21:293–311. [19] National Research Centre for the Working Environment. Kortlægningen af det psykiske arbejdsmiljø med AMI’s spørgeskema om psykisk arbejdsmiljø. Hvad betyder de forskellige dimensioner? [Mapping the psychosocial work environment with NRCWE’s questionnaire on psychosocial work environment. What do the different dimensions mean?]. Available at: http://www.arbejdsmiljoforskning.dk/upload/ 3d-ii-dimensioner.pdf (accessed 28 January 2009).

[20] Rugulies R, Aust B, Syme S. Epidemiology of health and illness. A socio-psycho-physiological perspective. In: Sutton S, Baum A, Johnston M, editors. The Sage handbook of health psychology. London: Sage; 2004. pp. 27–68. [21] Mackenbach JP. Health inequalities: Europe in profile. An independent, expert report commissioned by the UK Presidency of the EU (February 2006). Available at: http://www.dh.gov.uk/assetRoot/04/12/15/84/04121584.pdf (accessed 28 November 2006). [22] Siegrist J, Ro¨del A. Work stress and health risk behavior. Scand J Work Environ Health 2006;32:473–81. [23] Mokdad AH, Marks JS, Stroup DF, Gerberding JL. Actual causes of death in the United States, 2000. JAMA 2004;291:1238–45. [24] Olesen KG, Thoft E, Hasle P, Kristensen TS. Virksomhedens sociale kapital. Hvidbog [White paper on organizational social capital]. København: Arbejdsmiljøra˚det og Det Nationale Forskningscenter for Arbejdsmiljø; 2008. [25] Oksanen T, Kouvonen A, Kivimaki M, Pentti J, Virtanen M, Linna A, et al. Social capital at work as a predictor of employee health: multilevel evidence from work units in Finland. Soc Sci Med 2008;66:637–49.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 51–58

ORIGINAL ARTICLE

Positive work-related states and long-term sickness absence: A study of register-based outcomes

THOMAS CLAUSEN1, KARL BANG CHRISTENSEN2 & VILHELM BORG1 1

National Research Centre for the Working Environment, Copenhagen, Denmark, and 2Department of Biostatistics, Institute of Public Health, University of Copenhagen, Øster Farimagsgade 5, Copenhagen, Denmark

Abstract Aims: To investigate the association between positive work-related states and long-term sickness absence (LTSA). The positive states that were investigated were commitment to the work-place (CW) and experience of meaning of work (MW). Methods: This association was investigated using Poisson regression analysis. Data consisted of a merge between Danish register data on sickness absence compensation and survey data collected among 9,560 employees in the Danish eldercare sector. Results: CW and MW were significantly associated with LTSA. Employees experiencing low MW had a significantly increased risk of LTSA for more than two and eight weeks, when adjusted for psychosocial work characteristics, work-time arrangements and physical workload. Compared to employees with low and high CW, employees with medium CW had a significantly decreased risk of LTSA for more than eight weeks, when adjusted for psychosocial work characteristics, work-time arrangements and physical workload. Furthermore, employees with low CW had an increased risk of LTSA for more than two weeks, but this association became borderline insignificant when adjusted for psychosocial work characteristics, work-time arrangements and physical workload. The analyses also revealed an interaction effect between CW and MW in predicting LTSA for more than eight weeks. Conclusions: CW and MW are associated with LTSA. Against our expectations, however, we found that high levels of CW and MW were not protective against LTSA. Instead, low levels of MW proved decisive in predicting LTSA, and medium levels of CW had a protective effect on LTSA for more than eight weeks.

Key Words: Copenhagen Psychosocial Questionnaire (COPSOQ), eldercare, meaning at work, organizational commitment, register data, sickness absence, work environment

Introduction Over the past decade increasing attention has been devoted to the study of positive states, behaviours and outcomes within the field of work and organizational psychology [1–3] and some research has focused on the association between positive workrelated states and work-related absence. In some studies positive work-related states, such as organizational commitment [3–6] and experience of meaning at work [7] were found to be inversely associated with work-related absence, whereas other studies were unable to confirm such associations [8,9].

Theoretical work by Steers and Rhodes [10], however, suggest that positive work-related states must be considered crucial in order to understand work-related absence. I would be very grateful if you could substitute the marked sentence with the following sentence: Steers & Rhodes regard motivation as a central element in understanding work-related absence and conceptualize motivation as a function of (a) an employee’s affective responses to the job situation and (b) various internal and external pressures to attend (p.393).

As a complement to this motivation-based account, other studies offer a more health-oriented

Correspondence: Thomas Clausen, National Research Centre for the Working Environment, Lersø Parkalle´ 105, DK-2100 Copenhagen, Denmark. Tel: þ45 39165368. Fax: þ45 39165201. E-mail: [email protected] (Accepted 23 September 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809352105

52

T. Clausen et al.

perspective on the association between positive workrelated states and sickness absence. These studies indicate that positive emotions and states may have a positive impact on physical and mental health [11,12], which seems to indicate that experiences of positive affect in the work-place may prove important resources in reducing sickness absence. Accordingly, the literature provides evidence for expecting an association between positive workrelated states and sickness absence, and, hence, the aim of this study is to investigate whether positive work-related states are associated with long-term sickness absence (LTSA) as registered in a national Danish register. According to Meyer and colleagues [13] positive work-related states can be construed as psychological attachments of employees towards various workrelated foci. The common denominator for these positive work-related states is that they are characterized by positive arousal and that they are driven by organizational (i.e. work-related) phenomena. The positive work-related states that will be investigated are commitment to the work-place and experience of meaning of work. According to Meyer and Allen [14] the concept of affective organizational commitment refers to the employee’s identification with and emotional attachment to the organization. For the affectively committed employee, participation in organizational life becomes an end in itself and can increasingly be characterized by intrinsic motivation [15]. Accordingly, the concept of affective organizational commitment taps into employees’ psychological attachment to the work-place. In this study, affective organizational commitment will be measured using the scale ‘‘commitment to the work-place’’ from the Copenhagen Psychosocial Questionnaire (COPSOQ) [16,17]. The concept of meaning at work describes the sense that individuals subjectively make of their work situation [18]. A central feature in experiencing meaning at work is the experience of the possibilities of expressing oneself through work-related activities [19], and, accordingly, experiences of congruency between personal values and work activities may contribute to enhancing individual identification with working activities [18,20]. The concept of meaning at work thus gauges employees’ psychological attachment to more specific job tasks, and meaning at work will be measured using the ‘‘meaning of work’’ scale from COPSOQ [16,17]. In the literature, however, abundant evidence demonstrates the association between psychosocial work characteristics and sickness absence [7,21–24]. As positive work-related states have also been found to be positively associated with psychosocial

work characteristics such as influence and quality of leadership and negatively associated with role ambiguity [3,8,25,26], it may well be that the above cited associations between positive work-related states and absence from work are of a spurious nature, as the detected associations between positive states and sickness absence in reality may be mirroring associations between sickness absence and psychosocial work characteristics. It therefore appears pertinent to adjust for psychosocial work characteristics when investigating the association between positive work-related states and sickness absence. Furthermore, other studies have shown that risk of LTSA is also associated with work-time arrangements among staff in the eldercare services [27] and with physical work environment exposures in the general population [28]. Therefore, the aim of this paper is to investigate the association between positive work-related states – commitment to the work-place and experience of meaning of work – and registered LTSA amongst employees in the Danish eldercare-services. Hypothesis: Employees with high levels of commitment to the work-place and meaning of work have lower levels of registered LTSA than employees with medium or low levels of commitment to the workplace and meaning of work. We define LTSA as absence periods lasting two weeks or more and compare rates of LTSA for those with low, medium, and high levels of meaning and commitment. We also conduct additional analyses where we define LTSA as absence periods lasting eight weeks or more. Even though these extended periods affect a relatively small amount of people, they are of particular importance, as they account for a large part of the total amount of sickness absence days. It will furthermore be assessed whether positive work-related states significantly influence risk of LTSA when we adjust for psychosocial work characteristics, work-time arrangements and physical workload.

Methods Population This study is based on a merger of survey-data collected among employees in the eldercare services in 35 Danish municipalities in 2004–5 and a national register, DREAM, on social transfer payments [23]. The aim of the survey was to investigate associations between physical and psychosocial work characteristics, health, and well-being among employees in the Danish eldercare services. The survey included 12,746 employees and yielded a response rate of

Positive work-related states and long-term sickness absence 78% (9,949 persons). An analysis of non-response shows that non-respondents had higher absence rates than respondents. Respondents were followed in the DREAM register for one year after completion of the survey. Three hundred and eighty nine respondents had missing values on the scales measuring commitment to the work-place or the scale on meaning of work and the dataset used in the analyses therefore contains information on 9,560 persons. DREAM contains weekly information on granted sickness absence compensation for all citizens and residents in Denmark. However, the DREAM register contains no diagnostic information on the underlying health reasons on which sickness absence is grounded. Sickness absence compensation is given to the employer, who can apply for a refund from the State for employees after two weeks of sickness absence. In this study we have decided to investigate the association between positive work-related states and LTSA for absence periods lasting for two or more weeks because of this. An additional analysis focusing on absence lasting eight or more weeks was also conducted. This has been applied because the municipal social security departments are obliged to initiate action plans for sickness absent persons after eight weeks of absence. Outcome: Sickness absence Sickness absence was defined as (a) two or more consecutive weeks of absence in the one year followup period and (b) eight or more consecutive weeks of sickness absence in the one year follow-up period that started upon completion of the survey. Accordingly, the absence periods in (b) are a subset of absence periods in (a).

Positive work-related states and work characteristics Positive work-related states and psychosocial work characteristics were measured using 10 scales from the COPSOQ [16]. Table I provides an overview of these scales and the applied measures of physical workload, work-time arrangements, and self-rated health. Except for the items measuring physical workload and work-time arrangements, response was scored on five point Likert-scales, which were scored from 0–100 with 100 representing the highest degree of the measured state or work characteristic. Although this study is based on a sample of employees in the Danish eldercare sector, we wished to apply an external standard in assessing

53

low, medium, and high levels of commitment to the work-place and experience of meaning of work. These scales were therefore divided into three levels – low, medium, and high – on the basis of cut-points calculated from the upper and lower quartile-values from data from the COPSOQ-survey conducted in 2004–5 among a representative sample of Danes aged 20–59 years [17]. Demographics, occupation, and health behaviour All analyses were adjusted for age, gender, cohabitation (living with spouse/partner: yes/no), children living at home (yes/no), tenure, job function (care work/other work), smoking status (smoker/nonsmoker), and body mass index (BMI). Analyses We examined the sickness absence rates in the population, i.e. the total number of sickness absence episodes in the population divided by the total risk time in the population. The risk time was calculated as time from answering the questionnaire until first onset of sickness absence or end of the study period. Furthermore, subjects were censored at the time of death, immigration or retirement. Poisson regression analysis was used to calculate the rate ratios (RR) and 95% confidence intervals (95% CI). We only considered the first sickness absence episode for each person for two reasons: (i) persons with more than one absence period will influence the analysis more than those with a single sickness absence episode, and (ii) it is likely that the risk of sickness absence changes after a period of LTSA, and a model taking this into account is beyond the scope of the present study. Employees who were absent from work at baseline were excluded from the analysis. Our analyses were adjusted in three steps. In the first step we adjusted for potential confounders regarding demographics, occupation, and health behaviour. In the second step we adjusted for psychosocial work characteristics, and in the third step we furthermore adjusted for physical workload and work-time arrangements. Models were reduced using backwards stepwise elimination of the least significant variables – excepting the confoundervariables on demographics, occupation, and health behaviour. To test for interaction effects the scales measuring commitment to the work-place and experience of meaning of work were simultaneously entered into our explanatory models in conjunction with an interaction term consisting of the two scales multiplied with each other. Data were analyzed using SAS 9.1.

54

T. Clausen et al.

Results Of the 9,560 persons who entered the study 1,529 were absent for more than two consecutive weeks in the one year follow-up period and 568 were absent for more than eight consecutive weeks during followup. Table II shows baseline descriptive statistics and Spearman correlations among study variables. For descriptive purposes, Table II also includes baseline information on self-rated health. Table III shows the distribution of the respondents on the three levels of commitment to the work-place

(CW) and meaning of work (MW), along with rate ratios and 95% confidence intervals showing their association with LTSA. Model 1 in Table III shows that CW is significantly associated with long-term sickness absence. Employees who exhibit low CW have a significantly increased risk for being absent for more than two consecutive weeks and more than eight consecutive weeks in the follow-up period. Model 1 in Table III shows that there are no significant differences between groups exhibiting medium and high CW.

Table I. Overview of scales and sample items.

Scale name

Number of items

Commitment to the work-place Meaning of work Emotional demands Work pace Quantitative demands Role clarity Influence Possibilities for development Quality of leadership

4 3 4 1 2 3 4 4 4

Predictability

2

Physical workload Work-time arrangements Self-rated health

1 1 1

Cronbach’s alpha

Sample item Do you enjoy telling others about your place of work? Do you feel that the work you do is important? Is your work emotionally demanding? Is it necessary for you to work very fast? Do you have enough time for your work tasks? Do you know exactly how much say you have at work? Do you have a large degree of influence concerning your work? Do you have the possibility of learning new things through your work? To what extent would you say that your immediate superior is good at work planning? At your place of work, are you informed well in advance about, for example, important decisions, and changes of plans for the future? How do you estimate the total physical load in working with the users? At which hours of the day do you normally work? In general, how would you describe your health?

Table II. Baseline descriptive statistics and Spearman correlations among study variables. Scale name

Mean

SD

1

2

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

60.7 77.8 44.6 64.7 37.9 73.9 45.9 71.9 57.1 56.8 53.5 62.0 65.7 45.4 95.6 80.7 52.3 81.8

19.2 14.0 19.2 20.0 21.2 15.0 20.7 14.5 21.9 19.9 18.4 21.5

* 0.52*** 0.15*** 0.20*** 0.15*** 0.37*** 0.35*** 0.44*** 0.50*** 0.51*** 0.25*** 0.20*** 0.04** 0.10*** 0.01 0.04*** 0.04*** 0.09***

0.52*** * 0.06*** 0.01 0.08*** 0.52*** 0.25*** 0.60*** 0.29*** 0.36*** 0.11*** 0.18*** 0.03* 0.03** 0.03** 0.05*** 0.02 0.02

0.01 0.05*** 0.00

0.02 0.02 0.05***

19 20 21

Commitment to the workplace Meaning of work Emotional demands Work pace Quantitative demands Role clarity Influence Possibilities for development Quality of leadership Predictability Physical workload Self-rated health Working day-shift (percent) Age Women (percent) Cohabiting with spouse, partner (percent) Children living in the household (percent) Engaged in the provision of care services (percent) Tenure (years) Body mass index Regular smoker (percent)

*p < 0.05, **p < 0.01, ***p < 0.0001.

9.1 25.0 35.6

10.0

7.2 4.4

0.74 0.69 0.81 – 0.67 0.75 0.75 0.58 0.89 0.76 – –

Positive work-related states and long-term sickness absence

55

eight week absence period we found that the interaction term was significantly associated with LTSA (Model 1: p ¼ 0.0001, Model 2: p ¼ 0.0006, and Model 3: p ¼ 0.0002), when the individual scales measuring CW and MW were included in the models. The interaction effect for Model 1 for the eight week absence period is illustrated in Table IV. Table IV shows that combinations of low CW and low MW on the one hand and low CW and high MW on the other entail an increased risk of LTSA for more than eight weeks. The table furthermore shows that the combination of medium CW and high MW leads to a decrease in the risk of being absent from the work-place for more than eight consecutive weeks in the one year follow-up period. Finally, the results in Table IV were by and large reproduced when we adjusted for psychosocial work characteristics, work-time arrangements, and physical workload (results not shown).

In Model 2 we further adjust our model with eight psychosocial work characteristics. The group exhibiting low CW has an increased risk of being absent for two consecutive weeks in the follow-up period. For the eight-week period the group with low CW does not differ significantly from employees with high CW when we adjust for psychosocial work characteristics. The group exhibiting a medium level of CW is faced with a significantly lower risk of recording an eight week absence period, when compared to the group exhibiting high and low levels of CW. These results are by and large reproduced in Model 3 where we also adjust for physical workload and work-time arrangements. With regard to meaning of work (MW), Model 1 in Table III shows that experiences of MW are significantly associated with long-term sickness absence. Employees who experience low MW have a significantly increased risk of being absent for more than two and eight consecutive weeks during follow-up. These results are reproduced in Model 2 and Model 3. We also investigated whether we could observe an interaction effect between CW and MW in predicting LTSA. The results showed that the interaction term was not significantly associated with LTSA in the two-week absence period (Model 1: p ¼ 0.0625, Model 2: p ¼ 0.0976, and Model 3: p ¼ 0.0674), when the individual scales measuring CW and MW were included in the models. When analysing the

Discussion The aim of this study was to investigate whether positive work-related states, such as commitment to the work-place and experience of meaning of work, were associated with long-term sickness absence as registered in the Danish DREAM register. The analyses showed that CW and experience of MW were significantly associated with risk of LTSA for

Table III. Rate ratios (RR) and 95% confidence intervals (95% CI) for onset of long-term sickness absence during the 12 months of followup for low, medium and high levels of commitment to the work-place and experience of meaning of work. Model 1a Absence period More than two weeks

Commitment to the workplaced

More than two weeks

Experience of meaning of workd

More than eight weeks

Commitment to the workplaced

More than eight weeks

Experience of meaning of workd

a

Model 2b

Model 3c

Level

%

RR

(95% CI)

RR

(95% CI)

RR

(95% CI)

High Medium Low High Medium Low High Medium Low High Medium Low

18.2 49.2 32.6 42.4 52.3 5.3 18.2 49.2 32.6 42.4 52.3 5.3

1.00 0.98 1.36 1.00 1.02 1.49 1.00 0.86 1.40 1.00 0.94 1.51

– (0.86–1.11) (1.19–1.55) – (0.93–1.13) (1.24–1.79) – (0.74–1.00) (1.20–1.63) – (0.84–1.06) (1.21–1.87)

1.00 0.96 1.25 1.00 1.04 1.45 1.00 0.79 1.17 1.00 0.89 1.34

– (0.84–1.10) (1.07–1.45) – (0.94–1.15) (1.19–1.78) – (0.67–0.93) (0.98–1.41) – (0.79–1.01) (1.05–1.70)

1.00 0.95 1.17 1.00 1.03 1.33 1.00 0.78 1.10 1.00 0.90 1.35

– (0.83–1.09) (1.00–1.37) – (0.92–1.14) (1.07–1.65) – (0.66–0.92) (0.91–1.32) – (0.79–1.02) (1.06–1.72)

Rate ratios are adjusted for age, gender, cohabitation, children living at home, job function, tenure, BMI, and smoking status. Rate ratios are adjusted for eight psychosocial work characteristics (work pace, quantitative demands, emotional demands, role clarity, influence, possibilities for development, predictability, and quality of leadership), age, gender, cohabitation, children living at home, job function, tenure, BMI, and smoking status. cRate ratios are adjusted for eight psychosocial work characteristics (work pace, quantitative demands, emotional demands, role clarity, influence, possibilities for development, predictability, and quality of leadership), age, gender, cohabitation, children living at home, job function, tenure, BMI, smoking status, physical workload, and work-time arrangements. d n ¼ 9,560. b

56

T. Clausen et al. Table IV. Rate ratios for interaction between commitment to the workplace and experience of meaning of work on sickness absence periods for more than eight weeksa.

Low meaning of work Medium meaning of work High meaning of work

Low commitment to the workplace

Medium commitment to the workplace

High commitment to the workplace

1.71 (1.3–2.20) n ¼ 426 1.13 (0.94–1.35) n ¼ 1,975 1.78 (1.45–2.18) n ¼ 715

0.81 (0.40–1.67) n ¼ 76 0.87 (0.72–1.04) n ¼ 2,648 0.78 (0.64–0.95) n ¼ 1,980

– n¼6 0.85 (0.62–1.19) n ¼ 374 1 (reference) n ¼ 1,360

a

Rate ratios are adjusted for age, gender, cohabitation, children living at home, job function, tenure, BMI, and smoking status.

absence periods of over two weeks and eight weeks and overall these associations remained statistically significant when adjusted for a series of psychosocial work characteristics, physical workload and worktime arrangements. As stated in our hypothesis we expected that employees with high levels of commitment to the work-place and meaning of work would have lower rates of LTSA than their colleagues. The results, however, did not support our hypothesis. On the contrary, the general trend in our results show that low levels of CW and MW are associated with higher risk of sickness absence, whereas high levels of CW and MW are not associated with lower risk of LTSA than medium levels of CW and MW. Our hypothesis regarding the salience of positive work-related states in predicting LTSA was informed by two theoretical perspectives. On the one hand, Steers and Rhodes’ theory [10] would suggest that high levels of CW and MW would increase employee motivation and thereby decrease the level of absence recorded by the given employee. An alternative perspective stressed the positive health-related consequences of positive states. Firstly, Keyes [11] stated that the satisfaction of a series of psychological needs, such as experiences of purpose and meaningfulness (MW) and a sense of social belongingness (CW) contributed to bringing individuals into a state of ‘‘complete mental health’’ (p.100), which again should reduce the risk of LTSA. Secondly, Taylor and colleagues [12] stated that positive emotions and states may in themselves directly and indirectly protect against physical disease, thereby also reducing the risk of LTSA. These two explanatory models thus stress the importance of positive states in explaining LTSA. However, to the extent that better-than-average levels of CW and MW can be characterized as positive work-related states, the present study could not affirm the salience of positive work-related states.

Interestingly, medium levels of CW are associated with significantly lower risk of sickness absence for eight or more weeks in Models 2 and 3 in Table III, while for MW the association is borderline significant. Although this is not ‘‘classic’’ overcommitment [30] it may imply that employees who are strongly committed to their work organization or who experience their work as very meaningful feel indispensable at their work-place to a higher extent than colleagues with more moderate psychological attachments to their work and their work-place. This may entail that employees with high CW and MW overtax their physical or mental resources, thus increasing the risks of long-term physical or mental illness. This may particularly be accentuated for the occupational group that form the basis of this study, because employees in the eldercare sector are inherently engaged in the provision of human services and feelings of loyalty towards colleagues and clients may entice employees to attend work in spite of ill health. The analyses also uncovered an interaction effect of CW and experience of MW on LTSA of more than eight weeks. Employees who simultaneously experienced low MW and low CW were at an increased risk of LTSA of more than eight weeks. We were more surprised to note that the group that exhibited low CW and high MW were faced with a similar risk of LTSA of more than eight weeks. One explanation for this may be that the efforts of these employees to perform their highly meaningful work tasks are frustrated by the lack of organizational capacity to support these efforts, which again may increase the risk of burnout [29]. The concepts of affective organizational commitment and meaning at work were operationalized using the COPSOQ scales ‘‘commitment to the work-place’’ and ‘‘meaning of work’’. The items in the commitment to the work-place scale have a clear affinity with Meyer et al’s scale on affective organizational commitment [30] and orient themselves towards the relationship between the employee and

Positive work-related states and long-term sickness absence the work organization. The meaning of work scale orients itself towards the relation between the individual and the work tasks and therefore also appears to be in accordance with the reflections on the concept presented in the introduction. The two scales also show acceptable Cronbach’s alpha values and their intercorrelations with the COPSOQ measures of psychosocial work characteristics also indicate acceptable discriminant validity. On this background, this study adds to the existing knowledge on the consequences of affective organizational commitment and experience of meaning at work.

Limitations and strengths The DREAM register contains no diagnostic information on the underlying health reasons that sickness absence is grounded upon. It can be considered a weakness of the study that we are not able to identify more specific types of diagnoses that are associated with low levels of positive work-related states. However, this lack of diagnostic information should not remove our attention from the findings that demonstrate an association between CW, MW and LTSA. We also consider it a weakness of this study that we are not able to provide evidence on the processual aspects of the association between CW, MW and LTSA. It should furthermore be noticed that this study is based on survey data collected among staff in the Danish eldercare service and the results are therefore not necessarily representative of the working population as a whole. Also, non-respondents had higher risk ratios for LTSA than did respondents. As the non-respondents differ from the respondents in this respect it is therefore not possible to ascertain whether the associations observed in this study are valid for the entire group of employees in the Danish eldercare sector. However, we have no reason to expect that the observed associations should be fundamentally different in the group of non-respondents, and in the view of the authors the differences between respondents and non-respondents should not invalidate the general tendencies indicated by our findings. Indeed, the findings of this study should be regarded with a high degree of credibility. Firstly, the study is based on a dataset with a large number of observations. Secondly, we have adjusted our analyses for a series of background factors, psychosocial work characteristics, work-time arrangements, and physical workload that could have proven potential confounders. By taking these factors into account we

57

were able to demonstrate the robustness of the association between positive work-related states and LTSA. Also, our explanatory variables and our outcome variables stem from different sources of data, which serves to underline the credibility of the results as this precludes the risk of observing spurious associations that can be ascribed to common methods biases [31]. Finally, although this study is based on employees in the elderly care sector, the division of our sample into high, medium and low levels of positive work-related states was based on an external criteria, as we deployed upper and lower quartile values from the representative COPSOQ-survey from 2004–5 as cut-points for the three groups of respondents. This implies that the levels of positive work-related states that we use in our analyses reflect general levels in the Danish working population, which seems preferable to a division of the respondents in e.g. upper and lower quartiles of the sample under investigation.

Conclusion This study shows that commitment to the work-place and experiences of meaning of work are significantly associated with risk of long-term sickness absence. We expected that employees with high levels of commitment to the work-place and meaning at work would have lower risk of long-term sickness absence than employees with medium or low levels of commitment to the work-place and meaning at work. The results, however, show that low levels of positive work-related states are associated with risk of longterm sickness absence, whereas high levels of positive work-related states do not seem to have a protective effect, when compared with medium levels.

Acknowledgements This research is part of a Ph.D. project funded by the National Research Centre for the Working Environment (Denmark), Department of Psychology, University of Copenhagen and the Danish Research School of Psychology.

References [1] Linley PA, Joseph S, Harrington S, Wood AM. Positive psychology: past, present, and (possible) future. J Positive Psychol 2006;1(1):3–16. [2] Bakker AB, Schaufeli WB, Leiter MP, Taris TW. Work engagement: an emerging concept in occupational health psychology. Work & Stress 2008;22(3):187–200.

58

T. Clausen et al.

[3] Meyer JP, Stanley DJ, Herscovitch L, Topolnytsky L. Affective, continuance, and normative commitment to the organization: a meta-analysis of antecedents, correlates, and consequences. J Voc Behav 2002;61(1):20–52. [4] Burton JP, Lee TW, Holtom BC. The influence of motivation to attend, ability to attend, and organizational commitment on different types of absence behaviors. Journal of Managerial Issues 2002;14(2):181–97. [5] Mathieu JE, Zajac DM. A review and meta-analysis of the antecedents, correlates, and consequnces of organizational commitment. Psycholog Bull 1990;108(2):171–94. [6] Mathieu JE, Kohler SS. A test of the interactive effects of organizational commitment and job involvement on various types of absence. J Voc Behav 1990;36(1):33–44. [7] Nielsen ML, Rugulies R, Christensen KB, Smith-Hansen L, Bjorner JB, Kristensen TS. Impact of the psychosocial work environment on registered absence from work: a two-year longitudinal study using the IPAW cohort. Work & Stress 2004;18(4):323–35. [8] Brown SP. A meta-analysis and review of organizational research on job involvement. Psycholog Bull 1996; 120(2):235–55. [9] Riketta M. Organizational identification: a meta-analysis. J Voc Behav 2005;66(2):358–84. [10] Steers RM, Rhodes SR. Major influences on employee attendance: a process model. J Appl Psychol 1978; 63(4):391–407. [11] Keyes CLM. Mental illness and/or mental health? Investigating axioms of the complete state model of health. J Consulting Clin Psychol 2005;73(3):539–48. [12] Taylor SE, Kemeny ME, Reed GM, Bower JE, Gruenewald TL. Positive resources, positive illusions, and health. Am Psychologist 2000;55(1):99–109. [13] Meyer JP, Becker TE, Van Dick R. Social identities and commitments at work: toward an integrative model. J Organ Behav 2006;27(5):665–83. [14] Meyer JP, Allen NJ. Commitment in the work-place. Theory, research, and application. Thousand Oaks: SAGE Publications; 1997. [15] Deci EL, Ryan RM. Intrinsic motivation and self-determination in human behavior. New York: Plenum; 1985. [16] Kristensen TS, Hannerz H, Hogh A, Borg V. The Copenhagen Psychosocial Questionnaire – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work, Environ Health 2005;31(6):438–49. [17] Pejtersen JH, Kristensen TS, Borg V, Bjørner JB. The second version of Copenhagen Psychosocial Questionnaire (COPSOQ II). Scand J Public Health 2010;38(Suppl 3):8–24. [18] Wrzeniewski A. Finding positive meaning in work. In: Cameron KS, Dutton JE, Quinn RE, editors. Positive organizational scholarship. Foundations of a new discipline. San Francisco: Berett-Kohler Publishers, Inc.; 2003. pp. 296–308.

[19] Chalofsky N. An emerging construct of meaningful work. Hum Res Develop Int 2003;6(1):69–83. [20] Sagiv L, Roccas S, Hazan O. Value pathways to well-Bbing: healthy values, valued goal attainment, and environmental congruence. In: Linley PA, Joseph S, editors. Positive psychology in practice. Hoboken: John Wiley & Sons Inc; 2004. pp. 68–85. [21] Rugulies R, Christensen KB, Borritz M, Villadsen E, Bu¨ltmann U, Kristensen TS. The contribution of the psychosocial work environment to sickness absence in human service workers: results of a 3-year follow-up study. Work & Stress 2007;21(4):293–311. [22] Nielsen ML, Rugulies R, Christensen KB, Smith-Hansen L, Kristensen TS. Psychosocial work environment predictors of short and long spells of registered sickness absence during a 2-year follow up. J Occup Environ Med 2006; 48(6):591–8. [23] Lund T, Labriola M, Christensen KB, Bu¨ltmann U, Villadsen E, Burr H. Psychosocial work environment exposures as risk factors for long-term sickness absence among Danish employees: results from DWECS/DREAM. J Occup Environ Med 2005;47(11):1141–7. [24] Christensen KB, Nielsen ML, Rugulies R, Smith-Hansen L, Kristensen TS. Work-place levels of psychosocial factors as prospective predictors of registered sickness absence. J Occup Environ Med 2005;47(9):933–40. [25] Schaufeli WB, Bakker AB. Job demands, job resources, and their relationship with burnout and engagement: a multisample study. J Organ Behav 2004;25(3):293–315. [26] Arnold KA, Turner N, Barling J, Kelloway EK, McKee MC. Transformational leadership and psychological well-being: the mediating role of meaningful work. J Occ Health Psychol 2007;12(3):193–203. [27] Tu¨chsen F, Christensen KB, Nabe-Nielsen K, Lund T. Does evening work predict sickness absence among female carers of the elderly? Scand J Work, Environ Health 2008; 34(6):483–6. [28] Lund T, Labriola M, Christensen KB, Bu¨ltmann U, Villadsen E. Physical work environment risk factors for long term sickness absence: prospective findings among a cohort of 5357 employees in Denmark. BMJ 2006; 332(7539):449–52. [29] Leiter MP, Harvie P, Frizzel C. The correspondence of patient satisfaction and nurse burnout. Soc Sci Med 1998;47(10):1611–17. [30] Meyer JP, Allen NJ, Smith CA. Commitment to organizations and occupations: extension and test of a three-component conceptualization. J Appl Psychol 1993; 78(4):538–51. [31] Podsakoff PM, MacKenzie SB, Lee JY, Podsakoff NP. Common method biases in behavioral research: a critical review of the literature and recommended remedies. J Appl Psychol 2003;88(5):879–903.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 59–68

ORIGINAL ARTICLE

Do dimensions from the Copenhagen Psychosocial Questionnaire predict vitality and mental health over and above the job strain and effort–reward imbalance models?

HERMANN BURR, KAREN ALBERTSEN, REINER RUGULIES & HARALD HANNERZ National Research Centre for the Working Environment, Copenhagen, Denmark

Abstract Aims: The Copenhagen Psychosocial Questionnaire (COPSOQ) comprises dimensions (emotional demands, demands of hiding emotions, meaning of work, quality of leadership, and predictability) that are not in the job strain or the effort–reward imbalance (ERI) models. The study aim was to investigate whether these dimensions explain changes in vitality and mental health over and above the job strain and ERI models. Methods: A cohort of 3552 employees in 2000 were followed up in 2005 (cohort participation of 51%). Regression analyses were carried out with mental health and vitality as dependent variables. A significance level of 0.01 was applied when comparing regression models. Results: Regarding mental health, both the full COPSOQ–ERI model (p ¼ 0.005) and the full job strain–COPSOQ model (p ¼ 0.01) were significantly better than the ERI and the job strain models. Regarding vitality, none of the full COPSOQ models (i.e. with new COPSOQ dimensions together with job strain or ERI respectively) was significantly better than the ERI (p ¼ 0.03) or the job strain (p ¼ 0.04) models. Emotional demands and meaning of work predicted poor mental health and low vitality. Conclusions: In relation to mental health, new psychosocial risk factors have the potential to add to the predictive power of the job strain and ERI models. The current practice of including only items from the ERI and job strain models in public health studies should be reconsidered. Theories regarding the status of, for example, emotional demands and meaning of work should be developed and tested.

Key Words: COPSOQ, demands for hiding emotions, emotional demands, ERI, fatigue, JCQ, meaning of work, predictability, quality of leadership, quality of management

Background A number of general psychosocial questionnaires, such as the Copenhagen Psychosocial Questionnaire (COPSOQ) [1], differ from the Job Content Questionnaire (JCQ) [2] and the effort–reward imbalance (ERI) questionnaire [3], as they are not based on one specific theoretical model regarding the relationship between psychosocial work environment and health [4]. Consequently, these questionnaires contain dimensions that are not covered by the job strain model and the ERI model. In the present study, we determined to what extent such dimensions predict changes in vitality and mental health over and above the job strain and the ERI models.

In the 2000 round of the Danish Work Environment Cohort Study (DWECS) [5,6], some dimensions from COPSOQ, version 1 [1] were included. Among these, emotional demands, demands of hiding emotions, meaning of work, predictability and quality of leadership did not form part of the job strain and the ERI models. Only a few prospective studies have looked at the health effects of these factors [7–11]. Emotional demands [7], meaning of work [11] and leadership fairness [8] have been found to be predictors of fatigue. Emotional demands [10] and quality of leadership [9] have been found to be risk factors for mental health. Our research question was whether emotional demands, demands for hiding emotions, meaning of

Correspondence: Hermann Burr, National Research Centre for the Working Environment, Lersø Parkalle´ 105, 2100 Copenhagen Ø, Denmark. Tel: þ45 3916 5364. Fax: þ45 3916 5201. E-mail: [email protected] (Accepted 7 October 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809353436

60

H. Burr et al.

work, quality of leadership and predictability explain changes in vitality and mental health over and above the job strain and ERI models. On the basis of the few findings mentioned above, we expected these psychosocial working conditions to do so regarding both outcomes.

Material and methods Population Among 11,437 randomly selected residents in Denmark from the Central Population Register, 8583 participated (75%) in the 2000 round of the DWESC [5,6]. Of these, 256 died or emigrated up to 2005. Of the remaining population, 5625 participated (68%, n ¼ 8327); the cohort participation rate was 51%, and of these, 3856 were employees. Among the employees, there was no association between attrition at follow-up and baseline vitality or mental health. In the cohort, 304 people did not have sufficient information on the relevant variables, so the analyses are based on 3552 people. Variables Sex and age were register data as per date of data extraction (1 October 2000). All other variables were, in 2000, based on telephone interviews on the respondents’ home phone – or, if this was not possible, on face-to-face interviews in the respondents’ homes. In 2005, most respondents responded to postal questionnaires; the rest were interviewed by telephone. In the following description, all scales except ERI had values from 0 to 100. Vitality and mental health. Vitality was measured by means of the following SF-36 items [12]: ‘‘How much of the time during the past 4 weeks ‘‘did you feel full of energy?’’ (item 1), ‘‘did you have a lot of energy?’’ (item 2), ‘‘did you feel worn out?’’ (item 3), and ‘‘did you feel tired?’’ (item 4), with the response options ‘‘All of the time’’, ‘‘Most of the time’’, ‘‘A good bit of the time’’, ‘‘Some of the time’’, ‘‘A little of the time’’, and ‘‘None of the time’’. The items were combined into a scale where the values of items 3 and 4 were reversed. Cronbach’s alpha of scales was 0.82; inter-item correlations were 0.42–0.70. Mental health was measured by means of the following SF-36 items [12]: ‘‘How much of the time during the past 4 weeks ‘‘have you been a very nervous person?’’ (item 1), ‘‘have you felt so down in the dumps that nothing could cheer you up?’’ (item 2), ‘‘have you felt calm and peaceful?’’ (item 3), ‘‘have you felt downhearted and blue?’’ (item 4), and ‘‘have you been a happy person?’’ (item 5), with the

same response options as for vitality. The items were combined into a scale where items 1, 2 and 4 were reversed. Cronbach’s alpha was 0.81; inter-item correlations were 0.37–0.57. ERI. ERI was based on measurements with proxy measures [13]. The DWECS did not contain items regarding work-related overcommitment. Effort was measured by means of four items from COPSOQ I [1]. Three items were the same as items WP1, QD1 and QD2 from COPSOQ II [14] – for an explanation of COPSOQ II items, see also Table I. The fourth item was ‘‘Do you have to do overtime?’’. All of the items had the following response options (and values for the scale): ‘‘Always’’ (5), ‘‘Often’’ (4), ‘‘Sometimes’’ (3), ‘‘Seldom’’ (2), and ‘‘Never/hardly ever’’ (1). Cronbach’s alpha for the scale was 0.64; inter-item correlations were 0.22–0.40. Reward was measured by means of three non-COPSOQ items (‘‘Have you good prospects for the future in your job?’’, ‘‘Is your work recognized and appreciated by management?’’, and ‘‘How would you assess your salary with regard to your effort and your qualification?’’, with response options (and values for the scale) for the two first items being ‘‘To a very large extent’’ (5), ‘‘To a large extent’’ (4), ‘‘Somewhat’’ (3), ‘‘To a small extent’’ (2), and ‘‘To a very small extent’’ (1), and response options for the third item being ‘‘Too high’’ (3), ‘‘Appropriate’’ (3), ‘‘A little bit too low’’ (2), and ‘‘Much too low’’ (1)) and four COPSOQ I items [1], being the same as SC1, SS1, JI1 and JI4 from COPSOQ II [14] (the response options for the two first items (and values for the sum scale) were ‘‘Always’’ (5), ‘‘Often’’ (4), ‘‘Sometimes’’ (3), ‘‘Seldom’’ (2), and ‘‘Never/hardly ever’’ (1), and the response options for the two latter items were ‘‘Yes’’ (1) and ‘‘No’’ (2)). Cronbach’s alpha for the scale was 0.53; inter-item correlations were 0.00–0.42. An ERI ratio was constructed [13] with the effort score in the nominator and the reward score in the denominator, with a correction factor in the denominator adjusting for the higher number of reward items. Hence, higher values of the ratio expressed a higher level of imbalance between high effort and low reward. Demand–control. Job strain was based on two dichotomized scales, psychological demands and control, resulting in four categories: ‘‘No strain’’ – low demands and high control; ‘‘Active’’ – high demands and low control; ‘‘Passive’’ – low demands and low control; and ‘‘Strain’’ – high demands and low control. The scales were constructed accurately by means of items from COPSOQ [1] and the DWECS corresponding to JCQ items [2].

Do dimensions from the Copenhagen Psychosocial Questionnaire predict vitality and mental health

61

Table I. Copenhagen Psychosocial Questionnaire (COPSOQ) I items in the Danish Work Environment Cohort Study (DWECS) used in scales or variables of the present article.

COPSOQ II dimension

Used in scale or variable of the present article

COPSOQ II item no. [14]

Work pace Quantitative demands

ERI, job strain ERI, job strain

WP1 QD1

ERI, job strain

QD2

ERI, social support

SC1

Social support

SC2

ERI, social support

SS1

Social support

SS2

Job insecurity

ERI ERI

JI1 JI4

Influence

Job strain

IN1

Job strain

IN2

Job strain

IN3

Job strain

IN4

Job strain Job strain

VA1 PD1

Job strain

PD2

Emotional demands

ED1

Social support from colleagues

Social support from supervisors

Variation Possibilities for development

Emotional demands

Demands for hiding emotions Meaning of work

Demands for hiding emotions Meaning of work

Quality of leadership

Quality of leadership

ED3 ED4 HE2 MW1 MW2 MW3

QL1

Predictability

Predictability

QL2 QL3 QL4 PR1

PR2

Question

Response optionsa

Do you have to work very fast? Is your workload unevenly distributed, so that it piles up? How often do you not have time to complete all your work tasks? How often do you get help and support from your colleagues? How often are your colleagues willing to listen to your problems at work? How often is your nearest superior willing to listen to your problems at work? How often do you get help and support from your nearest superior? Are you worried about becoming unemployed? Are you worried about being transferred to another job against your will? Do you have a large degree of influence concerning your work? Do you have a say in choosing who you work with? Can you influence the amount of work assigned to you? Do you have any influence on what you do at work? Is your work varied? Does your work require you to take the initiative?

a a

Do you have the possibility of learning new things through your work? Does your work put you in emotionally disturbing situations? Is your work emotionally demanding? Do you get emotionally involved in your work? Does your work require that you hide your feelings? Is your work meaningful? Do you feel that the work you do is important? Do you feel motivated and involved in your work? To what extent would you say that your immediate superior: Makes sure that the individual member of staff has good development opportunities? Gives high priority to job satisfaction? Is good at work-planning? Is good at solving conflicts? At your place of work, are you informed well in advance concerning, for example, important decisions, changes, or plans for the future? Do you receive all the information that you need in order to do your work well?

d

a b b a a c c a a a a e d

a d d d d d d

d d d d d

d

ERI: effort–reward imbalance. All items listed were taken from COPSOQ I and are still used in COPSOQ II. Item numbers refer to the appendix of Pejtersen et al. in the present issue [14]. aResponse options: a – Always; Often; Sometimes; Seldom; Never/hardly ever; b – Always; Often; Sometimes; Seldom; Never/hardly ever; Not relevant; c (COPSOQ I response options) – Yes; No; d – To a very large extent; To a large extent; Somewhat; To a small extent; To a very small extent; e (non-COPSOQ response options [28]) – ‘‘To a high degree’’; ‘‘To some degree’’; ‘‘Only to a lesser degree’’; ‘‘No, or only to a slight degree’’.

62

H. Burr et al.

The psychological demands scale was based on a mean of three items from the medium-length COPSOQ I quantitative demand scale (the same as items WP1, QD1 and QD2 in COPSOQ II [14]; Table I) and one DWECS item, ‘‘How correct or incorrect are the following statements about your role in your work? – Conflicting demands are placed on me in my work’’, with the response options ‘‘Completely correct’’, ‘‘Sometimes correct’’, and ‘‘Completely incorrect’’. Cronbach’s alpha was 0.55; inter-item correlations were 0.11–0.40. In the calculation of the mean, the first item counted as six individual items, so that this question regarding work pace had the same weight in the scale as in the JCQ. Control was – as in the JCQ – calculated as a mean of the items of the medium-length scales influence at work (i.e. decision authority) and possibilities for development (i.e. skill discretion) from COPSOQ 1. The items [1] were the same as IN1-IN4, VA1, PD1 and PD2 in COPSOQ II [14]. Cronbach’s alpha of the total control scale was 0.80; inter-item correlations were 0.32–0.50. Social support was measured by means of four COPSOQ I items [1], which were the same as SC1, SC2, SS1 and SS2 in COPSOQ II [14]. Cronbach’s alpha was 0.74; inter-item correlations were 0.32– 0.60. New COPSOQ dimensions. All new COPSOQ scales were the medium-length scales from COPSOQ I [1], most items of which are still used in COPSOQ II [14] (Table I). The three emotional demands items were the same as items ED1–ED3 in COPSOQ II [14]. Cronbach’s alpha of the scale was 0.88; inter-item correlations were 0.68–0.73. The two demands for hiding emotions items were the same as item HE2 in COPSOQ II [14] and the item ‘‘Does your work require that you do not state your opinion?’’. Cronbach’s alpha was 0.53; the inter-item correlation was 0.36. The three meaning of work items were the same as items MW1–MW3 in COPSOQ II [14]. Cronbach’s alpha of the scale was 0.78; inter-item correlations were 0.52–0.56. The four quality of leadership items were the same as items QL1–QL4 in COPSOQ II [14]. Cronbach’s alpha of the scale was 0.84; inter-item correlations were 0.50–0.63. The two predictability items were the same as items PR1–PR2 in COPSOQ II [14]. Cronbach’s alpha of the scale was 0.70; the inter-item correlation was 0.53. Other variables. Occupational physical activity was a dichotomization of a scale with the following categories: ‘‘Low’’ (0–35) and ‘‘High’’ (435–100).

The scale was a mean of three questions and one subscale. The questions were ‘‘Does your job require that you sit down?’’ (reversed), ‘‘Does your job require that you kneel or squat?’’, and ‘‘How much of your time at work do you push or pull something?’’, with the responses (given the following values) ‘‘Almost all the time’’ (100), ‘‘Approximately 3/4 of the time’’ (75), ‘‘Approximately 1/2 of the time’’ (50), ‘‘Approximately 1/4 of the time’’ (25), ‘‘Rarely/very little’’ (6), and ‘‘Never’’ (0). The subscale was based on a mean of two questions (given the following values): ‘‘For how many of your working hours do you carry or lift things/people?’’, with the response options as above; and the question ‘‘What does the load you carry normally weigh?’’ with the response options ‘‘Less than 3 kg’’ (2.5), ‘‘3–10 kg’’ (10.8), ‘‘11–29 kg’’ (33.3), ‘‘30–49 kg’’ (65.8), and ‘‘50 kg or more’’ (100). If respondents answered ‘‘Never’’ to the former question, we coded the latter question as 0. Smoking status was divided into two categories: ‘‘Never or former smoker’’ and ‘‘Currently smoking’’. Analysis Regression analyses were carried out in which ERI, job strain, emotional demands, demands for hiding emotions, influence at work, possibilities for development, meaning of work, quality of leadership and predictability were independent variables. The distribution of the COPSOQ scales and the ERI scales can be seen in Table II. In the analysis, all scales were transformed so that their standard deviation was equal to 1. A correlation analysis (Table III) was carried out. A number of bivariate correlations were 0.35 or above – (a) emotional demands were correlated with demands for hiding emotions; (b) meaning of work, quality of leadership and predictability of work were all correlated with each

Table II. Distribution of psychosocial variables among employees in the baseline of the 2000–2005 cohort. Standard Minimum Maximum Mean deviation ERI Social support Emotional demands Demands for hiding emotions Meaning of work Quality of leadership Predictability of work

0.15 0 0 0

1.60 100 100 100

0.53 76 29 22

0.19 19 25 22

0 0 0

100 100 100

82 60 66

15 21 22

ERI: effort–reward imbalance.

Do dimensions from the Copenhagen Psychosocial Questionnaire predict vitality and mental health

63

Table III. Correlations between psychosocial variables among employees in the baseline of the 2000–2005 cohort; Pearson correlations.

Social support Emotional demands Demands for hiding emotions Meaning of work Quality of leadership Predictability of work

ERI

Social support

Emotional demands

Demands for hiding emotions

Meaning of work

Quality of leadership

0.48 0.26 0.28 0.14 0.42 0.35

0.08 0.19 0.27 0.51 0.40

0.48 0.11 0.08 0.10

0.06 0.21 0.19

0.35 0.36

0.59

ERI: effort–reward imbalance. All correlations: p ¼ 0.000.

other – the highest correlation was between quality of leadership and predictability (0.59); and (c) both social support and ERI were correlated with quality of leadership and predictability, and to a lesser extent with meaning of work – and with each other. A total of 26% were categorized as having ‘‘job strain’’, 23% as having ‘‘no strain’’, 26% as being ‘‘active’’, and 25% as being ‘‘passive’’. People with job strain had – as compared with other people – higher ERI, more emotional demands, more demands for hiding emotions and lower social support, less meaning of work, lower management quality, and lower predictability of work (p ¼ 0.000–0.003). For each outcome, two regression analyses were carried out, one with ERI as independent variable, and another with job strain as independent variable in the first step (model 1). These two dimensions could not be entered in the same analysis, as they share quantitative demand items and some social support items. In the second step (model 2), the COPSOQ scales emotional demands, demands for hiding emotions, meaning of work, quality of leadership and predictability were included. In the second step, ERI and job strain, respectively, were kept in the model, without taking into account whether these variables predicted the outcome. All analyses were controlled for gender, age, occupational physical activity, smoking, and mode of interview [15]. Vitality and mental health at follow-up were estimated in multiple linear regression analyses controlled for baseline mental health and vitality, respectively. F-tests were used to determine whether model 2 explained the outcome over and above model 1. When the models were compared, a significance level of 0.01 was applied, as four comparisons were carried out.

Results Vitality When a significance level of 0.01 was applied in comparison of regression models, none of the full

COPSOQ models (i.e. where the new COPSOQ dimensions were in the model together with job strain or ERI, respectively) was significantly better at predicting vitality at follow-up – controlled for baseline vitality – than the ERI (p ¼ 0.03) or the job strain (p ¼ 0.04) models (Table IV). In the full COPSOQ models, we could observe that vitality decreased with more emotional demands and increased with more meaning of work. For example, an employee with an increase of one standard deviation of emotional demands would have a decrease of 0.71 (1.37; 0.04) points in the vitality score. Mental health Both the full COPSOQ–ERI model (p ¼ 0.005) and the full job strain–COPSOQ model (p ¼ 0.01) were significantly better at predicting mental health at follow-up – controlled for baseline mental health – than the ERI or the job strain model alone (Table V). In both full COPSOQ models, mental health deteriorated with more emotional demands and improved with more meaning of work. Neither demands for hiding emotions, quality of leadership nor predictability of work was associated with any of the outcomes.

Discussion Regarding mental health, the full COPSOQ models (i.e. where the new COPSOQ dimensions were in the model together with job strain or ERI, respectively) were better than both the ERI model and the job strain model. Regarding vitality, the full COPSOQ models were not significantly better than the ERI or job strain model (p < 0.01). Therefore, our hypothesis was partly confirmed. Emotional demands predicted deterioration in mental health and vitality. Also, meaning of work predicted both outcomes, but was only marginally significant when predicting mental health together with job strain. Demands for hiding emotions,

64

H. Burr et al.

Table IVa. Vitality at 5-year follow-up among employees aged 18–69 years at baseline; N ¼ 3552. Model 1a

ERI Emotional demands Demands for hiding emotions Meaning of work Quality of leadership Predictability of work

Model 2b

p

Beta

95% CI

p

Beta

95% CI

0.20

0.37

0.94 to 0.19

0.83 0.04 0.72 0.02 0.62 0.30

0.07 0.71 0.11 0.70 0.18 0.37

0.70 to 0.67 1.37 to 0.04 0.75 to 0.52 0.09–1.32 0.90 to 0.54 0.33 to 1.07

CI, confidence interval. Multiple linear regression. Beta, unstandardized. Effort–reward imbalance (ERI) model compared with an enhanced COPSOQ item model. p ¼ 0.027 for comparison of model 2 with model 1 (F-test). a Sum of squares ¼ 237 857; degrees of freedom ¼ 7. F ¼ 125. R2 ¼ 0.20. b Sum of squares ¼ 241 287; degrees of freedom ¼ 12. F ¼ 74. R2 ¼ 0.20. The psychosocial risk factor scales were transformed so that their standard deviation was equal to 1. Bold values indicate associations where p < 0.05. The vitality scale ranged from 100 (high) to 0 (low). Controlled for gender, age, mode of interview, baseline occupational physical activity, baseline smoking, and baseline vitality.

Table IVb. Vitality at 5-year follow-up among employees aged 18–69 years at baseline; N ¼ 3552. Model 1a p Job strain Active vs. no strain Passive vs. no strain Strain vs. no strain Social support Emotional demands Demands for hiding emotions Meaning of work Quality of leadership Predictability of work

Model 2b

Beta

95% CI

0.43 0.15 2.19 0.09

2.00 to 1.14 1.45 to 1.75 3.82 to 0.56 0.65 to 0.48

Beta

95% CI

0.20 0.38 1.67 0.34 0.70 0.08 0.70 0.13 0.34

1.79–1.38 1.28 to 2.05 3.39 to 0.05 0.99 to 0.30 1.37 to 0.03 0.72 to 0.56 0.07–1.34 0.87 to 0.61 0.37 to 1.04

0.10

0.02

0.76

p

0.30 0.04 0.81 0.03 0.74 0.35

CI, confidence interval; ERI, effort–reward imbalance. Multiple linear regression. Beta, unstandardized. Job strain model compared with an enhanced COPSOQ item model. p ¼ 0.038 for comparison of model 2 with model 1 (F-test). a Sum of squares ¼ 240 372; degrees of freedom ¼ 10. F ¼ 89. R2 ¼ 0.20. b Sum of squares ¼ 243 570; degrees of freedom ¼ 15. F ¼ 60. R2 ¼ 0.20. The psychosocial risk factor scales were transformed so that their standard deviation was equal to 1. Bold values indicate associations where p < 0.05. The vitality scale ranged from 100 (high) to 0 (low). Controlled for gender, age, mode of interview, baseline occupational physical activity, baseline smoking, and baseline vitality.

Table Va. Mental health at 5-year follow-up among employees aged 18–69 years at baseline; N ¼ 3552. Model 1a

ERI Emotional demands Demands for hiding emotions Meaning of work Quality of leadership Predictability of work

Model 2b

p

Beta

95% CI

p

Beta

95% CI

0.07

0.40

0.83 to 0.03

0.59 0.04 0.10 0.04 0.57 0.42

0.13 0.53 0.41 0.48 0.16 0.22

0.62 to 0.35 1.03 to 0.03 0.89 to 0.07 0.02–0.94 0.70 to 0.39 0.31 to 0.75

CI, confidence interval. Multiple linear regression. Beta, unstandardized. Effort–reward imbalance (ERI) model compared with an enhanced COPSOQ item model. p ¼ 0.0050 for comparison of model 2 with model 1 (F-test). a Sum of squares ¼ 130 694; degrees of freedom ¼ 7. F ¼ 120. R2 ¼ 0.19. b Sum of squares ¼ 133 296; degrees of freedom ¼ 12. F ¼ 72. R2 ¼ 0.20. The psychosocial risk factor scales were transformed so that their standard deviation was equal to 1. Bold values indicate associations where p < 0.05. The mental health scale was from 100 (best) to 0 (worst). Controlled for gender, age, mode of interview, baseline occupational physical activity, baseline smoking, and baseline mental health.

Do dimensions from the Copenhagen Psychosocial Questionnaire predict vitality and mental health

65

Table Vb. Mental health at 5-year follow-up among employees aged 18–69 years at baseline; N ¼ 3552. Model 1a p Job strain Active vs. no strain Passive vs. no strain Strain vs. no strain Social support Emotional demands Demands for hiding emotions Meaning of work Quality of leadership Predictability of work

Beta

Model 2b 95% CI

0.10

0.14

p

Beta

95% CI

0.10 0.14 0.89 0.21 0.57 0.37 0.42 0.25 0.16

1.10 to 1.30 1.11 to 1.40 2.19 to 0.41 0.28 to 0.71 1.08 to 0.07 0.85 to 0.11 0.06 to 0.90 0.81 to 0.31 0.37 to 0.70

0.20 0.13 0.13 1.16 0.32

1.32 1.08 2.39 0.01

to 1.06 to 1.33 to 0.07 to 0.04

0.39 0.03 0.13 0.08 0.38 0.55

CI, confidence interval; ERI, effort–reward imbalance. Multiple linear regression. Beta, unstandardized. Job strain model compared with an enhanced COPSOQ item model. p ¼ 0.0093 for comparison of model 2 with model 1 (F-test). a Sum of squares ¼ 131 560; degrees of freedom ¼ 10. F ¼ 85. R2 ¼ 0.19. b Sum of squares ¼ 133 927; degrees of freedom ¼ 15. F ¼ 58. R2 ¼ 0.20. The psychosocial risk factor scales were transformed so that their standard deviation was equal to 1. Bold values indicate associations where p<0.05. The mental health scale was from 100 (best) to 0 (worst). Controlled for gender, age, mode of interview, baseline occupational physical activity, baseline smoking, and baseline mental health.

quality of leadership and predictability did not predict any of the outcomes.

Strengths and weaknesses The strengths of the present study are that it is: (a) prospective; (b) relies on a representative sample of employees in a country; (c) uses validated scales of psychosocial factors; and (d) controls for relevant potential confounders such as smoking and occupational physical activity. Regarding point (a), the number of prospective studies on associations between psychosocial factors other than job strain and ERI with vitality and mental health are scarce [7–11], and only two of them have been based on general populations [7,10]. Regarding point (d), job strain is correlated positively with smoking, but also occupational physical activity, which has been shown to have adverse health effects [16]. The potential weaknesses of the study are that: (a) the follow-up time is relatively long (5 years); (b) the proportion in the cohort is low (51%); and (c) the measurements of job strain and ERI are not based on questions from the original JCQ (job strain) and ERI questionnaires. Regarding point (a), a recent study [17] of psychosocial risk factors for self-rated health (general health, distress, and depression) has shown somewhat stronger associations with distress and depression in a 2-year follow-up as compared with a 6-year follow-up. We might therefore suspect that the 5-year follow-up period is too long to detect all relevant associations. Regarding point (b), we do not know to what extent attrition at baseline was related to vitality and mental health in the study population. Surprisingly, attrition from baseline to follow-up was

not related to vitality and mental health among baseline participants. Regarding point (c), we have been able to measure job strain with items corresponding to the items in the JCQ [2]. However, we have removed occupational physical activity items from the demand scale, and instead used these items as a general control variable. Regarding ERI, the wording of the original questions in the ERI questionnaire have been criticized for mixing the measurement of exposures (stressors) and outcomes (stress reactions) – a criticism that does not apply to the items used in this article [13]. On the other hand, the DWECS did not measure overcommitment, so the measurement of ERI would potentially have predicted more of the outcome, had this measurement been part of the ERI measure. Regarding the job strain and ERI measures, the order of the proxy questions in the present article differs from the order of the original questions in the JCQ (job strain) and ERI questionnaires. We do not think that question order does affect the validity of our proxy measures considerably. The Cronbach alpha for demands in the proxy job strain measure was relatively low – this has also been found in the original measures of demands [18]. The Cronbach alpha for rewards was lower than in the original ERI measure [3]. One reason for this might be that the wording of proxy items – as mentioned above – was more neutral than in the original ERI measure, leading to a lower correlation between the proxy items. It might also be that Cronbach’s alpha is not the proper measure for validity of these scales if one assumes that the items in the scales have common effects.

66

H. Burr et al.

Comparison with other studies The results indicate that the total health effects of the psychosocial work environment will be underestimated in some cases if one only includes the ERI and job strain models. One should, however, note, as mentioned above, that the risks associated with ERI might have been higher had we been able to include overcommitment in the analysis. It seems that the new COPSOQ dimensions decrease the effects of ERI and job strain (Tables IV and V). In the vitality analyses, perhaps the effects of job strain decreased (model 2, Table IV) because emotional demands were higher among people with job strain (p ¼ 0.003). Perhaps the effect of job strain was lowered because meaning of work might mediate the association between job strain and vitality and mental health. In the following, we will discuss the results in detail. Emotional demands have previously been found to predict low vitality (i.e. fatigue) [7–10] and mental health [10]. A 1-year follow-up study of risk factors of fatigue in a heterogeneous Dutch population found that emotional demands increased fatigue in males. A case-control study in the Danish population using occupational information 1 year before treatment as a proxy for psychosocial working conditions found – among women – that high emotional demands on an occupational level were associated with a higher risk for hospitalizations for depression [10]. Emotional demands might cause poor mental health because emotional work potentially challenges the psychological vulnerability of the employee. Another explanation for a part of this finding might be that occupations with high emotional demands also have a higher incidence of violence [19]. One prospective study has found that violence at work predicts fatigue [20]. Another study, based on the same data as in the case-control study mentioned above [10], found that people working in occupations with a high risk of violence had a higher risk for hospitalizations for both depressive and anxiety disorders [21]. However, we still need more studies to know whether emotional demands actually cause deterioration in mental health. To our knowledge, only one published prospective study has looked at possible associations between meaning of work and vitality and mental health [11]. This study, on a population of human service workers, found that meaning of work was associated with increased vitality cross-sectionally, but – unexpectedly – decreased vitality prospectively. The vitality measure in the study (labelled burnout) was based on four vitality/fatigue items, one general health item, and one mental health item. The question remains of

whether meaning of work also predicts vitality and mental health among employees not working with costumers or clients. The COPSOQ research team categorizes meaning of work as a measure of ‘‘work organization and job content’’ [1,14]. Meaning of work correlated with other ‘‘work organization and job content’’ variables (possibilities for development (Spearman correlation ¼ 0.54) and decision authority (0.43)), but also with some ‘‘Interpersonal relations and leadership’’ variables (role clarity (0.46) and quality of leadership (0.37)) [1]. It has been suggested that there is a close link between meaning of work, trust, and leadership [22]. Meaning might be considered to be a mediator between psychosocial work environment and health [23], rather than an independent predictor of health [24]. In none of the models (Tables IV and V) did quality of leadership predict vitality and mental health. This was also so when meaning of work was removed from the model (data not shown) or when quality of leadership was analysed alone, i.e. without any adjustment for ERI, job strain, or any other psychosocial dimension. The COPSOQ quality of leadership scale included items dealing with the degree to which the immediate superior ensured good development opportunities, job satisfaction, work planning, and conflict-solving [14]. In a recent study of nurses [8], fairness of immediate superior’s leadership predicted a decreased risk for persistent fatigue. Here, leadership fairness was measured by means of items regarding justice (whether the immediate supervisor distributes the work fairly and impartially, and treats the workers fairly and equally) and one item regarding the possible stressful relationship with the immediate supervisor [25]. A study of British civil servants [26] found that relational justice (two items on predictability, one item on social support, and two items on fairness of feedback) predicted cases of psychiatric morbidity. So, it might be that quality of leadership predicts fatigue and mental health when one addresses the experienced degree of fairness of leadership.

Conclusion The analyses in the present article indicate that there are a number of relevant psychosocial risk factors for mental health that are not included in the existing job strain and ERI models. Thus, the tradition of including only items from the ERI and job strain models in studies of possible health effects [27] should be reconsidered. Today, we know very little of the public health consequences of these non-model dimensions. Such dimensions should be

Do dimensions from the Copenhagen Psychosocial Questionnaire predict vitality and mental health investigated with regard to other outcomes and in more populations. Other outcomes could be ‘‘hard’’ outcomes, such as clinical depression, use of prescribed medicine (e.g. antidepressants), sickness absence, cardiovascular disease, or mortality. Furthermore, dimensions of the psychosocial work environment, such as justice and fairness [8], should be investigated. Theories regarding the status of dimensions such as emotional demands and meaning of work, together with the dimensions from the ERI and job strain models, should be developed and tested empirically. For example, it should be explored whether it is reasonable to regard meaning of work as an exposure [1] or as an outcome [24] of the psychosocial work environment. This also applies to the status of other psychosocial variables. Theoretical and empirical development would make it possible to suggest which core variables should be included: (a) when describing the psychosocial work environment; and (b) when analysing its possible health effects.

Acknowledgements The 2000 round of the DWECS was funded by the Danish Working Environment Authority and the Ministry of Labour. The 2005 round was funded by the Ministry of Employment as a part of a surveillance programme on occupational health. The writing of this article was, in part, supported by a grant from the Danish Working Environment Research Fund (grant number: 5-2006-04). No conflict of interest is declared. The study has been notified to and registered by the Danish Data Protection Agency (Datatilsynet; see http://www.datatilsynet.dk/english for details). Questionnaire- and register-based studies do not need approval from the Danish National Committee on Biomedical Research Ethics (Den Centrale Videnskabetiske komite´; see http:// www.cvk.im.dk/cvk/site.aspx?p¼119 for details).

References [1] Kristensen TS, Hannerz H, Hogh A, Borg V. The Copenhagen Psychosocial Questionnaire – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31:438–49. [2] Karasek R. Job content questionnaire and user’s guide, rev. 1.7. Lowell: University of Massachusetts-Lowell; 1997. [3] Siegrist J, Starke D, Chandola T, Godin I, Marmot M, Niedhammer I, et al. The measurement of effort–reward imbalance at work: European comparisons. Soc Sci Med 2004;58:1483–99.

67

[4] Kompier M. Job design and well-being. In: Schabracq MJ, Winnubst JAM, Cooper CL, editors. The handbook of work and health psychology, 2nd edn. Chichester: Wiley; 2002. pp. 429–54. [5] Burr H, Bjorner JB, Kristensen TS, Tuchsen F, Bach E. Trends in the Danish work environment in 1990–2000 and their associations with labor-force changes. Scand J Work Environ Health 2003;29:270–9. [6] Feveile H, Olsen O, Burr H, Bach E. The Danish Work Environment Cohort Study 2005: from idea to sampling design. Statistics Transition 2007;8:441–58. [7] Bultmann U, Kant IJ, Van den Brandt PA, Kasl SV. Psychosocial work characteristics as risk factors for the onset of fatigue and psychological distress: prospective results from the Maastricht Cohort Study. Psychol Med 2002;32:333–45. [8] Eriksen W. Work factors as predictors of persistent fatigue: a prospective study of nurses’ aides. Occup Environ Med 2006;63:428–34. [9] Ylipaavalniemi J, Kivimaki M, Elovainio M, Virtanen M, Keltikangas-Jarvinen L, Vahtera J. Psychosocial work characteristics and incidence of newly diagnosed depression: a prospective cohort study of three different models. Soc Sci Med 2005;61:111–22. [10] Wieclaw J, Agerbo E, Mortensen PB, Burr H, Tuchsen F, Bonde JP. Psychosocial working conditions and the risk of depression and anxiety disorders in the Danish workforce. BMC Public Health 2008;8:280. [11] Borritz M, Bultmann U, Rugulies R, Christensen KB, Villadsen E, Kristensen TS. Psychosocial work characteristics as predictors for burnout: findings from 3-year follow up of the PUMA Study. J Occup Environ Med 2005;47:1015–25. [12] Ware Jr, JE, Sherbourne CD. The MOS 36-item short-form health survey (SF-36). I. Conceptual framework and item selection. Med Care 1992;30:473–83. [13] Rugulies R, Norborg M, Sorensen TS, Knudsen LE, Burr H. Effort–reward imbalance at work and risk of sleep disturbances. Cross-sectional and prospective results from the Danish Work Environment Cohort Study. J Psychosom Res 2009;66:75–83. [14] Pejtersen JH, Kristensen TS, Borg V, Bjorner JB. The second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II). Scand J Public Health 2010; 38(Suppl 3):8–24. [15] Feveile H, Olsen O, Hogh A. A randomized trial of mailed questionnaires versus telephone interviews: response patterns in a survey. BMC Med Res Methodol 2007;7:27. [16] Krause N, Brand RJ, Kaplan GA, Kauhanen J, Malla S, Tuomainen TP, et al. Occupational physical activity, energy expenditure and 11-year progression of carotid atherosclerosis. Scand J Work Environ Health 2007; 33:405–24. [17] Ibrahim S, Smith P, Muntaner C. A multi-group cross-lagged analyses of work stressors and health using Canadian National sample. Soc Sci Med 2008;68:49–59. [18] Choi B, Kawakami N, Chang S, Koh S, Bjorner J, Punnett L, et al. A cross-national study on the multidimensional characteristics of the five-item psychological demands scale of the Job Content Questionnaire. Int J Behav Med 2008;15:120–32. [19] Hogh A. Aggression at work. Bullying, nasty teasing and violence. Prevalence, mediating factors and consequences. Copenhagen: National Institute of Occupational Health, University of Copenhagen; 2005.

68

H. Burr et al.

[20] Hogh A, Borg V, Mikkelsen KL. Work-related violence as a predictor of fatigue: a 5-year follow-up of the Danish work environment cohort study. Work Stress 2003; 17:182–94. [21] Wieclaw J, Agerbo E, Mortensen PB, Burr H, Tuchsen F, Bonde JP. Work related violence and threats and the risk of depression and stress disorders. J Epidemiol Community Health 2006;60:771–5. [22] Cartwright S, Holmes N. The meaning of work: the challenge of regaining employee engagement and reducing cynicism. Human Resource Manag Rev 2006; 16:199–208. [23] Arnold KA, Turner N, Barling J, Kelloway EK, McKee MC. Transformational leadership and psychological well-being: the mediating role of meaningful work. J Occup Health Psychol 2007;12:193–203.

[24] Chalofsky N. An emerging construct for meaningful work. Human Resource Dev Int 2003;6:69–83. [25] Dallner M, Elo A-L, Gamberale F. Validation of the General Nordic Questionnaire (QPSNordic) for Psychological and Social Factors at Work. Copenhagen: Nord; 2000. [26] Ferrie JE, Head J, Shipley MJ, Vahtera J, Marmot MG, Kivimaki M. Injustice at work and incidence of psychiatric morbidity: the Whitehall II study. Occup Environ Med 2006;63:443–50. [27] Netterstrom B, Conrad N, Bech P, Fink P, Olsen O, Rugulies R, et al. The relation between work-related psychosocial factors and the development of depression. Epidemiol Rev 2008;30:118–32. [28] Wikman A, Ørhede E. Pilotprojekterne i Danmark og Sverige [The pilot projects in Denmark and Sweden]. Stockholm: Statistics Sweden; 1988.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 69–80

ORIGINAL ARTICLE

Psychosocial work environment and intention to leave the nursing profession: Results from the longitudinal Chinese NEXT study

JIAN LI1,2,3, HUA FU2, YAN HU4, LI SHANG3, YINGHUI WU5, TAGE S. KRISTENSEN6, BERND H. MUELLER1 & HANS MARTIN HASSELHORN1 1

Department of Safety Engineering, University of Wuppertal, Wuppertal, Germany, 2School of Public Health, Fudan University, Shanghai, China, 3School of Public Health, Kunming Medical University, Kunming, China, 4 School of Nursing, Fudan University, Shanghai, China, 5School of Nursing, Kunming Medical University, Kunming, China, and 6Task-Consult, Denmark

Abstract Aims: A shortage of nurses happens not only in developed countries, but also in developing countries, such as in China, but the nurse turnover here makes the situation worse. Why do Chinese nurses want to leave the nursing profession? Our hypothesis is that unfavourable psychosocial work environment could predict nurses’ intention to leave (ITL). Methods: Collaborating with the EU NEXT study (Nurses’ Early eXit sTudy), the longitudinal study was conducted in China, and the psychosocial work environment was measured with the Copenhagen Psychosocial Questionnaire (COPSOQ). A total of 3,088 registered female nurses working in hospitals were eligible for the baseline analyses by multivariate logistic regression, and 1,521 for the one-year follow-up analyses by multivariate Poisson regression. Results: A wide range of psychosocial factors at work – in particular, increased emotional demands, decreased meaning of work, decreased commitment to the workplace, and decreased job satisfaction – were associated with ITL in both baseline analyses and prospective analyses after adjusting for numerous confounders. Conclusions: The findings suggest that unfavourable psychosocial work environment predicts ITL in Chinese nurses. Improvements in the psychosocial work environment may be helpful in retention of the nursing workforce.

Key Words: Copenhagen Psychosocial Questionnaire, intention to leave, longitudinal study, psychosocial work environment

Introduction A shortage of nurses is reported to be an increasing problem worldwide. In 2002 it was reported that most industrialized countries are or will be facing nursing shortages [1]. Estimations from the Bureau of Labour Statistics in 2001 indicate that, in the USA, more than one million new nurses will be needed by the year 2010, and shortages are expected to grow to 30% by the year 2020 [2]. In a report from the Organisation for Economic Cooperation and Development (OECD, 2005) examining nurse shortages in the OECD countries the following underlying reasons for the increased demands were listed: economic expansion, population growth, an

ageing population, technological advances, and higher patient expectations [3]. Nursing shortage and understaffing have been linked to a number of indicators for insufficient care, such as patient mortality, adverse events, accidents and nosocomial infections [4]. However, it might be surprising to many people that the nurse shortage happens not only in developed countries, but also in developing countries, such as in China. According to the World Health Report 2006 (WHO, 2006), the average density of nurses per 1,000 throughout the world is 4.06, whereas the density of nurses per 1,000 in China is only 1.06, ranking 133rd out of 191 WHO member countries [5]. The Nursing Development Plan in

Correspondence: Jian Li, Department of Safety Engineering, University of Wuppertal, Gaussstrasse 20, 42119 Wuppertal, Germany. Tel: þ49 202 439-3253. Fax: þ49 202 439-3828. E-mail: [email protected] (Accepted 14 October 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809354361

70

J. Li et al.

China (2005–2010) released from the Ministry of Health [6] indicated clearly: ‘‘the nurse shortage has great influence on clinical nursing quality and the development of nursing workforce.’’ One of the key aims in the future is ‘‘to increase the number of nursing positions, and to promote the retention of existing nursing staff’’. Theoretically, there are four ways in which the pool of active nurses might be increased [7,8]. 1) The input may be increased by providing more education facilities at nursing schools. However, currently, it seems unlikely that an increase in provision of nursing training alone will solve the future demands for nursing staff. Among other reasons this is due to the relative unattractiveness of the nursing professions to young people in many countries [9]. 2) Another way of increased input would be through immigration of nursing staff from other countries. Currently nurse migration mainly occurs in an East to West direction. The phenomenon has been observed that more and more Chinese nurses go abroad to work in Singapore, Hong Kong, and even in Europe [10]. It seems that the input side in China could not improve the situation much. 3) On the output side, raising the retirement age may be regarded by some as a solution to the problem of a shortage of nurses. However, the rising unemployment rate in the Chinese labour market makes late retirement impossible. 4) As a result, the most effective way of assuring nursing in the future therefore seems to promote the retention of existing nursing staff. Collaborating with the European Nurses’ Early eXit sTudy (NEXT Study), the Chinese NEXT Study investigates the working conditions, health, and intention to leave (ITL) in registered hospital nurses. According to the research findings from the European NEXT Study, some psychosocial work characteristics, such as effort, reward, overcommitment (based on the Effort-Reward Imbalance model), job demand, and influence at work (based on the Demand-Control model) are the big reasons of professional turnover intention among European nurses [11,12]. Yet, some other psychosocial work characteristics, like emotional demands, meaning of work, predictability, quality of leadership, feedback, social community at work, and job satisfaction etc, are not well tested. In particular, client-specific work characteristics (such as emotional demands) are regarded as very important psychosocial stressors, especially in the area of human service work, for example, healthcare/nursing work [13,14]. These psychosocial characteristics in the workplace are all included in a recent questionnaire, the Copenhagen Psychosocial Questionnaire (COPSOQ) [15].

Furthermore, global changes in the economies during the past decades have heavily impacted the healthcare organizations, and accordingly, the working conditions of healthcare workers have changed enormously [16]. More recently, it has been pointed out that the accurate assessment of change in psychosocial work characteristics is crucial when investigating the causal relations between work stress and outcomes with longitudinal design [12,17]. Aims This article therefore has two aims. Firstly, we examine the impact of a broad range of psychosocial work characteristics measured by COPSOQ on ITL among a sample of Chinese registered hospital female nurses, in cross-sectional and one-year follow-up analysis. Secondly, we analyze the prospective association between change of psychosocial work characteristics and ITL. Methods Study design and population The Chinese NEXT is a one-year longitudinal study. It was approved by the Ethics Committee of Fudan University, China. In the baseline survey during January to May 2007, 3,800 nurses working in inpatient wards from 12 hospitals were recruited, of whom 3,418 filled in the questionnaire and returned it (response rate: 89.95%). We excluded 330 respondents (89 questionnaires with logical errors, 219 questionnaires with missing values of key measures, and 22 male nurses), resulting in a sample of 3,088 respondents at baseline for cross-sectional analysis. In the follow-up survey during January to May 2008, 3,800 questionnaires were sent to the same 12 hospitals, and 3,342 nurses responded (response rate: 87.95%). We further excluded 488 respondents (123 questionnaires with logical errors, 335 questionnaires with missing values of key measures, and 30 male nurses), resulting in a sample of 2,854 respondents at follow up. Of the 3,088 female respondents from the baseline survey, 1,791 nurses responded to the follow-up questionnaire (follow-up rate: 58.00%). One thousand, five hundred and twenty one female nurses who had no intention of leaving the nursing profession at baseline were included in the prospective follow-up analysis. Measurements Psychosocial work environment: Fourteen different psychosocial work characteristics were measured at

COPSOQ and ITL in Chinese nurses both baseline and follow up by the short version of the COPSOQ [15], whose validated Chinese version is available [18]. The following 14 characteristics of the psychosocial work environment were measured: quantitative demands, emotional demands, influence, possibilities for development, degree of freedom at work, meaning of work, commitment to the workplace, predictability, quality of leadership, social support, feedback, social community at work, job insecurity, and job satisfaction. For most items, five response categories were used either with intensity (from ‘‘to a very large extent’’ to ‘‘to a very small extent’’) or frequency (from ‘‘always’’ to ‘‘never/ hardly ever’’). All the scores for each psychosocial work characteristic were ranged from 0 to 100, the high value representing a high level of the concept being measured. Psychosocial work environment index: Besides each psychosocial work characteristic, a psychosocial work environment index was created to evaluate the combined effect of total characteristics [19]. The values of the four overall domains (demands at work, work organization and job contents, interpersonal relations and leadership, and work-individual interface, see Table III) were added together after adjusting for scoring direction, resulting in a possible score from 0 to 400, with a higher score indicating a more unfavourable psychosocial work environment. Intention to leave (ITL) the nursing profession: At both baseline and follow up, the intention to leave the nursing profession was measured by a single item (‘‘How often during the course of the past year have you thought about leaving nursing?’’). The response categories were ‘‘never’’, ‘‘sometimes a year’’, ‘‘sometimes a month’’, ‘‘sometimes a week’’, ‘‘every day’’. Nurses indicating that they wanted to leave the profession ‘‘sometimes a month’’ or ‘‘sometimes a week’’ or ‘‘every day’’ were considered to ‘‘have ITL’’ [11,12]. In addition, information on age, education, marital status, economic status, life style and work history were collected at baseline. Data analysis Descriptive statistics were generated in the first phase. Means and standard deviations (SDs) were investigated for continuous variables, and relative frequencies were examined for categorical variables. We calculated Pearson’s correlation coefficients among 14 psychosocial work characteristics at baseline and among 14 changes of psychosocial work characteristics between baseline and follow up. In addition, we calculated the correlation

71

coefficients between each pair of psychosocial work characteristics at baseline and at follow up. Next, we analyzed: (1) cross-sectional associations between psychosocial work characteristics and ITL at baseline in 3,088 nurses, using multivariate logistic regression. The results are shown as odds ratios (ORs) with 95% confidence intervals (CIs); due to weakness of drawing causal relations of cross-sectional analyses, we then explored (2) prospective associations between psychosocial work characteristics at baseline and new onset of ITL at follow up in 1,521 nurses, and furthermore (3) prospective associations between changes of psychosocial work characteristics and new onset of ITL at follow-up in 1,521 nurses. To take the ‘‘ceiling effect’’ and the ‘‘floor effect’’ into account, we followed a measurement procedures described by Twisk [20]. The increased changes were defined as [(scores at follow-up – scores at baseline) / (maximal score – scores at baseline)], and the decreased changes were defined as [(scores at follow-up – scores at baseline) / (scores at baseline – minimal score)]. Multivariate Poisson regression was applied for prospective associations to calculate relative risks (RRs) and 95% CIs. The Poisson regression modelling has been suggested for the estimation of RRs for binary data when the outcome is common (incidence410%) [21]. The analyses were adjusted in five steps: In model I, the psychosocial work characteristics were not adjusted for any confounding factors, i.e., the crude model; in model II, the psychosocial work characteristics were adjusted for age, birthplace, marital status, education, smoking, alcohol drinking, physical exercise, work tenure, work hours per week, shift work, contract status, position rank, professional title, and departments at baseline. In both model I and II, the 14 psychosocial work characteristics were not adjusted for each other. In model III, the 14 psychosocial work characteristics were mutually adjusted to examine the independent effects of each psychosocial work characteristic. In model IV, one psychosocial work characteristic (social support) was excluded from the mutually adjusted analyses due to its high correlation with other psychosocial work characteristics [22]. In model V, two more psychosocial work characteristics (commitment to the workplace and job satisfaction) were further excluded from the mutually adjusted analyses because they might be considered as ‘‘intermediate’’ variables between the work environment and outcomes [15,19]. Besides, the psychosocial work environment index was regressed only in models I and II without adjusting for the 14 psychosocial work characteristics.

72

J. Li et al.

Table I. Characteristics of study participants at baseline and at follow up. Baseline Continuous variable

Follow-up

n

Mean  SD

n

Mean  SD

Age (years) Work hours per week (hours)

3,088 3,088

30.94  8.29 41.71  3.99

1,521 1,521

30.79  7.97 41.70  4.00

Categorical variables

n

%

n

%

Birthplace Local Outside

2,504 584

81.09 18.91

1,261 260

82.91 17.09

Marital status Unmarried Married Others

1,193 1,836 59

38.63 59.46 1.91

579 911 31

38.07 59.89 2.04

Educational level Nursing school Nursing college Nursing university and higher

1,512 1,337 239

48.96 43.30 7.74

769 625 127

50.56 41.09 8.35

Smoking Never Ex-smoker Current smoker

3,021 41 26

97.83 1.33 0.84

1,494 21 6

98.22 1.38 0.40

Alcohol drinking Never Ex-drinker Current drinker

2,591 227 270

83.91 7.35 8.74

1,293 110 118

85.01 7.23 7.76

Physical exercise Never Ex-exerciser Current exerciser

931 1,644 513

30.15 53.24 16.61

462 815 244

30.37 53.58 16.05

Work tenure < 5 years 5–9 years 10–14 years 414 years

875 642 762 809

28.33 20.79 24.68 26.20

416 323 392 390

27.35 21.24 25.77 25.64

Shift work Regular day shift Irregular day shift Rotating shift (day þ night)

725 311 2,052

23.48 10.07 66.45

393 156 972

25.84 10.26 63.90

Position rank Low High

2,446 642

79.21 20.79

1,182 339

77.71 22.29

Professional title Junior Senior

1,361 1,727

44.07 55.93

630 891

41.42 58.58

Contract status Permanent Temporary

2,312 776

74.87 25.13

1,170 351

76.92 23.08

Departments Internal medicine Surgery Gynecology & Obstetrics Operation room Emergency Other clinics ICU Others

896 756 204 204 182 340 233 273

29.03 24.48 6.60 6.60 5.89 11.01 7.55 8.84

451 378 88 113 63 165 117 146

29.65 24.85 5.79 7.43 4.14 10.85 7.69 9.60

COPSOQ and ITL in Chinese nurses All scores of psychosocial work characteristics were ranged from 0 to 100. Rather than reporting the effect of an increase by 1 point, all the scores were Z-transformed in order to achieve parameter estimates on a scale that is easier to interpret. Thus, the ORs and RRs were presented for an increase by 1 SD in the independent variables. All analyses were conducted by the program SAS 9.2.

Results Characteristics of study participants at baseline and at follow up Table I gives information on the sample composition (means and percentages of socio-demographic and occupational characteristics) at baseline and at follow up. Of the 3,088 baseline female respondents in this study, the mean age was 30.94 years, 81.09% were from local area and 59.46% were married. Most of the nurses (92.26%) graduated from nursing schools or colleges. Few of them smoked, drank alcohol, or took regular exercise. The mean working hours per week was 41.71, and the mean duration of healthcare practice was 11.26 years; 66.45% of the nurses had the duty of rotating shift (day þ night), and 79.21% occupied a low position rank. Nearly three quarters had a permanent work contract, and more than half served in the departments of internal medicine and surgery. The characteristics of the follow-up sample were quite comparable to the baseline sample. Intention to leave the nursing profession At baseline, the prevalence of intention to leave the nursing profession was 16.26% (502 out of 3,088 nurses). After one-year follow-up, among 1,521 nurses who were initially free of intention to leave the nursing profession at baseline, 220 nurses developed ITL, and the incidence rate was 14.46%. Correlations among psychosocial work characteristics at baseline and at follow up As shown in Table II, at baseline, the highest correlation between the characteristics on work organization and job contents was between ‘‘meaning of work’’ and ‘‘commitment to the workplace’’ (r ¼ 0.50, p < 0.001); among the five characteristics on interpersonal relations and leadership the highest correlation occurred between ‘‘social support’’ and ‘‘feedback’’ (r ¼ 0.72, p < 0.001). This was the highest correlation among the 14 psychosocial work characteristics at baseline. The similar pattern was

73

found among changes of psychosocial work characteristics between baseline and follow-up, the highest correlation between the changes of characteristics on work organization and job contents was between ‘‘change of meaning of work’’ and ‘‘change of commitment to the workplace’’ (r ¼ 0.37, p < 0.001); among the five changes of characteristics on interpersonal relations and leadership the highest correlation occurred between ‘‘change of social support’’ and ‘‘change of feedback’’ (r ¼ 0.67, p < 0.001). This was the highest correlation among the 14 change variables. Moreover, each pair of psychosocial work characteristics at baseline and at follow-up was significantly correlated (r ¼ 0.28–0.45, p < 0.001) Baseline associations between psychosocial work characteristics and ITL At baseline, even after adjustment for a number of confounding factors (model II), most psychosocial work characteristics (except influence and job insecurity) were strongly associated with ITL (p < 0.001). In the mutually adjusted model III, social support showed a reverse association with ITL (OR 1.27, 95% CI 1.04–1.56). In the final mutually adjusted model V, high emotional demands (OR 1.99, 95% CI 1.73–2.30), low meaning of work (OR 0.78, 95% CI 0.69–0.87), low predictability (OR 0.72, 95% CI 0.63–0.82), and low possibilities for development (OR 0.79, 95% CI 0.70–0.90) were the strongest psychosocial work characteristics related to ITL (Table III). Prospective associations between psychosocial work characteristics at baseline and new onset of ITL at follow up As can be seen from Table IV, in model II, many baseline psychosocial work characteristics predicted new onset of ITL after one-year follow-up (p < 0.05). However, all significant associations (except commitment to the workplace at baseline and new onset of ITL) disappeared when making mutual adjustment (model III). In the final mutually adjusted model V, low meaning of work was the only significant predictor of new onset of ITL (RR 0.86, 95% CI 0.75–0.98), while high quantitative demands and high emotional demands had marginal associations with new onset of ITL. Prospective associations between changes of psychosocial work characteristics and new onset of ITL at follow up Table V shows that, in model II, most changes of psychosocial work characteristics between baseline

QD, quantitative demands; ED, emotional demands; IN, influence; PD, possibilities for development; DF, degree of freedom at work; MW, meaning of work; CW, commitment to the workplace; PR, predictability; QL, quality of leadership; SS, social support; FE, feedback; SW, social community at work; JI, job insecurity; JS, job satisfaction. Area in light grey is for correlation matrix among psychosocial work characteristics at baseline (n ¼ 3,088); area in clear is for correlation matrix among changes of psychosocial work characteristics between baseline and follow up (n ¼ 1,521); area in dark grey is for correlation matrix between each pair of psychosocial work characteristics at baseline and at follow-up (n ¼ 1,521). *p < 0.05, **p < 0.01, ***p < 0.001

JS

0.26*** 0.46*** 0.03 0.22*** 0.17*** 0.35*** 0.46*** 0.36*** 0.31*** 0.27*** 0.27*** 0.28*** 0.02 0.36***

JI SW FE

0.06** 0.12*** 0.14*** 0.25*** 0.16*** 0.23*** 0.23*** 0.33*** 0.53*** 0.72*** 0.33*** 0.35*** 0.00 0.13***

SS

0.09*** 0.14*** 0.09*** 0.27*** 0.13*** 0.28*** 0.27*** 0.30*** 0.68*** 0.33*** 0.67*** 0.45*** 0.01 0.13***

QL

0.15*** 0.19*** 0.02 0.25*** 0.09*** 0.28*** 0.28*** 0.27*** 0.38*** 0.49*** 0.41*** 0.33*** 0.02*** 0.16***

PR

0.08*** 0.21*** 0.27*** 0.38*** 0.28*** 0.36*** 0.43*** 0.32*** 0.18*** 0.16*** 0.23*** 0.10*** 0.09*** 0.20***

CW

0.15*** 0.29*** 0.10*** 0.33*** 0.12*** 0.50*** 0.45*** 0.37*** 0.16*** 0.16*** 0.18*** 0.18*** 0.09*** 0.26*** 0.15*** 0.25*** 0.14*** 0.44*** 0.10*** 0.39*** 0.37*** 0.23*** 0.16*** 0.17*** 0.12*** 0.17*** 0.05 0.13***

MW DF

0.02 0.08*** 0.26*** 0.15*** 0.28*** 0.08** 0.09*** 0.20*** 0.05 0.03 0.06* 0.02 0.01 0.11*** 0.02 0.08*** 0.31*** 0.36*** 0.07** 0.34*** 0.25*** 0.30*** 0.12*** 0.10*** 0.09*** 0.16*** 0.01 0.10***

PD IN

0.19*** 0.17*** 0.39*** 0.18*** 0.21*** 0.05 0.06* 0.22*** 0.03 0.04 0.07** 0.01 0.01 0.01 0.48*** 0.41*** 0.18*** 0.02 0.06* 0.09*** 0.20*** 0.10*** 0.08** 0.09*** 0.06* 0.10*** 0.02 0.27***

ED QD

0.33*** 0.33*** 0.13*** 0.03 0.01 0.08** 0.11*** 0.01 0.06* 0.02 0.01 0.09*** 0.01 0.16***

Table II. Correlation matrix among psychosocial work characteristics at baseline and at follow up.

0.15*** 0.19*** 0.05** 0.28*** 0.05** 0.35*** 0.32*** 0.26*** 0.44*** 0.59*** 0.46*** 0.32*** 0.01 0.16***

0.02 0.04* 0.05* 0.03 0.01 0.03 0.13*** 0.02 0.05** 0.01 0.01 0.04* 0.41*** 0.02

J. Li et al.

QD ED IN PD DF MW CW PR QL SS FE SW JI JS

74

and follow up (except change of influence and change of job insecurity) were significantly associated with new onset of ITL (p < 0.05). In the mutually adjusted model III, only increased emotional demands, decreased commitment to the workplace, and decreased job satisfaction remained significant (p < 0.01). In the final mutually adjusted model V, it was found that increased emotional demands (RR 1.47, 95% CI 1.28–1.68), and decreased meaning of work (RR 0.80, 95% CI 0.69–0.94) were the only psychosocial work factors that significantly predicted new onset of ITL, while increased high quantitative demands had marginal association with new onset of ITL. The psychosocial work environment index Not surprisingly, the psychosocial work environment index was associated with ITL, in crosssectional analysis, in prospective analysis, and analysis of change score as predictor of new onset of ITL (p < 0.001) (details see model I and II in Tables III–V).

Discussion The main aim of this paper was to explore the impact of psychosocial work environment measured by COPSOQ on ITL. To our knowledge, this is the first study addressing the association between the COPSOQ factors and ITL worldwide. The results indicated that a wide range of psychosocial work characteristics, particularly high emotional demands, low meaning of work, low commitment to the workplace, and low job satisfaction were constantly predictive. The findings were in line with other studies [23]. In European nurses, the first three factors strongly associated with ITL were poor professional opportunities, unpleasant work organization (including quantitative demands, influence at work, role conflict and ambiguity, and possibilities for development), and low health status, whereas work content (including emotional demands, and lifting and bending) just played a minor role [8]. A recent Swedish study investigated the factors contributing to the decision to leave nursing care, and it was found that unsatisfactory salary contributed most to the nursing personnel’s decision to leave, followed by lack of professional opportunities and restricted professional autonomy [24]. Investigating 787 nurses in the USA showed the work/professional satisfaction and organizational commitment had implications for ITL [25], while another USA study also confirmed such

COPSOQ and ITL in Chinese nurses

75

Table III. Baseline associations between psychosocial work characteristics and ITL (n ¼ 3,088). Model I OR (95% CI) Demands at work Quantitative demands 1.65 (1.50, Emotional demands 2.49 (2.21, Work organization and job contents Influence 0.99 (0.90, Possibilities for 0.61 (0.55, development Degree of freedom at work 0.75 (0.67, Meaning of work 0.54 (0.49, Commitment to the 0.31 (0.28, workplace Interpersonal relations and leadership Predictability 0.52 (0.47, Quality of leadership 0.63 (0.58, Social support 0.70 (0.64, Feedback 0.73 (0.66, Social community at work 0.66 (0.60, Work-individual interface Job insecurity 0.94 (0.86, Job satisfaction 0.26 (0.22, Psychosocial Work 3.27 (2.89, Environment Index

Model II OR (95% CI)

Model III OR (95% CI)

Model IV OR (95% CI)

Model V OR (95% CI)

1.82)*** 2.80)***

1.63 (1.48, 1.80)*** 2.49 (2.20, 2.82)***

1.19 (1.04, 1.35)* 1.52 (1.30, 1.78)***

1.19 (1.04, 1.36)** 1.52 (1.30, 1.77)***

1.16 (1.03, 1.31)* 1.99 (1.73, 2.30)***

1.09) 0.67)***

1.00 (0.90, 1.10) 0.61 (0.56, 0.68)***

1.11 (0.98, 1.26) 0.83 (0.73, 0.94)**

1.11 (0.98, 1.26) 0.84 (0.73, 0.95)**

1.14 (0.99, 1.27) 0.79 (0.70, 0.90)***

0.83)*** 0.59)*** 0.35)***

0.75 (0.67, 0.83)*** 0.54 (0.49, 0.60)*** 0.31 (0.27, 0.35)***

0.94 (0.83, 1.07) 1.03 (0.91, 1.17) 0.44 (0.38, 0.52)***

0.94 (0.83, 1.07) 1.03 (0.91, 1.17) 0.45 (0.39, 0.52)***

0.88 (0.78, 0.99)* 0.78 (0.69, 0.87)***

0.58)*** 0.69)*** 0.77)*** 0.80)*** 0.72)***

0.52 (0.46, 0.63 (0.58, 0.70 (0.64, 0.72 (0.65, 0.67 (0.61,

0.99 (0.86, 0.87 (0.75, 1.27 (1.04, 1.00 (0.84, 0.94 (0.81,

0.99 (0.85, 1.15) 0.95 (0.83, 1.08)

0.72 (0.63, 0.82)*** 0.85 (0.75, 0.96)*

1.10 (0.95, 1.27) 0.99 (0.87, 1.13)

1.08 (0.94, 1.24) 0.93 (0.83, 1.06)

1.04) 0.30)*** 3.70)***

0.95 (0.86, 1.05) 0.25 (0.22, 0.29)*** 3.28 (2.89, 3.72)***

1.01 (0.90, 1.14) 0.40 (0.34, 0.48)***

0.92 (0.82, 1.02)

0.58)*** 0.70)*** 0.78)*** 0.80)*** 0.74)***

1.15) 1.01) 1.56)* 1.18) 1.07)

1.01 (0.90, 1.14) 0.40 (0.34, 0.47)***

*p < 0.05, **p < 0.01, ***p < 0.001. Model I: non-adjustment. Model II: adjustment for age, birthplace, marital status, education, smoking, alcohol drinking, physical exercise, work tenure, work hours per week, shift work, contract status, position rank, professional title, and departments. Model III: Model II þ adjustment for all psychosocial work characteristics mutually. Model IV: Model II þ adjustment for psychosocial work characteristics mutually, except social support. Model V: model II þ adjustment for psychosocial work characteristics mutually, except social support, commitment to the workplace, and job satisfaction.

findings [26]. In Taiwan, two studies showed that ITL-related psychosocial factors at work were job satisfaction, professional commitment, salary and promotion [27,28]. Interestingly, two cross-sectional studies available from China indicated low pay and high workload out of seven work-related characteristics were the main reasons for nurses’ ITL [29,30]. One of the highlights of the present study was to use the psychosocial work environment index, which was produced by combining four overall domains of psychosocial work characteristics. We found better risk estimations in association with ITL. In a recent Danish longitudinal study, the psychosocial work environment index was also used, showing a gradual association with future sickness absence days [19]. The findings from our study further emphasized the importance of changes of psychosocial work characteristics in prediction to developing ITL, since the baseline psychosocial work characteristics showed limited predictive power for new onset of ITL. Such accurate exposure assessment in work stress research has drawn more attention in recent years [12,17]. The prospective facts of reduced risks of ITL by improved psychosocial work characteristics provide solid evidence to policy makers and mangers of nursing institutions.

Concerning the outcome of intention to leave the nursing profession, in our study, the prevalence of ITL at baseline was 16.26%, and the incidence rate of developing ITL at follow up was 14.46%. The results were comparable to Western countries. The average prevalence of ITL in 10 European countries was 16.50% (9.92–36.59%), while in the USA it was 22.7% and in Canada it was 16.6% [31]. The one-year incidence rate of developing ITL in eight European countries was 8.73% (5.09–14.11%) [7]. Another outcome from the European NEXT Study was actual turnover, with a one-year rate of 7.99% (5.33–13.99% across eight countries) [7]. According to a report from China [32], the nurses’ actual turnover rate in Shanghai during 2001 to 2005 was 12.8%. All the substantial numbers suggest the critical situation of retaining existing nursing workforce in China. Methodological considerations The measurement of outcome: ITL vs. actual turnover. In general, ITL is regarded as the immediate precursor of actual turnover [33], in that it is closely linked to the subsequent steps in the decision process leading to turnover in nursing. In the Chinese NEXT Study,

76

J. Li et al.

Table IV. Prospective associations between psychosocial work characteristics at baseline and new onset of ITL at follow up (n ¼ 1,521).

Demands at work Quantitative demands Emotional demands

Model I RR (95% CI)

Model II RR (95% CI)

1.26 (1.12, 1.41)*** 1.33 (1.18, 1.51)***

Work organization and job contents Influence 0.98 (0.87, Possibilities for development 0.81 (0.72, Degree of freedom at work 0.94 (0.83, Meaning of work 0.76 (0.68, Commitment to the 0.61 (0.54, workplace Interpersonal relations and leadership Predictability 0.79 (0.70, Quality of leadership 0.90 (0.79, Social support 0.93 (0.82, Feedback 0.93 (0.83, Social community at work 0.84 (0.76, Work-individual interface Job insecurity Job satisfaction Psychosocial Work Environment Index

Model III RR (95% CI)

Model IV RR (95% CI)

1.25 (1.11, 1.41)*** 1.29 (1.13, 1.48)***

1.12 (0.97, 1.30) 1.07 (0.91, 1.25)

1.13 (0.98, 1.30) 1.07 (0.91, 1.25)

1.15 (0.99, 1.32) 1.15 (0.98, 1.36)

1.11) 0.91)*** 1.06) 0.85)*** 0.68)***

1.02 0.85 0.93 0.77 0.63

(0.89, 1.16) (0.75, 0.96)* (0.82, 1.05) (0.69, 0.87)*** (0.56, 0.72)***

1.04 (0.90, 1.19) 0.95 (0.83, 1.08) 1.00 (0.88, 1.14) 0.96 (0.84, 1.09) 0.70 (0.60, 0.82)***

1.04 (0.90, 1.20) 0.95 (0.83, 1.08) 1.00 (0.88, 1.14) 0.96 (0.84, 1.09) 0.70 (0.60, 0.82)***

1.04 (0.90, 0.93 (0.82, 0.99 (0.87, 0.86 (0.75,

0.89)*** 1.01) 1.04) 1.05) 0.94)**

0.81 0.90 0.92 0.93 0.86

(0.71, 0.92)*** (0.79, 1.02) (0.82, 1.04) (0.82, 1.05) (0.77, 0.95)**

0.99 (0.85, 1.16) 1.02 (0.87, 1.19) 1.10 (0.90, 1.33) 1.02 (0.85, 1.22) 0.94 (0.82, 1.09)

0.99 (0.85, 1.16) 1.05 (0.90, 1.22)

0.89 (0.76, 1.04) 1.01 (0.86, 1.18)

1.05 (0.90, 1.23) 0.96 (0.84, 1.10)

1.04 (0.89, 1.22) 0.95 (0.82, 1.08)

0.91 (0.80, 1.04) 0.72 (0.63, 0.82)*** 1.39 (1.23, 1.57)***

0.95 (0.83, 1.08) 0.87 (0.75, 1.02)

0.95 (0.83, 1.08) 0.87 (0.75, 1.02)

0.89 (0.78, 1.02)

0.91 (0.80, 1.03) 0.70 (0.62, 0.79)*** 1.44 (1.28, 1.62)***

Model V RR (95% CI)

1.20) 1.07) 1.13) 0.98)*

*p < 0.05, **p < 0.01, ***p < 0.001. Model I: non-adjustment. Model II: adjustment for age, birthplace, marital status, education, smoking, alcohol drinking, physical exercise, work tenure, work hours per week, shift work, contract status, position rank, professional title, and departments. Model III: Model II þ adjustment for all psychosocial work characteristics mutually. Model IV: Model II þ adjustment for psychosocial work characteristics mutually, except social support. Model V: model II þ adjustment for psychosocial work characteristics mutually, except social support, commitment to the workplace, and job satisfaction.

we did not follow up the actual turnover, but both outcomes were explored in the European NEXT Study, where 53.2% of all those who later left the profession to work in another profession had ‘‘frequently considered’’ leaving nursing vs. 13.7% of those who remained in their institution during the following 12 months. In addition, the final decision to leave the profession is usually made within six months prior to departure among 83% of all actual leavers, while 80% of all leavers had started seriously considering leaving the profession within the 12 months prior to departure [7]. Moreover, the use of ITL instead of actual turnover in our study could reveal a target population for early intervention before the turnover process becomes too advanced and perhaps irreversible [33].

Multicollinearity High correlations between explanatory variables (multicollinearity) can cause instability of item parameter estimates in multivariate regression modelling. However, in many social and psychological studies, particularly the questionnaire surveys, the explanatory variables are usually correlated so that

multicollinearity occurs frequently [22,34]. As can be seen in Table II, most of the 14 psychosocial work characteristics were correlated at baseline and at follow up (particularly social support with other psychosocial factors at work). Multicollinearity could seriously distort the interpretation of a regression model. Often, the sign of regression coefficients might have large standard errors or might be contrary to expectation, which is the initial indication of problematic multicollinearity. In our study, when adjusting for the 14 psychosocial work characteristics mutually, the associations between ITL and many psychosocial work characteristics were not significant any longer (see model III in Tables III–V), even the social support turned out to be a risk factor to ITL (see model III in Table III). It has been pointed out that the simplest method to eliminate problematic multicollinearity is to delete at least one offending explanatory variable from the regression model. Therefore, social support was excluded from the mutually adjusted analyses (see Model IV in Tables III–V), and the sign of regression coefficients had smaller standard errors afterwards. An alternative approach is centering of explanatory variables, which could reduce correlations between them, as we did by

0.85)*** 0.90)** 0.94)** 0.97)* 0.90)**

0.92 (0.81, 1.05) 0.57 (0.50, 0.64)*** 1.80 (1.59, 2.03)***

(0.65, (0.66, (0.70, (0.75, (0.66,

(0.66, (0.65, (0.70, (0.75, (0.66,

(0.89, (0.67, (0.76, (0.57, (0.46, 0.86)*** 0.89)*** 0.94)** 0.97)* 0.91)**

1.16) 0.90)** 0.96)* 0.98)*** 0.62)***

0.91 (0.80, 1.04) 0.59 (0.52, 0.67)*** 1.76 (1.54, 1.99)***

0.75 0.76 0.82 0.85 0.78

1.02 0.78 0.85 0.67 0.53

1.29 (1.14, 1.46)*** 1.62 (1.42, 1.83)***

Model II RR (95% CI)

(0.90, (0.81, (0.83, (0.90, (0.79,

(0.91, (0.81, (0.85, (0.77, (0.57, 1.19) 1.07) 1.17) 1.23) 1.08)

1.17) 1.06) 1.06) 1.04) 0.82)***

0.94 (0.82, 1.07) 0.75 (0.66, 0.86)***

1.04 0.93 0.98 1.05 0.92

1.03 0.93 0.95 0.89 0.68

1.10 (0.97, 1.24) 1.30 (1.14, 1.49)***

Model III RR (95% CI)

(0.90, (0.79, (0.83, (0.69,

1.16) 1.03) 1.05) 0.94)**

0.94 (0.82, 1.07) 0.75 (0.66, 0.86)***

0.91 (0.80, 1.04)

1.01 (0.89, 1.16) 0.89 (0.76, 1.04)

1.02 0.90 0.93 0.80

1.04 (0.92, 1.18) 0.92 (0.79, 1.07)

1.17) 1.06) 1.06) 1.04) 0.82)***

0.91 (0.79, 1.05) 0.89 (0.77, 1.03)

(0.91, (0.81, (0.85, (0.77, (0.57,

1.12 (0.99, 1.27) 1.47 (1.28, 1.68)***

Model V RR (95% CI)

1.04 (0.90, 1.19) 0.93 (0.81, 1.07)

1.03 0.93 0.95 0.89 0.68

1.10 (0.97, 1.24) 1.30 (1.14, 1.49)***

Model IV RR (95% CI)

*p < 0.05, **p < 0.01, ***p < 0.001. Model I: non-adjustment. Model II: adjustment for age, birthplace, marital status, education, smoking, alcohol drinking, physical exercise, work tenure, work hours per week, shift work, contract status, position rank, professional title, and departments. Model III: Model II þ adjustment for all changes of psychosocial work characteristics mutually. Model IV: Model II þ adjustment for changes of psychosocial work characteristics mutually, except change of social support. Model V: model II þ adjustment for changes of psychosocial work characteristics mutually, except change of social support, change of commitment to the workplace, and change of job satisfaction.

Change of work-individual interface Change of job insecurity Change of job satisfaction Change of Psychosocial Work Environment Index

0.75 0.77 0.81 0.85 0.77

Change of interpersonal relations and leadership Change of predictability Change of quality of leadership Change of social support Change of feedback Change of social community at work

1.15) 0.88)*** 0.96)* 0.77)*** 0.61)***

1.01 0.76 0.85 0.66 0.52

Change of work organization and job contents Change of influence Change of possibilities for development Change of degree of freedom at work Change of meaning of work Change of commitment to the workplace (0.88, (0.66, (0.75, (0.56, (0.44,

1.32 (1.17, 1.48)*** 1.65 (1.46, 1.87)***

Change of demands at work Change of quantitative demands Change of emotional demands

Model I RR (95% CI)

Table V. Prospective associations between changes of psychosocial work characteristics and new onset of ITL at follow-up (n ¼ 1,521).

COPSOQ and ITL in Chinese nurses 77

78

J. Li et al.

using Z-transformation. Another solution to multicollinearity is to reduce the number of explanatory variables by grouping them [22,34]. The psychosocial work environment index calculated by combining all 14 psychosocial work characteristics showed better explanatory power, as seen from all our regression analyses. Mediating effects of intermediate variables Conceptually, commitment to the workplace and job satisfaction could very well be considered as intermediate variables between the work environment and outcomes. As such, they are not really essential work environment variables. Thus, to treat these two variables equally with other psychosocial factors in the workplace would make the causal inference unclear and make the effect size decrease [15,19]. A number of studies point out clearly that commitment to the workplace and job satisfaction are the intermediate steps between work conditions and ITL [25–29,35]. For this reason, commitment to the workplace and job satisfaction were excluded from the mutually adjusted regression analyses to explore the mediating effects of them. Compared to model IV, there were weak to modest increases in the effect sizes of most psychosocial work characteristics in model V (see Tables III–V). Measuring changes of psychosocial work characteristics The measurement of change between two time points using repeated measures is often more complicated than is usually realized. The common methods are absolute change, relative change, and absolute change correcting for baseline value; while other techniques are also available, such as analysis of residual change and analysis of covariance [20]. However, it is often found that the conclusions from the same dataset about change measured by different analyses are inconsistent. This situation is referred to as Lord’s paradox, which occurs where baseline differences cannot be attributed to chance alone [17,36]. Another solution might be putting both the baseline score and the follow-up score as explanatory variables into same regression model, and the coefficient of follow-up score could be interpreted as a change over time. But since the baseline score and the follow-up score are usually highly correlated (see Table II) this approach runs the risk of multicollinearity. Importantly, the psychosocial work characteristics measured with COPSOQ have a maximal score of 100 and a minimal score of 0. The participants with very high scores at baseline would not be able to increase at follow up and those

with very low scores could not really decrease, that is, the ‘‘ceiling effect’’ and the ‘‘floor effect’’, which should be considered seriously. Thus, we applied the method introduced by Twisk to measure changes of psychosocial work characteristics while taking into account ‘‘ceiling/floor’’ [20]. While the recommendation from Twisk concerns the outcome variable, we have applied the approach to the explanatory variables. Basically, our transformation might give more weight to an upwards change that occurs near the ceiling and a downwards change that occurs near the floor. Compared with other methods of measuring changes, such a technique demonstrated plausible predictive improvement (data not shown). Further debates on this issue are worth developing. Strengths of the study The main strengths of the study are the longitudinal design and the large sample size. Many studies on working conditions, ITL, and turnover were cross-sectional with relatively small samples [37]. More importantly, this is the first longitudinal study in China on work stress research, to the best of our knowledge [38]. Given the nature of research design, it is possible to draw conclusions on causality of the observed associations between work stress and ITL. The comprehensive measurement of psychosocial work characteristics by a standardized questionnaire is another strength of our study. Even by using the COPSOQ short version, 14 main psychosocial work characteristics could be covered [15]. The third strength is the repeated measure of psychosocial work characteristics over time, as mentioned above, to assess accurate changes [17]. We suggest further studies taking this into account. Limitations of the study Firstly, the follow-up rate of this Chinese NEXT Study was 58.0%. Compared to the European NEXT Study (40.5%), it seems to be acceptable. When looking at the differences between the nurses who participated in the baseline survey only and those who participated in both baseline and follow-up surveys, it is found that the prevalence of ITL in the former nurses was higher (17.89% vs. 15.08%). Therefore, the relatively low follow-up rate might partly be due to actual turnover. Secondly, since a self-reported questionnaire survey was used to measure both explanatory and outcome variables, common method variance might bias the associations in this study. However, it is suggested that common method variance would be less problematic when the study design is longitudinal [39]. Thirdly, the healthy

COPSOQ and ITL in Chinese nurses worker effect should also be considered. The nurses who suffered from extremely unfavourable psychosocial work environment factors might already have left the hospital so that they could not participate in our surveys. Thus, the observed associations may be underestimated. Lastly, it has been stated by many studies that money/payment is an important factor to predict ITL [24,29,30,37], however, it is not measured by the current COPSOQ I short version. Notably, in the newly developed COPSOQ II, monetary factor is included. Besides, a broader concept of reward from work should also be considered, such as Effort-Reward Imbalance, for further analysis [11]. In conclusion, it is crucial for China to retain its existing nursing staff when facing the critical situation of nurse shortages in the country. Our research findings indicate that a broad range of psychosocial work characteristics are associated with intention to leave the nursing profession cross-sectionally and prospectively. The fact of considerable variability of the psychosocial work environment assessed during the 12-month period may imply potential for improvement, which makes the stress intervention in the workplace warranted.

Acknowledgements This research is supported by a Marie Curie International Incoming Fellowship within the 7th European Community Framework Programme (PIIF-GA-2008-220641). The authors would like to thank all the nurses of the Chinese NEXT Study for their continuous participation, and we also gratefully acknowledge all the nursing department directors for their kind assistance during data collection. Last but not least, the authors thank the anonymous reviewers and editors for their invaluable suggestions.

Conflict of interest None declared.

References [1] International Council of Nurses (ICN). Workforce forum 2002. Available at: http://www.icn.ch/forum2002overview.pdf (accessed 16 January 2009). [2] Kim VE. Women and international migration in the health sector: final report of public services international’s participatory action research 2003. Available at: http:// www.world-psi.org/Content/ContentGroups/English7/ Publications1/Final_Report_Migration.pdf (accessed 16 January 2009).

79

[3] Simoens S, Villeneuve M, Hurst J. Tackling nurse shortages in OECD countries. OECD health working papers no. 19, DELSA/ELSA/WD/HEA(2005)1. Available at: http:// www.oecd.org/dataoecd/11/10/34571365.pdf (accessed 16 January 2009). [4] Buchan J, Calman L. The global shortage of registered nurses: an overview of issues and actions. Geneva: ICN; 2004. Available at: http://www.icn.ch/global/shortage.pdf (accessed 16 January 2009). [5] World Health Report. Annex table 4, global distribution of health workers in WHO member states, the World Health Report 2006 – Working Together for Health, Geneva. Available at: http://www.who.int/whr/2006/annex/ 06_annex4_en.pdf (accessed 16 January 2009). [6] Ministry of Health. The nursing development plan in China (2005–2010) [in Chinese]. Chinese Hospitals 2005;9(9):26–8. [7] Hasselhorn HM, Muller BH, Tackenberg P. University of Wuppertal: NEXT Scientific Report; 2005. Available at: http://www.next.uni-wuppertal.de/EN/download.php?f¼ 133131f960ff1150b2e0192c51d18f53&target¼0 (accessed 9 September 2009). [8] Hasselhorn HM, Tackenberg P, Kuemmerling A, Wittenberg J, Simon M, Conway PM, et al. Nurses’ health, age and the wish to leave the profession – findings from the European NEXT-Study. Med Lav 2006;97(2):207–14. [9] Ungos K, Thomas E. Lessons learned from China’s healthcare system and nursing profession. J Nurs Scholarsh 2008;40(3):275–81. [10] Fang ZZ. Potential of China in global nurse migration. Health Serv Res 2007;42(3 Pt 2):1419–28. [11] Hasselhorn HM, Tackenberg P, Peter R, Next-Study Group. Effort-reward imbalance among nurses in stable countries and in countries in transition. Int J Occup Environ Health 2004;10(4):401–8. [12] Hasselhorn HM, Conway PM, Widerszal-Bazyl M, Simon M, Tackenberg P, Schmidt S, Camerino D, Mu¨ller BH, NEXT study group. Contribution of job strain to nurses’ consideration of leaving the profession – results from the longitudinal European nurses’ early exit study. Scand J Work Environ Health 2008; Suppl(6):75–82. [13] van Vegchel N, de Jonge J, Meijer T, Hamers JP. Different effort constructs and effort-reward imbalance: effects on employee well-being in ancillary health care workers. J Adv Nurs 2001;34(1):128–36. [14] Rugulies R, Braff J, Frank JW, Aust B, Gillen M, Yen IH, et al. The psychosocial work environment and musculoskeletal disorders: design of a comprehensive intervieweradministered questionnaire. Am J Ind Med 2004;45 (5):428–39. [15] Kristensen TS, Hannerz H, Hogh A, Borg V. The Copenhagen Psychosocial Questionnaire – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31(6):438–49. [16] Ostry AS, Spiegel JM. Labor markets and employment insecurity: impacts of globalization on service and healthcare-sector workforces. Int J Occup Environ Health 2004;10(4):368–74. [17] Smith P, Beaton D. Measuring change in psychosocial working conditions: methodological issues to consider when data are collected at baseline and one follow-up time point. Occup Environ Med 2008;65(4):288–96. [18] Shang L, Liu P, Fan LB, Gu HK, Li J. Psychometric properties of the Chinese version of Copenhagen

80

[19]

[20]

[21]

[22] [23]

[24]

[25]

[26]

[27]

[28]

J. Li et al. Psychosocial Questionnaire [in Chinese]. J Environ Occup Med 2008;25(6):572–6. Rugulies R, Christensen KB, Borritz M, Villadsen E, Bu¨ltmann U, Kristensen TS. The contribution of the psychosocial work environment to sickness absence in human service workers: results of a 3-year follow-up study. Work & Stress 2007;21(4):293–311. Twisk J. Longitudinal studies with two measurements: the definition and analysis of change. In: Twisk J, ed. Applied longitudinal data analysis for epidemiology: a practical guide. New York: Cambridge University Press; 2003. pp. 167–78. Zou G. A modified poisson regression approach to prospective studies with binary data. Am J Epidemiol 2004;159 (7):702–6. Schroeder MA. Diagnosing and dealing with multicollinearity. West J Nurs Res 1990;12(2):175–84. O’Brien-Pallas L, Duffield C, Hayes L. Do we really understand how to retain nurses? J Nurs Manag 2006;14 (4):262–70. Fochsen G, Sjo¨gren K, Josephson M, Lagerstro¨m M. Factors contributing to the decision to leave nursing care: a study among Swedish nursing personnel. J Nurs Manag 2005;13 (4):338–44. Lynn MR, Redman RW. Faces of the nursing shortage: influences on staff nurses’ intentions to leave their positions or nursing. J Nurs Adm 2005;35(5):264–70. Ingersoll GL, Olsan T, Drew-Cates J, DeVinney BC, Davies J. Nurses’ job satisfaction, organizational commitment, and career intent. J Nurs Adm 2002;32(5):250–63. Lu KY, Lin PL, Wu CM, Hsieh YL, Chang YY. The relationships among turnover intentions, professional commitment, and job satisfaction of hospital nurses. J Prof Nurs 2002;18(4):214–19. Tzeng HM. The influence of nurses’ working motivation and job satisfaction on intention to quit: an empirical investigation in Taiwan. Int J Nurs Stud 2002;39(8):867–78.

[29] Wu XJ, Zhang XJ, Gao FL. A study on the relationship between nurses’ intention to quit and work stress [in Chinese]. Chinese J Nurs 2000;35(4):197–9. [30] Yang ML, Wang RR, Hou SX. Investigation and analysis of relative factors of nurses’ quit willing [in Chinese]. Modern Nursing 2006;12(18):1667–9. [31] Aiken LH, Clarke SP, Sloane DM, Sochalski JA, Busse R, Clarke H, et al. Nurses’ reports on hospital care in five countries. Health Aff (Millwood) 2001;20(3):43–53. [32] Shi Y, Gao QY. Establishment and operation of surveillance system for nursing quality in Shanghai [in Chinese]. Chinese Nurs Man 2007;7(11):8–10. [33] Alfonso S, Fred H. Analyzing job mobility with job turnover intention: an international comparative study. J Econ Issue 2004;38(1):130–7. [34] Tu YK, Kellett M, Clerehugh V, Gilthorpe MS. Problems of correlations between explanatory variables in multiple regression analyses in the dental literature. Br Dent J 2005;199(7):457–61. [35] Coomber B, Barriball KL. Impact of job satisfaction components on intent to leave and turnover for hospitalbased nurses: a review of the research literature. Int J Nurs Stud 2007;44(2):297–314. [36] Lord FM. A paradox in the interpretation of group comparisons. Psychol Bull 1967;68(5):304–5. [37] Hayes LJ, O’Brien-Pallas L, Duffield C, Shamian J, Buchan J, Hughes F, et al. Nurse turnover: a literature review. Int J Nurs Stud 2006;43(2):237–63. [38] Li J, Jin TY. Work stress and health – current research activities and implications in China. WHO: WHO Global Occupational Health Network (GOHNET) Newsletter; 2007 Special. p 25–8. [39] Podsakoff PM, MacKenzie SB, Lee JY, Podsakoff NP. Common method biases in behavioral research: a critical review of the literature and recommended remedies. J Appl Psychol 2003;88(5):879–903.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 81–89

ORIGINAL ARTICLE

The effect of the work environment and performance-based self-esteem on cognitive stress symptoms among Danish knowledge workers

KAREN ALBERTSEN1,2, REINER RUGULIES1,2, ANNE H. GARDE1 & HERMANN BURR1 1

National Research Centre for the Working Environment, Copenhagen, Denmark, and 2Affiliated at the University of Copenhagen, Denmark

Abstract Aims: Interpersonal relations at work as well as individual factors seem to play prominent roles in the modern labour market, and arguably also for the change in stress symptoms. The aim was to examine whether exposures in the psychosocial work environment predicted symptoms of cognitive stress in a sample of Danish knowledge workers (i.e. employees working with sign, communication or change in knowledge) and whether performance-based self-esteem had a main effect, over and above the work environmental factors. Methods: 349 knowledge workers, selected from a national, representative cohort study, were followed up with two data collections, 12 months apart. We used data on psychosocial work environment factors and cognitive stress symptoms measured with the Copenhagen Psychosocial Questionnaire (COPSOQ), and a measurement of performance-based self-esteem. Effects on cognitive stress symptoms were analyzed with a GLM procedure with and without adjustment for baseline level. Results: Measures at baseline of quantitative demands, role conflicts, lack of role clarity, recognition, predictability, influence and social support from management were positively associated with cognitive stress symptoms 12 months later. After adjustment for baseline level of cognitive stress symptoms, follow-up level was only predicted by lack of predictability. Performance-based self-esteem was prospectively associated with cognitive stress symptoms and had an independent effect above the psychosocial work environment factors on the level of and changes in cognitive stress symptoms. Conclusions: The results suggest that both work environmental and individual characteristics should be taken into account in order to capture sources of stress in modern working life.

Key Words: COPSOQ, demands, flexibility, knowledge work, predictability, psychosocial, role ambiguity, work without boundaries

Background Symptoms of stress and burnout have become major problems in most Western countries, not only in the lowest socioeconomic groups, but at all societal levels. In order to contribute to preventive efforts, we try in this study to identify possible causal factors in the work environment as well as at the individual level. Over the past decades Denmark, as well as many other industrialized countries, has undergone major transitions of work life [1]. These transitions have involved a high occurrence of organizational changes, reorganizations, downsizings and changes in ownership [2] together with shifts in production systems away from manufacturing jobs and towards more service oriented jobs and jobs involving exchange of knowledge [3,4].

At the same time there has been a rapid development in the technological possibilities for exchange of information [5]. The development has been followed by a change from a more constitutive regulation with defined actions expected to be followed by the employees to a more regulatory regulation. A regulatory regulation implies that the rules direct action, and that the employees deliberately choose to follow the rules [6]. This development may have influenced working life in many different ways. Some work environmental stressors might still be important and the effects may even have been attenuated, while others may have lost importance. First of all, the concept of ‘‘boundaryless’’ work [6] stresses the potential endless demands in these kinds

Correspondence: Karen Albertsen, National Research Centre for the Working Environment, Lersø Parkalle´ 105, 2100 Copenhagen Ø, Denmark. Tel: þ45 3916 5466. Fax: þ45 3916 5201. E-mail: [email protected] (Accepted 23 September 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809352104

82

K. Albertsen et al.

of jobs, so there are reasons to believe that role overload – or quantitative demands, are still, as proposed by the job-demand-control model [7], an important stressor [8]. It is also likely that the regulatory regulation of work will sometimes lead to vaguely defined and contradictory tasks and expectations (role conflicts and role ambiguity (or low clarity)) [8], which may lead to stress as suggested by the role stress theory [9]. Secondly, high demands and increased individualization [10] may, in turn, lead to an increased call for more directly expressed recognition from the supervisory level. As part of the effort-reward-imbalance model, rewards have for a long time been identified as an important dimension in the psychosocial work environment [11], and may not be less important in this context. At the same time, high speed of organizational change [2] and increased global competition [12] may lead to diminished predictability at work [2,8]. Thirdly, in a context of unclear goals, control, autonomy or influence at work may not any longer, as presupposed by the job-demand-control model [7], only be beneficial for the employee, but may under certain circumstances even lead to increased stress [8,13]. In fact, recent studies have found no effect of influence at work on outcomes such as long-term sickness absence [14] and work–family conflict [15]. Fourthly, Allvin has described the development away from the formal network and toward more informal ones as a situation where ‘‘the individual is increasingly left to herself, while at the same time being more socially committed than ever’’ (p. 34 [6]). Social support at work was included in the iso-strain (isolated high strain) model around 1990 [16]. Since then, it may have diminished in importance, while interaction within informal work-related networks may have gained importance. A recent study among British civil servants found no association between social support at work and anxiety, and only a small preventive effect toward depression (odds ratio (OR) ¼ 0.96) [17], while a study conducted 10 years ago in a similar sample of British civil servants found strong support for a preventive effect of social support on psychiatric disorders [18]. Thus it may be hypothesized that social support may have a lesser effect on stress symptoms. Fifthly, the regulatory regulation requires high demands of individual responsibility, goal-orientation and self-directedness [8]. It has become more difficult to isolate the work environmental conditions from the personal abilities of the worker [10], and uncertainty about work and own competencies have become an important issue [8]. The concept of performance-based self-esteem has been introduced by Hallsten et al. as a factor that may influence how

susceptible employees are to burnout. Performancebased self-esteem is defined as self-esteem contingent on how well one performs in roles that are of vital significance for one’s self-realization [19]. In case of failure or overload at work, it is hypothesized that people with a high degree of performance-based selfesteem are more susceptible to burnout. Until now, only a few studies have explored the effects of performance-based self-esteem on psychological well-being [19–21], while more studies have explored contingent self-esteem [19]. It may be hypothesized that the tendency to be highly dependent on successful performance will play an important role for the change in stress symptoms among employees working under the conditions of a regulatory regulation of working life. Stress symptoms may take many different forms; physical (e.g. headache), behavioural (e.g. isolation), emotional (e.g. irritability), or cognitive (e.g. memory problems) [22]. We assumed that cognitive symptoms (problems with concentration, thinking clearly, taking decisions and remembering) would be particularly relevant among skilled employees with complex work tasks and with high demands to cognitive performance. We also assumed that these kind of symptoms would be a good indicator of the general stress level in the sample. Cognitive stress symptoms have in previous studies been associated with the psychosocial dimensions of psychological (quantitative) demands, lack of decision authority (influence), lack of meaning at work and high skill discretion [23]. Unfortunately, we were not able to include meaning and skills discretion in this study as they were not measured at baseline. Furthermore, cognitive stress symptoms have been associated with conflicts [23] and violence at work [24], and negatively with the sense of coherence [23,24], an integrated perception of one’s life as being comprehensible, manageable and meaningful.

Aim To explore the effect of quantitative demands, influence, role conflicts and role clarity, predictability, recognition and social support at work on the level and the change in cognitive stress symptoms in a sample of knowledge workers who were selected from a national, representative cohort study and followed over one year. Furthermore, we would examine whether performance-based self-esteem had a major effect on cognitive stress, independent of the effects of the psychosocial factors.

Work environment, performance-based self-esteem and stress symptoms Hypotheses Based on the considerations above, we hypothesized that the following work environmental variables would be prospectively associated with level and changes in cognitive stress symptoms: (1) Quantitative demands, role conflicts and lack of role clarity would be positively associated. (2) Predictability and recognition would be negatively associated. (3) Social support from supervisors and colleagues would be negatively, but only weakly associated. (4) Influence would be negatively, but only weakly associated. We further hypothesized that performance-based selfesteem: (5) Would be positively associated. (6) Would have an independent positive association above the psychosocial work environment factors.

Material and methods Participants Participants were recruited from the second National Danish Psychosocial Work Environment Study, encompassing 3,517 employees, representative for Danish wage earners between 20 and 59 years in the years 2004–2005. The response rate was 63%, and 52% were women. See the article from Pejtersen et al. in this issue for details [25]. From this initial survey, all respondents who were employed as knowledge workers were invited to participate (n ¼ 853). Knowledge work was broadly defined as working with signs, communication, or change in knowledge, and thereby making it possible to perform some part of the work via information technology equipment. Participants were employed, for example, as professionals, physicians, dentists, engineers, architects, staff within IT and media, teachers, researchers, managers and leaders, social workers, librarians, accountants, bank clerks, or salespersons. Data were collected in two rounds with 12 months (range 11–13 months) in between. In the first round in September/October 2006 (baseline), questionnaires were sent out to the 853 knowledge workers. In addition saliva was collected for cortisol determination, but will not be further dealt with in this paper. The response rate after the first round was 46% (n ¼ 396). At follow up, questionnaires were sent out to all 396 respondents from the baseline round. Three hundred and forty-nine questionnaires were returned, yielding a response rate at follow up of 86% (n ¼ 349) of the responders from baseline. Two reminders were sent out, after three weeks and

83

again after two weeks. The last reminder included a new copy of the questionnaire. Because of the low response rate in the first data collection round, we examined to what degree responders and non-responders differed from each other, by using data from the initial survey from 2004/2005. Responders and non-responders did not differ with regard to cognitive stress, self-rated health, general stress, quantitative demands, influence at work, or number of working hours. However, the responders were slightly older than the nonresponders (45.69 vs. 44.23 years) and there were more women among the responders than among the non-responders (56.2% vs. 44.3%). The study was approved by the local ethical committee (journal no.11510). Measures Stress symptoms: The following four items from the The COpenhagen PsychoSOcial Questionnaire (COPSOQ-II) [26] were used to assess cognitive stress symptoms: ‘‘How often have you had problems concentrating?’’, ‘‘How often have you found it difficult to think clearly?’’, ‘‘How often have you had difficulty in taking decisions?’’ and ‘‘How often have you had difficulty with remembering?’’. At the top of the page was the explanation ‘‘These questions are about how you have been during the last 4 weeks’’, and the response categories were: all the time; a large part of the time; part of the time; a small part of the time; not at all. For scale characteristics see Table I. Psychosocial work environment: The following long versions of scales from the COPSOQ-II were used to assess psychosocial working conditions: quantitative demands, influence, social support from supervisor, social support from colleagues, predictability, recognition, role clarity, role conflicts. For a description of the items and scales see Pejtersen et al. [25]. All scales were standardized from 0–100. A high score on a scale always indicates a high degree of the actual dimension, that is, a high score on quantitative demands indicates many quantitative demands; a high score on influence indicates a higher level of influence and so on. Performance-based self-esteem: This was measured by four items developed by Hallsten and colleagues [19]: (i) ‘‘I think that I sometimes try to prove my worth through my work’’ (ii) ‘‘My self-esteem is far too dependent on my work achievements’’ (iii) ‘‘At times, I have to be better than others to be good enough myself’’ and (iv) ‘‘Occasionally I feel obsessed to accomplish something of value through my work’’. Each item had four response

84

K. Albertsen et al.

Table I. Descriptive statistics.

Variable (no items)

n

Min

Max

Mean (SD)

Cronbach’s alpha

Inter-item correlations

Working hours/week Quantitative demands (4) Influence (4) Predictability (2) Recognition (3) Role conflicts (4) Role clarity (3) Social support from colleagues (3) Social support from supervisor (3) Performance based self-esteem (4) Cognitive stress t1 Cognitive stress t2

374 375 375 376 375 376 376 369 369 375 375 344

20 0 0 0 8 0 17 8 0 0 0 0

80 94 100 100 100 94 100 100 100 100 75 100

42 (8) 50 (18) 55 (18) 57 (19) 65 (17) 45 (16) 68 (17) 58 (16) 61 (22) 43(24) 22 (16) 22 (17)

0.80 0.71 0.72 0.80 0.71 0.81 0.60 0.80 0.78 0.82 0.85

0.37–0.69 0.27–0.52 0.56 0.47–0.71 0.32–0.46 0.52–0.68 0.23–0.46 0.49–0.64 0.42–0.58 0.43–0.68 0.48–0.68

Correlations between scale-scores at baseline and follow up 0.69 0.66 0.65 0.47 0.56 0.56 0.59 0.58 0.52 0.71 0.64

Table II. Pearson’s correlation coefficients.

1. Quantitative demands 2. Influence 3. Predictability 4. Recognition 5. Role clarity 6. Role conflicts 7. Social support from supervisors 8. Social support from colleagues 9. Performance based self-esteem 10. Cognitive stress

1.

2.

3.

4.

5.

6.

7.

8.

9.

0.09 0.23 0.29 0.23 0.42 0.26 0.11 0.38 0.44

0.35 0.41 0.32 0.03 0.18 0.07 0.08 0.21

0.56 0.55 0.24 0.42 0.19 0.16 0.25

0.51 0.32 0.57 0.27 0.23 0.31

0.23 0.32 0.26 0.25 0.33

0.31 0.08 0.29 0.25

0.39 0.12 0.15

0.03 0.12

0.40

Correlations in bold are statistically significant.

categories: does not fit at all, fits a little bit, fits quite well, fits perfectly. Cronbach’s alpha for the scale is 0.78. Beside this, gender and age groups (26–29, 30–39, 40–49, 50–59, 60–62 years) were included in analyses. Age was divided into groups because there was no linear relationship between age as continuous variable and outcome. Statistical analysis Analyses were conducted with the SPSS version 17.0 software. Correlations between variables were analyzed with Pearson correlation coefficients. For each of the psychosocial work environmental dimensions a model was tested by use of a multiple GLM procedure. The model included the work environment dimension measured at baseline and cognitive stress symptoms at follow-up and was adjusted for gender and age groups (step1). Secondly we adjusted each of the models for cognitive stress symptoms at baseline (step 2). The effect of performance-based self-esteem at baseline on cognitive stress-symptoms

at follow up was further studied in three steps: Model 1 with adjustment for gender and age-groups, Model 2 with further adjustment for the psychosocial workenvironmental factors and Model 3 with adjustment for cognitive stress-symptoms at baseline.

Results Correlations between all scale variables are shown in Table II. Many of the psychosocial variables were inter-correlated with moderately high correlations (0.30 < r < 0.60). Particularly, there were tendencies toward clustering between predictability, recognition, role clarity and social support from supervisors. Quantitative demands also correlated moderately high with role conflicts. All the psychosocial scales measured at baseline, except for social support from colleagues, were correlated with cognitive stress symptoms at follow up (Table III, step 1). All estimates were in the expected direction, but low to moderate of size. The scale measuring influence and the two scales

Work environment, performance-based self-esteem and stress symptoms

85

Table III. Associations between psychosocial work environment dimensions at baseline and cognitive stress symptoms at follow up. Step 1

Quantitative demands Influence Social support from management Social support from collegues Role conflicts Role clarity Recognition Predictability

Step 2

B (95% CI)

p-values

B (95% CI)

p-values

0.27(0.17;0.37) 0.14(0.25;0.37) 0.12(0.20;0.03) 0.07(0.19;0.05) 0.21(0.10;0.32) 0.28(0.39;0.17) 0.27(0.38;0.16) 0.23(0.33;0.13)

p < 0.000 p ¼ 0.011 p ¼ 0.007 p ¼ 0.257 p < 0.000 p < 0.000 p < 0.000 p < 0.000

0.01(0.08;0.10) 0.00 (0.09;0.09) 0.02(0.09;0.04) 0.01(0.08;0.10) 0.04(0.05;0.01) 0.08(0.17;0.02) 0.08(0.17;0.01) 0.10(0.18;0.01)

p ¼ 0.832 p ¼ 0.968 p ¼ 0.486 p ¼ 0.814 p ¼ 0.952 p ¼ 0.102 p ¼ 0.097 p ¼ 0.023

Individual GLM analyses of each of the psychosocial dimensions on cognitive stress symptoms. Model 1: adjusted for gender and age-groups Model 2: adjusted for gender, age-groups and cognitive stress at baseline.

Table IV. Associations between performance-based self-esteem at baseline and cognitive stress symptoms at follow up, with and without control for work environmental factors and cognitive stress symptoms at baseline. Model 1

Performance based self-esteem baseline R2 for the total model (N)

Model 2

Model 3

B (95% CI)

p-values

B (95% CI)

p-values

B (95% CI)

p-values

0.26 (0.18–0.33) 0.18 (309)

p < 0.000

0.19 (0.11–0.27) 0.26 (299)

p < 0.000

0.10 (0.03–0.17) 0.46 (298)

p ¼ 0.006

Model 1: Adjusted for age groups and gender. Model 2: Adjusted for age groups, gender, quantitative demands, influence, predictability, recognition, role conflicts, role clarity social support from supervisors and social support from colleagues. Model 3: Adjusted additionally for cognitive stress symptoms at baseline.

measuring social support had estimates below 0.20 and the rest of the scales had estimates between 0.20 and 0.30. After adjustment for cognitive stress symptoms at baseline, the only scale that was still associated with outcome was low predictability (Table III, step 2). Performance-based self-esteem at baseline was significantly associated with cognitive stress symptoms at follow up; before adjustment for workenvironmental variables (Table IV, Model 1), after adjustment for work-environment (Model 2) and after further adjustment for cognitive stress symptoms at baseline. The estimate before control for work environmental factors and baseline level was of moderate size (B ¼ 0.26). The full model explained 46% of the variation in cognitive stress symptoms, while the model without control for baseline value explained 26%. A model with all the work environmental variables and cognitive stress at baseline, (but without performance-based self-esteem) had an R2 of 0.45, while R2 was 0.20 in the model without baseline cognitive stress. Finally, R2 was 0.44 for a model that only controlled for gender, age and baseline cognitive stress. Thus, the largest part of the variation was explained by baseline cognitive stress symptoms.

Discussion We found cognitive stress symptoms at follow up associated with work environmental factors (quantitative demands, role conflicts, lack of role clarity, recognition, predictability, influence and social support from management) as well as performancebased self-esteem at baseline. In the following sections we will discuss implications in relation to the psychosocial work environmental factors and to the concept of performance-based self-esteem. Furthermore, we will take up some methodological and theoretical considerations.

Psychosocial work environmental factors as predictors of cognitive stress All the hypotheses 1–4 regarding the psychosocial work environmental factors were partly confirmed. As expected, the baseline measures of demands, role conflicts, role clarity, influence, predictability, recognition and social support from management were all associated with cognitive stress symptoms at follow up, only social support from colleagues was unrelated, while we had expected a small association. In contrary to our expectations – in the analyses

86

K. Albertsen et al.

adjusted for cognitive stress symptoms at baseline, only lack of predictability was significantly associated with cognitive stress symptoms at follow up. Hence, all these psychosocial variables seem to be of importance for the level of cognitive stress symptoms, while only the scale measuring predictability was associated with the change in symptoms. We performed, however, 16 independent tests of associations between the psychosocial work environment and outcome, which imply a high risk for random significant findings. The finding that lack of predictability was an antecedent of the development of cognitive stress symptoms may be a mere coincidence. There have been relatively few previous studies of predictability. Lack of predictability has been associated with burnout [27], absence [28] and low leadership quality [26]. The dimension of predictability reflects to what degree the employee gets sufficient information about decisions, future developments and changes in work situation. The importance of the dimension in relation to cognitive stress cannot be settled by this study, but may deserve attention in future research. Baseline influence was negatively, but relatively weakly, associated with cognitive stress symptoms at follow up, suggesting that the degree of influence still has a positive, although small, buffering effect against cognitive stress symptoms among knowledge workers. After adjustment for baseline level, no association was found. A recent Swedish study tested the hypothesis that the effect of job control (influence) on psychological well-being should vary with the level of demands and between groups confronted with different kinds of demands. None of the hypotheses could be confirmed [13]. The scale measuring social support from management was, as hypothesized, weakly and negatively associated with the level of cognitive stress symptoms at follow up, while the scale measuring social support from colleagues was unrelated to cognitive stress symptoms. After adjustment for baseline levels, both scales were unrelated. Thus, social support at work did not seem to have any strong preventive effect toward stress symptoms in this group of employees. An interesting question for future research would be whether the associations between influence and social support at work on the one hand and health and well-being on the other, have changed over the last decades, and whether the effects differ for different occupational groups. Performance-based self-esteem Hypotheses 5 and 6 were both confirmed. Performance-based self-esteem was positively and

prospectively associated with cognitive stress symptoms and had an independent effect over and above the psychosocial work environment factors on the level of and changes in cognitive stress symptoms. The size of the estimate was comparable with the size of the estimates for the scales measuring quantitative demands, role conflicts, role clarity, recognition and predictability. It seems likely that people who have a high need to perform well at work and whose selfesteem is highly dependent on performance and success may run an increased risk for overload and accordingly increase in cognitive stress symptoms. Hallsten et al. have described how a vicious circle may arise in which strain-distress and lowered selfesteem are followed and aggravated by further coping attempts [19]. The risk for overload may be higher under conditions of a regulatory regulation of work, with no externally set limits for the performance. Along these lines previous studies have found high performance-based self-esteem associated with increased risk for burnout and psychiatric morbidity [20,21].

Theoretical considerations Two mainly theoretical issues deserve further attention. First, the high inter-correlation we found in this study between the different psychosocial scales from the COPSOQ instrument. This is not unique to the present study or instrument, but is an inherent challenge when studying the psychosocial work environment. It does, however, constitute a serious problem both with regard to theory and statistical methods. In order to gain detailed knowledge about the psychosocial work environment it is preferable to draw a comprehensible picture that includes many different aspects. On the other hand, it is not very surprising that these different aspects are highly interrelated with each other, which makes the use of statistical multiple regression models problematic. All scales were, however, conceptually different and assumed to contribute independently to the outcome, so we decided to explore the relationship between each of the work environmental variables and cognitive stress symptoms. Accordingly, we chose to make separate analyses for the psychosocial work factors rather than include them all in the same models. The results from the present study suggested that all the included psychosocial dimensions were of relevance for cognitive stress symptoms. However, it seems necessary with developmental work in order to understand the relation between the dimensions of the psychosocial work environment, both of

Work environment, performance-based self-esteem and stress symptoms

87

theoretical and empirical nature (see also Burr et al. [29] in this issue). Second, all of the estimates for the psychosocial work environmental variables were of small or moderate size. The model including the psychosocial variables, without baseline level of cognitive stress symptoms and without performance-based selfesteem, only explained 20% of the total variance in cognitive stress symptoms. One interpretation could be that the psychosocial work environment only contributes 20% to cognitive stress symptoms, whereas the rest of the variance could be explained by, e.g., other social relationships with family or friends or by disease giving rise to the same symptoms. It is, however, relevant to ask, whether our measurement of the psychosocial work environment in this sample was adequate. Did we miss aspects of importance for the psychosocial work environment among knowledge workers? In a study by Hallgren and colleagues, scales measuring task completion ambiguity (uncertainty whether a job is finished or not) and task quality ambiguity (uncertainty about the quality of own work) explained variation in mental health [8]. Uncertainty about performance is, in the work by Hallgren et al., viewed more as a consequence of the organization and management of work than as a consequence of a personality disposition. Inclusions of such factors in our study might have increased the explanatory value of the work environmental factors and might have reduced the effect of performance-based self-esteem. Particularly in relation to a new and more regulatory regulation (see background section) of work it may be quite difficult to distinguish between the work environmental aspects surrounding the work and the capacities of the individual to meet the requirements in this environment. An individual tendency to be uncertain about own performance may, for example, be highly challenged under conditions where you have to set the goals and standards for performance partly by your self, and where feedback is limited. A theoretical and empirical clarification of the potential relationship between performance-based self-esteem and other, more established personality factors (e.g. overcommitment, neuroticism, and negative affectivity) as well as new dimensions in the psychosocial work environment (e.g. task completion and quality ambiguity) seems warranted.

change in cognitive stress level. Changes in a variable are difficult to demonstrate when the level at baseline is already high. Moreover, baseline levels might have been, at least partly, caused by the exposure variables of interest. When, for example, quantitative demands at baseline are associated with cognitive stress symptoms at follow up, then one can speculate that quantitative demands in the past had caused the onset of cognitive stress symptoms at baseline. However, the level of quantitative demands might not lead to further increases in the level of cognitive stress symptoms. On the other hand, adjusting for cognitive stress symptoms at baseline might be important for controlling reporting bias. Cognitive stress symptoms at baseline might influence both the report of work environment factors at baseline and the level of cognitive stress symptoms at follow up. The full model did in fact only explain 2% more of the variation than the model only adjusted for cognitive stress at baseline. For a more detailed discussion about the advantages and disadvantages of adjusting for baseline values see Rugulies et al. [30]. In order to provide the most information for the reader, we chose to present the models both with and without control for baseline levels.

Methodological considerations

Conclusion

There are both pros and cons for adjusting for baseline values of cognitive stress symptoms. When no adjustment is made, the outcome is future cognitive stress. When adjustment is made, the outcome is

In this prospective study in a sample of Danish knowledge workers, we found that baseline quantitative demands, role conflicts, lack of role clarity, lack of recognition, and lack of predictability were

Strengths and limitations It is a strength of the study that the sample was not restricted to specific worksites, but was selected from a national, representative sample of wage earners, making the results more possible to generalize to other Danish knowledge workers. The prospective design of the study makes it possible to draw conclusions about development over time. However, it is unknown how long the ideal followup period should be in relation to change in cognitive stress symptoms. The major limitation of the study is the relatively low response rate from the first survey. We were able to explore differences between responders and nonresponders, and did not find indications of selection processes, but we cannot totally rule out selection processes having taken place. A recent study found that role overload increased the likelihood of nonresponse, while role ambiguity, against the hypothesis, decreased the likelihood [31].

88

K. Albertsen et al.

positively associated with cognitive stress symptoms at follow up, when tested individually. Influence and social support from management were only to a small extent negatively associated with cognitive stress symptoms, and social support from colleagues was unrelated. Furthermore, we found that the tendency to be dependent on performance and success in the job in order to maintain a high self-esteem was positively associated with level and change in cognitive stress symptoms. Results suggest that work environmental and individual characteristics should be taken into account in order to capture the sources of stress in modern working life.

[9]

[10]

[11]

[12]

[13]

Acknowledgements We would like to thank Tage Søndergaard Kristensen and Jan Pejtersen for letting us re-contact respondents from the Second National Danish Psychosocial Work Environment Study.

[14]

[15]

Funding This work was supported by the Danish Working Environment Research Fund, [Grant no: 20050072489] as part of the project ‘‘Boundaryless work, stress, sleep and private life’’.

[16]

[17]

References [1] Na¨swall K, Hellgren J, Sverke M. The individual in the changing working life, 1st edn. Cambridge: Cambridge University Press; 2008. [2] Olsen O, Albertsen K, Nielsen ML, Poulsen KB, Gron SM, Brunnberg HL. Workplace restructurings in intervention studies – a challenge for design, analysis and interpretation. BMC Med Res Methodol 2008;8:39. [3] Burr H, Bjorner JB, Kristensen TS, Tu¨chsen F, Bach E. Trends in the Danish work environment in 1990–2000 and their associations with labor-force changes. Scand J Work Environ Health 2003;29(4):270–9. [4] Fura˚ker B. Post-industrial profiles: North American, Scandinavian and other Western labour markets. In: Fura˚ker B, Boje TP, editors. Post-industrial labour markets. Profiles of North America and Scandinavia. London: Routledge; 2003. pp. 241–61. [5] Eriksen TH. Øyeblikkets Tyranni. Rask og langsom tid i informasjonsalderen [Fast and slow time in the age of the information society]. Oslo: Aschehough & Co; 2007. [6] Allvin M. New rules of work: exploring the boundaryless job. In: Na¨swall K, Hellgren J, Sverke M, editors. The individual in the changing working life, 1st edn. Cambridge: Cambridge University Press; 2008. pp. 19–45. [7] Karasek R, Theorell T. Healthy work. Stress, productivity, and the reconstruction of working life, Basic Books; 1990. [8] Hellgren J, Sverke M, Na¨swall K. Changing work roles: new demands and challenges. In: Na¨sswall K, Hellgren J, Sverke M, editors. The individual in the changing working

[18]

[19]

[20]

[21]

[22]

[23]

[24]

[25]

life, 1st edn. New York: Cambridge University Press; 2008. pp. 46–66. Rizzo JR, House RJ, Lirtzman SI. Role conflict and ambiguity in complex organizations. Administrative Sciences Quarterly 1970;15:150–63. Allvin M, Aronsson G. The future of work environment reforms: does the concept of work environment apply within the new economy? Int J Health Serv 2003;33(1): 99–111. Siegrist J, Klein D, Voigt KH. Linking sociological with physiological data: the model of effort-reward imbalance at work. Acta Physiol Scand Suppl 1997;640:112–16. Major DA, Germano LM. The changing nature of work and its impact on the work-home interference. In: Jones F, Burke RJ, Westman M, editors. Work-life balance. A psychological perspective. New York: Psychology Press; 2006. pp. 13–38. Gro¨nlund A. Employee control in the era of flexibility – a stress buffer or a stress amplifier? European Societies 2007;9(3):409–28. Lund T, Labriola M, Christensen KB, Bultmann U, Villadsen E, Burr H. Psychosocial work environment exposures as risk factors for long-term sickness absence among Danish employees: results from DWECS/DREAM. J Occup Environ Med 2005;47(11):1141–7. Voydanoff P. The effect of work demands and resources on work-to-family conflict and facilitation. J Marriage Family 2004;66:398–412. Johnson JV, Stewart W, Hall EM, Fredlund P, Theorell T. Long-term psychosocial work environment and cardiovascular mortality among Swedish men [see comments]. Am J Public Health 1996;86(3):324–31. Griffin JM, Greiner BA, Stansfeld SA, Marmot M. The effect of self-reported and observed job conditions on depression and anxiety symptoms: a comparison of theoretical models. J Occup Health Psychol 2007;12(4): 334–49. Stansfeld SA, Fuhrer R, Shipley MJ, Marmot MG. Work characteristics predict psychiatric disorder: prospective results from the Whitehall II Study. Occup Environ Med 1999;56(5):302–7. Hallsten L, Josephson M, Torgen M. Performance-based self-esteem. A driving force in burnout processes and its assessment. Stockholm: National Institute for Working Life; 2005. Report No.: 2005:4. Dahlin M, Joneborg N, Runeson B. Performance-based selfesteem and burnout in a cross-sectional study of medical students. Med Teach 2007;29(1):43–8. Dahlin ME, Runeson B. Burnout and psychiatric morbidity among medical students entering clinical training: a three year prospective questionnaire and interview-based study. BMC Med Educ 2007;7:6. Setterlind S, Larsson G. The Stress Profile: a psychosocial approach to measuring stress. Stress Medicine 1995;11(2):85–92. Albertsen K, Nielsen ML, Borg V. The Danish psychosocial work environment and symptoms of stress: the main, mediating and moderating role of sense of coherence. Work & Stress 2001;15(3):241–53. Hogh A, Mikkelsen EG. Is sense of coherence a mediator or moderator of relationships between violence at work and stress reactions? Scand J Psychol 2005;46(5):429–37. Pejtersen JH, Kristensen TS, Borg V, Bjørner JB. The second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II). Scand J Public Health 2010;38(Suppl 3):8–24.

Work environment, performance-based self-esteem and stress symptoms [26] Kristensen TS, Hannerz H, Hogh A, Borg V. The Copenhagen Psychosocial Questionnaire – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31(6): 438–49. [27] Borritz M, Bu¨ltmann U, Rugulies R, Christensen KB, Villadsen E, Kristensen TS. Psychosocial work characteristics as predictors for burnout: findings from 3-year follow up of the PUMA Study. J Occup Environ Med 2005;47(10): 1015–25. [28] Nielsen ML, Rugulies R, Christensen KB, Smith-Hansen L, Bjørner JB, Kristensen TS. Impact of the psychosocial work environment on registered absence from work: a two-year longitudinal study using the IPAW cohort. Work & Stress 2004;18(4):323–35.

89

[29] Burr H, Albertsen K, Rugulies R, Hannerz H. Do dimensions from the Copenhagen Psychosocial Questionnaire predict vitality and mental health over and above the job-strain and effortreward-imbalance models? Scand J Public Health 2010; 38(Suppl 3):59–68. [30] Rugulies R, Christensen KB, Borritz M, Villadsen E, Bu¨ltmann U, Kristensen TS. The contribution of the psychosocial work environment to sickness absence in human service workers: results from a 3-year follow-up. Work & Stress 2007;21(4):293–311. [31] Barr CD, Spitzmuller C, Stuebing KK. Too stressed out to participate? Examining the relation between stressors and survey response behaviour. J Occup Health Psychol 2008;13(3):232–43.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 90–105

ORIGINAL ARTICLE

Evaluating construct validity of the second version of the Copenhagen Psychosocial Questionnaire through analysis of differential item functioning and differential item effect

JAKOB BUE BJORNER1,2 & JAN HYLD PEJTERSEN1 1

National Research Centre for the Working Environment, Copenhagen, Denmark, and 2Institute of Public Health University of Copenhagen, Copenhagen, Denmark

Abstract Aims: To evaluate the construct validity of the Copenhagen Psychosocial Questionnaire II (long version) by means of tests for differential item functioning (DIF) and differential item effect (DIE). Methods: We used a Danish general population postal survey (n ¼ 4,732 with 3,517 wage earners) with a one-year register based follow up for long-term sickness absence. DIF was evaluated against age, gender, education, social class, public/private sector employment, and job type using ordinal logistic regression. DIE was evaluated against job satisfaction and self-rated health (using ordinal logistic regression), against depressive symptoms, burnout, and stress (using multiple linear regression), and against long-term sick leave (using a proportional hazards model). We used a cross-validation approach to counter the risk of significant results due to multiple testing. Results: Out of 1,052 tests, we found 599 significant instances of DIF/DIE, 69 of which showed both practical and statistical significance across two independent samples. Most DIF occurred for job type (in 20 cases), while we found little DIF for age, gender, education, social class and sector. DIE seemed to pertain to particular items, which showed DIE in the same direction for several outcome variables. Discussion: The results allowed a preliminary identification of items that have a positive impact on construct validity and items that have negative impact on construct validity. These results can be used to develop better shortform measures and to improve the conceptual framework, items and scales of the COPSOQ II. Conclusions: We conclude that tests of DIF and DIE are useful for evaluating construct validity.

Key Words: Differentiel item functioning, differentiel item effect, psychosocial work environment, questionnaire, construct validity

Introduction The construct validity of a scale concerns whether the scale or items in the scale perform in a way concordant with the theory about the construct measured by the scale. To evaluate the construct validity of a scale, researchers should specify hypotheses about the relations between the construct measured by the scale and other variables. To allow empirical evaluation, at least some of these variables need to be observable. The system of specific hypotheses is referred to as a nomological network [1]. Using this framework, studies by Ruguelis et al. [2] and Pejtersen et al [3] generally supported the validity of COPSOQ II scales. Register-based long term sickness absence was

significantly related to high cognitive demands, high emotional demands, and high role conflict, although the effect of high cognitive demands was no longer statistically significant after adjustment for multiple [2] testing. Long-term sickness absence was also significantly related to higher scale values of Stress and Depression [3]. At the workplace level, self-reported sickness absence was significantly related to mean score values of Justice and Recognition. The job-related scales Work pace, Variation and Work–family conflicts were able to discriminate among job groups as hypothesized [3]. The current paper further investigates construct validity of the COPSOQ II, in particular whether each scale item is a measure of the construct it is hypothesized to measure.

Correspondence: Jakob Bue Bjorner, National Research Centre for the Working Environment, Lersø Parkalle´ 105, DK-2100 Copenhagen, Denmark. Tel: þ45 39 16 54 76. Fax: þ45 39 16 52 02. E-mail: [email protected] (Accepted 28 September 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809352533

Construct validity of COPSOQ

…Required to treat everyone equally…

Evaluation of the construct validity of scales concerning the psychosocial working environment has routinely evoked the ‘‘internal consistency assumption’’, i.e. the assumption that items belonging to the same scale have to have a positive correlation [4]. The implicit model is that items are ‘‘effect indicators’’ of the latent construct measured by the scale. For example, consider the COPSOQ II scale on depressive symptoms consisting of items about mental states such as: ‘‘sadness’’ (DS1),’’ ‘‘lack of selfconfidence’’ (DS2), ‘‘bad conscience’’ (DS3), ‘‘lack of interest’’ (DS4) (see appendix 2 in [3]). It seems plausible that high scores on these items are all effects of one common state of depression and thus that the internal consistency assumption can be applied. However, as pointed out by Bollen [4,5], not all questionnaire scales can be conceived as consisting of effect indicator items. Some items must be seen as causes of the latent construct rather than effects. Consider for example the COPSOQ II scale on demands for hiding emotions at work containing items like: ‘‘required to treat everyone equally, even if you do not feel like it’’ (HE1), ‘‘your work requires that you hide your feelings’’ (HE2), ‘‘required to be kind and open towards everyone – regardless of how they behave towards you’’ (HE3). According to the effects indicator model, jobs that require the employee to treat everyone equally should also require the employee to hide their feelings. However, as shown in Figure 1, these two items are not strongly associated on job level: in the COPSOQ II study (see [3] and below) the correlation between the two item scores across 476 jobs was only 0.19 (and 0.23 on individual level). For example, many technical jobs and crafts require the employees to treat everyone equally but do not require them to hide their feelings. Thus, the scale does not have

5

4

3

2

1 1

2

3

4

5

…Work requires that you hide your feelings?

Figure 1. Association between the mean item score per job group for two items on hiding emotions across 476 jobs [3]. Bubble size indicates number of people in job group.

91

construct validity under an effect indicator model but may be valid if regarded as causes of the same effect: restricting the employees’ options in coping with events at work. It is interesting that Cronbach and Meehl in their seminal paper on construct validity noted that: ‘‘Only if the underlying theory of the trait being measured calls for high item intercorrelations, do the correlations support construct validity.’’ ‘‘High internal consistency may lower validity’’ [1]. If the internal consistency assumption is not relevant for a scale, analysis of item intercorrelations, item-total correlations, or item-level factor analyses may be irrelevant as evidence of construct validity or lack hereof. While this has been understood for more than 50 years [1,6–8], specific strategies for evaluation of item level construct validity for causal indicator scales have been scarce. However, some suggestions within modern psychometrics offer a way forward. Kreiner [9] has argued that all multiitem scales – regardless of assumptions about their latent structure – need to satisfy requirements about lack of differential item functioning: for a given level of the scale score, there should be no association between the items and important external variables such as age, gender, job group, health, or sickness absence. This requirement can be motivated based on measurement efficiency. For example, if an analysis of the demands for hiding emotions scale across jobs finds that healthcare professionals have higher demands for hiding emotions than service workers, we would expect each item on the scale to show roughly the same trend across healthcare professionals and service workers. If it turns out that healthcare professionals to a greater degree than service workers endorse the item ‘‘hide your feelings’’, but service workers to a greater degree than healthcare professionals have to ‘‘treat everyone equally’’, we would conclude that a demand for the hiding emotions scale that includes these two items is not a good way to summarize the information, at least with regard to the comparison of healthcare professionals and service workers. In the language of psychometrics, the demands for the hiding emotions scale would exhibit differential item functioning (DIF) with regard to job type. Similarly, if the demands for the hiding emotions scale predict low job satisfaction, we would assume that each item in the scale is associated with low job satisfaction (and roughly to the same extent as the other items in the scale). If it turns out that one item, ‘‘hide your feelings’’, is associated with low job satisfaction, but the other items are not, we would conclude that this item should not be combined with the other items on demands for hiding emotions, at least not for analyses of job satisfaction.

92

J. B. Bjorner & J. H. Pejtersen

Since we regard job satisfaction as an effect of demands for hiding emotions (amongst other things), we would term this problem differential item effect (DIE) [10]. Demands for no DIF and no DIE are relevant both for effect indicator and causal indicator scales [10], but tests for DIF and DIE are particularly important for evaluating the construct validity of causal indicator scales, because traditional psychometric criteria based on internal consistency do not apply. Tests for DIF for job group in the COPSOQ I quantitative demands scale led to a separation of this scale into two scales concerning intensity of work (work pace) and extensive work (quantitative demands) [11]. Health-related outcome variables were included in the analyses, but DIE was not formally tested. Similar analyses have also been used in the evaluation of the Job Content Questionnaire [12]. In the development of COPSOQ II, scales were tested for DIF against gender, age and occupational status [3], but this work has not yet been thoroughly documented. The purpose of this paper is to evaluate the construct validity of the COPSOQ II by testing for: (1) DIF with regards to the background variables age, gender, education, social class, employment in the private or public sector, and job group, (2) DIE with regards to the outcome variables job satisfaction, symptoms of depression, burnout, stress, general health perception, and sickness absence. By testing for DIF and DIE against a broad range of external variables we build upon the idea of a broad nomological network [1] but extend this idea to evaluating item-level construct validity. In all analyses, the basic hypothesis to be tested is that the COPSOQ II scales do not show DIF or DIE. In practice, the hypotheses to be tested are that given the scale value, all items in the scale are independent of the background variables (DIF criterion) and of the outcome variables (DIE criterion – see appendix 1).

Methods Sample The sample included 8,000 adult respondents randomly selected from the Danish general population through the Centralized Civil Register. The sample and the data collection procedure have been described in detail by Pejtersen et al. [3]. In brief, data was collected by postal questionnaires, using two written reminders and a final follow up by phone. Respondents also had the opportunity to respond through the internet. The study took place in autumn and winter 2004/2005 and achieved a total of 4,732 valid responses (60.4%), of which 3,517 indicated

they were wage earners. In the statistical analyses of the scales we used all 4,732 respondents for the scales on health whereas the sample of 3,517 wage earners was used for analyzing the work environment scales. Variables We evaluated the long version of the COPSOQ II [3], with three exceptions: (1) we did not evaluate the scale for self-efficacy, since we wanted to focus our attention on work-place risk factors and individual health outcomes; (2) we did not evaluate the four scales concerning the social capital of the workplace (Trust regarding management, Mutual trust between employees, Justice and Social inclusiveness) since these scales are intended to describe general properties of the company rather than individual factors. Thus, validation of the social capital scales would require another approach; (3) we did not evaluate single item scales, since the concepts of differential item functioning and differential item effect applies to multi-item scales only. The Civil Register provided data on the respondents’ age and gender. Education, social class, private or public sector employment, and job type was measured by self-report. Social class was classified according to the European Classification of Social Class based on the three digits ISCO88 code [13] (for details, see [14]). Job groups were classified on the basis of self-reported information on occupation, industry, education and socioeconomic status using the 1986 Danish extended version of the International Standard Classification of Occupations (ISCO) [15] and further combined into 14 homogenous job groups, based on the three digit modified ISCO codes. The 14 main job groups comprised: academics, teaching professionals, technicians, health professionals and healthcare workers, social work professionals and pre-school teachers, managers, clerks, service workers, protective service workers, skilled industrial workers, unskilled industrial workers, drivers and mail carriers (for details, see [16]). The outcome variables included self-report single items on general health perception (also called selfrated health (SRH)) and overall job satisfaction, the COPSOQ II scales on depression, burnout, and stress, each consisting of four items (see [3]) and register data on long-term (more than two weeks) sickness absence. Sickness absence data was gathered from the DREAM register (see [2]). We counted time (in weeks) to first sickness absence with censoring after one year or in case of pensioning or death. The correlations between the three scales on depression, burnout, and stress were high (range 0.67–0.72).

Construct validity of COPSOQ The correlation between years of school education and social class was 0.48. All other correlations between background or outcome variables were below 0.40. Data analysis Several tests for DIF have been proposed in the literature (for recent overviews, see [17,18]). We used ordinal logistic regression DIF methods [19], which have gained general acceptance. A separate analysis was performed for each combination of an item and a background variable. In each analysis, the dependent variable was the item in question, while the independent predictors were the scale and the background variable in question. A verdict of DIF for an item in relation to a background variable required a statistically significant association of sufficient magnitude between the item and the background variable when controlling for the scale score. In line with previous studies [20], the criteria for sufficient magnitude for the association required that the background variable explained at least an additional 2% of the item variance (using the difference in Nagelkerke’s Pseudo R2 [21]). The variables social class and job group were treated as class variables, while age and length of education were treated as interval scaled variables. In all analyses, we tested for both uniform DIF (an impact of the background variable when controlling for the scale score) and non-uniform DIF (an interaction between the scale score and the background variable). However, to simplify presentation, we do not distinguish between uniform and non-uniform DIF in the results. Tests for DIE have not been discussed in the literature to a great extent, but similar to DIF tests, DIE can be evaluated by testing conditional independence between each item and the outcome variable in question, when controlling for the scale score (see Appendix 1). Following the logic of causal order, this independence should be tested in a regression model using the outcome variable as dependent variable, the item in question as independent variable and the scale score as covariate. For job satisfaction and SRH, which was assessed by rank scaled single items, we used ordinal logistic regression. For depression, burnout and stress, which were measured as multi-item scales, we used multiple linear regression. Sickness absence data was analyzed using a proportional hazards model with late entry depending on the actual time of answering the questionnaire. All DIE tests were carried out for the scales concerning workplace characteristics, but the COPSOQ II scales for health outcomes (depression, burnout, sleeping troubles and stress) were only tested for DIE against sickness absence. Furthermore, the job

93

satisfaction scales were not tested against the overall job satisfaction item, since this item is part of the scale. We tested both uniform and non-uniform DIE. A verdict of DIE required a statistically significant association of sufficient magnitude. Since no guidelines for sufficient magnitude have been suggested in the literature, we propose the following criteria: the association has to be at least half of the magnitude (absolute value) of the association between the scale and the outcome variable (when the scale and the item are scaled to the same range). See appendix 1 for discussion of this criterion. Due to the large number of items and external variables, the number of planned tests is huge (1,052 in total) and the risk of getting significant results purely due to chance is huge. The specification of thresholds for minimally important associations reduces but does not eliminate the risk of spurious results. For this reason, we adopted a cross-validation strategy. We split the sample into two random halves and carried out the tests for DIF and DIE in each half. We only accepted results on DIF and DIE for items showing significant associations that exceeded the thresholds for minimal importance in both subsamples. We compared this approach to more traditional ways of adjusting for multiple testing through the Bonferroni correction. After identifying the items with DIF or DIE, we performed a final evaluation of the magnitude of DIF or DIE using the total sample. In order to ease interpretation of DIE across outcome variables and scales, we reverse scored the outcome variables for job satisfaction and SRH so that high score meant a poor outcome for all external variables. Furthermore, we reverse coded some scales and items so that high score meant poor working environment for all scales and items.

Results We tested 28 scales, 588 possible cases of DIF and 464 possible cases of DIE for a total of 1,052 combinations of items and external variables. In DIF tests in the total sample, 354 tests were significant (Figure 2) and 268 remained significant after Bonferroni correction. Of the 354 significant tests, only 20 satisfied our requirements of DIF of a sufficient magnitude in the total sample; 13 of these instances were also statistically significant after Bonferroni adjustment. Applying our criteria of DIF for two random subsamples identified 23 cases of DIF, 21 of which would also have been identified through significance testing with Bonferroni correction. We found substantial agreement between our cross-validation approach and a total sample analysis

94

J. B. Bjorner & J. H. Pejtersen Item level analyses

All DIF tests (588 tests) 234

Significant in overall sample (354)

All DIE tests (464 tests) 219

Significant in overall sample (245)

DIF in overall sample1 (20)

DIE in overall sample1 (103)

182

90 DIF in crossvalidation2 (23)

Significant with Bonferroni adjustment (168)

144

DIE in crossvalidation2 (46)

2

1

1

3

14

7

Significant with Bonferroni adjustment (113)

52

35

7

0

22

39

0

Scale level analyses All scales evaluated for DIE (28 scales)

All scales evaluated for DIF (28 scales) 0

Significant in overall sample (28)

3

Significant in overall sample (25) DIE in overall sample1 (21)

DIF in overall sample1 (11)

2

3

DIE in crossvalidation2 (16)

DIF in crossvalidation2 (13) Significant with Bonferroni adjustment (24)

11

1DIF/DIE 2DIF/DIE

0

1

0

1

9

3

Significant with Bonferroni adjustment (21)

2

2

0

0

3

16

0

in overall sample means that the criteria for sufficient magnitude of DIF/DIE was satisfied in the overall sample in cross-validation means that the criteria for sufficient magnitude of DIF/DIE was satisfied in both subsamples

Figure 2. Instances of DIF/DIE by different criteria.

with Bonferroni correction and test of DIF magnitude (Cohens Kappa ¼ 0.69). The 23 cases of DIF are the ones we regard as true findings in the current paper. From the perspective of the scales, the crossvalidation approach identified DIF in 13 scales. A Bonferroni correction combined with a test of DIF magnitude in the total sample would have identified DIF in nine of these scales and would have identified DIF in one additional scale. Out of 464 DIE tests in the total sample, 245 tests were significant (Figure 2) and 113 remained significant after Bonferroni correction. Of the 245 significant tests, 103 satisfied our requirements of DIE of a sufficient magnitude in the total sample; 61 of these instances were also statistically significant after Bonferroni adjustment. Applying our criteria of DIE for two random subsamples identified 46 cases of DIE, 39 of which would also have been identified through significance testing with Bonferroni correction. We found substantial agreement between our

cross-validation approach and a total sample analysis with Bonferroni correction and test of DIE magnitude (Cohens Kappa ¼ 0.69). The 46 cases of DIE are the ones we regard as true findings in the current paper. From the perspective of the scales, the crossvalidation approach identified DIE in 16 scales. Bonferroni correction combined with a test of DIE magnitude in the total sample would also have identified DIE in these 16 scales as well as in three additional scales. Table I provides an overview of the results. DIF or DIE was found in 19 scales that mostly concerned demands at work, work organization and job content, and the work–individual interface. For half of the scales concerning interpersonal relations and leadership we found evidence of DIE for one or more items. For the scales on health and well-being, we only identified one instance of DIF (against job group in the burnout scale). No DIF was found for the external variables gender, education, and

Instances of DIF/DIE for variable

Health and well-being Burnout Stress Sleeping troubles Depressive symptoms Somatic stress Cognitive stress

Work-individual interface Job insecurity Job satisfaction Work-family conflict Family-work conflict

Interpersonal relations and leadership Predictability Recognition Role clarity Role conflicts Quality of leadership Social community at work Social support from supervisor Social support from colleagues

Work organization and job content Influence Possibilities for development Variation Meaning of work Commitment to the workplace

Demands at work Quantitative demands Work pace Cognitive demands Emotional demands Demands for hiding emotions

1

1

Age

0

Gender

0

Education

2

1 1

Social class

DIF for variable

0

Sector

Table I. Number of instances of DIF/DIE within each scale for each external variable.

20

1

3 1 1

1 3

14

– – – – – –

1 – 1

1 2 1 2

2 2 1

3 2 1

2 1

1

Job satisfaction

1

Job group

9

– – – – – –

2

2

1

1 1

2

Depression

6

– – – – – –

1

1

1

1

2

Burnout

6

– – – – – –

1

1

2

2

Stress

DIE for variable

7

– – – – – –

1

1

1

2

1 1

SRH

4

1

1

2

Sickness absence

69

1 0 0 0 0 0

4 1 9 1

1 2 0 0 0 0 1 3

3 3 2 5 8

5 1 5 12 2

Instances of DIF/DIE for scale

Construct validity of COPSOQ 95

96

J. B. Bjorner & J. H. Pejtersen

public/private sector. For age, we found one case of DIF in the scale regarding commitment to the workplace, since older employees significantly less often thought about seeking work elsewhere. This DIF leads the scale to overestimate the commitment to the workplace of older employees. Two cases of DIF were found for social class. The scale for work–family conflict had DIF with regards to social class for the item WF1 ‘‘Do you often feel a conflict between your work and your private life, making you want to be in both places at the same time?’’. For a given level of work–family conflict, persons in upper social classes tended to endorse this item more. The scale for family–work conflict also had DIF for social class. This DIF involved the item FW1 ‘‘Do you feel that your private life takes so much of your energy that it has a negative effect on your work?’’ For a given level for family–work conflict, persons in lower social classes tended to endorse this item more often. Most instances of DIF were found in relation to job group (20 cases of DIF in all, Table I). More detailed information on DIF in relation to job group is given in Appendix 2. No job group DIF was found for scales concerning interpersonal relations and leadership and only one instance of DIF was found for scales on health and well-being. This instance concerned the item ‘‘How often have you been emotionally exhausted?’’ from the burnout scale. For a given level of burnout, emotional exhaustion was more frequent among academics, teaching professionals and social work professionals, while it was less frequent among industrial workers, drivers, mail carriers and agricultural workers (Appendix 2). We found 46 cases of DIE involving 25 items in 16 scales (Table I). The items could be classified into positive DIE, i.e. items that have a stronger association with the outcome variables than indicated by the overall scale (Table II), and negative DIE, i.e. items that have less strong associations with the outcome variable than indicated by the overall scale or had association in the opposite direction than the one hypothesized (Table III). In total, we identified 16 items with positive DIE in 15 scales. If an item had positive DIE with regard to one outcome variable it tended also to have positive DIE for other outcome variables, although not always satisfying our formal requirements for magnitude of DIE. For example, for people with the same level of quantitative demands, respondents who indicated that they did not have enough time for their work tasks (item QD4), had higher risk of low job satisfaction, poor SRH, and long-term sick leave (Table II). Similarly, if an item had negative DIE with regard to one outcome variable it tended to have negative DIE for other

outcome variables. For example, the emotional demands item ED2 ‘‘Do you have to relate to other people’s personal problems as part of your work?’’ had less strong associations with both low job satisfaction, depressive symptoms, burnout, stress and poor SRH than would be predicted from the total scale (for job satisfaction, the association was actually in the opposite direction than the one predicted by the scale model, since people having to relate to other people’s personal problems as part of their work rated their job satisfaction as higher). We identified nine items with negative DIE in nine scales (Table III). The two-item scale on variation was the only scale containing items with positive and negative DIE at the same time. The positively formulated item VA1, ‘‘Is your work varied?’’ had stronger association with job satisfaction than predicted by the scale, but less strong association with sickness absence than predicted by the scale. The negatively formulated item VA2 showed the opposite pattern.

Discussion The initial development of the COPSOQ [3,22] used traditional psychometric techniques such as factor analyses and reliability assessment through Cronbach’s alpha. However, we have become concerned that such techniques may not be appropriate for many COPSOQ II scales for which the items are combined based on a hypothesized common effect rather than a hypothesized common cause. This concern prompted a strategy for construct validation that did not rely on the ‘‘internal consistency assumption’’ [4,5]. Tests for differential item functioning have applied in testing cross-language adaptations [23] and construct validity [24] of different occupational health questionnaires. In particular, DIF analyses have been used to evaluate scales from COPSOQ I [11] and in the development of COPSOQ II [3] as well as other scales [12,24]. The current paper extends upon this work by including tests for more variables – in particular a test for DIE with regard to important outcome variables. While this approach is new, we believe that it is based soundly in the basic logic of construct validity [1,6,7]. We chose to apply DIF and DIE testing to all the COPSOQ II scale to use a coherent evaluation framework because these tests apply to both effect indicator and causal indicator scales. This does not mean that we regard all COPSOQ II scales as causal indicator scales. Detailed hypotheses regarding causal and effect indicator relations for COPSOQ II scales are found in [25]. Tests for DIF and DIE are important because they have direct implications for the interpretation of

HE2 Does your work require that you hide your feelings? IN1 Do you have a large degree of influence concerning your work? (R) PD4 Does your work give you the opportunity to develop your skills? (R) VA1 Is your work varied? (R)

Emotional demands

Demands for hiding emotions

Influence (R)

CW4 How often do you consider looking for work elsewhere? PR2 Do you receive all the information you need in order to do your work well? (R) RE3 Are you treated fairly at your workplace? (R) SS1 How often is your nearest superior willing to listen to your problems at work? (R) SC2 How often are your colleagues willing to listen to your problems at work? (R) JI1 Are you worried about becoming unemployed? WF2 Do you feel that your work drains so much of your energy that it has a negative effect on your private life?

Commitment to the workplace (R)

Predictability (R)

Social support from supervisor (R)

Work–family conflict

Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) Est (SD) 0.38 (0.06) 0.47 (0.06) 0.69 (0.06) 0.45 (0.05) 0.37 (0.05) 0.72 (0.08) 0.48 (0.06) 0.48 (0.06) 1.14 (0.08) 0.50 (0.06) 0.30 (0.09) 0.39 (0.08) 0.46 (0.08) 0.17 (0.06) 0.24 (0.06) 0.78 (0.07)

1.00 (0.41) 2.97 (0.43) 5.46 (0.47) 4.05 (0.36) 1.64 (0.39) 3.11 (0.57) 0.15 (0.47) 0.15 (0.47) 4.69 (0.57) 4.07 (0.36) 4.55 (0.65) 1.94 (0.56) 2.75 (0.57) 1.55 (0.44) 1.87 (0.45) 6.43 (0.47)

Depressive symptomsd 2.51 (0.45) 2.71 (0.47) 5.19 (0.51) 2.62 (0.40) 0.81 (0.43) 3.88 (0.63) 0.88 (0.51) 0.88 (0.51) 4.30 (0.64) 3.89 (0.40) 2.77 (0.72) 3.90 (0.62) 3.20 (0.64) 1.77 (0.48) 1.32 (0.50) 8.24 (0.50)

Burnoutd 2.45 (0.42) 3.03 (0.45) 6.02 (0.49) 4.18 (0.39) 0.58 (0.42) 3.88 (0.61) 0.14 (0.51) 0.14 (0.51) 4.80 (0.62) 4.38 (0.39) 4.20 (0.70) 3.37 (0.60) 3.76 (0.61) 1.67 (0.47) 1.33 (0.49) 6.14 (0.48)

Stressd

0.24 (0.05) 0.00 (0.05) 0.35 (0.06) 0.16 (0.04) 0.13 (0.05) 0.23 (0.07) 0.03 (0.05) 0.03 (0.05) 0.34 (0.07) 0.06 (0.04) 0.21 (0.08) 0.35 (0.07) 0.17 (0.07) 0.09 (0.05) 0.09 (0.05) 0.64 (0.06)

Poor SRHc

0.26 (0.06) 0.13 (0.06) 0.05 (0.07) 0.00 (0.05) 0.00 (0.05) 0.28 (0.08) 0.27 (0.07) 0.27 (0.07) 0.11 (0.08) 0.03 (0.05) 0.16 (0.09) 0.16 (0.08) 0.03 (0.07) 0.04 (0.06) 0.03 (0.06) 0.32 (0.07)

Risk of long-term sick leavee

(R) indicates that the scale has been reverse coded so that high score is poor work environment. b(R) indicates that the item has been reverse coded so that high score is poor work environment. Logistic regression b parameter. dLinear regression b parameter. eProportional hazards model b parameter.

c

a

Job insecurity

Social support from colleagues (R)

Recognition (R)

MW3 Do you feel motivated and involved in your work? (R)

Meaning of work (R)

Variation (R) VA2 Do you have to do the same thing over and over again?

CD1 Do you have to keep your eyes on lots of things while you work? ED1 Does your work put you in emotionally disturbing situations?

Cognitive demands

Possibilities for development (R)

QD4 Do you have enough time for your work tasks? (R)

Itemb

Quantitative demands

Scalea

Low job satisfactionc

Table II. DIE parameter estimates for items that show positive differential item effect. Significant results that satisfy the magnitude criteria are in bold with grey background.

Construct validity of COPSOQ 97

98

J. B. Bjorner & J. H. Pejtersen

Table III. DIE parameter estimates for items that show negative differential item effect. Significant results that satisfy the magnitude criteria are in bold with grey background.

Scalea Quantitative demands Work pace Cognitive demands

Emotional demands

Possibilities for development (R) Meaning of work (R)

Commitment to the workplace (R) Social support from colleagues (R) Work–family conflict

Stressd

Poor SRHc

Risk of long-term sick leavee

2.71 (0.58) 3.20 (0.77) 1.73 (0.45)

1.87 (0.55) 2.23 (0.74) 1.84 (0.44)

0.09 (0.06) 0.31 (0.08) 0.05 (0.05)

0.26 (0.07) 0.27 (0.10) 0.06 (0.06)

4.31 (0.36)

4.06 (0.39)

4.37 (0.38)

0.27 (0.04)

0.05 (0.05)

0.63 (0.06)

3.31 (0.44)

3.60 (0.49)

4.30 (0.47)

0.23 (0.05)

0.18 (0.06)

Est (SD)

0.84 (0.09)

2.64 (0.62)

4.02 (0.70)

3.60 (0.68)

0.26 (0.07)

0.30 (0.09)

Est (SD)

0.13 (0.06)

4.04 (0.41)

3.82 (0.46)

5.83 (0.44)

0.15 (0.05)

0.03 (0.06)

Est (SD)

0.31 (0.06)

3.01 (0.43)

1.87 (0.48)

2.62 (0.46)

0.13 (0.05)

0.03 (0.06)

Est (SD)

0.45 (0.06)

4.14 (0.40)

3.27 (0.43)

2.32 (0.41)

0.15 (0.05)

0.04 (0.06)

Low job satisfactionc

Depressive symptomsd

Burnoutd

Est (SD) Est (SD) Est (SD)

0.44 (0.07) 0.18 (0.09) 0.22 (0.05)

1.36 (0.53) 2.12 (0.71) 1.44 (0.41)

Est (SD)

0.46 (0.05)

Est (SD)

Itemb QD3 Do you get behind with your work? WP3 Is it necessary to keep working at a high pace? CD3 Does your work demand that you are good at coming up with new ideas? ED2 Do you have to relate to other people’s personal problems as part of your work? PD1 Does your work require you to take the initiative? (R) MW2 Do you feel that the work you do is important? (R) CW2 Do you feel that your place of work is of great importance to you? (R) CW1 How often do you get help and support from your colleagues? (R) WF4 Do your friends or family tell you that you work too much?

a (R) indicates that the scale has been reverse coded so that high score is poor work environment. b(R) indicates that the item has been reverse coded so that high score is poor work environment. cLogistic regression b parameter. dLinear regression b parameter. eProportional hazards model b parameter.

analytic results. For example, the scale on variation showed strong DIE with regard to sick leave. To evaluate the practical implications of this result, we analyzed two separate models where each single item was used as a predictor for long-term sick leave (and controlling for gender, age and social status). In these analyses, the item VA2 (‘‘Do you have to do the same thing over and over again?’’) strongly predicted sick leave. Compared to the best score (‘‘never/hardly ever’’), people with the worst score (‘‘always’’) had a relative risk of 1.39 for sick leave (95% confidence interval (95% CI) ¼ 1.04–1.85). On the other hand, the item VA1 (‘‘Is your work varied?’’) had a nonsignificant association with sick leave in the opposite direction. Compared to the best score (‘‘always’’) people with the worst score (‘‘never/hardly ever’’) had a relative risk of 0.97 for sick leave (95% CI ¼ 0.72– 1.31). Using the ‘‘variation’’ scale obscured this picture: compared to the best scale score, people with the worst scale score had a non-significantly

increased risk of sick leave (RR ¼ 1.30, 95% CI ¼ 0.91–1.86). While the possible explanations for this pattern (e.g. that VA2 might pertain more to manual labour) are speculative so far, the difference in prediction is important, because the scale will underestimate the impact of repetitive work. One may ask (as did one reviewer) whether tests for DIE are always relevant for evaluating construct validity. Maybe scales that do not predict health or other outcomes or that show high DIE in their association with outcomes might nevertheless be important for describing the psychosocial work environment? It is true that if a scale is used only as an outcome variable in analyses, testing for DIE is not relevant (but testing for DIF is still relevant). This might be the case for the scales regarding job satisfaction, health and well-being. For scales regarding other workplace factors, we would question the necessity of measuring a workplace factor that is not related to any outcome. However, it is a relevant

Construct validity of COPSOQ question, whether our selection of outcome variables was the most relevant for all scales. The selection partly reflected what we consider important outcomes in occupational health studies, and partly reflected what was possible given the available data. Some relevant outcome variables, in particular related to work productivity, could not be examined in the current study. Since testing for DIF/DIE involves a large number of statistical tests, we face a considerable risk of achieving significant results purely by chance. To counter this problem we took a fairly conservative approach of defining criteria for magnitude of DIF and requiring these criteria to be fulfilled also in a cross-validation sample. This strengthens our confidence that the results on DIF/DIE items would generalize to other datasets. However, due to the conservative nature, less faith can be put in the ‘‘negative’’ results for items that did not show DIF/ DIE. We found few or no instances of DIF for the variables age, gender and social class. This is unsurprising, because DIF for these variables were checked before finalizing the scales for COPSOQ II. However, in line with researchers studying other questionnaires [24] we found multiple instances of DIF for job type. Many of these instances of DIF have immediate face validity, for example that emotional fatigue (for a given level of burnout) is most common for people working with people [26,27] and that at a given level of possibilities for development, teachers and managers in particular feel that their work requires them to take the initiative [28]. While many of the results make intuitive sense, it is not certain that these mechanisms would be considered to explain differences between job groups on the scale level. Therefore, we believe that such DIF analyses are important to enable correct interpretation of results on the differences in the psychosocial working environment between job groups. Some of our exogenous variables were highly correlated, in particular the three scales concerning depression, burnout and stress. While the three concepts are conceptually different, the high correlation causes each additional scale to provide less independent information. Due to this fact, the strong concordance in the DIE results for these three scales (Tables II and III) should be interpreted cautiously. On the other hand, since we did employ an elaborate strategy for dealing with significance due to multiple testing, we opted to test for DIF and DIE in relation to many different variables to get as detailed a validation as possible. We note that the concordance in Tables II and III also pertains to job satisfaction,

99

SRH and sick leave, variables that are not strongly correlated with the other outcomes. Another potential problem in the DIE analyses is the risk of multicollinearity problems because of high correlation between each item and the scale it is part of. Such multicollinearity will tend to reduce the power of the statistical tests and make the parameter estimates less robust [29]. This may be the reason that far more instances of DIE were identified if a magnitude criterion was applied in the overall sample than if we required the magnitude criterion to crossvalidate in two independent samples (Figure 2). This was one of reasons to prefer the cross-validation approach over other approaches to correct for multiple testing. The fact that our DIE results replicated over several different outcome variables makes us confident that the cross-validation approach provides robust results. Nevertheless, the potential problem of multicollinearity should be examined in simulation studies. Another limitation of the paper is that all outcome variables except for sickness absence are measured cross-sectionally by self-report. For this reason, the true causal direction might be the opposite of the hypothesized one. This is less of a problem for evaluations of DIF and DIE than for analysis of causal hypotheses. Basically, DIF and DIE involve hypotheses about conditional independence between items and exogenous variables given the scale score. Thus, if the causal direction is opposite, we would just have to test for DIF instead of DIE. In fact, we can test the robustness of our approach by performing such tests. For example, a test of DIE for the quantitative demands scale with regards to job satisfaction found DIE concerning the items QD4 ‘‘enough time’’. If instead we test for DIF (assuming that job satisfaction affects evaluation of quantitative demands), we find significant DIF for the same item. However, because of the self-report nature of some of our exogenous variables we cannot rule out that some DIE results may be caused by similarities in wording between the item in question and the outcome variables. On the other hand, it is a strength of the paper that we include a broad array of outcome variables and do include one register-based variable measured longitudinally. The task of interpreting the results regarding DIE is somewhat simplified by the observation that positive DIE in respect of one outcome variable increases the probability of finding positive DIE in respect of other outcome variables. This allows a summarization of the DIE results (Table IV) that can be used as a background for preliminary interpretation of the

100

J. B. Bjorner & J. H. Pejtersen

possible reasons for DIE. We have identified four such possible reasons that can be regarded as working hypotheses for future studies: (1) Core items. For many domains, the item most central to the domain seems to work best in terms of having the strongest association with the outcomes and thus positive DIE. For example, the item QD4,‘‘enough time’’, in the quantitative demands scale seems to capture the essence of quantitative demands, while the item QD3, ‘‘get behind’’, is not assessing quantitative demands in itself, but a consequence of demands. Also, the item IN1, ‘‘influence concerning your work’’, is a direct measure of influence at work, while the other items, focusing on subdimensions of influence, seem to be less efficient. Similarly, item PD4, ‘‘opportunity to develop your skills’’, seems central in the possibilities for development scale, item JI1, ‘‘worried about becoming unemployed’’, seems central in the job insecurity scale, and item HE2, ‘‘work requires that you hide your feelings’’, seems central in the demands for hiding emotions scale. Finally, the effectiveness of the items SC2 and SS1, ‘‘willing to listen to problems’’, in the two social support scales suggests that this aspect of social support is the most important one in the workplace. (2) Ambiguous items. Some items may show negative DIE because they are ambiguous in relation to the construct measured. For example, while most items in the commitment to the workplace scale imply a favourable evaluation of one’s place of work, the item CW2, ‘‘place of work is of great importance to you?’’, may also be endorsed by a very frustrated employee. In the two scales for social support, the items on ‘‘help and support from colleagues’’ and ‘‘help and support from nearest supervisor’’ could be interpreted as practical help and standard collaboration rather than social support. (3) Items with outcome ‘‘flavour’’. Some scales combine items on particular work situations with items on reactions to these situations. For example, in the emotional demands scale, an item such as ED2, ‘‘have to relate to other people’s personal problems as part of your work?’’, is a neutral description of a work situation, while item ED1, ‘‘Does your work put you in emotionally disturbing situations?’’ includes a description of a emotional reaction. Thus, it is not surprising that this latter item has stronger associations with selfevaluations of job satisfaction, mental symptoms and health. This may be a situation where the item with the least strong association with the external variables (ED2) may still be the most valid one as a measure of the construct in question. In analyses of DIF for job type (Appendix 2), ED2 also shows a pattern of associations that makes sense from the

perspective of job content, while the DIF pattern for ED1 does not. Similarly, the item MW3, ‘‘feel motivated’’, in the meaning of work scale and the item WF2, ‘‘drains your energy’’, in the work– family conflicts scale seem to describe intraindividual states that have outcome ‘‘flavour’’. (4) Heterogenous domains. For some domains, the most reasonable interpretation of the results is that the domains are heterogeneous and may need to be reconceptualized. For example, the item CD1, ‘‘have to keep your eyes on lots of things’’, fits into the overall construct of demands as a health risk, but the item CD3, ‘‘work demand that you are good at coming up with new ideas’’, seem to imply more positive constructs such as influence, variation and possibilities for development. Also, the item RE3 ‘‘treated fairly at your workplace’’ may have more to do with the construct of justice than with the construct of recognition.

All of the interpretations above are preliminary at best and can be challenged. We state them to illustrate that construct validity can be discussed on item level even in the absence of an internal consistency assumption. Further work on the COPSOQ can take several different routes. First, more efficient short forms can be developed based on the items that performed best in the current analyses. However, such short forms should be evaluated based on new data. Second, abandoning attempts to construct internally consistent scales provides more freedom to construct scales from items that show the same pattern of predictions of outcome variables. Again, such scales should be tested with new data. Finally, qualitative studies using e.g. cognitive interviews [30] could be used to further analyze the items with DIF and DIE. Such analyses could lead to item improvements and reconceptualization of domains, leading to improvements in the next version of the COPSOQ. Until such further validation studies are available, we advise researchers to take our current results into consideration when planning studies or interpreting results. In some cases (e.g. the emotional demands scale) association with self-assessed outcomes such as depression, burnout or stress may be due to single items with an emotional ‘‘flavour’’. In these cases, the robustness of results may be evaluated by supplementary single item analyses. In other cases, (e.g. cognitive demands) it might be best to plan analyses based on reduced scales that have more homogenous content. Such evaluations should preferably be made while planning the statistical analyses, to avoid capitalizing on change.

Construct validity of COPSOQ

101

Table IV. Scales with positive DIE items (bold) and negative DIE items (grey background). Scale

Item

Text

Quantitative demands

QD1 QD2 QD3 QD4 WP1 WP2 WP3 CD1 CD2 CD3 CD4 ED1 ED2 ED3 ED4 HE1 HE2 HE3

Is your workload unevenly distributed so it piles up? How often do you not have time to complete all your work tasks? Do you get behind with your work? Do you have enough time for your work tasks? Do you have to work very fast? Do you work at a high pace throughout the day? Is it necessary to keep working at a high pace? Do you have to keep your eyes on lots of things while you work? Does your work require that you remember a lot of things? Does your work demand that you are good at coming up with new ideas? Does your work require you to make difficult decisions? Does your work put you in emotionally disturbing situations? . . . have to relate to other people’s personal problems as part of your work? Is your work emotionally demanding? Do you get emotionally involved in your work? Are you required to treat everyone equally, even if you do not feel like it? Does your work require that you hide your feelings? . . . required to be kind and open towards everyone – regardless of how they behave? Do you have a large degree of influence concerning your work? Do you have a say in choosing who you work with? Can you influence the amount of work assigned to you? Do you have any influence on what you do at work? Does your work require you to take the initiative? Do you have the possibility of learning new things through your work? Can you use your skills or expertise in your work? Does your work give you the opportunity to develop your skills? Is your work varied? Do you have to do the same thing over and over again? Is your work meaningful? Do you feel that the work you do is important? Do you feel motivated and involved in your work? Do you enjoy telling others about your place of work? Do you feel that your place of work is of great importance to you? Would you recommend a good friend to apply for a position at your workplace? How often do you consider looking for work elsewhere? At your place of work, are you informed well in advance concerning for example important decisions, changes, or plans for the future? Do you receive all the information you need in order to do your work well? Is your work recognised and appreciated by the management? Does the management at your workplace respect you? Are you treated fairly at your workplace? How often do you get help and support from your colleagues? . . . are your colleagues willing to listen to your problems at work? . . . do your colleagues talk with you about how well you carry out your work? . . . is your nearest superior willing to listen to your problems at work? . . . do you get help and support from your nearest superior? . . . does your nearest superior talk with you about how well you carry out your work? Are you worried about becoming unemployed? Are you worried about new technology making you redundant? . . . worried about it being difficult for you to find another job if . . . unemployed? Are you worried about being transferred to another job against your will? Do you often feel a conflict between your work and your private life, making you want to be in both places at the same time? Do you feel that your work drains so much of your energy that it has a negative effect on your private life? Do you feel that your work takes so much of your time that it has a negative effect . . .? Do your friends or family tell you that you work too much?

Work pace

Cognitive demands

Emotional demands

Demands for hiding emotions

Influence

Possibilities for development

Variation Meaning of work

Commitment to the workplace

Predictability

Recognition

Social support from colleagues

Social support from supervisors*

Job insecurity

Work–family conflict

IN1 IN2 IN3 IN4 PD1 PD2 PD3 PD4 VA1 VA2 MW1 MW2 MW3 CW1 CW2 CW3 CW4 PR1 PR2 RE1 RE2 RE3 SC1 SC2 SC3 SS1 SS2 SS3 JI1 JI2 JI3 JI4 WF1 WF2 WF3 WF4

102

J. B. Bjorner & J. H. Pejtersen

Acknowledgements The authors would like to thank Svend Kreiner for discussions regarding the DIE concept, and Reiner Rugulies and two anonymous reviewers for valuable comments and suggestions to improve the manuscript.

[18]

References

[20]

[1] Cronbach LJ, Meehl PE. Construct validity in psychological testing. Psychol Bull 1955;52:281–302. [2] Rugulies R, Aust B, Pejtersen JH. Do psychosocial work environment factors measured with scales from the COPSOQ predict register-based sickness absence of three weeks or more in Denmark? Scand J Public Health 2010; 38(Suppl 3):42–50. [3] Pejtersen JH, Kristensen T, Borg V, Bjorner JB. The second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II). Scand J Public Health 2010;38(Suppl 3): 8–24. [4] Bollen KA. Multiple indicators: internal consistency or no necessary relationship? Qual Quant 1984;18:377–85. [5] Bollen KA, Lennox R. Conventional wisdom on measurement: a structural equation perspective. Psychol Bull 1991; 110(2):305–14. [6] Cronbach LJ. Essentials of psychological testing, 5th edn. New York: Harper & Row; 1990. [7] Messick S. Validity. In: Linn RL, editor. Educational measurement 3rd, edn. New York: Macmillan; 1989. pp. 13–103. [8] Carver CS. How should multifaceted personality constructs be tested? Issues illustrated by self-monitoring, attributional style, and hardiness. J Pers Soc Psychol 1989;56(4):577–85. [9] Kreiner S. Validation of index scales for analysis of survey data – the Symptom Index. In: Dean K, editor. Population Health research – linking theory and methods. Beverly Hills: Sage; 1993. pp. 116–44. [10] Kreiner S. Validity and objectivity: reflections on the role and nature of Rasch models. Nordic Psychol 2007;59(3):268–98. [11] Kristensen TS, Bjorner JB, Christensen KB, Borg V. The distinction between work pace and working hours in the measurement of quantitative demands at work. Work & Stress 2004;18(4):305–22. [12] Choi B, Kawakami N, Chang S, Koh S, Bjorner J, Punnett L, et al. A cross-national study on the multidimensional characteristics of the five-item psychological demands scale of the Job Content Questionnaire. Int J Behav Med 2008;15(2):120–32. [13] Rose D, Pevalin DJ, Elias P, Martin J. Towards a European socio-economic classification: final report to Eurostat of the Expert Group. London and Colchester: ONS and ISER, University of Essex; 2001. [14] Moncada S, Pejtersen JH, Navarro A, Llorens C, Burr H, Hasle P, et al. Psychosocial work environment and its association with socioeconomic status. A comparison of Spain and Denmark 2010;38(Suppl 3):42–50. [15] The Directorate of Labor. The Danish job classification [in Danish]. Copenhagen: The Directorate of Labor; 1986. [16] Ortega A, Hogh A, Pejtersen JH, Olsen O. Prevalence of workplace bullying and risk groups: a representative population study. Int Arch Occup Environ Health 2009;82(3):417–26. [17] Groenvold M, Petersen MA. The role and use of differential item functioning analysis in clinical trials. In: Fayers PM, Hay RD, editors. Assessing quality of life in clinical

[19]

[21] [22]

[23]

[24]

[25]

[26]

[27]

[28] [29]

[30] [31]

trials: methods and practice, 2nd edn. Oxford: Oxford University Press; 2005. pp. 195–208. Teresi JA, Fleishman JA. Differential item functioning and health assessment. Qual Life Res 2007;16(Suppl 1):33–42. Zumbo BD. A handbook on the theory and methods of differential item functioning (DIF): logistic regression modeling as a unitary framework for binary and likert-type (ordinal) item scores. Ottawa, ON: Directorate of Human Resources Research and Evaluation, Department of National Defense; 1999. Bjorner JB, Kosinski M, Ware Jr, JE. Calibration of an item pool for assessing the burden of headaches: an application of item response theory to the headache impact test (HIT). Qual Life Res 2003;12(8):913–33. Nagelkerke NJD. A note on a general definition of the coefficient of determination. Biometrika 1991;78:691–2. Kristensen TS, Borg V, Hannerz H. Socioeconomic status and psychosocial work environment: results from a Danish national study. Scand J Public Health 2002;(Suppl 59):41–8. Choi B, Bjorner JB, Ostergren PO, Clays E, Houtman I, Punett L, et al. Cross-language differential item functioning of the Job Content Questionnaire among European countries: the JACE study. Int J Behavioral Med, 16(2):136–147, 2009. Orhede E, Kreiner S. Item bias in indices measuring psychosocial work environment and health. Scand J Work Environ Health 2000;26(3):263–72. Thorsen SV, Bjorner JB. Reliability of the Copenhagen Psychosocial Questionnaire. Scand J Public Health 2010; 38(Suppl 3):25–32. Brenninkmeijer V, VanYperen N. How to conduct research on burnout: advantages and disadvantages of a unidimensional approach in burnout research. Occup Environ Med 2003;60(Suppl 1):i16–i20. Cordes CL, Dougherty TW, Blum M. Patterns of burnout among managers and professionals: a comparison of models. J Organ Beh 1997;18(6):685–701. Fay D, Frese M. The concept of personal initiative: an overview of validity studies. Human Performance 2001;14(1):97–124. Goldberger AS. Multicollinearity. A course in econometrics. Cambridge, MA: Harvard University Press; 2009. pp. 245–53. Jobe JB. Cognitive psychology and self-reports: models and methods. Qual Life Res 2003;12(3):219–27. Maldonado G, Greenland S. Simulation study of confounderselection strategies. Am J Epidemiol 1993;138(11):923–36.

Appendix 1. Notes on differential item effect (DIE) This appendix discusses the criteria for magnitude of DIE and shows the equivalence between two formulations of the requirement of no DIE: (a) All items should be associated with the outcome variable to roughly the same extent. (b) Given the simple sum scale score, the outcome variable should be independent of the item score on each item.

Consider an outcome Y that is predicted by a set of items x1xn: Y ¼  0 þ  1 x1 þ  2 x2 þ    þ  n xn

ð1Þ

Construct validity of COPSOQ In a linear regression model, Y represents the expected score on the outcome variable. In a logistic regression model, Y is the log odds. In a survival analysis, Y represents the log of the hazards ratio. Formulation (a) of the no DIE criteria can be specified as 1 ¼ 2 ¼    ¼ n ¼  m Y ¼ 0 þ x1 þ x2 þ    þ xn

ð2Þ

¼ 0 þ ðx1 þ x2 þ    þ xn Þ ¼ 0 þ S where S is the simple sum of the items. DIE for x1 implies that 0 1 ¼ 1   6¼ 0 m   Y ¼ 0 þ 01 þ  x1 þ x2 þ    þ xn

ð3Þ

¼0 þ S þ 01 x1 Equation 3, Y ¼ 0 þ S þ 01 x1 , is the most convenient way to test DIE for a single item. However, formulation (b) of conditional independence (i.e. 01 ¼ 0) can be derived from formulation (a) and vice versa. The two criteria are really two different formulations of the same requirement. Since we use formulation (b) when testing for DIE, a reasonable criterion for the magnitude of DIE must be specified using the parameter 01 in Equation 3. However, formulation (a) provides a more intuitive starting point for deriving a criterion for DIE magnitude. Let us first consider a scale that is the sum of two items, x1 and x2, where x2 influences an outcome Y, while x1 does not. In a test of DIE for item x1 (using Equation 3) the expected value of the DIE parameter 01 is . If the two items have approximately the same variance, ignoring the DIE for this scale will result in an estimated effect of the scale on Yof approximately 0.5 , since an increase of one point in S in 50% of cases will come from an increase in x1, which will have no effect on Y and in 50% of cases from an increase in x2, which will lead to an increase in Y of . If instead x1 has a strong influence on Y (double the magnitude of x2), the expected value of the DIE parameter 01 will be þ. Ignoring a DIE of this magnitude will lead to an estimated total effect of the scale of 1.5. If we have a scale formed as the sum of three items, x1x3, of which x1 does not influence Y, but x2 and x3 do, the expected value of the DIE parameter 01 will again be . However, the effect of ignoring the DIE will be a 33% underestimation of . Thus, the effect of a single DIE item will be diluted in long scales.

103

This paper takes the perspective that the magnitude of DIE should be evaluated based on its effect on the total scale score. In other words, an item with a DIE of a certain magnitude is a bigger problem for a short scale than for a long scale. To achieve an evaluation of DIE impact that is independent of scale length, it is convenient to rescale S to the same range as the items by taking the average: S ¼

S ,  ¼ n n

In the current paper, the criterion for important magnitude of 01 was chosen as at least half the magnitude of  . This means that ignoring a DIE of this magnitude would lead to an underestimation of  of at least 50% (in case of negative DIE) or an overestimation of  of at least 50% (in case of positive DIE). As seen in the first example, such a DIE would be found in a two-item scale where one item affected the outcome, while the other item did not or had the opposite effect (when controlling for the other item). We consider this a large effect. In evaluation of confounding, a change in a parameter of 10% as been proposed as criterion for noteworthy confounding [31]. However, due to the unresolved issues regarding multicollinearity, we opted for a conservative criterion. For a practical example, in the DIE analyses of the Workplace commitment item CW4 ‘‘consider looking for work elsewhere’’ against depression, the regression coefficient  was 2.7 for the regression of depression on the workplace commitment scale (when the scale had a range of 1 to 5 similar to the item). This means that a statistically significant 01 parameter below 1.35 or above 1.35 was considered of sufficient magnitude to indicate DIE. The analysis found a 01 of 4.07 – clearly of sufficient magnitude. In an analysis similar to formulation (a) where each item was used as an independent predictor of depression (controlled for the other items), the following regression coefficients were found:  for CW1 ¼ 1.34,  for CW2 ¼ 1.24,  for CW3 ¼ 1.65,  for CW4 ¼ 4.75. In an analysis where each item was entered as a single predictor (without control for the other items), the following regression coefficients were found:  for CW1 ¼ 3.93,  for CW2 ¼ 2.69,  for CW3 ¼ 4.11,  for CW4 ¼ 5.45. Thus, although all analyses point to CW4 as being a stronger predictor of depressive symptoms than the other items, this example also demonstrates that the parameters in a regression that controls for the other items can differ significantly from the parameters in an analysis where each item is regarded independently.

0.27 (0.11)

0.26 (0.12)

Est (SD)

0.27 (0.12)

0.26 (0.12)

0.33 (0.12)

0.47 (0.12)

0.27 (0.12)

0.24 (0.12)

0.20 (0.12)

0.15 (0.12)

Academics

Est (SD)

Est (SD)

Influence IN4 Do you have any influence on what you do at work?

IN3 Can you influence the amount of work assigned to you? Possibilities for development PD1 Does your work require you to take the initiative?

Est (SD)

Est (SD)

Demands for hiding emotions HE2 Does your work require that you hide your feelings?

ED2 Do you have to relate to other people’s personal problems as part of your work?

Est (SD)

Est (SD)

CD2 Does your work require that you remember a lot of things?

Emotional demands ED1 Does your work put you in emotionally disturbing situations?

Est (SD)

Est (SD)

Cognitive demands CD1 Do you have to keep your eyes on lots of things while you work?

CD3 Does your work demand that you are good at coming up with new ideas?

Est (SD)

Quantitative demands QD1 Is your workload unevenly distributed so it piles up?

Scale Item

0.77 (0.14)

0.13 (0.13)

0.34 (0.14)

0.69 (0.13)

0.41 (0.14)

0.50 (0.15)

0.31 (0.14)

1.16 (0.14)

0.17 (0.14)

0.44 (0.14)

Teaching professionals

0.02 (0.18)

0.03 (0.17)

0.13 (0.18)

0.13 (0.17)

0.41 (0.18)

0.78 (0.19)

0.20 (0.18)

0.44 (0.18)

0.57 (0.18)

0.38 (0.18)

Technicians

0.13 (0.11)

0.15 (0.10)

0.10 (0.11)

0.41 (0.11)

0.62 (0.12)

0.65 (0.12)

0.39 (0.11)

1.12 (0.12)

0.28 (0.11)

0.51 (0.12)

0.95 (0.13)

0.90 (0.12)

0.18 (0.11)

0.81 (0.12)

0.85 (0.12)

0.57 (0.12)

0.74 (0.12)

Social work professionals and pre-school teachers

0.47 (0.11)

0.08 (0.11)

0.44 (0.11)

Health professionals and healthcare workers

0.85 (0.18)

0.91 (0.16)

0.01 (0.17)

0.60 (0.16)

0.60 (0.16)

0.10 (0.17)

0.62 (0.18)

0.20 (0.18)

0.51 (0.20)

0.60 (0.17)

Managers

0.80 (0.10)

0.11 (0.09)

0.03 (0.10)

0.12 (0.09)

0.13 (0.10)

0.40 (0.10)

0.77 (0.10)

1.21 (0.10)

0.62 (0.10)

0.28 (0.10)

Clerks

0.44 (0.13)

0.07 (0.12)

0.36 (0.12)

0.22 (0.12)

0.40 (0.12)

0.35 (0.13)

0.16 (0.13)

0.03 (0.12)

0.06 (0.12)

0.47 (0.12)

Shop and market sales workers

0.03 (0.14)

0.24 (0.13)

0.35 (0.14)

0.48 (0.14)

0.14 (0.14)

0.18 (0.15)

0.08 (0.14)

0.17 (0.14)

0.04 (0.14)

0.15 (0.14)

Service workers

0.54 (0.25)

0.80 (0.23)

0.14 (0.23)

0.02 (0.23)

0.59 (0.24)

0.12 (0.25)

0.32 (0.24)

0.01 (0.24)

0.96 (0.24)

0.32 (0.24)

Firefighters, police, prison guards, armed forces

0.13 (0.11)

0.16 (0.10)

0.67 (0.11)

0.43 (0.11)

0.11 (0.11)

0.44 (0.12)

0.12 (0.11)

0.37 (0.11)

0.42 (0.11)

0.11 (0.11)

Skilled industrial workers

0.43 (0.13)

0.45 (0.13)

0.45 (0.13)

0.32 (0.13)

0.66 (0.15)

0.45 (0.15)

0.69 (0.13)

0.58 (0.13)

0.31 (0.13)

0.09 (0.13)

Unskilled industrial workers

(Continued)

0.36 (0.14)

0.19 (0.14)

0.22 (0.14)

0.91 (0.14)

0.70 (0.14)

0.34 (0.15)

0.90 (0.14)

0.92 (0.14)

0.16 (0.14)

0.15 (0.12)

Drivers, mail carriers and agricultural workers

Appendix 2. DIF by job group. Estimates are logistic regression b parameters, with the average effect across job groups as reference value. Significant estimates for the particular job group are in bold with grey background.

104 J. B. Bjorner & J. H. Pejtersen

Est (SD)

Est (SD)

0.25 (0.12)

0.75 (0.13)

0.40 (0.15)

0.23 (0.12)

0.11 (0.13)

0.45 (0.14)

0.39 (0.12)

0.10 (0.12)

0.36 (0.14)

0.34 (0.15)

0.43 (0.17)

0.01 (0.15)

0.34 (0.16)

0.64 (0.19)

0.56 (0.14)

0.12 (0.14)

0.35 (0.13)

0.53 (0.17)

Teaching professionals

0.26 (0.18)

0.14 (0.19)

0.04 (0.24)

0.31 (0.18)

0.25 (0.19)

0.14 (0.19)

0.15 (0.18)

0.14 (0.18)

0.10 (0.18)

0.19 (0.21)

Technicians

0.16 (0.11)

0.27 (0.12)

0.49 (0.14)

0.66 (0.13)

0.89 (0.13)

0.16 (0.14)

0.11 (0.11)

0.43 (0.11)

0.03 (0.11)

0.57 (0.13)

Health professionals and healthcare workers

0.36 (0.12)

0.19 (0.13)

0.03 (0.16)

0.23 (0.13)

0.03 (0.13)

0.52 (0.14)

0.31 (0.12)

0.33 (0.12)

0.45 (0.12)

0.51 (0.15)

Social work professionals and pre-school teachers

(R) Indicates that the item has been reverse coded to fit the general direction of the scale.

Burnout BO3 How often have you been emotionally exhausted?

Work-family conflict WF1 Do you often feel a conflict between your work and your private life, making you want to be in both places at the same time?

Est (SD)

Est (SD)

JI3 Are you worried about it being difficult for you to find another job if you became unemployed?

Job satisfaction JS2 How pleased are you with the physical working conditions?

Est (SD)

JI4 Are you worried about being transferred to another job against your will?

Est (SD)

Est (SD)

CW2 Do you feel that your place of work is of great importance to you?

Job insecurity JI2 Are you worried about new technology making you redundant?

Est (SD)

0.55 (0.12)

Commitment to the workplace CW4 How often do you Est consider looking for (SD) work elsewhere? (R)

CW1 Do you enjoy telling others about your place of work?

0.63 (0.14)

Academics

Est (SD)

Meaning of work MW1 Is your work meaningful?

Scale Item

Appendix 2. Continued.

0.01 (0.18)

0.46 (0.17)

0.09 (0.23)

0.06 (0.18)

0.22 (0.20)

0.34 (0.22)

0.53 (0.17)

0.80 (0.17)

0.70 (0.17)

0.08 (0.21)

Managers

0.10 (0.10)

0.21 (0.11)

0.15 (0.13)

0.17 (0.10)

0.03 (0.10)

0.34 (0.11)

0.17 (0.10)

0.29 (0.10)

0.13 (0.10)

0.03 (0.12)

Clerks

0.16 (0.13)

0.13 (0.13)

0.26 (0.16)

0.23 (0.13)

0.24 (0.14)

0.33 (0.14)

0.12 (0.12)

0.43 (0.13)

0.31 (0.12)

0.53 (0.15)

Shop and market sales workers

0.12 (0.14)

0.46 (0.15)

0.31 (0.19)

0.17 (0.15)

0.46 (0.15)

0.24 (0.15)

0.05 (0.14)

0.31 (0.14)

0.19 (0.14)

0.55 (0.16)

Service workers

0.06 (0.25)

1.00 (0.26)

0.68 (0.29)

0.04 (0.26)

1.09 (0.26)

0.37 (0.30)

0.31 (0.24)

0.67 (0.24)

0.43 (0.24)

0.11 (0.28)

Firefighters, police, prison guards, armed forces

0.65 (0.11)

0.37 (0.12)

0.06 (0.14)

0.32 (0.11)

0.07 (0.12)

0.28 (0.12)

0.52 (0.11)

0.06 (0.11)

0.37 (0.11)

0.23 (0.13)

Skilled industrial workers

0.51 (0.14)

0.53 (0.14)

0.09 (0.17)

0.18 (0.14)

0.39 (0.14)

0.58 (0.14)

0.65 (0.13)

0.16 (0.13)

0.51 (0.13)

0.50 (0.15)

Unskilled industrial workers

0.79 (0.14)

0.87 (0.15)

0.23 (0.18)

0.27 (0.14)

0.33 (0.15)

0.36 (0.15)

0.62 (0.14)

0.14 (0.14)

0.67 (0.14)

0.35 (0.16)

Drivers, mail carriers and agricultural workers

Construct validity of COPSOQ 105

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 106–119

ORIGINAL ARTICLE

When workplace interventions lead to negative effects: Learning from failures

BIRGIT AUST1, REINER RUGULIES1,2,3, ANNETT FINKEN1 & CHRIS JENSEN4 1

National Research Centre for the Working Environment, Copenhagen, Denmark, 2Institute of Public Health, University of Copenhagen, Denmark, 3Department of Psychology, University of Copenhagen, Denmark, 4 ˚ rhus, Denmark MarselisborgCentret, Danish Centre for Rehabilitation, Research and Development, A

Abstract Aims: To investigate if workplace interventions resulted in changes in the psychosocial work environment. Process evaluation was conducted to study the implementation process and to use this knowledge to understand the results. Methods: Seven intervention units (n ¼ 128) and seven non-randomized reference units (n ¼ 103) of a large hospital in Denmark participated in an intervention project with the goal of improving the psychosocial working conditions. The intervention consisted of discussion days for all staff, employee working groups, leader coaching, and activities to improve communication and cooperation. Measures of the psychosocial work environment were conducted before the start of the intervention and again after 16 months using 13 scales from the Copenhagen Psychosocial Questionnaire, version I (COPSOQ-I). Results: In the intervention units there was a statistically significant worsening in six out of 13 work environment scales. The decrease was most pronounced for three scales that measure aspects of interpersonal relations and leadership. In addition, all three scales that measure aspects of work organization and job content decreased. In comparison, the reference group showed statistically significant changes in only two scales. Process evaluation revealed that a large part of the implementation failed and that different implicit theories were at play. Conclusions: Without the insights gained from process data the negative effects of this intervention could not be understood. Sometimes – as it seems happened in this study – more harm can be done by disappointing expectations than by not conducting an intervention.

Key Words: Interpersonal relations, intervention studies, occupational health, organizational innovation, stress, psychological, workplace

Background In addition to having developed the Copenhagen Psychosocial Questionnaire (COPSOQ), which is honoured in this special issue of the journal, Tage S. Kristensen has stressed the importance of critical reflections about occupational intervention studies [1–3]. This article draws on both of these contributions to the field of occupational health research. Tage S. Kristensen and others have pointed out that the effectiveness of workplace interventions cannot be assessed by looking only at final outcomes such as, for example, employees’ health [1,4]. If an intervention fails, a sole focus on outcomes does not

reveal if the failure occurred because of unsuccessful implementation (also called programme failure) or because the underlying assumptions about the intervention were wrong (theory failure). Without this distinction, it will remain unclear if implementation needs to be improved or if the underlying assumptions of the interventions have to be questioned, or both. The lack of sufficient process and context information in intervention studies has often been criticized [5,6]. Only if this information is available can researchers and practitioners better understand which interventions may have which effects, in what circumstances, and by what mechanism [7,8] – and therefore make learning from failure possible [9].

Correspondence: Birgit Aust, National Research Centre for the Working Environment, Lersø Parkalle´ 105, DK-2100 Copenhagen, Denmark. Tel: þ45 39 16 54 64. Fax: þ45 39 16 52 01. E-mail: [email protected] (Accepted 14 October 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809354362

When workplace interventions lead to negative effects Aims The aim of this paper is to investigate if workplace interventions with the goal of improving psychosocial working conditions resulted in changes in the psychosocial work environment, measured with the COPSOQ. A process evaluation was conducted to study the implementation process and to use this knowledge to understand the study results.

Methods Study design and sample This is a controlled workplace intervention study, with measurements at baseline and at 16 months of follow up. In the autumn of 2002, a large hospital in Denmark decided to conduct an intervention project with the goal of improving psychosocial working conditions and to decrease sickness absence. The hospital project group, based in the hospital’s personnel department, initiated and led the project. Seven units were asked by the hospital project group to act as intervention units. When researchers from the National Research Centre for the Working Environment (NRCWE) were invited to evaluate the project, they requested the inclusion of reference units. Therefore, seven reference units, which matched the specialty of the intervention units as far as possible, were chosen to participate in the study. Because reference groups were recruited for the study after the intervention units had been chosen, randomization was not possible. The study started with a baseline questionnaire on employees’ working conditions and health, conducted in all 14 units. Next, workplace interventions to improve the psychosocial work environment were carried out by three consultants from a public consulting company in the seven intervention units. About 16 months after baseline measurement, the participants received a follow-up questionnaire. Baseline data of the individuals were linked with their follow-up data via the questionnaire identification number. Baseline and follow-up questionnaires were distributed, collected and analyzed by the researchers from NRCWE. The researchers acted as an external evaluation group, that is, they were not involved in conducting the intervention. Likewise, the consultants were not involved in analyzing the data of the effect evaluation. They were, however, responsible for parts of the process documentation and contributed their own evaluation of the project. Employees at the 14 units were eligible for the study if they were on regular duty at the time of the baseline survey. Physicians were excluded because

107

they were usually assigned to more than one unit. A total of 450 employees fulfilled the eligibility criteria. Of the 450 eligible employees, 399 participated in the baseline survey (response rate: 89%). Of these, 97 had left the unit at follow up, reducing the sample to 302 employees, of which 231 (76%) responded to the follow-up questionnaire: 128 in the intervention and 103 in the reference group. When we compared the 231 study participants with the 97 employees who had left the unit before follow up, we found that those who had left were younger (36 vs. 41 years, p < 0.001) and had worked for less years at the unit (4 vs. 8 years, p < 0.001). There was no difference with regard to gender, type of job, mental health score, and vitality score (see ‘‘Measurement of covariates’’ for a description of how these variables were measured). When we compared the 231 study participants with the 71 employees who had not responded to the follow-up questionnaire, we found that the non-responders had lower mental health (72 vs. 79, p < 0.001) and vitality scores (55 vs. 65, p < 0.001). There was no difference with regard to gender, age, years at unit, or type of job. Measurement of the psychosocial work environment To measure the psychosocial work environment, we used 13 scales from the Copenhagen Psychosocial Questionnaire, version I (COPSOQ-I), which covered the domains ‘‘demands at work’’, ‘‘work organization and job content’’, and ‘‘interpersonal relations and leadership’’ [10]. The 13 scales are listed in Table IV. All scales are scored from 0 to 100. Scoring follows the label of the scale, i.e. a higher score on the emotional demand scale means more emotional demands, a higher score on the role clarity scale means more role clarity and so on. Therefore, depending on the type of the scale, a high score can be either undesirable (e.g., emotional demands) or desirable (e.g., role clarity). We have reported earlier that the internal consistency was satisfactory for most of the scales in the present sample (Cronbach’s alphas between 0.73 to 0.87). Only for the scales on demands for hiding emotions (0.47) and possibilities for development (0.65) were Cronbach’s alphas below 0.70 [11]. See Kristensen et al. [10] for a more detailed description of the use of the COPSOQ-I in general and Aust et al. [11] for a more detailed description of the use of the COPSOQ in the present sample. Measurement of covariates As covariates we included, gender, age, years of employment in the unit, occupational group, mental

108

B. Aust et al.

health, and vitality. Mental health and vitality were measured with the two corresponding scales from the Danish version of the Short-form 36 item (SF-36) questionnaire [12]. These scales were included to control for differences in psychological health between intervention and reference group at baseline that might have influenced the effects of the interventions. Statistical analysis of changes between baseline and follow up We first analyzed changes in the mean score for each of the 13 psychosocial work environment scales, separately for intervention and reference group. Since interventions were made within units, we may expect a clustering effect within units. To take this effect into account, we analyzed data using a mixed model for repeated measures and included unit as a random effect [13]. Next, we analyzed if the changes in the intervention and reference group differed statistically significantly from each other by analyzing the combined data, testing a time*intervention effect and including unit as a random effect [13] (for analysis of these type of studies see Dahl-Jørgensen and Saksvik [14]). Analyses were adjusted for gender, age, years at unit, job group, and baseline values of mental health and vitality (as fixed effects). In addition, we calculated for each scale, how many participants had experienced a substantial change (either worsening or improvement). ‘‘Substantial’’ was defined as a change of 10 points or more. We used logistic regression analyses to calculate if the proportion of participants that had either experienced an improvement or a worsening differed between intervention and reference group, after adjustment for the covariates. These analyses used generalized estimation equation (GEE) to handle the potential clustering effect within units [15,16]. Content of the intervention The hospital conducted several activities to bring the intervention project to the attention of the employees, for example with T-shirts showing the project’s slogan and information on the hospital’s homepage. The intervention started by providing detailed written reports about results from the baseline questionnaire to all intervention and reference units. Thereafter, the reference units did not get any further input from the project team until they received a final report about the project. In the intervention units the baseline results were used as a starting point for a discussion about their psychosocial work environment. The consultants met with each unit leader to

discuss the results of the survey and to find out which issues the unit leaders thought were most important. All employees were invited to a kick-off day in their respective units. Selected results of the baseline survey were presented by the unit leader. Under the guidance of the consultants, employees were asked to comment on the results and add further information about potential areas for work environment improvements. After discussing the issues, the units decided on which topics they were going to focus during the next months. All units decided to establish working groups to discuss the chosen issues further. In addition, the consultants offered leadership-coaching sessions to all unit leaders. In all units the intervention phases ended with a final meeting where the intervention and its effects were discussed. More details about the meetings, other activities conducted during the intervention phase and participants’ evaluation of the intervention are presented in the results section and are summarized in Table II. Process evaluation of the interventions Project resources did not allow for a sophisticated qualitative analysis of the intervention processes. For example, there were no interviews with the specific goal of assessing how the participants perceived the intervention. Instead, those involved in conducting the study (researchers, consultants, hospital project group) documented the intervention processes with rather simple methods. The function of this qualitative part of the study was to complement the quantitative part of the study. To use a category from Needleman and Needleman [17], the qualitative data in this study serves as an interpreter for the quantitative results, that is they are used to try to understand why the quantitative results turned out as they did. Qualitative and quantitative methods were used for the process evaluation. The qualitative part of the process evaluation included meeting notes and a log-book, in which relevant context information were registered. A student research assistant took detailed notes of all staff meetings at the beginning and at the end of the intervention. All of these meeting notes followed the same systematic structure: It was documented how many participants attended the meetings (including information about their roles, i.e. leader or regular employee), followed by a detailed description of the course of the meeting (i.e. consultants presentation or for example group work), the topics discussed, including a more detailed description if controversial topics were discussed, and finally a complete listing of all decisions at the end of the meetings. In addition, observations that might be important to get a better sense of the atmosphere at

When workplace interventions lead to negative effects the meetings were also noted, for example if participants seemed highly motivated or frustrated during the meetings. The consultants kept short notes about their meetings with the unit leaders (participants, topic and outcome of the meeting). This included all preparatory meetings, leader-coaching sessions as well as other meetings or contacts. At the end of the intervention the consultants wrote a detailed report that included a description of their approach to the intervention, their evaluation of all meetings (including staff meetings), a description of the intervention process in each unit, and the consultants’ reflections about the entire intervention project. The intervention process was also assessed quantitatively by including a number of questions about the intervention process in the follow-up questionnaire.

109

report and the researchers’ own notes from project team meetings.

Results Characteristics of study participants in intervention and reference group Table I shows key characteristics of the intervention and the reference group. The groups did not differ in gender distribution, age, and years of employment in the unit. There was a difference in the distribution of occupational position, with the intervention group including a higher proportion of laboratory technicians than the reference group. Participants in the two groups did not differ in mental health and vitality at baseline.

Analyses of the qualitative process evaluation data Meeting notes, the consultants report and the log-book were analyzed for all relevant information contributing to a better understanding of how the intervention was conducted. Basic information about how many meetings were held, for which purpose and with how many participants was systematically collected from the different data sources. In addition, the available data was analyzed for content, to collect information about why meetings were discontinued or cancelled. The most detailed and systematic meeting notes were from the kick-off days and the meetings at the end of the intervention. Here, information about planned and actually conducted activities as well as information about why activities were not conducted was retrieved. Also the employees’ and leaders’ general evaluation of the intervention was retrieved from this data. Relevant context information and information about discussions within the entire project team (researchers, consultants and hospital project group) was retrieved from the consultants’

Process evaluation of the interventions Table II gives an overview about the intervention activities and the qualitative evaluation about the entire intervention project in each unit. In general, the interventions consisted of a starting phase with a kick-off day, leadership coaching, working group meetings and other activities and ended with a final discussion day.

Starting phase of the intervention The content analysis of the notes from the kick-off days revealed that in most units about three to four topics were identified, for example cooperation between colleagues, specific work organizational issues, and relation to and expectations towards unit leaders. Most employees participated in the kick-off days and seemed motivated to participate in a process to improve their psychosocial work environment.

Table I. Baseline characteristics of participants in intervention and reference groups. Intervention group (n ¼ 128) Gender Women Men Age Job group Nurses/Midwifes Nurse assistants/healthcare assistants Laboratory technicians Other Years at unit Mental health score Vitality score

Reference group (n ¼ 103) 2 ¼ 1.25, p ¼ 0.26

n (%) n (%) Mean (SD)

124 4 40.6

(96.9) (3.1) (9.4)

102 1 42.2

(99.0) (1.0) (8.6)

n (%) n (%)

79 12

(61.7) (9.4)

75 12

(72.8) (11.7)

n (%) n (%) Mean (SD) Mean (SD) Mean (SD)

35 2 7.5 79.1 64.9

(27.3) (1.6) (7.4) (13.9) (17.5)

11 5 8.3 79.8 65.0

(10.7) (4.9) (7.6) (13.6) (16.3)

t ¼ 1.36 p ¼ 0.18 2 ¼ 11.34, p ¼ 0.01

t ¼ 0.78 p ¼ 0.44 t ¼ 0.38 p ¼ 0.71 t ¼ 0.05 p ¼ 0.96

6 sessions, 1–4 participants (unit leader þ 2 other unit leaders þ department leader)

2 sessions, unit leader

1 session, unit leader

3 sessions, unit leader

P

D

F

G

Unit

No. of leader coaching meetings & participantsa

Communication group very active, other groups stopped because of too many demands and too many other projects

No working groups. Meetings and the whole project was not prioritized because of other activities in the unit (for example the unit had to move)

Working groups were not continued because of high demands at work

Working groups stopped because employees were unclear about the meaning and goal of the project

No. of working group meetings and comments about these meetingsa,c

Table II. Description of the interventions in the seven units.















  







4 hour meeting about communication with 6 participants Introduction of a newsletter to improve communication Meeting about respective expectations from leaders and employees and about which projects should be prioritized

Closer cooperation with a similar unit to cover vacations Notice board to improve cooperation with physicians Meetings between unit leader and physicians to discuss cooperation

All-day staff meeting (with consultants) to discuss communication and reduction of sickness absence Common shift-schedule for all three units in department The three unit leaders and the department leader agreed on how they will work with reducing sickness absence Salary for additional shifts (short time solution) No more special arrangements for some Several small organizational changes were implemented to improve cooperation, communication and planning Discussion about the employees and the leaders’ roles

Other activities, implemented changesa,b,c

there was the impression that leadership roles and cooperation between the leaders had improved there was more flexibility and better cooperation between the units and better overview for all (leaders and employees) employees would have needed a clear structure and help in prioritizing one topic

(continued)

the small organizational changes led to some improvements and a discussion process was started  the high motivation after the kick-off day was followed by almost no activities, which led to frustrations  employees had been unsure about their roles/resources and would have needed more help from consultants  employees appreciated that their unit leader worked on role clarity, but could not notice any effects so far The unit leader appreciated that a discussion about the leader role was started, but she did not succeed in involving her supervisors in a new definition of roles  good experience with cooperation with the other unit during summer vacation time (improved relationship)  the notice board to improve communication was appreciated  cooperation with physicians was not improved despite meeting (physicians were not motivated to change)  discussion days with consultants were appreciated, but led to almost no further activities because of too many other projects (mostly physicians’ projects)  disappointment about few changes/improvements, despite enthusiastic start of the project  employees appreciate that a discussion about the leaders’ role was started The unit leader appreciated the coaching sessions. The unit leader mentioned that the discussions and activities with regard to communication seem to have had an effect; for example, she regarded the last staff meeting as a positive experience. 







Final evaluation of the entire intervention projectb,c In the final meeting about the project (in most units between the consultants and the entire staff and their leader, in unit P and U between consultants and representatives of the unit and in unit N between the leader of the hospital project group and the unit staff) employees mentioned that:

110 B. Aust et al.

2 sessions, unit leader

8 sessions, 1–8 leaders (leader group)

M

N

Working groups stopped shortly after kick-off day because employees felt meetings were not relevant

Agreed on 7 working groups at the kick-off meeting, but because of confusion and lack of solutions and support all were stopped

Working groups never started because of unclear roles, resources and responsibilities

No. of working group meetings and comments about these meetingsa,c

 





Workshop for group leaders: 11 participants Presentation about communication

Staff meeting about status of working groups

Staff meeting about respective expectations between leaders and employees

Other activities, implemented changesa,b,c

 the meeting about respective expectations went well and led to small improvements  employees in general were dissatisfied because of no working groups and very few intervention activities  employees did not know what was expected of them, they expected more help from the consultants  there were too many other projects, unclear about resources for this project while dealing with major budget cuts  discussion days were appreciated, but employees remained unclear about their roles and resources  working group about group leaders’ role stopped because it was difficult for employees to tell each other how things should be. Help would have been needed for this task  the unit leader had improved communication and information flow (i.e. newsletter) No final evaluation day with consultants because employees felt there was nothing to talk about – project was regarded as irrelevant for the unit The leader group of this unit appreciated the coaching sessions and felt that some development was achieved.

Final evaluation of the entire intervention projectb,c In the final meeting about the project (in most units between the consultants and the entire staff and their leader, in unit P and U between consultants and representatives of the unit and in unit N between the leader of the hospital project group and the unit staff) employees mentioned that:

a

Information retrieved from consultants’ meeting notes. bInformation retrieved from student research assistant notes of staff meetings about the intervention project. cInformation retrieved from consultants’ final report.

3 sessions, unit leader

U

Unit

No. of leader coaching meetings & participantsa

Table II. Continued.

When workplace interventions lead to negative effects 111

112

B. Aust et al.

Leader coaching

Implementation of changes

Early on in the project the consultants mentioned in project group meetings that they assumed that a development of leadership qualities would be important for improving the psychosocial work environment. This assumption, they said, was based on their preparatory meetings with the unit leaders and confirmed by the results from the baseline study. A meeting was held in which the hospital project group agreed that the consultants could offer leadership-coaching sessions as part of the intervention activities. All unit leaders participated in leader-coaching sessions (Table II). The sessions lasted about two hours each and focused on the development of the unit leaders’ role and on issues the leaders wanted to discuss. It was up to the leaders to decide how many coaching sessions they wanted to participate in. The leaders of unit F, D and M had a need for only one or two sessions, while the leaders of unit G and U had three sessions each. The leaders of unit P and N had six and eight sessions respectively and used some of the sessions to discuss leadership issues with other leaders. On average the unit leaders received about seven hours of leadership coaching.

Several units arranged a staff meeting in order to discuss issues such as communication, sickness absence, expectations from leaders and employees or about which project should be prioritized. In some units organizational changes were established. For example one unit (P) managed to establish a common shift-schedule for all three units in the department and another unit (F) found a better solution to manage summer vacation time by improving the cooperation between two similar units. Overall, however, few changes were implemented (Table II).

Working group meetings At the kick-off days it was agreed that employee working groups should be formed at each unit to discuss the issues further, find solutions and start implementing them as far as possible. To support this process, employees could contact the consultants when needed. Analysis of process data showed, however, that these working groups either never met or met only very few times before they stopped completely (Table II). Early in the project, the hospital project group criticized the consultants for focusing too much on leadership development, while leaving the working groups to themselves. In meetings about this topic it was agreed that the consultants should offer more help to the working groups. Although the consultants thereafter contacted the units again to hear if they needed help, the situation basically did not change. During project meetings and again in their report, the consultants said that, in their view, the intervention focused too much on the involvement of employees and not enough on the important role of the unit and of the department leaders. They argued that hospital employees do not have time to organize and conduct group meetings.

End of the intervention project In most units the intervention phase ended after about nine months with a staff meeting where the entire intervention process was discussed and evaluated under the guidance of the consultants. In unit P and U the consultants only met with a group of representatives of the unit and in unit N the leader of the hospital project group held a short meeting with the unit. To give more time for implementation of changes and to ensure that already implemented changes could take effect, the follow-up survey was conducted about six months after the official end (final staff meeting) of the intervention project, resulting in about 16 months between baseline and follow-up survey. When asked about their evaluation of the intervention project at the final meetings, employees in almost all units mentioned improvements with regard to communication, cooperation and/or work organizational issues. In the units P, D, G, M and N employees and/or unit leaders pointed out that the work with the leaders’ role had been appreciated and in some cases started to show an effect. Some problems could not be solved because discussions with other relevant partners failed (unit D and F). In five of the seven units (P, D, F, U, M) employees criticized that they had been unsure about their roles and resources during the project, that they would have needed a clearer structure and more help from the consultants and that they were frustrated about the few implemented changes.

The quantitative process evaluation Table III shows how the participants rated the intervention projects in the follow-up questionnaire. Asked about the kick-off day, 76% of the participants answered that this day was at least somewhat useful

When workplace interventions lead to negative effects

113

Table III. Responses from the participants in the intervention group to selected process evaluation items in the follow-up questionnaire. Kick-off day Was the kick-off day useful for identifying the core problems in your unit?

Yes, to a high degree

Yes, to some degree

Yes, but only a little bit

No, not at all

5 (4.5%)

41 (36.6%)

39 (34.8%)

27 (24.1%)

To what degree did your group use the help offered by the consultants?

To a high degree

To some degree

Only a little bit

Not at all

13 (10.9%)

26 (21.9%)

28 (23.5%)

How helpful were the consultants during the course of the project?

Very much

Somewhat

Only a little bit

Not at all

8 (7.1%)

21 (18.6%)

9 (8.0%)

9 (8.0%)

I am not aware that help was offered 66 (58.4%)

To a high degree

To some degree

Only a little bit

Not at all

Do not know

26 (21.3%)

36 (29.5%)

34 (27.9%)

2 (9.8%)

14 (11.5%)

The activities were more positive

The activities were neither positive nor negative 44 (36.4%)

The activities were more negative

I am not aware of any activities

20 (16.5%)

39 (32.2%)

Help from the consultants

0

I am not aware that help was offered 52 (43.7%)

Priority of leaders To what degree was the project given high priority by the leader of your unit? Changes at the workplace Overall, how do you rate the activities that had been initiated in relation with this project?

18 (14.9%)

for identifying core problems. In both questions regarding receiving help from consultants, a large part of the participants (44% and 58%) answered that they had not been aware that help was offered. Around one third of the participants answered that they (at least a little bit) used the help offered by the consultants and that this help was (at least a little bit) helpful. With regard to the unit leaders, 21% of the employees felt that their leaders had prioritized the project highly. When asked about the activities that the project had initiated, 15% reported that the activities were more positive, whereas 17% found that the activities were more negative. The vast majority was either not aware of any activities (32%) or rated the activities as neither positive nor negative (36%).

Changes in psychosocial work environment factors during follow up Table IV shows baseline and follow-up values of the 13 work environment scales for the intervention and the reference groups. At baseline, the intervention group had in general a less favourable psychosocial work environment than the reference group, although only for high work pace (p ¼ 0.005) was this difference statistically significant (p-values not shown in table).

After 16 months of follow up, the participants of the intervention group showed a statistically significant decrease in the scales of emotional demands, influence, possibilities for development, meaning of work, supervisor support, predictability, and quality of leadership (see Table IV for details). The decrease was most pronounced for quality of leadership with a mean decline of 9.2 points, followed by supervisor support (minus 8.3 points) and predictability (minus 6.1 points). The reference group showed statistically significant changes on two scales – an increase in work pace and a decrease in predictability. A combined analysis showed that the changes in the two groups were significantly different from each other for the variables meaning of work, social support from supervisors, and quality of leadership (Table IV). The analyses were adjusted for gender, age, years at unit, job group, and baseline values of mental health and vitality. Table V compares the percentage of participants in the intervention and reference groups with a worsening of 10 points or more on the work environment scales. Compared to the reference group, participants in the intervention group were more likely to experience a worsening in possibilities for development, supervisor support, and quality of leadership, after adjustment for covariates. We also compared percentages for participants with an improvement of 10 points or more on the scales (data not shown in table).

49.5 70.1 56.4 42.8 37.5 (17.3) 69.6 (13.5) 76.7 (14.3) 70.0 (16.3) 51.8 (21.2) 68.4 (12.5) 40.8 (18.5) 48.3 (18.0) 45.9 (17.2)

47.8 (15.5) 68.4 (15.8) 58.7 (18.0) 40.6 (17.0) 41.7 (17.3) 72.3 (14.1) 81.1 (13.7) 71.4 (16.8) 60.1 (22.1) 70.3 (13.0) 40.7 (16.4) 54.4 (17.9) 55.1 (16.2)

(14.0) (15.1) (17.6) (16.2)

Follow up Mean (SD)

1.4 8.3 1.9 þ0.1 6.1 9.2

4.2 2.7 4.4

þ1.7 þ1.7 2.3 þ2.2

change

0.99 3.90 1.49 0.12 3.74 5.41

3.53 2.63 4.17

1.68 1.19 2.28 1.80

t

0.32 <0.001 0.14 0.90 <0.001 <0.001

0.001 0.01 <0.001

0.10 0.24 0.02 0.07

p (14.9) (14.1) (20.4) (14.7)

73.8 (14.6) 67.6 (19.2) 74.8 (12.8) 35.9 (14.7) 62.1 (19.7) 67.8 (20.1)

48.8 (16.1) 76.2 (12.1) 83.3 (12.7)

48.5 65.6 56.8 35.0

Baseline Mean (SD) (11.9) (14.3) (17.7) (15.5)

73.9 (16.0) 69.3 (19.6) 72.9 (11.4) 37.8 15.5) 56.8 (17.6) 67.0 (15.7)

47.4 (17.0) 76.0 (10.4) 82.5 (11.8)

49.6 69.8 55.7 37.6

Follow up Mean (SD)

þ0.1 þ1.7 1.9 þ1.9 5.3 0.8

1.4 0.2 0.8

þ1.1 þ4.2 1.1 þ2.6

change

t

0.23 0.85 1.68 1.55 3.15 0.66

1.32 0.15 0.56

0.89 3.37 0.71 1.64

Reference group (n ¼ 103)

0.82 0.40 0.10 0.13 0.002 0.51

0.19 0.88 0.58

0.38 0.001 0.48 0.11

p

1.7 9.9 0.1 2.0 0.7 8.2

2.7 2.5 3.7

0.4 3.0 1.3 0.3

Est

0.82 3.40 0.04 1.04 0.32 3.52

1.57 1.80 2.36

0.23 1.44 0.85 0.14

t

0.41 0.001 0.97 0.30 0.75 0.001

0.12 0.07 0.02

0.82 0.15 0.40 0.89

p

Interaction change  group

Changes within groups are analyzed with mixed model for repeated measures including unit as a random effect. Interaction effects for change x group are calculated with combined analyses with unit as a random effect and adjusted for gender, age, years at unit, job group (nurses and midwifes, nurse assistants and healthcare assistants, laboratory workers, others), and baseline values of mental health and vitality (as fixed effects). p-values of <0.05 are printed in bold.

Demands at work Quantitative demands High work pace Emotional demands Demands for hiding emotions Work organization and job content Influence Possibilities for development Meaning of work Interpersonal relations and leadership Social support from colleagues Social support from supervisor Role clarity Role conflicts Predictability Quality of leadership

Baseline Mean (SD)

Intervention group (n ¼ 128)

Table IV. Changes in psychosocial work environment factors in the intervention and the reference groups during 16 months of follow up.

114 B. Aust et al.

When workplace interventions lead to negative effects

115

Table V. Percentage of participants with a substantial worsening in psychosocial work environment factors in the intervention and the reference group during 16 months of follow up. Intervention group

Reference group

Worsening

Demands at work Quantitative demands High work pace Emotional demands Demands for hiding emotions Work organization and job content Influence Possibilities for development Meaning of work Interpersonal relations and leadership Social support from colleagues Social support from supervisor Role clarity Role conflicts Predictability Quality of leadership

Worsening

Odds ratio (OR) for worsening in intervention group

At risk

n

(%)

At risk

n

(%)

OR

(95% CI)

p

128 128 127 127

33 31 11 48

(25.8) (24.2) (8.7) (37.8)

103 101 103 103

23 23 14 38

(22.3) (22.8) (13.6) (36.9)

1.15 0.81 0.61 0.97

(0.73–1.82) (0.56–1.19) (0.29–1.30) (0.65–1.47)

0.54 0.29 0.20 0.89

128 127 128

45 41 32

(35.2) (32.3) (25.0)

101 103 102

21 17 13

(20.8) (16.5) (12.8)

1.86 2.51 2.16

(0.99–3.49) (1.19–5.29) (0.98–4.74)

0.06 0.02 0.06

128 128 128 128 127 128

47 64 34 30 58 62

(36.7) (50.0) (25.6) (23.4) (45.7) (48.4)

103 103 103 103 101 102

37 31 29 27 40 24

(35.9) (30.1) (28.2) (26.2) (39.6) (23.5)

1.17 2.31 0.95 0.78 1.12 3.25

(0.76–1.78) (1.10–4.85) (0.57–1.58) (0.53–1.14) (0.38–3.34) (1.03–10.23)

0.48 0.03 0.85 0.20 0.84 0.04

Analyses are adjusted for gender, age, years at unit, job group (nurses and midwifes, nurse assistants and healthcare assistants, laboratory workers, others), and baseline values of mental health and vitality. For the four demand factors and for role conflict worsening is defined as an increase of 10 points or more. For the other eight factors worsening is defined as a decrease of 10 points or more. p-values of <0.05 are printed in bold.

Here, we found differences for only one scale: intervention group participants were more likely to experience a reduction in work pace (OR ¼ 3.56, p ¼ 0.010) than participants in the reference group. Finally, we analyzed changes in the scores of mental health and vitality. In the intervention group, the mental health score decreased from 79.11 (SD: 13.91) to 78.64 (13.71; difference ¼ 0.47, t ¼ 0.38, p ¼ 0.71) and the vitality score decreased from 64.92 (SD: 17.54) to 63.11 (SD: 18.66; difference ¼ 1.81, t ¼ 1.31, p ¼ 0.19). In the reference group, the mental health score decreased from 79.64 (SD: 13.59) to 79.44 (SD: 13.69; difference ¼ 0.21; t ¼ 0.13, p ¼ 0.89) and the vitality score increased from 65.03 (SD: 16.28) to 67.02 (SD: 15.29; difference þ1.99, t ¼ 1.27, p ¼ 0.21). Analysis of the combined sample showed, after adjustment for covariates, that there was no difference in the change of mental health between the two groups (t ¼ 0.13, p ¼ 0.90). With regard to the vitality score, however, the differences in the changes approached statistical significance (t ¼ 1.82, p ¼ 0.07).

Discussion In the intervention group the psychosocial work environment became worse after the intervention:

there was a statistically significant worsening in six out of 13 work environment scales. The decrease was most pronounced for three scales that measure aspects of interpersonal relations at work and leadership. All three scales that measure aspects of work organization and content decreased, whereas none of the scales that measure demands at work showed a worsening. In comparison, the reference group showed statistically significant changes in only two scales (increase in work pace and a decrease in predictability). Further analyses showed that the changes in intervention and reference groups were statistically significant for meaning for work, social support from supervisor, and quality of leadership, even after adjustment for differences in socio-demographic (gender, age), work-related (job group, years at unit) and psychological health characteristics (mental health and vitality score) of the participants in the two groups. However, one positive change was found in the intervention group: emotional demands became less. It is difficult to say if this can be viewed as an improvement or as an indication of burnout. Because of the poor work environment, people in the intervention group might have become emotionally exhausted and therefore the lower level of emotional demands at follow up might indicate that participants had started to distance themselves from their work

116

B. Aust et al.

and consequently have felt emotionally less involved. Considering the general trend towards a less favourable work environment in the intervention group, this interpretation seems at least possible. The interpretation is further supported by the finding that vitality scores had decreased in the intervention group during followup. The results of this intervention study are in contrast to the general perception that workplace interventions – even though they are far from always showing positive results – seldom lead to negative results [18], although some exceptions exist [19,20]. However, publication bias, that is, the reluctance to publish negative results, is also a likely reason for this general perception. In 1999, Briner and Reynolds warned that the all too optimistic impression of organizational interventions acted as a barrier to further development and elaboration of research and practice [21]. The following discussion attempts to overcome this barrier by analyzing in detail what might have caused the negative effects of this intervention. However, because of the rather limited resources for a more in-depth process evaluation, a certain degree of speculation cannot be avoided in the following discussion. The aim is to discuss a number of possible explanations in order to find out which seems most plausible considering all the information available. First, randomization was not possible. Participants in the intervention group reported in general a less favourable work environment compared to the reference units. Because of this selection bias, one could argue that the psychosocial work environment was maybe on its way down in the intervention units and that the intervention project was just not able to stop this downward spiral. Against the argument of the unstoppable downward trend, one could argue that the lower baseline values in the intervention group would have made it easier to find positive effects because there was ‘‘room for improvement’’. But even if the downward trend plays a part in this development, the implementation processes should be analyzed in detail, because it is usually here where intervention studies fail [18]. As can be seen in Table II, there were some changes regarding improving cooperation, communication, and work organizational aspects. All in all though, only a few changes were implemented. Two attempts to improve the leaders’ role (unit D) and cooperation (unit F) failed, because the other relevant parts from a higher hierarchical level (unit D: supervisor; unit F: physicians) were not willing or motivated to find a better solution.

Almost all unit leaders participated in leader coaching sessions offered by the consultants (Table II), i.e. this part of the implementation was successful. On the other hand, it could be argued that the intervention intensity was too low, i.e. that there were too few coaching sessions. The consultants themselves noted in their report that the few coaching sessions were only a start and that the development of leadership qualities should be continued. However, the low intensity of that part of the intervention does not explain why leadership quality and social support from supervisor were those scales that decreased the most in the intervention units. Is it possible that the few coaching sessions with the unit leaders were the reason for the profound worsening in the scales measuring leadership quality? This would be an unusual powerful intervention effect – although in the wrong direction. One could also argue that the focus on leadership coaching raised employees’ expectations about improved leadership behaviour. If that is the case, the more negative evaluation of leadership qualities in the follow-up survey could be interpreted as a sign of disappointment among employees. However, looking at employees’ comments at the end of the intervention (Table II) there was no critique with regard to this aspect of the intervention. Instead, it was mentioned several times that employees appreciated that their leaders had started to develop leadership styles further. Assuming that the leaders changed their leadership style as a result of the coaching sessions, a possible explanation for the negative developments in leadership qualities could be that employees regretted this new leadership style. Maybe the leaders’ role became more distinct and led to more distance between employees and leaders. Also the change process itself can be regarded as stressful and therefore cause negative effects [22,23]. But again the rather positive evaluation of this part of the intervention in the final meetings does not support this hypothesis. The perspective of the consultants, who wrote their report before the follow-up survey, provides another possible explanation. In their view, the intervention focused too much on the involvement of employees and not enough on the important role of the unit and of the department leaders. They argue that hospital employees do not have time to organize and conduct group meetings and point out that despite the fact that they repeatedly offered their help, employees only contacted them twice during the entire project. This argument is supported by the fact that in almost all units the working groups failed (Table II). Furthermore, the consultants pointed out that the

When workplace interventions lead to negative effects leaders were unsure about their role in the intervention project and therefore never fully took ownership for it. If the consultants are right in their evaluation of the intervention process, the following explanation for the decrease of leadership qualities and social support from leaders seems possible: employees were not able to be the driving force in the intervention process. Instead they might have felt left alone with a task too big for them and were disappointed about the fact that their leaders did not get involved more in the intervention process. Only 21% of the participants answered that the leaders had given high priority to the project (Table III). The worsening of leadership qualities (as well as the worsening in the other scales) in the follow-up survey could therefore reflect the employees’ disappointment about their leaders not getting more involved in the intervention process, rather than the negative effect of a few coaching sessions. When asked to evaluate the project, employees in most units raised critique about how the project was conducted (Table II). Although this seems like a possible explanation, it also needs to be pointed out that involvement of employees in intervention processes has been successfully done in other intervention projects, including those in hospitals [24,25]. Nevertheless, the consultants are right in pointing out that leadership support for the project was too weak. Many studies have shown that the full support of leaders is essential for the success of workplace interventions [1,26,27]. However, another important element for participative interventions in the workplace is a clear structure of the intervention process, for example helping a working group to define goals, set deadlines and follow up on them. It seems as if the working groups in this study were never offered this kind of active help – neither from their leaders nor from the consultants. In most units, it was mentioned in the final meetings that working groups failed because employees needed help in prioritizing and structuring their activities (Table II) and the quantitative process evaluation revealed that a large percentage of employees were not aware that the consultants were available for help (Table III). The failure to implement working groups that would develop and later implement improvements at the workplace seems therefore not only due to the unclear role of the leaders, but also to a too passive approach from the consultants. In a recent intervention study involving 14 workplaces, the lack of structure, clear roles and responsibilities was identified as the major reason why some of the intervention processes came to a halt even after an enthusiastic start [26]. Others have highlighted that a variety of skills are needed to implement

117

organizational change processes [28]. Nytrø et al. [9] in accordance with Smith [29], suggest that organizational change projects should consist of five small indispensable overlapping steps: telling, selling, testing, consulting and co-creating. These steps require employee involvement, and especially in the co-creating phase the original plans might need to be modified. It seems that the parties involved in this study were not able to manage this challenge. The passive approach of the consultants was noticed by the hospital project group and raised critique, but the situation did not change. The consultants continued their coaching sessions with the unit leaders and, as they wrote in their report, they believed that this was the key to a better psychosocial work environment. This documented conflict reveals that the consultants and the hospital project group had different implicit theories about what will lead to meaningful improvements in the psychosocial work environment. The hospital project group argued for more involvement of employees and thereby implicitly assumed that this bottom-up approach and the experience of involvement would be an important part of the development towards a more favourable psychosocial work environment. In contrast, the consultants assumed that a top-down approach would be more successful. It is difficult to say how much employees were aware of this conflict. It seems, however, plausible that the difference between how the project was introduced to them by the hospital project group and the actual implementation of the project by the consultants was noticed. In most units employees mentioned their disappointment (Table II), which might have been the reason for more unfavourable ratings of the psychosocial work environment at follow up. In a recent study it was found that employees’ appraisal of the quality of an intervention and their own influence in it mediates how they rate the outcome of this intervention [30]. The participants’ negative appraisal of the intervention activities might have led to the negative intervention outcomes. To summarize, it can be said that there are indications that the negative effects found in the COPSOQ scales are at least partly due to the employees’ disappointment about unfulfilled expectations. While the kick-off days more or less worked out as intended, employees’ activities came to a halt shortly thereafter. The working groups did not know how to tackle their task, the unit leaders did not know if and how they should help and the consultants were too passive in offering help – resulting in an almost complete failure of this part of the intervention. Furthermore, the implementation process was disturbed because the

118

B. Aust et al.

hospital project group and the consultants had different implicit theories about the intervention and did not manage to find a constructive solution to this conflict. Summing up, this intervention seems to have failed not because the concepts were not sound but because they were not well implemented. In Kristensen’s terms, this would be a programme failure (in contrast to a theory failure) [1]. On the positive side, through this project the hospital became aware of several problem areas (including the unclear role of unit leaders) and started activities to improve the situation. Also, from a research perspective, this study has its good side, because it shows how things can go wrong when implementation fails and thereby highlights how important it is to plan and conduct workplace interventions carefully, including an agreement about the implicit theories at play.

Conclusions The study shows how important it is to follow workplace interventions with a close process documentation and evaluation. The insights gained from process data help to understand the negative effects of this intervention. The study also highlights how important it is to think about employees’ expectations about improved work environments when starting a workplace intervention project. Sometimes – as it seems happened in this study – more harm can be done by disappointing expectations than by not conducting an intervention.

Acknowledgements The write-up of this study was partly funded by a grant from the Danish Working Environment Research Fund (grant number: 34-2005-09). We are indebted to Jakob B. Bjørner, Helene Feveile and Jørgen Vinsløv Hansen for their invaluable help with the statistical analysis.

References [1] Kristensen TS. Intervention studies in occupational epidemiology. Occup Environ Med 2005;62:205–10. [2] Skov T, Kristensen TS. Etiologic and prevention effectiveness intervention studies in occupational health. Am J Ind Med 1996;29:378–81. [3] Kompier M, Kristensen TS. Organizational work stress interventions in a theoretical, methodological and practical context. In: Dunham J, editor. Stress in the workplace. Past, present and future. London: Whurr Publishers Ltd; 2000. pp. 164–90.

[4] Saksvik PØ, Nytrø K, Dahl-Jørgensen C, Mikkelsen A. A process evaluation of individual and organizational occupational stress and health interventions. Work & Stress 2002;16:37–57. [5] Goldenhar LM, Schulte PA. Methodological issues for intervention research in occupational health and safety. Am J Ind Med 1996;29:289–94. [6] Egan M, Bambra C, Petticrew M, Whitehead M. Reviewing evidence on complex social interventions: appraising implementation in systematic reviews of the health effects of organisational-level workplace interventions. J Epidemiol Community Health 2009;63:4–11. [7] Kompier M. Research must look at what interventions work as well as when and why. Healthcare Pap 2004;5:45–8. [8] Murta SG, Sanderson K, Oldenburg B. Process evaluation in occupational stress management programs: a systematic review. Am J Health Promot 2007;21:248–54. [9] Nytrø K, Saksvik PØ, Mikkelsen A, Bohle P, Quinlan M. An appraisal of key factors in the implementation of occupational stress interventions. Work & Stress 2000;14:213–25. [10] Kristensen T, Hannerz H, Høgh A, Borg V. The Copenhagen Psychosocial Questionnaire. A tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31:438–49. [11] Aust B, Rugulies R, Skakon J, Scherzer T, Jensen C. Psychosocial work environment of hospital workers: validation of a comprehensive assessment scale. Int J Nurs Stud 2007;44:814–25. [12] Bjorner JB, Damsgaard MT, Watt T, Groenvold M. Tests of data quality, scaling assumptions, and reliability of the Danish SF-36. J Clin Epidemiol 1998;51:1001–11. [13] Feng Z, Diehr P, Peterson A, McLerran D. Selected statistical issues in group randomized trials. Annu Rev Public Health 2001;22:167–87. [14] Dahl-Jørgensen C, Saksvik PØ. The impact of two organizational interventions on the health of service sector workers. Int J Health Serv 2005;35:529–49. [15] Liang K-Y, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika 1986;73:13–22. [16] Zeger SL, Liang K-Y. Longitudinal data analysis for discrete and continuous outcomes. Biometrics 1986; 42:121–30. [17] Needleman C, Needleman ML. Qualitative methods for intervention research. Am J Ind Med 1996;29:329–37. [18] Semmer NK. Job stress interventions and the organization of work. Scand J Work Environ Health 2006;32:515–27. [19] Landsbergis PA, Vivona-Vaughan E. Evaluation of an occupational stress intervention in a public agency. J Organiz Beh 1995;16:29–48. [20] Nielsen K, Fredslund H, Christensen KB, Albertsen K. Success or failure? Interpreting and understanding the impact of interventions in four similar worksites. Work & Stress 2006;20:272–87. [21] Briner RB, Reynolds S. The costs, benefits, and limitations of organizational level stress interventions. J Organiz Beh 1999;20:647–64. [22] Korunka C, Weiss A, Karetta B. Effects of new technologies with special regard for the implementation process per-se. J Organiz Beh 1993;14:331–48. [23] Dent EB, Goldberg SG. Challenging ‘‘Resistance to change’’. Journal Appl Behav Sci 1999;35:25–41. [24] Aust B, Ducki A. Comprehensive health promotion interventions at the workplace: experiences with health circles in Germany. J Occup Health Psychol 2004;9:258–70.

When workplace interventions lead to negative effects [25] Beermann B, Kuhn K, Kompier M. Germany: reduction of stress by health circles. In: Kompier M, Cooper C, editors. Preventing stress, improving productivity: European case studies in the workplace. London; 1999. pp. 222–41. [26] Nielsen K, Hvenegaard H, Aust B, Møller N, Kristensen TS. Metodeudvikling vedrørende behandling af psykisk arbejdsmiljø i APV-arbejdet. Hovedrapport [Development of methods for addressing the psychosocial work environment in workplace assessments. Main report]. Available at: http://www.at.dk/ sw5728.asp. (accessed November 3, 2009). [27] Kompier MAJ, Aust B, van den Berg A-M, Siegrist J. Stress prevention in bus drivers: evaluation of 13

119

natural experiments. J Occup Health Psychol 2000;5:11–31. [28] Grossman R, Scala K. Health promotion and organisational development: developing settings for health. Vienna: World Health Organization, Regional Office for Europe; 1993. [29] Smith B. Building shared vision: how to begin. In: Senge PM, Kleiner A, Roberts C, Ross RB, Smith BJ, editors. The fifth discipline fieldbook. London: Nicholas Brealey; 1994. pp. 312–27. [30] Nielsen K, Randall R, Albertsen K. Participants’ appraisals of process issues and the effects of stress management interventions. J Organiz Beh 2007;28:793–810.

Scandinavian Journal of Public Health, 2010; 38: 120–124

SHORT COMMUNICATION

The Copenhagen Psychosocial Questionnaire (COPSOQ) in Germany: From the validation of the instrument to the formation of a job-specific database of psychosocial factors at work

MATTHIAS NUEBLING1 & HANS MARTIN HASSELHORN2 1

FFAS – Freiburg Research Centre of Occupational and Social Medicine (Freiburger Forschungsstelle Arbeits- und Sozialmedizin) Freiburg, Germany, and 2University of Wuppertal, FB – D, Empirical Work Research Group, Wuppertal, Germany

Abstract The German version of the Copenhagen Psychosocial Questionnaire (COPSOQ) was established and tested in a sample of 2561 employees in order to: (a) assess the questionnaires’ psychometric properties; and (b) develop an appropriate instrument to use in the assessment of psychosocial risk factors. A shortened version of the instrument was developed, reducing the number of items from 141 to 87. With this, a database has been established since 2005. In a cooperation model between science (Freiburg Research Centre of Occupational and Social Medicine) and companies or organizations, new COPSOQ data are added to the dynamically growing database with profession-specific profiles of psychosocial factors at work. In return, companies can compare their results with job-related data in the database, facilitating the interpretation of their results and the implementation of improvement measures. The COPSOQ database has reached425,000 respondents. Ongoing projects will expand the German COPSOQ database and include representative samples. Furthermore, a job exposure matrix for psychosocial factors at work will be constructed in 2009. Finally, in several projects, a first assessment has been followed by efforts to improve the problematic areas of psychosocial working conditions.

Key Words: Psychosocial factors, assessment, reference data

The German COPSOQ validation study The Copenhagen Psychosocial Questionnaire (COPSOQ) is a comprehensive questionnaire for the assessment of psychosocial factors at work, originally developed in Denmark by Kristensen and Borg (Danish and English version) [1,2]. In a validation study, a German version was established in a translation–retranslation process and tested in a sample of 2561 employees. The scientific goal was the detailed examination and assessment of the questionnaire’s psychometric properties and measuring qualities; the practical goal was the development of a shortened instrument for the enterprises to use in their risk assessment for psychosocial factors. The German standard version includes 87 items and 25 aspects (mostly scales) (Figure 1).

Supplementary scales can be added to cover profession-specific psychosocial aspects (such as shift work or conflicts with clients). Detailed information on the assessment of psychometric properties and the content of the questionnaire are given by Nuebling et al. [3,4]. Further information, the questionnaire and an online version including, individual feedback. are available at the German COPSOQ website: www.copsoq.de (in German). COPSOQ surveys and COPSOQ database Since the end of the validation study in 2005, the standard questionnaire has found widespread use as a paper and pencil questionnaire and as an online tool. In a model of cooperation between science (Freiburg

Correspondence: Matthias Nuebling, FFAS – Freiburg Research Centre of Occupational and Social Medicine (Freiburger Forschungsstelle Arbeits- und Sozialmedizin), Bertoldstraße 27, D – 79098 Freiburg, Germany. Tel: þ49 761 / 894421. Fax: þ 49 0761 / 83432. E-mail: [email protected] (Accepted 9 October 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809353652

COPSOQ in Germany Demands –Quantitative demands –Emotional demands –Demands for hiding emotions –Work –privacy conflict

Influence and development –Influence at work –Degree of freedom at work –Possibilities of development –Meaning of work –Workplace commitment

121

Interpersonal relations and leadership –Predictability –Role clarity –Role conflicts –Quality of leadership –Social support –Feedback –Social relations –Sense of community –Mobbing/ Bullying

Further parameters –Job insecurity

Supplementary scales –conflicts with clients –shift work –teacher items –…

Strain (effects, outcomes) –Job satisfaction –Intention to leave –General health –Burnout –Cognitive stress –Satisfaction with life

Figure 1. Content of the German standard COPSOQ; differences from the Danish/English original questionnaire in italics.

Research Centre of Occupational and Social Medicine (FFAS)) and institutions, all interested organizations may use the COPSOQ. Thus, new survey data contribute to a dynamically growing database with profession-specific profiles of psychosocial factors at work. In return, the institutions receive feedback on their results as compared with the results from other professions and with their profession-specific reference values in the COPSOQ database. This facilitates the interpretation of results and the determination of targets for preventive action. All data records are held and processed outside the enterprises/organizations, and this (at least in Germany) is an advantage in guaranteeing anonymity and professionalism. Only group-level results covering at least 10 respondents (normally) are reported to the enterprises and their employees. The process of the assessment in the cooperation model is shown in Figure 2. As of today (February 2009), the COPSOQ database has reached 425,000 respondents. However, since the COPSOQ is also used independently of this cooperation model, we have no information on the total use of the COPSOQ in Germany.

100,000 teachers are being assessed from 2008 to 2010 as a part of the compulsory work risk assessment. The questionnaire is a combined instrument consisting of COPSOQ and FASS (Questionnaire on the working situation in schools, Kaempf and Krause [5]); the psychometric properties of the combined instrument were assessed recently [6]. Another ongoing project is the development of a job exposure matrix (JEM) for psychosocial factors at work until the end of 2009. Aggregated job-specific (and age- and gender-specific) psychosocial work load profiles will be provided in this JEM and can be used by science and policy. The occupations are classified according to the classification of the Federal Statistical Office ‘‘KdB92’’ (as always in the German COPSOQ database). A third study using the COPSOQ is the Gutenberg Heart Study (GHS), a population-based representative prospective study with 15,000 participants enrolled. The COPSOQ is used half and half with ERI [7,8] and combined with the FIT (German version of the Demand-Control Questionnaire [9,10]). Data of the baseline survey will be available in 2010 and allow methodological comparison of these instruments.

Ongoing projects and further development of the database The largest currently ongoing COPSOQ survey is the online assessment of psychosocial factors at work among all teachers of all schools in Baden-Wu¨rttemberg (region in south-west Germany). About 4200 schools and more than

Strengths and limitations of the German approach A strength of the German COPSOQ approach is the thorough methodological transformation of the

122

M. Nuebling & H. M. Hasselhorn 4. Data analysis

2. Performance survey

Enterprise / Organisation Employees 1 2 3 …

Science / FFAS

1. materials (online Q, p&p Q-sets) 3. Q or online-data 3a. Direct feedback (online only)

5. Comparison with reference data Inclusion in database

7. Interpretation, Improvement actions, Further process

COPSOQ data base: Profession specific reference values

6. report + CD (4 weeks)

Figure 2. Cooperation model: science enterprises and COPSOQ database.

Danish and English version of the COPSOQ instrument, including the systematic investigation of generalizability with respect to occupational groups, gender and age [3]. Even if the final standard version is not a mere copy of the Danish/English original, but has some addenda, where this was plausible, the degree of international comparability to other COPSOQ versions remains high, since all items of the middle Danish/English version are included. A limitation of the German COPSOQ database and the derived reference values, however, is that the data are not based on a representative sample of the employed population – as is the case in Denmark or in Spain. Thus, the distribution of participants in the data set is the result of the initial scientific project and the (uncontrolled) data collection from enterprises using the COPSOQ in the cooperation model. Some occupational groups are therefore overrepresented (office clerks, physicians, and teachers), while others show too low numbers as compared with their proportion in the working population (e.g. agriculture and blue-collar workers). This situation will improve with the integration of the COPSOQ in a large representative population study (GHS; see above). However, the overrepresentation of certain professions also has advantages: we are able to analyse psychosocial work profiles for some professions where population-based representative studies do not reach a sufficient number of cases, e.g. the school headmasters (see above) or catholic priests and protestant pastors. Figure 3 gives an example of how the data of the COPSOQ database are presented. Specific assessment of certain work groups also reveals that workers within the same profession may

show different degrees of psychosocial workload, depending on their professional specialty: as an example, physicians working in hospitals are exposed to substantially higher emotional demands (mean 66 points) than occupational health physicians or anaesthesiologists (56 and 57 points, respectively). Collapsing all members of one profession (here, all are physicians by profession) into one group may thus run the risk of dilution. Like the whole COPSOQ community, we are aware of the subjectivity of mere questionnaire assessments [11]. A general limitation of the assessment of workload as well as strain with self-reported data only is the so-called ‘‘triviality trap’’ [12,13]. Correlations found could be due to (methodological) artefacts. To avoid this risk, the linkage of COPSOQ data with independent measures of exposure would be desirable, e.g. with data from different sources such as occupational health physicians, enterprise records or experts’ job valuation, enabling triangulation. Similarly, linkage to objective outcomes such as objective health data or economic outcomes might lead to additional insights. Some projects are ongoing in this field, e.g. a survey with policemen in which COPSOQ data are matched to ‘‘objective’’ data from an in depth check-up by a physician. Another example is the German 3Q-Study (www.3q.uni-wuppertal.de), in which staff questionnaire data are linked to quality of care and economic outcomes in more than 50 nursing homes. However, the number of these scientifically desirable analyses is limited, owing to widespread reluctance of enterprises to share internal data, the fear of lack of privacy protection for employees, and the higher costs of such projects.

COPSOQ in Germany

123

Scale: emotional demands

Mean (95% confidence interval)

100 90

Value group

80

Average COPSOQ (all profs)

70 60 50 40 66

30

55

20

55

54

54

48

40

56

56

57

67

64 56

57

50

66

70

75

67

54 45

44

47

10

s, Ph ar tis Ph ys. t , ys ho s . Ph , oc spi ys c. tal . A He ne al Em st th er he N ge u rs tist nc s e y m ry c ed ar .S e Ph er ys vic O e th i er oth he era p. So alth ca ci al re w G o er rke ia tri rs c c Te are C a at c Pr hol her ic s ot p es ta ries O nt p ts t W her ast as ed ors te uc m an atio O th n ag er e se me rv nt ic e p r O th of. er pr of .

g

rit y

in

cu

ad

se th

or

n, tio

Au

di ng

le ec ot

m

in

Pr

Ad

ot

,n

in

,l

ea

ic

hn

ec

Ad

:t

ch

m

in ng

:e ch

Te

Te

M

an

uf

ac

ee

te

rs

rs ia ns

0

Occupation (according to kdB 92) Figure 3. Emotional demands in different occupations.

So far, the strength of the use of the COPSOQ in Germany lies in the following: (1) it reasonably categorizes the complex nature of the psychosocial work environment and the individual reactions to it (2) it quantifies and visualizes comprehensible constructs of the psychosocial work environment and potential targets for action (3) it provides profession-specific reference data (benchmarks) allowing identification of needs and areas of improvement (4) and it may thus create awareness for and initiate discussions about the psychosocial work environment.

From the scientific assessment to action Within a few years, the COPSOQ has become a well-established instrument in psychosocial work research. Most probably, the reasons are the free access to the instrument, the provision of benchmarks, and especially the comprehensibility of the COPSOQ. Currently, we observe that, in German enterprises, the COPSOQ is increasingly becoming a tool not only for work risk assessments but also for workplace health promotion. The number of people using this

tool for the exploration of psychosocial work risks and for the evaluation of intervention effects seems to be increasing. However, we have also learned that the transmission process from measurement into organizational action is not easy. Often, the COPSOQ assessment raises too high expectations (and disappointments) among management of institutions with respect to definite steps to be taken. Here, clarification in advance that measures for action and prevention need to be developed in the process within the institution is essential. Another recent issue discussed in Germany is the educational effect of the COPSOQ assessment. In contrast to the situation in the Scandinavian countries, the awareness of the relevance of a good psychosocial work environment is rather limited among the management of many organizations. But increasingly, pressure is coming from outside to address this topic – either because of legal requirements for risk assessments, or owing to collective moral pressure. Two specific sectors where this is the case are the German healthcare sector and public authorities. This often results in the performance of a staff survey with limited support from management. It is well known and understandable that surveys without management support are likely to fail and even to further degrade the psychosocial work environment [14]. However, what we need to know are

124

M. Nuebling & H. M. Hasselhorn

the key factors and key roles for successful transmission of measurement results into action. But we also need to know more about the effect of a questionnaire assessment per se. We know that individuals filling in the COPSOQ and receiving a benchmarking feedback (more than 15,000 individuals have filled in the online version of the COPSOQ) may start contemplating their work and life. Similarly, we see indications that the performance of the COPSOQ assessment may have an educational effect in organizations. Currently, we are investigating whether this postulated effect is due to the assessment or to the feedback of benchmarking results. The findings will have implications for the future use of the COPSOQ: is it enough to feed back job- or organization-related COPSOQ findings and their benchmarks, or is the (extensive) process of a questionnaire assessment necessary for developing an openness for psychosocial issues at work?

Websites www.copsoq.de (German COPSOQ; most material in German, some English material) www.3q.uni-wuppertal.de

Acknowledgements We want to express our gratitude to T. S. Kristensen, who is, along with Vilhelm Borg, one of the two ‘‘fathers of COPSOQ’’. His continued readiness to support, his sympathy for international collaboration and, not least, his devoted philosophy of free access to questionnaires have been of invaluable help for our work with the COPSOQ and for the COPSOQ work in Germany in general.

References [1] Kristensen TS, Borg V. AMI’s spørgeskema om psykisk arbejdsmiljø. Copenhagen: National Institute of Occupational Health; 2000.

[2] Kristensen TS, Hannerz H, Høgh A, Borg V. The Copenhagen Psychosocial Questionnaire (COPSOQ) – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31:438–49. [3] Nuebling M, Sto¨ßel U, Hasselhorn H-M, Michaelis M, Hofmann F. Methoden zur Erfassung psychischer Belastungen – Erprobung eines Messinstrumentes (COPSOQ). Schriftenreihe der Bundesanstalt fu¨r Arbeitsschutz und Arbeitsmedizin, Fb 1058. Bremerhaven: Wirtschaftsverlag NW; 2005. [4] Nuebling M, Sto¨ßel U, Hasselhorn H-M, Michaelis M, Hofmann F. Measuring psychological stress and strain at work: evaluation of the COPSOQ Questionnaire in Germany. GMS Psychosoc Med 2006;3:Doc05. Available at: http:// www.egms.de/en/journals/psm/2006-3/psm000025.shtml. [5] Kaempf S, Krause A. Gefa¨hrdungsbeurteilungen zur Analyse psychischer Belastungen am Arbeitsort Schule. In: Bungard W, Koop B, Liebig C, editors. Psychologie und Wirtschaft leben – Aktuelle Themen der Wirtschaftspsychologie in Forschung und Praxis. Mu¨nchen: Rainer Hampp; 2004. pp. 314–19. [6] Nuebling M, Wirtz M, Neuner R, Krause A. Ermittlung psychischer Belastungen bei Lehrkra¨ften. Entwicklung eines Instruments fu¨r die Vollerhebung in Baden-Wu¨rttemberg. Zbl Arbeitsmed 2008;58:312–13. [7] Siegrist J. Adverse health effects of high effort–low reward conditions at work. J Occup Health Psychol 1996;1:27–43. [8] Siegrist J. A theory of occupational stress. In: Dunham J, editor. Stress in the workplace. London: Whurr Publishers; 2001. pp. 52–66. [9] Karasek RA. Job demands, job decision latitude and mental strain: implications for job redesign. Admin Sci Q 1979;24:285–308. [10] Karasek RA, Theorell T. Healthy work. Stress, productivity, and the reconstruction of working life. New York: Basic Books; 1990. [11] Theorell T, Hasselhorn HM. On cross-sectional questionnaire studies of relationships between psychosocial conditions at work and health – are they reliable? Int Arch Occup Environ Health 2005;78:517–22. [12] Frese M, Zapf D. Methodological issues in the study of work stress: objective vs. subjective measurement of work stress and the question of longitudinal studies. In: Cooper CL, Payne R, editors. Causes, coping and consequences of stress at work. New York: Wiley; 1988. pp. 375–411. [13] Kristensen TS. Job stress and cardiovascular disease: a theoretic critical review. J Occup Health Psychol 1996;1:246–60. [14] Kompier MAJ, Cooper CL, editors. Preventing stress, improving productivity. European case studies in the workplace. London: Routledge; 1999. pp. 33–51.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 125–136

ORIGINAL ARTICLE

Psychosocial risk exposures and labour management practices. An exploratory approach ´ S2, ERNEST CANO3, ARIADNA FONT1, PERE JO ´ DAR4, CLARA LLORENS1,2, RAMON ALO 1 5 3 5 ´ PEZ , ALBERT NAVARRO , AMAT SA ´ NCHEZ , MIREIA UTZET & VICENTE LO 1 SALVADOR MONCADA 1

Union Institute of Work Environment and Health (ISTAS), Barcelona and Valencia, Spain, 2Department of Sociology, School of Political Science and Sociology, Autonomous University of Barcelona (UAB), Spain, 3Department of Economics, School of Social Sciences, University of Valencia (UV), Spain, 4Department of Political Sciences and Sociology, Pompeu Fabra University (UPF), Barcelona, Spain, and 5Biostatistics Unit, School of Medicine, Autonomous University of Barcelona, Spain

Abstract Aim: The purpose was to explore the relationship between psychosocial risk exposures and labour management practices (LMP), as indicators of work organization and pertinent features for primary preventive intervention. Methods: Cross-sectional study of a representative sample of salaried working population in Spain (n ¼ 7,612). Information was obtained in 2004-2005 using a standardized questionnaire administered through personal interviews at the household. Questions on working conditions were used to establish LMP indicators and the psychosocial exposures data were obtained on the basis of the Copenhagen Psychosocial Questionnaire (COPSOQ) I (ISTAS21). A multivariate description was performed through multiple correspondence analysis, and associations between LMPs and psychosocial exposures were assessed by ordinal logistic analysis adjusting for age and sex. Results: Correspondence analysis showed a good-bad coherent pattern regarding both psychosocial dimension and LMPs, though several LMPs categories were placed in the centre. Among the 14 possible associations of each psychosocial scale with LMP variables, several scales showed significant associations with more than eight LMP variables. Most relevant results referred to the LMP variable ‘‘Consultative and delegative participation in methods’’. Conclusions: In line with previous research, psychosocial exposures were associated with LMP. LMP may constitute a step on a pathway from work organization to health. Our exploratory work suggested that good psychosocial exposures were related to participatory working methods, being hired with a permanent labour contract, not being made to feel easily replaceable, having superiors with non-authoritarian and non-aggressive manners, not being threatened with dismissal, upward functional mobility, being paid according to the number of working hours and occupation, working between 31 and 40 hours per week and in regular morning shifts. Hence, the more these features became part of LMP in the workplace, the better the psychosocial work environment would be.

Key Words: COPSOQ, employment status, labour management practices, occupational health, pay, primary prevention, psychosocial work environment, stress, work organization, work process, working time

Background and aims An extensive body of scientific evidence based on Karasek-Theorell-Johnson-Hall’s demand-controlsupport model and Siegrist’s effort-reward model has documented labour psychosocial risk exposures as work organization conditions that can damage

workers’ health, within a range of adverse effects, from cardiovascular diseases to mental illness[1,2]. Numerous studies have taken economic sectors, occupations, or socio-demographic features as independent variables, showing important exposure inequalities regarding class, gender, age or ethnic group.

Correspondence: Clara Llorens, Salvador Moncada, Instituto Sindical de Trabajo, Ambiente y Salud (ISTAS), Vı´a Laietana, 16, 6a. 08003 Barcelona. Tel: (þ34) 93 4812835. Fax: +34934812770. E-mail: [email protected]; [email protected] (Accepted 14 October 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809354363

126

C. Llorens et al.

Such evidence proved useful to describe exposures. However, fewer studies have examined the impact of work organization on these psychosocial risk exposures [3], thus hindering the possibilities for researchers to propound effective and lasting preventive action [4]. Work organization includes many aspects of the way work is designed, organized and managed, which have been the subject of numerous research studies. Occupational health researchers agree on targeting the organizational context when working conditions are explored [5]. In this paper, following the segmentation theory [6], labour management practices (LMP) are understood as a set of strategic actions at company level aimed at recruiting, promoting, rewarding, using, developing and keeping or dismissing workers (i.e. work process design, working time, employment, pay, training or communication practices). LMP are the result of management strategies used to achieve workers’ flexibility, adequate productive performance and profitability [7], being factors that could determine psychosocial exposures as intermediate outcomes that may in turn lead to ill health. From this theoretical perspective, LMP are both influenced by social, institutional and economic systems (macro-level features) and in turn are key factors in shaping the very systems’ settings. In addition, LMP vary according to occupation, sex, age or ethnic group. Thus, LMP may use, reproduce and strengthen social inequalities. Nonetheless, this paper will not explore LMP segregation and will only focus on the workplace – in contrast to macro level aspects. We put LMP at the centre of the analysis of organizational working conditions because they help us to enforce the perspective of the organizations, the workplaces and the social and technical aspects of the jobs in which people are employed. This approach reverses the usual perspective of preventive practice focused on individual issues [8]. Although research is still limited, increasing evidence from intervention, longitudinal and cross-sectional studies [9,10] suggests that many LPM factors predict exposure to psychosocial risks. Such LPM factors include work process design, working time, type of employment contract, wages and pay structure. However, the number of available studies varies depending on the subject. In spite of such efforts, research is still necessary for a better understanding of trends in organizational practices [4,11] and their influence on psychosocial exposures. This is a fact in a field where the majority of researchers agree that organizational changes ought to be fostered from the efficiency point of view regarding psychosocial risk prevention. However, prevalent in-company practice continues to be

dominated by individual-focused strategies supported by a commercial boom of the stress industry. Work process design (tasks and methods) is associated with psychosocial risks. Furthermore, it has been shown that task restructuring interventions that ignore the potential health impact of the interventions may eventually be damaging [12]. On the one hand, most published studies have documented adverse effects of Taylorism on psychosocial risks and health [1,13]. On the other hand, research on task variety and participatory formulas has found them to be open situations that can be solved either in an instrumental-competitive way (which involves a deterioration of working conditions) or in a democratic solidarity-based and fair approach, which leads to an improvement of working conditions that includes job enrichment and cooperative relations [14]. Research has suggested that the contractual relationship affects health and psychosocial exposures. Temporary workers suffer higher psychological distress [15] and more musculoskeletal disorders, fatal injuries, non-fatal injuries and premature mortality in comparison with permanent workers. Studies have suggested that the relationship between temporary employment and health may reflect the adverse effect of job insecurity [16] and other psychosocial exposures related to underemployment [17]. Studies on working time (length, schedules and changes) and health are far more abundant than studies on the relation between working time and psychosocial risks. Shift work has been intensively researched, compared with regular daytime work, and it has been associated with an increasing number of diseases [18,19]. Working more than 40 hours per week has been associated with CVD, anxiety and depression, which might hint at dose-response relationship [20]. Also, factors such as influencing work schedules, decreasing working hours at own request or having the possibility to take a day off, have been proved to reduce work-home interference [21] and even cardiovascular risk factors [22]. We could not find many occupational health studies dealing with salaries and pay structure. Research on piece-rate work was found to be associated with long-term work disability due to CVD, musculoskeletal diseases and injuries [23], compared with fixed pay work, and also with higher job demands. The lack of studies of more modern performance-based pay (i.e variable pay) in this field is remarkable. Low salaries have been related to poor health and psychosocial well-being [24] but with the associations may be confounded by poverty factors. To conclude, we can rely on the plentiful empirical evidence to suggest that labour management practices, considered as indicators of work organization,

Psychosocial risk exposures and labour management practices may be determinants for psychosocial exposures and health. However, knowledge on the topic is not sufficient to state which LMPs are important for psychosocial risk exposures because the relationship is complex, with many levels and no research agreement, and thus it remains a vast field to explore. Joining recent calls to incorporate work organization into occupational health research [4], the purpose of this paper is to explore the association between psychosocial risk exposures and labour management practices among the salaried working population in Spain. We are interested in labour management practices as pertinent factors to design primary preventive interventions, to ‘‘stop stress at its origins’’ [25], addressing ‘‘the way we work’’ [10].

Methods Design, study population and sampling This is a cross-sectional study carried out on the basis of a representative sample of the salaried working population in Spain. The main purposes of the study were to obtain Spanish reference values for all Copenhagen Psychosocial Questionnaire (COPSOQ) I scales and to analyze psychosocial exposures at workplaces in Spain. For these purposes, sample size was established in 7,650 individuals, and included individuals resident in Spain, aged 16–65, who had worked at least one hour for a wage the week before they were contacted at home. Detailed description of the sampling and field work are available [26,29]. In short, it was a multi-stage sampling by conglomerates. Sample units were: municipalities (stratified by Autonomous Region and size), census units, households and individuals. The selection of households was made by random routing and that of individuals by random numbers. Information was obtained by a questionnaire handed out during personal interviews in the respondents’ own homes. The whole fieldwork was carried out in three phases between October 2004 and July 2005. Fellow workers were excluded from this study so the study population was 7,612.

Measurements The questionnaire included closed questions to collect individuals’ data (age, sex, education and home town) as well as information about their family unit, occupation and working conditions and the 21 scales (73 Likert type items, each with five response categories) of the COPSOQ (ISTAS21) - i.e. the Spanish version of COPSOQ I [27,28].

127

Dependent and independent variables Questions on working conditions were used to construct the LMP indicators while COPSOQ I (ISTAS21) scales were used to describe psychosocial work environment. The data analyses used multiple correspondence analyses of all variables followed by ordinal logistic regression analyses with the LPM indicators as independent variables and the COPSOQ scales as dependent variables. LMP variables The aggregation of LMP indicators was theoretically motivated and started with the selection of the available working-conditions-related questions that could inform on different aspects of LMP as previously defined – management actions aimed at recruiting, promoting, rewarding, using, developing and keeping or dismissing workers. Twenty six questions were selected at this step. Then, a descriptive analysis of all these questions was performed, including the assessment of their discriminating capability and sensibility to show differences according to occupational class, gender and age groups in line with the segmentation theory. Then, a bivariate analysis explored the relationship between all these working conditions and the psychosocial COPSOQ I (ISTAS21) scales. In the second phase, information was analyzed by a Consensus Experts Panel who selected the most appropriate questions to set up the LMP variables. The panel comprised Economy, Sociology, Law and Psychology professors and professionals with expertise in the labour market, industrial relations and labour management, along with three researchers from ISTAS. The panel met three times and reviewed five versions of a consensus document following an echeloned process that started with the discussion of the descriptive and bivariate analysis, deciding on the questions that could eventually be considered and suggesting how to transform variables. Proposals included the reorganization of some response categories, the combination of different questions into new variables and the introduction of filters. Some original questions were dropped at this stage due to low discriminating power, inconsistency or high correlation with other questions. Twenty one original questions were finally used to build up 14 LMP variables that included: three variables on employment relationship (employment status; replacement; and occupation and educational level correspondence), four on working time (weekly working hours; daily working time schedule; weekly distribution of working days; and time availability demands),

128

C. Llorens et al.

three on pay (earnings and number of working hours match; salary structure; and earnings and job match), two on work design (consultative and delegative participation in methods and functional mobility), and two on supervision manners (authoritative and aggressive manners; and threatening with dismissal). One or two ‘‘Well-being LMP categories’’ (which theoretically represented the most favourable, healthy and well-being values) were defined for each LMP variable (see Table I). Variables were categorized as ‘‘good’’ or ‘‘poor’’ on this basis according to correspondence with well being criteria. COPSOQ I (ISTAS21) psychosocial variables Standardized scores of all 21 psychosocial scales were computed (score range 0–100). Double Presence, Quantitative, Sensorial, Emotional and Hiding Emotions Demands, Role Conflict and Insecurity are negative dimensions, so high scores meant poor health and well-being. All the other exposures were positive, so higher scores indicate better results. The original scores were classified into two new variables: first, they were categorized in three exposure levels which were defined and labelled as ‘‘good’’, ‘‘intermediate’’ and ‘‘poor’’ according to the Spanish COPSOQ I (ISTAS21) normative reference values [29], and secondly, we set up a variable with five categories, depending on the quintiles of the original distribution. Statistical analysis Analysis was carried out in two steps. First, a multivariate description of the three categories – poor, intermediate and good – of all 21 COPSOQ I (ISTAS21) psychosocial scales was performed through multiple correspondence analysis along with the projection of the two categories – good, poor – of the 14 LMP variables. This analysis supplies a graphical description of the relationship among the 63 (21  3) psychosocial categories with the 28 (14  2) LMP categories in a multivariate framework. The quantification of inertia was done using Greenacre’s adjustment [30]. In the second step, the 21 psychosocial scales were categorized in quintiles – the higher quintiles corresponding to better scale scores. Afterwards, they were analyzed as dependent variables in ordinal logistic models adjusting for age and sex, [31]. All the 14 LMP variables were included as independent variables using the ‘‘poor’’ category as reference. Associations were estimated by odds ratio, in terms of the odds of being in a higher quintile for each of the COPSOQ I (ISTAS21) scales.

Results Results from correspondence analysis are shown in Figure 1. In the graph, axes are described by psychosocial categories. The closer they are to the extreme of the axes the more they contribute to explaining them. Distance among categories refers to the relationship among them – the closer they are, the stronger the association among them is. A left-right general pattern is shown regarding psychosocial dimensions so that good psychosocial categories appear on the left, intermediate in the centre and the poor ones on the right of the graph. Vertical factor (factor 2) is mainly described by demands dimensions while the horizontal one (factor 1) describes control dimensions and social support ones. Rewards are distributed as follows: insecurity is in the vertical factor and esteem in the horizontal one. Inertia of factor 1 was 64.4% and factor 2 was 19.3%. LMP categories’ projection showed a similar left (good categories)– right (poor categories) pattern except for salary structure, time availability demands and earnings and job match, which showed the opposite pattern. Moreover, several categories were placed centrally. Consultative and delegative participation in methods; Earnings and number of working hours match; Replacement; Threatened with dismissal; Authoritative and aggressive manners; Occupation & educational level correspondence; and Functional mobility were those with the largest distance between good and poor categories, while the rest showed closer good and poor categories. LMP good category of Consultative and delegative participation in methods was the closest to the psychosocial dimensions’ good categories of Influence, Possibilities for development, Working time control, Commitment to the workplace, and Cognitive demands, whereas the poor Consultative and Delegative participation LMP category was close and in the same quadrant as the bad categories of the same dimensions. Threatened with dismissal and Replacement poor LMP categories were closer to the psychosocial dimensions’ poor categories of Role clarity, Esteem, Sense of community, and colleagues’ support. Furthermore, the good categories of the same LMP were in the same quadrant of the same good exposures. LMP good categories of Occupation and educational level match and employment status were close to Role clarity intermediate exposure, whereas Functional mobility was next to intermediate exposures of supervisors’ support and meaning of work and the poor category of Insecurity. The good category of Weekly distribution of working days was close to intermediate exposure of Esteem and colleagues’ support.

7 closed – all typologies of possible contractual relations in Spain

5 in Likert’s frequency scale (always to never) 8 (closed) – all typologies of Spanish educational levels

3 (closed): above/at the same level than/ below

Quantitative continuous response

7 (closed) – Morning shift / Split shift / Afternoon shift/ Night shift / Shift without night / shift including night / Irregular

5 closed - Monday to Friday / Monday to Saturday / Only week-ends & holidays / Monday to Friday & exceptionally Saturdays and holidays / Any working day or holy day

Do they make you feel easily replaceable at work?

What’s the highest degree of education you have completed?

The work you do. is your educational level?

How many hours did you work for your company last week?

What is your daily working time schedule?

What is your schedule regarding the days of the week you work?

Response categories

What kind of relation do you have with the company that hires you?

Question

Up to 20 21–30 31–35 36–40 41–45 46 or more Morning shift Split shift Afternoon shift Night shift 24 hrs. shifting Irregular Monday to Friday Monday to Saturday Only weekends & holidays Monday to Friday & exceptionally Saturdays and holidays Any working day or holy day

Daily working time schedule

Weekly distribution of working days

Middle level education & non qualified job Higher education & job Higher education & non qualified job

Elementary education and job Elementary education & non qualified job Middle level education & job

Make worker feel easily replaceable Do not make worker feel easily replaceable

Permanent labour relation (civil servant, permanent labour contract) Precarious labour relation (temporary contracts, no contract) Trade relation (dependent self-employment)

Variable categories

Weekly working hours

Occupation & educational level correspondence

Replacement

Employment status

Transformed variable

Table I. Questions, transformed variables, categories, and well-being categories to characterize labour management practices.

14.33 1088

(continued)

59.79 21.80 1.09 2.99 4540 1655 83 227

X

23.6 56.6 4.1 1.7 7.4 6.7

6.33 6.01 8.49 57.24 8.37 13.56

7.83 18.02 4.24

30.96 2.94 36.00

1789 4293 310 127 562 510

479 455 643 4335 634 1027

567 1305 307

2242 213 2607

14.1 85.9

3.84

292 1075 6523

28.52

2169

% 67.64

n 5144

X

X X

X

X

X

X

Well-being category

Psychosocial risk exposures and labour management practices 129

Earnings and job match

Consultative & delegative participation in methods

Quantitative continuous

11 closed. Ordinal scale

2 closed –Yes / No

3 closed – Fixed salary / partly variable / all variable Quantitative continuous

4 closed – Yes / No. Work above salary category/ No. Work below salary category / Do not know 5 in Likert’s frequency scale (always to never)

5 in Likert’s frequency scale (always to never)

How many hours did you work last week?

How much is your net monthly salary?

Are you paid enough for the job you do?

Do you have a fixed or variable salary?

If variable what percentage varies regardless of overtime pay?

Does the job you do correspond with your professional salary category?

During last year did managers or supervisors consult you about how to improve the way of developing products or giving services?

On a daily basis can you decide on the methods and tools you use?

Salary structure

Earnings and number of working hours match

4 closed –Yes (number of hours) / No. I never work overtime/ No I did not work overtime last week but I usually do.

If you worked overtime last week: how many hours did you work?

Time availability demands

Transformed variable

4 closed – My working time schedule or week days is not modified / I am notified of changes the day before / . . .with (number of days) notice / I usually know my work schedule but it can be modified with one day’s notice

Response categories

Are your working days or daily schedules changed? How long in advance are you notified about changes?

Question

Table I. Continued.

Can Can Can Can Can

decide sometimes & is consulted sometimes decide sometimes & is always consulted always decide & is not consulted always decide & is consulted sometimes always decide & is always consulted

Cannot decide & is not consulted Cannot decide & is consulted sometimes Can decide sometimes & is not consulted Cannot decide & is always consulted

Consistent Inconsistent

11–25 % variable 25% or more variable

Fixed salary Up to 10% variable

Inconsistent

Consistent

No availability demands Availability demands

Variable categories

X

X

X

X

X

X

Well-being category

89.81 10.19

19.30 10.60 5.86 10.20 6.25 5.88 11.06 10.84 20.01

1453 798 441 768 471 443 833 816 1507

2.77 8.52

85.72 2.99

70.79

29.21

56.31 43.69

%

6506 738

209 642

6461 225

4859

2005

3888 3017

n

130 C. Llorens et al.

82.0

Threatened with dismissal 5 in Likert’s frequency scale (always to never) Are you afraid of losing your job if you do not do what you are told to do?

Frightened by being threatened with dismissal for not doing what is requested Not frightened by being threatened with dismissal

X

6231

18.0 1364

7.80 92.20 Authoritative and aggressive manners Neither authoritative nor aggressive manners 5 in Likert’s frequency scale (always to never) Are you treated in an authoritative and aggressive manner at work?

Authoritative and aggressive manner 4 closed - higher / lower / of the same level / do not know (If above answered 1 or 3) The tasks you perform beyond your work post duties are of a level

Downward functional mobility Performs other tasks of the same level

X

592 6997

3.24 7.78 244 585

86.71 2.27 6522 171 X Functional mobility 3 (closed) – Yes but is not usual / No (skip next question) / I usually do more than one job/ task or I am in more than a work post During the last four weeks. did you perform any other tasks apart from those described above? – Previous two questions are on the description of the job and their usual tasks.

Does not perform any other tasks Upward functional mobility

Transformed variable Response categories Question

Table I. Continued.

Variable categories

Well-being category

n

%

Psychosocial risk exposures and labour management practices

131

Surprisingly, the poor LMP category of time availability demands was very close to good double presence category. Ordinal logistic regression results are shown in Table II. When interpreting Table II it must be borne in mind that the ORs are between LMP variables (with the poor category as the reference) and quintiles of COPSOQ I (ISTAS21) psychosocial scales for which the worst quintile is the reference. For instance, the association between Consultative and delegative participation in methods and Influence (OR ¼ 2.2) is a 120% increase in the odds of having better influence, but the OR would be 23.4 (2.24) if we consider the lowest and highest quintiles. Among the 14 possible associations of each psychosocial scale with LMP variables Influence showed 12 statistically significant associations; Role clarity and Control over working times 11; Insecurity and Esteem 10; Meaning of work, Commitment to the workplace, Predictability, and Sense of community nine. The psychosocial scales with fewest significant associations were Quantitative demands, with two associations and Demands for hiding emotions, with three. Consultative and delegative participation in methods was the LMP variable which showed most statistically significant associations with psychosocial scales – 19 out of 21; followed by Authoritative and aggressive manners and Threatened with dismissal with 17 and 15 respectively. The LMPs with the least number of significant associations were Daily working time schedule with four, and Functional mobility with five significant associations. Consultative and delegative participation in methods and Threatened with dismissal were the LMP variables that showed stronger associations with psychosocial scales – 14 and eight out of 21 had ORs higher than 1.4. Most of the results were in line with expectations but some results were unexpected. For instance, good weekly distribution of working days, fixed salary, and occupational and educational level match were not associated with a more favourable exposure in 10, six, and seven cases respectively. Among the psychosocial dimensions, double presence, insecurity and role clarity were not associated with good LMP in five and in four cases respectively.

Discussion and conclusions We assumed that psychosocial exposures were rooted in organizational conditions rather than in individual characteristics. Results support the hypothesis that psychosocial exposures were associated with LMP variables, but these associations varied across LMP

132

C. Llorens et al.

Factor 2 0.8

LMP variables categories: Employment Status (lmp1); Replacement (lmp2); Occupation & educational level correspondence (lmp3); Weekly working hours (lmp4); Daily working time schedule (lmp5); Weekly distribution of working days (lmp6); Time availability demands (lmp7); Earnings and number of working hours edP match (lmp8); Salary structure (lmp9); Earnings and job match (lmp10); Consultative & delegative participation in methods (lmp11); Functional mobility (lmp12); Authoritative and aggressive manners (lmp13); Threatening with dismissal (lmp14) Final “G”: Good, final “P”: Poor.

rcoP qcP

Imp13P

dheP

cdG wcl

sdl infG 0.4

cwtG

wcG

Imp10G qll

sssl

Imp11G

Imp12G

pdG

pdl cdl

Imp8G

Imp3G Imp9P

Imp4G mwG

estP

insl psrP cssl

scl

dhel

cwtl

scP

psrl

Imp1G Imp6G Imp5P rcll Imp7P estl Imp4P Imp7G Imp12P qcl Imp2G dpG edl Imp14G Imp10PImp9G Imp13P Imp8P Imp6P dpl Imp5G Imp1P

0

Imp14P

prl

insP mwl

sdP

rclP Imp2P

infl

psrP cssP

rcol

prP

Imp11P

qlP mwP sssP

glG Imp3P

estG cssG sssG scG

psrG

–0.prG rclG Psychosocial dimensions categories: Quantitative demands (qd); Emotional demands (ed); Demands for hiring emotions (dhe); Cognitive demands (cd); Sensory demands (sd); Meaning of work (mw); Influence (inf); Freedom at work (cwt); Possibilities for development (pd); Commitment to the workplace (wc); Predictability (pr); Role conflict (rco); Role clarity (rcl); Possibilities for social relations (psr); Social support from supervisors (sss); Social support from colleagues (css); Sense of community (sc); Quality of rcoG –0.8 (ql); Insecurity (ins); Esteem (est); Double presence (dp). Leadership Final “P”: Poor, final “I”: Intermediate, final “G”: Good.

cwtP insG wcP

infP sdG

gcG

dheG

pdP

edG

cdP –0.8

–0.4

0

0.4

0.8 Factor 1

Figure 1. Labour management practices and psychosocial work environment (results from correspondence analysis). Wage earner population, Spain 2005 (n ¼ 7,612).

variables and psychosocial scales. LMPs may constitute a step on a pathway from work organization to health. Although the 21 scales measure part of the psychosocial work environment and share the same basic theoretical framework, the causal pathways linking each of them to LMP could differ partially. Most relevant results refer to the LMP variable Consultative and delegative participation in methods. This variable showed the most numerous and the strongest associations with psychosocial dimensions concerning control (influence, possibilities for development, meaning of work, freedom at work and commitment), social support (supervisors’ and colleagues’ support and quality of leadership, sense of community, role conflict and role clarity) and rewards (esteem). These results were consistent with previous research showing how democracy at the workplace and the implementation of direct participation formulas could lead to better psychosocial working environment [1,12–14] (see introduction). LMPs should take into account workers’ abilities and knowledge, and their basic needs of learning and autonomy as direct participation formulas. Such LMPs could significantly reduce or

eliminate part of psychosocial hazards in Spain [32,33], where Taylorism is an ever-present phenomenon that ignores workers as professionals and human beings [34]. The same arguments could be suggested for the relation between functional mobility and possibilities for development, meaning of work, influence and role clarity. The association between Threatened with dismissal and Demands for hiding emotions showed that emotional demands could have to do not only with the nature of the tasks but with LMPs. The results suggested that a change from precarious to permanent labour relations or diminishing workers’ feelings of being easily replaceable may improve influence and freedom at work and the majority of social support psychosocial dimensions, as well as esteem. One possible explanation could be that in spite of the legal framework, the current use of precarious contractual relations by managers in Spain is arbitrary and aims to promote more availability in terms of working time and tasks and a cheaper workforce. These precarious conditions can take place since workers with such a contractual relationship are vulnerable to unilateral termination of contract by

*p < 0.05; **p < 0.001.

Double presence Employment status 0.989 Replacement 0.962 Occupational & educational level 0.901 (**) correspondence Weekly working hours 0.812 (**) Daily working time schedule 1.024 Weekly distribution of working 0.923 (*) days Time availability demands 1.078 (*) Earnings and number of working 0.992 hours match Salary structure 1.068 Earnings and job match 0.804 (**) Consultative & delegative partic- 0.935 ipation in methods Functional mobility 0.918 Threatened with dismissal 0.992 Authoritative and aggressive 0.836 (**) manners Meaning of work Employment status 1.000 Replacement 1.085 Occupational & educational level 1.130 (**) correspondence Weekly working hours 1.115 (*) Daily working time schedule 1.028 Weekly distribution of working 0.950 (*) days Time availability demands 0.994 Earnings and number of working 1.349 (**) hours match Salary structure 0.873 (**) Earnings and job match 0.989 Consultative & delegative partic- 1.631 (**) ipation in methods Functional mobility 1.305 (**) Threatened with dismissal 1.484 (**) Authoritative and aggressive 1.118 (**) manners 1.103 (*) 1.067 0.943 0.931 0.898 (*) 1.247 1.311 (**) 1.275 1.301 (**) 0.911 1.275 (**) 1.120 1.048 1.142 (*) Insecurity

Role clarity 0.926 (**) 0.983 1.051 0.945 1.079 (**) 0.882 (**) 0.997 1.177 (**) 1.074 (*) 0.980 1.068 (*) 1.058 (*) 1.088 (**) 0.878 (**) 0.918 (**) 0.954 (*) 1.025 0.938 (**) 0.883 (**) 1.271 (**) 1.140 (**) 1.073 (*) 0.871 (**) 0.948 1.135 (**) 0.907 (*) 0.987 1.069 (**) 1.486 (**) 1.467 (**) 0.952 1.431 (**) 0.996 0.902 (*) 2.293 (**) 1.291 (**) 0.759 (**) 1.145 (**) 1.153 (**)

1.029 1.022 1.077 1.189 (**) 1.115 (**) 1.283 1.011 0.922 Commitment to the workplace 1.085 (**) 1.246 (**) 1.145 (**) 1.125 (**) 0.986 0.943 (*) 0.978 1.270 (**) 0.927 (*) 1.030 1.687 (**) 1.057 0.983 1.081 (*)

1.109 1.458 (**) 1.134 (*)

0.933 0.940 0.970

1.026 1.033

0.958 1.054 1.015

Demands for hiding emotions 0.962 0.895 0.818 (**)

1.048 1.413 (**) 1.079 (*)

1.006 0.984 1.410 (**)

1.025 1.057 (*)

1.107 (**) 0.990 0.935 (**)

Predictability Social support from colleagues 1.014 1.080 (**) 1.245 (**) 1.181 (**) 0.994 1.080 (**)

0.844 (**) 0.897 1.148 (**) 1.029 1.523 (**) 1.135 (**)

1.009 0.968 1.374 (**) 0.937

0.974 1.152 (**) 1.073

1.119 1.052 1.193 (**) 0.953 0.868 (**) 1.024

Emotional demands 1.016 0.909 0.764 (**)

1.114 1.057 1.003

Cognitive demands 1.056 (*) 0.982 1.302 (**)

Sensory demands 0.958 1.029 0.883 (**)

Quantitative demands 1.017 1.009 0.972

1.027 1.418 (**) 1.252 (**)

0.960 0.963 1.684 (**)

0.924 (**) 1.045

1.036 1.031 0.944

Social support from supervisors 1.067 (*) 1.381 (**) 1.104 (**)

1.172 (*) 0.800 (**) 1.162 (**)

0.788 (**) 1.175 (**) 2.203 (**)

0.940 (*) 1.363 (**)

0.929 1.011 1.069 (**)

1.110 (**) 1.186 (**) 1.1124 (**)

Influence

0.958 1.427 (**) 1.136 (**)

1.048 0.964 1.1067 (**)

1.044 (*) 1.033

1.071 (*) 0.985 0.924 (**)

Possibilities for social relations 1.009 1.008 1.022

0.952 0.818 (**) 1.202 (**)

0.852 (**) 1.120 (**) 1.556 (**)

0.905 (**) 1.085 (**)

0.995 0.958 1.047 (*)

Freedom at work 1.108 (**) 1.132 (**) 1.087 (**)

1.163 1.941 (**) 1.158 (*)

1.074 0.862 (*) 1.601 (**)

0.992 1.241 (**)

1.045 0.968 0.880 (**)

Sense of community 1.125 (**) 1.250 (**) 1.138 (**)

0.890 1.031 1.336 (**)

1.038 1.152 (*) 1.502 (**)

0.932 0.930

0.964 1.015 1.016

Role conflict 0.916 1.196 (*) 0.901 (*)

1.020 1.117 1.164 (**)

1.054 0.921 1.637 (**)

0.946 (*) 0.987

1.037 1.056 0.951

Quality of leadership 1.062 (*) 1.540 (**) 1.097 (**)

1.465 (**) 1.277 (**) 1.062

0.833 (**) 1.164 (**) 1.709 (**)

1.028 1.375 (**)

1.061 0.977 1.012

Possibilities for development 1.009 1.022 1.231 (**)

1.051 1.469 (**) 1.263 (**)

1.049 0.881 (**) 1.589 (**)

0.926 (**) 1.107 (**)

0.997 1.034 0.935 (**)

1.086 (**) 1.584 (**) 1.105 (**)

Esteem

Table II. Association between COPOSQ ISTAS21 psychosocial dimensions and labour management practices (LMP). Odds ratios from age and sex adjusted ordinal logistic regression models. Dependent variables: psychosocial dimensions (quintiles). Independent variables: LMP. Wage earning population. Spain 2005 (n ¼ 7,612).

Psychosocial risk exposures and labour management practices 133

134

C. Llorens et al.

managers [35]. Associations of these three LMPs and social support dimensions suggest that support at work or interpersonal relations could have to do with LMPs. Management leadership styles and their organizational prerequisites are currently becoming a topic in occupational health research [36]. Studies have suggested that participatory, supporting and fair management may be predictors of healthy workplaces, while laissez faire, autocratic or abusive leadership styles may promote the opposite [37]. Our results regarding authoritative and aggressive manners are consistent with this previous research. The inconsistent associations related to working time and salary LMPs should be the object of further research. Our partial exploration has proved insufficient to understand such complex relationships. But some arguments could still be suggested. The inconsistent associations on working time LMPs and double presence (i.e changing to social schedules and to social number of working hours increased exposure) may be explained by the fact that double presence in this study measures not only synchronic demands but domestic and family workload, which increases with conciliating timetables. Inconsistent associations between employment status, time availability demands, earnings and number of hours match, threatened with dismissal and authoritative and aggressive manners and insecurity (the best LMP, the worst psychosocial exposure) are understandable. Since insecurity refers to worries about changes of valued working conditions (tasks, working time arrangements, wage, job or employment) [16] against workers’ will, this could be worse among those who actually have valued working conditions now and could lose them, due to the expansion of precarious working conditions in Spain over the past decades [35]. Inconsistent results on time availability demands and salary structure (i.e the best LMP, the worst psychosocial exposure) could have to do with the fact that time availability demands and variable pay are more frequently used among workers in higher SES jobs who in turn are subjected to good categories of all the other LMPs. Class adjustment would have improved these results but would have affected others. LMPs are class segregated [6] so any class adjustment would become an over-adjustment. Lower frequency of associations between LMP and demands may be explained by the fact that qualitative demands are more affected by the nature of tasks than by LMP and quantitative demands are probably associated with LMPs that were not analyzed due to lack of data.

Limitations The study has some limitations. Among the LMP original variables there was a lack of information on determinants of work pace, workers’ promotion, team-work, in-company training and communication. However, data included a significant part of the LMP components that are relevant from an occupational health perspective [9–23]: recruitment practices (type of contract, replacement, job and educational level consistency); working time practices (length, daily and weekly schedules and availability demands); pay practices (income and consistency with pay determinants and pay structure); working methods (direct participation formulas and functional mobility); and managers’ manners (see Table I). Companies’ contextual information was not included in the analysis because the purpose of this research was to study the relationship between psychosocial risk exposures and LMP and not to understand the determinants of the LMP. LMP and psychosocial scales could have been partially and inexactly measured and this could affect the results. For example, the LMP identifications could be biased by individual features since they derived from workers’ questionnaires. However, most of the 21 questions used referred either to objective facts with closed answers – i.e. type of contractual relationship – or to Likert-type scales, which are usually used to ask about intangible concepts. The population study was compared with the Spanish Active Population Survey (EPA) during the same period of time and no evidence of bias was detected [26]. We may assume that respondents’ perception probably had very little effect on the data used for research. The selection of one or two categories of each of the 14 LMP variables as the Well-being category could be argued. In any case, this decision was taken considering the conceptual framework [9–23], Health-related evidence and data. Each individual was classified into one of the two good/poor categories of each LMP variable. The absence of published evidence on the relationship among all these LMP variables and psychosocial exposures justifies this procedure. It should be accepted as an approximation although the obtained results support its usefulness. Psychosocial exposures were measured using a known, reliable and validated instrument. Then they were classified in three categories – poor, intermediate, and good according to their reference values, which follow a theoretical tertile distribution [29]. Another alternative would have been to group

Psychosocial risk exposures and labour management practices exposures in two sets, above or below the median, as done in many research studies. Nonetheless, this alternative would not have improved the understanding of results since the trend pattern would have been weaker. The cross-sectional design does not allow us to make any interpretation about the direction of the association. However, it is highly unlikely that results could be spurious or attributable to bias. Alternative methods would have influenced the magnitude of differences and the strength of associations but not the fact that LMP could be related to psychosocial risk exposures. Accordingly, we can state that good psychosocial exposures are related to participatory (consultative and delegative) working methods, being hired with a permanent labour contract, not being made to feel easily replaceable, non-authoritarian and non-aggressive management style, not being threatened with dismissal, with upward functional mobility, being paid an amount of money matching the number of working hours and occupation, working between 31 and 40 hours per week and in regular morning shifts. Thus, the more these features became part of LMP in the workplace, the better the psychosocial work environment would be. The results suggest that Taylorist methods (labour division between design and execution, fragmentation and standardization of tasks and individual performance-based pay); participative Taylorism characteristics (lean production, participation and development without standardization declining); and unsocial schedules, low salaries, non-permanent contracts, and demands for availability on working conditions (such as working time, wage, contract, or tasks), which turn into inflexibility and precariousness for workers, should be limited if we seek to reduce harmful labour psychosocial risk exposures at the workplace. In this study we explored how COPSOQ’s psychosocial exposures relate to key aspects of how work is organized and managed. These key aspects are decided by either employers or managers at workplace level. According to Spanish health and safety legislation, which is based on the European Framework Directive 89/391, if risk assessment proves that working conditions derived from work organization are hazardous to health, such working conditions must be changed at source and with the participation of workers’ representatives. Our rationale was that if adverse psychosocial exposures need to be reduced or eliminated then these kind of analyses may support workers’ representatives and occupational health professionals to propose more accurate and efficient primary preventive interventions at the workplace, taking into account work

135

organization aspects that are mainly decided at this level, and so could also be changed at this level. The theoretical and empirical question behind this research concerns whether and how certain LMPs are more likely to lead to psychosocial exposures that in turn, lead to poor health indicators. Further research is needed. Used LMP variable definitions and measurements need to be improved and missing relevant information must be obtained. More longitudinal and company-level intervention studies are needed for conceptual and evidence clarification on which LMPs are more determinant in fostering psychosocial exposures. Future research on psychosocial risks should include LMP data collection and analysis to better understand causal pathways linking psychosocial exposures to LMP. This will help to promote effective and lasting workplace preventive actions, that is to say, prevention at source, ‘‘addressing the way we work’’.

Acknowledgements This research was funded by the Fondo de Investigacio´n Sanitaria (Health Research Fund) of the Instituto de Salud Carlos III (Spanish Ministry of Health, PI031499 and PI06152 projects). We especially thank the co-editor and reviewers for their improvement recommendations and Agustin Gonza´lez, Rosa Carre´ and Bronagh Reade for reviewing the English version.

Conflict of interests There is no conflict of interest.

References [1] Belkic K, Landsbergis P, Schnall P, Baker D. Is job strain a major source of cardiovascular disease risk? Scand J Work Environ Health 2004;30(2):85–128. [2] Stansfeld S, Candy B. Psychosocial work environment and mental health – a meta-analytic review. Scand J Work Environ Health. 2006;32(6):443–62. [3] Harestam A, Marklund S, Berntson E, Bolin M, Ylander J. Understanding the organisational impact on working conditions and health. Stockholm: National Institute for Working Life; 2006. [4] MacDonald LA, Ha¨renstam A, Warren ND, Punnett L. Incorporating work organisation into occupational health research: an invitation for dialogue. Occup Environ Med 2008;65(1):1–3. [5] Ha¨renstam A. Organizational approach to studies of job demands, control and health. Scand J Work Environ Health 2008;Suppl(6):144–9. [6] Rubery J. Developing segmentation theory: a thirty years perspective. E´conomies et Socie´te´s 2007;6:941–64.

136

C. Llorens et al.

[7] Recio A. La segmentacio´n del mercado de trabajo en Espan˜a [The segmentation of labour market in Spain]. In: Migue´lez F. y Prieto C, editors. Las relaciones laborales en Espan˜a [The industrial relationships in Spain]. Madrid: Siglo XXI; 1991. [8] Kompier MAJ, Kristensen TS. Organizational work stress interventions in a theoretical, methodological and practical context. In: Stress in the workplace: past, present and future. London: Wurr Publishers Ltd; 2001. [9] Lamontagne AD, Keegel T, Louie AM, Ostry A, Landsbergis PA. A systematic review of the job stress intervention evaluation literature: 1990–2005. Int J Occup Environ Health 2007;13(3):268–80. [10] California Public Policy Forum: The way we work and its impact on our health. Scientific Summaries Paper. Los Angeles: UCLA 2004. [11] Sauter S.L. Brightwell WS, Colligan MJ, Hurrell JJ, Katz TM, LeGrande DE. The changing organisation of work and the safety and health of working people. Knowledge gaps and research directions. DHHS (NIOSH); 2002. p 116. [12] Bambra C, Egan M, Thomas S, Petticrew M, Whitehead M. The psychosocial and health effects of workplace reorganisation. 2. A systematic review of task restructuring interventions. J Epidemiol Community Health 2007;61:1028–37. [13] Johnson JV, Johansson G, editors. The psychosocial work environment: work organization, democratization, and health. Amityville (NY): Baywood Publishing Company; 1991. [14] Hvid H, Lund H, Pejtersen J. Control, flexibility and rhythms. Scand J Work Environ Health 2008;Suppl.6:83–90. [15] Virtanen M, Kivima¨ki M, Joensuu M, Virtanen P, Elovainio M, Vahtera. Temporary employment and health: a review. Int J Epidemiol 2005;34(3):610–22. [16] Ferrie JE, Westerlund H, Virtanen M, Vahtera J, Kivima¨ki M. Flexible labor markets and employee health. Scand J Work Environ Health 2008;Suppl.6:98–110. [17] Benach J, Amable M, Muntaner C, Benavides FG. The consequences of flexible work for health: are we looking at the right place? J Epidemiol Community Health 2002;56(6):405–6. [18] Knutsson A. Health disorders of shift workers. Occup Med (Lond). 2003;53(2):103–8. [19] Kolstad HA. Nightshift work and risk of breast cancer and other cancers – a critical review of the epidemiologic evidence. Scand J Work Environ Health 2008;34(1):5–22. [20] Keppla E, Sanne B, Tell, G. Working overtime is associated with anxiety and depression: the Hordaland Health Study. J Occup Environ Med 2008;50:658–66. [21] Jansen NW, Kant I, Nijhuis FJ, Swaen GM, Kristensen TS. Impact of worktime arrangements on work-home interference among Dutch employees. Scand J Work Environ Health 2004;30(2):139–48. [22] Viitsalo K, Kuosma E, Laitinen J, Ha¨rma¨ M. Effects of shift rotation and the flexibility of a shift system on daytime alertness and cardiovascular risk factors. Scan J Work Environ Health 2008;34(3):198–205.

[23] Punnett L, Brisson Ch. Piece-rate work. In: California Public Policy Forum: The way we work and its impact on our health. Scientific Summaries Paper. Los Angeles: UCLA 2004. [24] Karlqvist LK, Ha¨renstam A, Leijon O, ; Sche´ele P, Research Group MOA. Excessive physical demands in modern worklife and characteristics of work and living conditions of persons at risk. Scand J Work Environ Health 2003; 29(5):363–77. [25] Collins S, Landsbergis P, Warren N, LaMontagne AD. Stopping stress at its origins: addressing working conditions. Hypertension 2007;49:1–2. [26] Gimeno X, Llorens C, Moncada S, Navarro A. Incidencias durante el trabajo de campo con encuestas personales en estudios epidemiolo´gicos [Events during field work with personal questionnaires in epidemiologic studies]. Metodologı´a de encuestas 2006;7(2):61–74. [27] Kristensen TS, Hannerz H, Høgh A, Borg V. The Copenhagen Psychosocial Questionnaire – a tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005; 31(6):438–49. [28] Moncada S, Llorens C, Navarro A, Kristensen TS. ISTAS21 COPSOQ: versio´n en lengua castellana del cuestionario psicosocial de Copenhague [ISTAS21 COPSOQ: Spanish version of the Copenhagen Psychosocial Questionnaire]. Arch Preven Riesgos Laboral 2005;8(1):18–29. [29] Moncada S, Llorens C, Font A, Galte´s A, Navarro A. Exposicio´n a riesgos psicosociales entre la poblacio´n asalariada en Espan˜a (2004–05): valores de referencia de las 21 dimensiones del cuestionario COPSOQ ISTAS21. Rev Esp Salud Pu´blica 2008;82(6):667–76. [30] Greenacre MJ. Correspondence analysis in practice. London: Chapman & Hall/CRC; 2007. [31] Agresti A. Categorical data analysis. Hoboken: John Wiley & Sons; 2002. [32] Ferna´ndez Steinko A. El sabor agridulce de los grupos de trabajo’’ [The bittersweet flavour of working teams]. Cuaderno de Relaciones Laborales. 2001;18: 257–83. [33] Llorens C, Moncada S, Alo´s R, Cano E, Font A, Jo´dar P, Lo´pez V, Sa´nchez A. Psychosocial hazards and work process design in Spain: Taylorism revisited? In: Benach J, Muntaner C, with Solar O, Santana V, Quinlan M. Employment, work, and health inequalities: a global perspective. Geneva: WHO; 2009 (forthcoming). [34] Lahera Sa´nchez A. La participacio´n de los trabajadores en la calidad total: nuevos dispositivos disciplinarios de organizacio´n del trabajo [Workers participation in total quality: new disciplinary tools of work organization]. REIS 2004;106:63–102. [35] Cano E. Formas, percepciones y consecuencias de la precariedad. Mientras Tanto 2004;93:67–81. [36] Nyberg A, Bernin P, Theorell T. The impact of leadership on the health of subordinates. Report 1. Stockholm: SALTSA; 2005. [37] Nyberg A, Westerlund H, Magnusson LL, Theorell T. Managerial leadership is associated with self-reported sickness absence and sickness presenteeism among Swedish men and women. Scand J Public Health 2008;36(8):803–11.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 137–148

ORIGINAL ARTICLE

Psychosocial work environment and its association with socioeconomic status. A comparison of Spain and Denmark

SALVADOR MONCADA1, JAN HYLD PEJTERSEN2, ALBERT NAVARRO3, CLARA LLORENS1,4, HERMANN BURR2, PETER HASLE2 & JAKOB BUE BJORNER2 1

Union Institute of Work Environment and Health (ISTAS), Barcelona, Spain, 2National Research Centre for the Working Environment, Copenhagen, Denmark, 3Universitat Auto`noma de Barcelona (UAB), Biostatistics Unit, Faculty of Medicine, Bellaterra, Barcelona, Spain, and 4Universitat Auto`noma de Barcelona (UAB), Department of Sociology, Political Sciences and Sociology Faculty, Bellaterra, Barcelona, Spain

Abstract Aims: The purpose of this study was to describe psychosocial work environment inequalities among wage earners in Spain and Denmark. Methods: Data came from the Spanish COPSOQ (ISTAS 21) and the Danish COPSOQ II surveys both performed in 2004–05 and based on national representative samples of employees with a 60% response rate. Study population was 3,359 Danish and 6,685 Spanish women and men. Only identical items from both surveys were included to construct 18 psychosocial scales. Socioeconomic status was categorized according to the European Socioeconomic Classification System. Analysis included ordinal logistic regression and multiple correspondence analysis after categorizing all scales. Results: A relationship between socioeconomic status and psychosocial work environment in both Denmark and Spain was observed, with wider social inequalities in Spain for many scales, describing a strong interaction effect between socioeconomic status and country. Conclusions: Socioeconomic status is related to psychosocial work environment and some adverse psychosocial conditions tend to cluster in lower socioeconomic status groups in both Spain and Denmark. This effect could be modified by a country’s characteristics, such as economic and labour market structures, normative regulations and industrial relations including work organization. Hence, preventive strategies to reduce social inequalities in working conditions should consider the combination of actions at the macro and micro levels.

Key Words: Denmark, inequalities, international COPSOQ, occupational exposures, occupational health, psychosocial factors, Spain

Background and aims Psychosocial risk factors represent a field of increasing interest in occupational health, both for their impact on health and health inequalities and for the changes in the work environment that imply growing exposure to these risk factors. The focus on psychosocial risk factors also point towards new needs and priorities for research and prevention [1,2]. At the same time, reducing the gap in social inequalities in health has been set up as a priority action by the World Health Organization (WHO) and most European Union governments. This requires preventive strategies to be based on a comprehensive

understanding of the conditions determining hazardous exposures at workplaces. Scientific evidence on social health inequalities is not limited to health consequences of general living conditions. Working conditions, especially the psychosocial work environment [3] have been found to account for some of the social gradients in mortality, mental well-being and sickness absence [4–7]. Overall, work stress measurements tend to be higher in lower socioeconomic status (SES) occupations [8], even when greater effort-reward imbalance may affect mostly higher SES jobs [9]. The lower SES occupations are more exposed to adverse

Correspondence: Salvador Moncada, Union Institute of Work Environment and Health (ISTAS), Via Laietana,16, E-08003 Barcelona, Spain. Tel: þ34934812835. Fax:+34934812770. E-mail: [email protected] (Accepted 11 October 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809353825

138

S. Moncada et al.

psychosocial conditions [10], but some exceptions to this rule have been pointed out. While control over job and rewards tends to be lower among low SES occupations [11] the opposite figure has been documented regarding job demands [12]. Successive European Working Conditions Surveys show both inter and intra country social inequalities in psychosocial working conditions across the European Union. While Spanish workers appeared among the most exposed to psychosocial risks, an opposite figure has been shown for the Danish workforce [13,14]. Inequalities in the psychosocial work environment across SES groups have been documented both in Spain [15] and in Denmark [16]. An exploratory study reported that work environment in Denmark appeared to be more active, developmental and challenging than work environment in Spain while no differences were found regarding interpersonal relations and leadership [17]. These disparities in psychosocial exposures may be related to differences in several factors at macro and micro levels including economic and labour market structure, normative regulations and industrial relations, labour management practices and working conditions, and occupational health and well-being policies [18]. Denmark has gone further than Spain in the development of the welfare state and the promotion of healthier working conditions, including those related to work organization and labour management practices that constitute the core basis of psychosocial exposure in the workplace. However, the possible consequences of these differences have not been subject to more detailed comparative analysis. A good understanding of these relationships may help to identify key targets for preventive action [16,19]. The purpose of this study is to describe psychosocial work environment inequalities among wage earners in Spain and Denmark. We hypothesize that a SES gradient in psychosocial work environment will be found in both Denmark and Spain. We further hypothesize that the social gradient will be more pronounced in Spain compared with Denmark. This second hypothesis is, among other things, motivated by the fact that Denmark has much lower organizational power distance than Spain [20] and that research has shown wider health inequalities in late democracies (such as Spain) compared to countries with a long tradition of social democracy (such as Denmark) [21].

performed in 2004–05 and based on national representative samples of employees. Collected information included the psychosocial dimensions, socio-demographics, employment and working conditions’ variables. Detailed description of the sampling and the study population for the two studies is included in, respectively, Llorens et al. and Pejtersen et al. (both in this issue). The Spanish COPSOQ (ISTAS21) survey The Spanish study used the Spanish version of COPSOQ I previously adapted from the original Danish and validated in Spain [22]. Information was obtained through the administration of the questionnaire by personal interview in the household. In all, 7,650 wage earners aged 16–65 answered the questionnaire (response rate 60%). The Danish COPSOQ II survey The Danish study used the second version of the COPSOQ [23] and respondents completed a mail-out-mail-back questionnaire or answered over the internet (used by 10% of respondents). In all, 3,517 wage earners aged 20–59 completed the questionnaire (response rate 60%). The study sample For the purpose of this paper we combined the two data sets. The SES categories were established according to the European Socio-Economic Classification system (ESEC) which is based on the ISCO88 occupational codes [24]. Fellow workers, farmer employees, Spanish dependent self-employed, individuals with missing ISCO88 codes and the ESEC ‘‘Lower Supervisors and Technicians’’ category, with a very low Danish frequency (n ¼ 13) were excluded, so SES was finally characterized by six categories – Higher professionals and managers, Lower professionals and managers, Higher clerical, services and sales workers, Lower clerical, services and sales workers, Skilled workers, and Semi- and unskilled workers. The final sample consisted of 10,044 employees (3,359 Danish and 6,685 Spanish). The characteristics of the two samples with regard to gender, age, and socioeconomic status can be seen in Table I. Measurements

Methods Data was taken from the Spanish COPSOQ (ISTAS 21) and the Danish COPSOQ II surveys, both

Only identical items from Danish COPSOQ II and Spanish COPSOQ I (ISTAS 21) scales were included in the analysis. In order to validate the scales

Psychosocial work environment and socioeconomic status: Spain and Denmark Table I. Study population characteristics in Denmark (n ¼ 3,359) and Spain (n ¼ 6,685) by social class, sex and age group; 2005. Study population characteristics Social class – Denmark Higher professionals and managers Lower professionals and managers Higher clerical services and sales workers Lower clerical services and sales workers Skilled workers Semi- and unskilled workers Social class – Spain Higher professionals and managers Lower professionals and managers Higher clerical services and sales workers Lower clerical services and sales workers Skilled workers Semi- and unskilled workers Sex – Denmark Women Men Sex – Spain Women Men Age – Denmark <31 31–45 445 Age – Spain <31 31–45 445

n

%

451 630 750 582 345 601

13.4 18.8 22.3 17.3 10.3 17.9

450 565 1,203 1,456 1,028 1,983

6.7 8.5 18.0 21.8 15.4 29.7

1,759 1,600

52.4 47.6

3,305 3,359

49.6 50.4

511 1,458 1,390

15.2 43.4 41.4

2,270 3,019 1,384

34.0 45.2 20.7

translation, analysis of differential item functioning (DIF) [25] with respect to the exogenous variables country and SES was performed for all scales. Special attention was given to three scales that had been modified from the Danish COPSOQ to the Spanish context. Additional items were added to the Spanish scales of Job insecurity and Influence and one of the items in the scale of Quantitative demands was reversed and therefore had a negative formulation in Danish but a positive formulation in Spanish. DIF for an item required a significant association of sufficient magnitude between the item and the exogenous variable when controlling for the scale score. A sufficient magnitude for the association required that the exogenous variable explained at least an additional 4% of the item variance [26]. The analysis only showed DIF with respect to country for one of the items in the scale Job insecurity. The scale was therefore divided into two – Insecurity, with two items and Concerns about employability, with one item. So, a final set of 18 multi-item psychosocial scales that were comparable between the two countries was established (see Table II). The scales were scored on a 0–100 metric.

139

Analysis Analysis was performed in two phases. In the first phase, the scale score for each scale was categorized into five categories with approximately equal number of respondents in each category. This was done because many of the reduced scales had floor or ceiling effects which could bias results from standard linear regression. We used ordinal logistic regression analysis for all categorized scales, with the main objective of evaluating the interaction between country and SES. The categorized scale was used as the dependent variable, while country, SES and an interaction between SES and country were used as independent variables. Furthermore, the analyses were adjusted for sex, age and interactions between: sex and country, sex and SES, age and country and age and SES. The odds ratio was the odds of being in a higher category for each of the COPSOQ scales. For the independent variables we chose the grand mean as the reference as we did not have natural reference categories for the SES variables. Supplementary analyses were made separately for each country. Ordinal logistic regression assumes proportional odds, which implies that the analyses in principle should give the same results as analysis of dichotomized scales (although the analysis of dichotomized scales would be less robust and have less power). To test this assumption, we performed a parallel analysis using dichotomized scales and compared results. Since the results were similar, we only report results from the ordinal logistic regression. In a second multivariate phase we conducted multiple correspondence analysis to study the descriptive relationship among all psychosocial scales with SES, sex and country categories in order to obtain a view based in a multivariate framework. We dichotomized each of the 18 psychosocial scales into two categories (below or above the median of the Danish males) and labelled them as ‘‘good psychosocial work environment’’ and ‘‘poor psychosocial work environment’’ according to the hypotheses of the developer of the COPSOQ [27]. The quantification of inertia was performed by means of the Greenacre’s adjustment [28].

Results When controlling for socioeconomic status in the ordinal logistic regression analyses, the gender differences were generally small, except for the following differences: men indicated more cognitive demands than women (odds ratio (OR) ¼ 1.29), higher influence (OR ¼ 1.32), more possibilities for development

140

S. Moncada et al.

Table II. COPSOQ scales and items included in the analysis. Number of items

a Cronbach alpha

Work pace Do you have to work very fast?

1



Quantitative demands Is your workload unevenly distributed so it piles up? How often do you not have time to complete all your work tasks? Do you have enough time for your work tasks?

3

0.65

Cognitive demands Does your work require that you remember a lot of things? Does your work require you to make difficult decisions?

3

0.65

Emotional demands Does your work put you in emotionally disturbing situations? Is your work emotionally demanding? Do you get emotionally involved in your work?

3

0.87

Demands for hiding emotions Does your work require that you hide your feelings?

1



Influence at work Do you have a large degree of influence concerning your work? Can you influence the amount of work assigned to you? Do you have any influence on what you do at work?

3

0.78

Possibilities for development Does your work require you to take the initiative? Do you have the possibility of learning new things through your work? Can you use your skills or expertise in your work?

3

0.84

Meaning of work Is your work meaningful? Do you feel that the work you do is important? Do you feel motivated and involved in your work?

3

0.72

Commitment to the workplace Do you enjoy telling others about your place of work? Do you feel that your place of work is of great importance to you?

2

0.74

Predictability At your place of work, are you informed well in advance concerning for example important decisions, changes, or plans for the future? Do you receive all the information you need in order to do your work well?

2

0.83

Role-clarity Does your work have clear objectives? Do you know exactly which areas are your responsibility? Do you know exactly what is expected of you at work?

3

0.79

Role-conflicts Do you do things at work that are accepted by some people but not by others? Are contradictory demands placed on you at work? Do you sometimes have to do things that ought to have been done in a different way? Do you sometimes have to do things that seem to be unnecessary?

4

0.86

Quality of leadership To what extent would you say that your immediate superior . . . – makes sure that the individual member of staff has good development opportunities? – is good at work planning? – is good at solving conflicts?

3

0.82

Social support from colleagues How often do you get help and support from your colleagues? How often are your colleagues willing to listen to your problems at work?

2

0.81

Social support from supervisors How often do you get help and support from your nearest superior? How often is your nearest superior willing to listen to your problems at work?

2

0.82

COPSOQ dimensions & items

(continued)

Psychosocial work environment and socioeconomic status: Spain and Denmark

141

Table II. Continued. Number of items

a Cronbach alpha

Sense of community Is there a good atmosphere between you and your colleagues? Is there good co-operation between the colleagues at work? Do you feel part of a community at your place of work?

3

0.89

Insecurity at work Are you worried about being transferred to another job against your will? Are you worried about becoming unemployed?

2

0.80

Concerns about employability Are you worried about it being difficult for you to find another job if you became unemployed?

1



COPSOQ dimensions & items

(OR ¼ 1.24) and lower concerns about employability (OR ¼ 0.84). Since no significant gender*country interactions were found, the results refer to the common trend in Denmark and Spain (not shown in tables). Trends with age were also fairly small, except for the following differences that all showed a clear trend across age groups: compared to the youngest group (18–30 years), the older employees (46–59 years) reported more cognitive demands (OR ¼ 1.31), more emotional demands (OR ¼ 1.43), higher degree of influence (OR ¼ 1.59), higher meaning of work (OR ¼ 1.39), more commitment to the workplace (1.35), higher role clarity (OR ¼ 1.29), but less support from colleagues (OR ¼ 0.79). No age group*country interactions were found (results not shown in tables). As shown in Table III, statistically significant differences in odds ratios between countries were seen for all scales except Emotional demands (Table III). Denmark had higher demands (Quantitative demands, Work pace and Cognitive demands). Furthermore, Denmark had better psychosocial work environment in terms of higher job control (Influence, Possibilities for development) and higher Commitment to the workplace than Spain. Also, Denmark had lower Job insecurity and Concern about employability than Spain. Spain had a better psychosocial work environment than Denmark with regards to social relations, higher Predictability, Role clarity, Quality of leadership, better Support from colleagues and lower Role conflict. The explained variance for country ranged from 1.0% to 10.0% for these scales. For both countries, statistically significant SES differences in odds ratios were seen for most scales, and were most notable for Cognitive demands, Emotional demands, Influence at work, Possibilities for development, Meaning of work and Commitment to the work place. The explained variance for SES ranged from 4.5% to 10.9% for these scales. However, not all scales showed a clear trend consistent with the standard rank order of SES groups. Thus, in both countries work pace was greater for higher

professionals and managers and for semi- and unskilled workers compared to the neighbour categories. The opposite picture was seen for social support from colleagues, which was comparatively low for these two groups. Also, while job insecurity was high for semi- and unskilled workers in both countries, the pattern for other SES groups varied between countries. Highly statistically significant differences in the social gradient between countries (i.e. SES*country interactions) were seen for 13 out of 18 scales (the exceptions were: Commitment to the workplace, Predictability, Role conflicts, Quality of leadership, Social support from colleagues, and Job Insecurity). The explained variance for the interaction effect ranged from 0.2% to 1.7% for these scales. In Table III, these SES*country interactions are indicated by the country specific ORs for SES. These ORs are the combination of the socioeconomic status main effect and the SES*country interaction. For many scales, inequalities were more pronounced for Spain than for Denmark as witnessed by the stronger gradient in the odds for SES in Spain. This was in particular the case for Cognitive demands, Influence at work, Possibilities for Development, Meaning of work, Social support from supervisors, and Social community at work. In some cases, the SES*country interactions nullified the SES trend. Thus, in separate analyses by country, no significant SES differences were found for Denmark for the scales: Work pace, Role clarity, Role conflicts, Social support from supervisors, and Social community at work (data not shown). For Denmark, more pronounced socioeconomic status differences were found for Quantitative demands in particular, with the highest socioeconomic status having the strongest Quantitative demands. While separate analyses for Spain showed significant differences with SES, the trend was not clear, with the highest Quantitative demands found for skilled workers and the lowest found for lower professionals and managers.

1.49 1.65 0.67 0.61

1.18 1.88 0.94 1.42 1.01 1.34 0.85 1.08 0.98 0.60 1.08 0.43

1.09 2.53 0.80 2.15 0.82 0.95 0.88 0.68 1.21 0.65 1.32 0.43 *** *** 3.8% 10.0% *** *** 1.1% 7.3% *** *** 0.2% 0.5%

1.93

1.35

1.47

0.71

0.65 0.57

1.12

0.75

1.14

0.87

1.16 1.04

*** 3.3%

*** 1.6% ***

1.3%

0.7%

*** 4.5% ***

1.7%

0.73 0.60

0.93

0.99

1.58

1.59

0.48 0.56

1.74

1.17

1.76

1.05

0.97 1.03

0.2%

*** 2.4% ***

*** 1.0%

0.74 0.71

1.14

0.92

1.32

1.37

0.62 0.66

1.69

1.04

1.26

1.10

0.79 1.26

0.6%

*** 6.3% ***

*** 1.6%

0.71 0.43

0.66

1.11

2.00

2.24

0.63 0.63

1.01

1.22

1.36

1.52

1.12 0.89

Pace Demandsb Demands Emotions Influenceb

1.38 0.73

Demands

0.7%

*** 10.9% ***

*** 3.9%

0.68 0.33

0.58

1.10

3.23

2.21

0.78 0.44

0.94

1.08

1.80

1.61

1.26 0.79

mentb

1.7%

*** 5.7% ***

*** 0.2%

0.70 0.41

0.61

0.95

2.72

2.19

0.76 0.69

1.17

1.11

1.18

1.23

0.88 1.13

0.3%

*** 4.6%

*** 1.0%

0.70 0.51

0.79

1.12

1.70

1.88

0.66 0.68

0.96

1.25

1.31

1.41

1.08 0.93

of Workb Workplaceb

0.2%

*** 1.7% **

*** 3.8%

0.82 0.69

0.88

0.97

1.46

1.41

0.71 0.87

0.96

1.15

1.09

1.35

0.65 1.55

bilityb

Predicta-

Role

Quality of

from

Social from

Support

0.6%

*** 0.6% ***

*** 3.2%

0.83 0.87

0.80

0.93

1.70

1.11

0.95 1.15

1.11

0.99

0.95

0.88

0.69 1.44

0.1%

0.3%

*** 1.0%

0.94 0.99

0.93

1.06

0.88

1.23

0.86 1.02

1.16

1.00

0.97

1.02

1.19 0.84

0.2%

*** 1.8% **

*** 3.0%

0.77 0.65

0.93

1.04

1.36

1.51

0.69 0.84

1.13

1.14

1.10

1.21

0.68 1.47

0.5%

*** 2.0% ***

*** 0.8%

0.76 0.60

0.93

1.16

1.41

1.43

0.88 0.97

1.10

1.11

0.94

1.01

1.09 0.91

0.2%

*** 2.0% **

*** 2.4%

0.85 0.63

0.99

1.12

1.42

1.18

1.08 0.72

1.15

1.04

1.14

0.95

0.75 1.34

Clarityb Conflicts Leadershipb Supervisorsb Colleaguesb

Role

to the

Meaning

Develop-

Hiding

for

for

0.7%

*** 1.0% ***

*** 0.7%

0.83 0.63

0.92

1.13

1.39

1.32

1.15 0.97

1.10

1.04

0.86

0.92

1.11 0.90

at Workb

Community

Social

0.1%

*** 0.8% *

*** 9.8%

0.93 1.14

1.14

1.09

0.67

1.14

0.99 1.34

0.90

1.07

0.81

0.97

0.54 1.84

Insecurity

Job

0.3%

*** 1.5% ***

*** 2.6%

1.10 1.19

1.09

0.96

0.75

0.97

1.02 1.54

0.83

1.22

0.67

0.93

0.77 1.30

bility

Employa-

about

Concerns

All parameter estimates concern comparison to the grand mean. For comparisons between countries, the total effect of the psychosocial work environment for a specific job group in one of the two countries is calculated from the combined effect of country, and the country specific SES ORs. For instance, the total effect of Cognitive demands for Higher professionals & managers is calculated as 1.88*1.65 ¼ 3.10 for Denmark and as 2.53*0.61 ¼ 1.54 for Spain. *p < 0.05, **p < 0.01, ***p < 0.001. a The values for explained variance are calculated as R2 differences with and without the term in question (using Nagelkerkes pseudo-R2) [26]. b High scale values are considered to have a positive effect on health [23].

Country Denmark (DK) Spain (ESP) SES DK Higher professionals & managers Lower professionals & managers Higher clerical, services & sales Lower clerical, services & sales Skilled workers Semi- & unskilled workers ESP Higher professionals & managers Lower professionals & managers Higher clerical, services & sales Lower clerical, services & sales Skilled workers Semi- & unskilled workers Significance of country Variance explained by countrya Significance of SES Variance explained by SESa Significance of SES*country Variance expl. by SES*countrya

Effect

Quantitative Work Cognitive Emotional

Social Support

Commitment

Possibilities

Demands

Table III. Odds ratio for higher category in each COPSOQ scale (categorized in five categories). All analyses controlled for age, gender and all bivariate interactions.

142 S. Moncada et al.

Psychosocial work environment and socioeconomic status: Spain and Denmark Figures 1 and 2 show the three factors described by correspondence analyses. In the graphs, axes are described by psychosocial categories. The closer they are to the extreme of the axes the more they contribute to explain them. Distance among categories refers to the relationship among them – the closer they are, the stronger the association among them is. Factor 1 was described by the dimensions of Meaning of work (mw), Social support from supervisors (sss), Quality of Leadership (ql), Possibilities for development (pd), Predictability (pre), Social community at work (scw), Commitment to the workplace (cw), Role clarity (rcl), Social support from colleagues (ssc). Factor 2 by Emotional demands (ed), Role conflict (rco), Demands for hiring emotions (dhe), Quantitative demands (qd), Cognitive demands (cd), Work pace (wp), Commitment to the workplace (cw), and Influence (inf). These two factors explained 84.1% of variability. An additional 4% of variability was explained by a third factor described by Job Insecurity (ins) and Concerns about employability (emp). Categories’ distribution showed a general relationship of psychosocial dimensions with SES.

143

Differences were more evident for Spain. Lines connecting ESEC groups were straighter and longer in Spain, while the Danish results showed more central positions and crossovers of group 1, 2, 3 and 7. In Spain, ESEC 1 and 2 were closer to good psychosocial categories, while ESEC 7, 8 and 9 were closer to the poor, with ESEC 3 with more intermediate position but closer to lower ESEC. In Denmark, ESEC 1, 2, 3 and 7 were closer to good categories psychosocially; while ESEC 8 and 9 were closer to the poor. Danish ESEC 8 and 2 and Spanish 9 were the groups with highest gender differences. The third factor (Figure 2) defined by Job insecurity (ins) and Concerns about employability (emp) clearly discriminated both countries.

Discussion This is an international population-based comparative study that used the same validated instrument [22,23] in two countries to collect information during the same time period. Both studies were

edG

dheG

rcoG insG

0,5

Work pace (wp); Quantitative demands (qd); Cognitive demands (cd); Emotional demands (ed); Demands for hiring emotions (dhe); Influence (inf); Possibilities for development (pd); Meaning of work (mw); Commitment to the workplace (cw); Predictability (pre); Role clarity (rcl); Role conflict (rco); Quality of Leadership (ql); Social support from colleagues (ssc); Social support for supervisors (sss); Social community at work (scw); Job Insecurity (ins); Concerns about Employability (emp). Final “P”: Poor, final “G”: Good

qdG

mDK8

empG rdG

wpG

Factor 2

scwG

sscG

qlG

mDK9 cdP

0,0

cwP pdP

wSP8 mSP9 wDK9 mSP7 mSP8 wSP3 wSP7

preG sssG

wSP9 infP

wDK8

mwP sssP qlP

mSP3

sscP

mwG scwP wSP2 mSP2

mDK2 insP

mDK3 wDK3

wDK7 pdG

cwG infG

wSP1

rclP

empP

mDK7

mSP1

mDK1

wDK2

Women, Denmark (wDK) Men, Denmark (mDK) Women, Spain (wSP) Men, Spain (mSP)

wDK1 wpP

–0,5

dheP edP

qdP rcoP

cdG

–0,5

preP

0,0 Factor 1

The number following the label identifies ESEC group: 1: Higher professionals and managers 2: Lower professionals and managers 3: Higher clerical. services and sales workers 7: Lower clerical. services and sales workers 8: Skilled workers 9: Semi-skilled and unskilled workers

0,5

Figure 1. Psychosocial work environment and ESEC relationship by country and sex. Factors 1 & 2. Male and female employees, Denmark and Spain (n ¼ 10,044), 2005.

144

S. Moncada et al. empP

0,5

Work pace (wp); Quantitative demands (qd); Cognitive demands (cd); Emotional demands (ed); Demands for hiring emotions (dhe); Influence (inf); Possibilities for development (pd); Meaning of work (mw); Commitment to the workplace (cw); Predictability (pre); Role clarity (rcl); Role conflict (rco); Quality of Leadership (ql); Social support from colleagues (ssc); Social support for supervisors (sss); Social community at work (scw); Job Insecurity (ins); Concerns about Employability (emp). Final “P”: Poor, final “G”: Good

scwG wSP1

rclG mwG

wSP9

cdP mSP9

mwP

infP

wSP8

rclP

edG

rcoG dheG

mSP1

rcoP

dheP edP

infG

qdP

pdG

cwP wDK8

wDK9

mSP2

Factor 3

pdP

mSP8

qdG

cwG wSP2

wpG

mSP3

sscG

preG

0,0

wSP7

wSP3 mSP7

qlG sssG

insP

scwP

mDK9

wDK3

preP

sssP

wpP qlP wDK7 mDK1

–0,5

mDK3 cdG

mDK2

mDK7 wDK1

wDK2

sscP

mDK8

empG

Women, Denmark (wDK) Men, Denmark (mDK) Women, Spain (wSP) Men, Spain (mSP) The number following the label identifies ESEC group: 1: Higher professionals and managers 2: Lower professionals and managers 3: Higher clerical. services and sales workers 7: Lower clerical. services and sales workers 8: Skilled workers 9: Semi-skilled and unskilled workers

insG

–1,0 –0,5

0,0 Factor 1

0,5

Figure 2. Psychosocial work environment and ESEC relationship by country and sex. Factors 1 & 3. Male and female employees, Denmark and Spain (n ¼ 10,044), 2005.

representative national samples, inclusion criteria were similar, and socioeconomic status was classified with an internationally comparable system – ESEC [24]. Only identical items were included in the analysis. This may constitute the main strengths of this article. For most scales, we found a relationship between SES and psychosocial work environment in both Denmark and Spain. The picture is complex and there is not always a clear trend in working conditions across SES groups. However, many scales showed a pattern of wider social inequalities in Spain while only a few scales (most notably Quantitative demands) showed the opposite picture. The correspondence analysis showed that 84% of the variability could be explained by two factors. SES differences were more clear in Spain, in particular with regards to the first factor, which covered the domains of meaning, supervisor support, quality of leadership, development possibilities, predictability, social community, commitment to the workplace, role clarity, and support from colleagues. A social gradient

(although less clear) could also be seen in Denmark, in particular for the second factor, which covered the domains of quantitative, cognitive, emotional, and hiding emotions demands; work pace; role conflicts; commitment to the workplace; and influence. These results are consistent with other findings which report that poor psychosocial working conditions tended to cluster in lower SES occupations despite higher psychological demands and higher effort maybe characterizing the upper SES groups [3,10–12]. However, not all psychosocial dimensions showed the same relationship with SES and we did not find identical social gradients in both countries. Our results showed a strong interaction effect between SES and country for many scales, which may suggest that some differences in economic and labour market structure, normative regulations and industrial relations between Spain and Denmark could partially explain this relationship. As a WHO recent report on employment conditions pointed out, the Danish labour market is more egalitarian than the Spanish one [18].

Psychosocial work environment and socioeconomic status: Spain and Denmark It has been remarked that work organization in Spain is mainly based on Tayloristic principles [29,30], and that Denmark and Spain differ in several ways regarding organizational culture [20]. Denmark had the second lowest organizational power distance among European countries, whereas Spain together with France, Belgium, Portugal and Greece had the highest power distance. Thus, Taylorism-based [31] and hierarchical organizational cultures lead to decreased influence and possibilities for development among lower status employees. In contrast, collective direct participation formulas can improve psychosocial work environment by increasing job complexity (skill discretion) and autonomy (decision latitude), as long as it is recognized (rewarded) in terms of wage and accepted by workers. These are fundamental features of the Danish labour market, secured by collective agreements and legislation as a mean to achieve employees’ well-being at worksite [32] (as elsewhere in Scandinavia) [33,34] on the basis of a long tradition of employee involvement in terms of information and consensus-based decision-making [35–37]. Furthermore, Denmark has high union density, a high degree of collective agreements, and formalized systems for employee influence [38–39]. Nevertheless, this active work organization can also have its flaws. It can be characterized by project – management or objectives management, which may involve higher uncertainty (fluid work division: flattening hierarchies without sufficient distribution of responsibilities and information, new responsibilities and goals, no rules to restrict demands: new tasks, new products, with strict deadlines) and individualization (individual performance measurement, procedures that restrict working together), which may involve ‘‘chaotic differentiation’’ and could explain the higher Quantitative demands, Work pace, Cognitive demands, and lower Role clarity, Predictability and Support from colleagues in Danish work settings compared to Spanish ones [40]. Inter-country differences on Job insecurity and Concerns about employability are other striking results. The high Spanish scores on Job insecurity and Concerns about employability could have many explanations. Denmark has a low legal protection for workers being fired, but high compensation rates for unemployment. On the other hand Spain has high legal protection but low compensation. Denmark has had, since the late 1990s and up to the study period, very low unemployment, high turnover rates and a high labour force participation rate – whereas Spain has had high unemployment, low turnover rates, and a low labour force participation rate [41]. The combination of low legal protection, high

145

compensation, high turnover rates, and a highly active labour market policy in Denmark has been labelled ‘‘flexicurity’’ [42,43]. The ideology and practice of a neo-Darwinian global economy based in volatile financial markets rationality makes work life more insecure all over the world, but if competitiveness in a country is based on low workforce costs the consequences may be even worse. In the global division of labour, Spain is on the execution side and Denmark on the design side. This is evidenced by differences in the gross domestic expenditure on research and development (Spain: 1.12% of GDP as opposed to Denmark: 2.44% of GDP, in 2005); in employment in knowledge-intensive services (27% of employment in Spain and 43.8% in Denmark in 2006) [44] and in the trends in development of unskilled jobs, which is growing in Spain, in spite of the growth in educational level of workers, but decreasing in Denmark [45]. Training and promotion of employment policies are following a high-skill, high-added-value strategy in the Nordic countries. In contrast, Spanish companies look for comparative advantages based on lower cost in the short term. Spanish lack of investment in productive work organization and labour management at mid-term was promoted by government labour reforms during the 1990s, when working conditions suffered a deregulation process that empowered employers to demand more flexibility from workers [46]. This development took place despite the expansion of public education and lifelong training negotiated between Spanish employers’ organizationa and trade unions and supported by the state and EU [46]. Outsourcing or transferring production or service to other countries is easy for transnational and national companies in Spain, as production is based on cost reduction and unskilled labour. The unemployment threat is therefore experienced as real. Unemployment rates are much higher in Spain (9.2%) than in Denmark (4.8%) and the use of temporary contracts (without dismissal rights) is massive compared with Denmark [45]. For these reasons, fear of losing employment and fear of degradation of working conditions are at the centre of working life for everybody in Spain [47]. Although using different designs, measures and explanations we see our results as being in general agreement with the results by Hofstede on cultural differences between countries [20]. Compared with Spain, Denmark has less power distance, less uncertainty avoidance, somewhat more individualism, and less masculinity. In light of such diverse differences in economic and labour market structure, policies and culture between Spain and Denmark, the smaller Danish SES

146

S. Moncada et al.

differences on many aspects of the psychosocial working environment seem logical. Furthermore, these results on psychosocial work environment inequalities are consistent with other research that found wider health inequalities in ‘‘Late democracy’’ (that includes Spain) than ‘‘Social democracy’’ (that includes Denmark) political tradition countries [21]. The study had some limitations. The Spanish study used personal interview in the household whereas the Danish study used a mail questionnaire. Respondents in telephone questionnaires and face to face administered questionnaires generally report better health and well-being than respondents in mail administered questionnaires [48–50]. Such a methods effect could partly explain why Spain had better scores than Denmark on nine scales, but would not explain why Denmark had better scores on the remaining nine scales. We would not expect such a methods effect to have an impact on the SES differences nor on the SES*country interactions. Also, differences between countries may be due to national differences in response styles or simply to drift in ‘‘difficulty’’ of an item when it is translated. Again, this may impact the difference between countries, but is unlikely to impact the SES differences or the SES*country interactions. Both studies had a 60% response rate, which is usual in such studies. Comparison of the Spanish study population with the employee population of the Spanish National Active Population Survey (EPA) was done for the same period of time and no evidence of bias was observed except for an excess of 8% of women from the retail sector working a split shift, probably due to differences in sampling strategies of both surveys – we used information from respondents exclusively, while EPA included next of kin information, and reported that it is more likely to find split shift employed women at home than men [22,51]. A previous analysis of response rates in the Danish study found that the response rate was higher for women and increased with age [27]. No difference in response rate was found for urbanization. We did not have any information on response rate in relation to SES. Thus, we cannot completely rule out the possibility that differential non-response with respect to SES and work factors might bias the social gradient. We found that Denmark had a higher percentage of workers in the higher SES groups whereas Spain had a higher percentage in the lower groups. However, this is likely a true difference, since it has been replicated in other studies [33,52]. A higher percentage of the workers in Spain are employed in agriculture, construction, wholesale and retail trade and hotels and restaurants than in Denmark.

In Denmark a higher proportion are employed in the health sector, education and transport and communication [45]. Since SES is the central variable in our analyses, the validity of the ESEC classification in both Denmark and Spain is crucial for the interpretation of results. Kunst et al. [52] found that the ESEC classification showed differences in self-rated health by SES group, but the gradient was less clear in Spain than in Denmark. Since Kunst et al. used data from 1994, we repeated their analysis with our data, which included wage earners but not farmers or self-employed people, as in Kunst et al. We found that the ESEC classification system was more appropriate for Spain in 2004–2005. Thus, 99.6% of Spanish women could be classified according to ESEC in our study as compared to 59.4% in the previous study. We also found a SES gradient for self-rated health in both Denmark and Spain, although the gradient was stronger in Denmark. While we cannot completely rule out the possibility of cross-cultural differences in the applicability of the ESEC classification, our results support the relevance of ESEC for both Denmark and Spain. The Spanish translation process for the COPSOQ followed standard translation and adaptation procedures – including back-translation process and pilot testing [20]. Therefore, we assume a possible bias due to translation to be minor and more likely to affect the country level rather than the SES gradient within each country. In conclusion, we found an SES gradient in the psychosocial work environment so that many poor psychosocial conditions clustered in lower SES occupations in both Spain and Denmark, with social inequalities being wider in Spain for many scales. This effect could reflect characteristics of the countries such as economic and labour market structures, normative regulations and industrial relations including work organization. Preventive strategies to reduce social inequalities in working conditions should consider the combination of actions at these macro and micro levels.

Acknowledgements The Spanish COPSOQ survey was partially funded by the Fondo de Investigacio´n Sanitaria (Health Research Fund) of the Instituto de Salud Carlos III (Spanish Health Ministry, IP031499 project). We thank Agustı´n Gonza´lez for the revision of the manuscript.

Psychosocial work environment and socioeconomic status: Spain and Denmark Conflict of interest There is no conflict of interest. References [1] Iavicoli S, Rondinone B, Marinaccio A, Fingerhut M. Research priorities in occupational safety and health: a review. Industrial Health 2006;44:169–78. [2] European Agency for Health and Safety at Work Future occupational safety and health research needsand priorities in the member states of the European Union. Luxembourg: Office for Official Publicationsof the European Communities; 2001. [3] Head J, Chandola T. Psychosocial working conditions and social inequalities in health. In: Benach J, Muntaner C, Santana V, Chairs. Final report to the WHO Commission on Social Determinants of Health (CSDH). Employment Conditions Knowledge Network (EMCONET) Final Report; 20 September 2007. [4] Andersen I, Burr H, Kristensen TS, Gamborg M, Osler M, Prescott E, et al. Do factors in the psychosocial work environment mediate the effect of socioeconomic position on the risk of myocardial infarction? Study from the Copenhagen Centre for Prospective Population Studies. Occup Environ Med 2004;61:886–92. [5] Chandola T, Siegrist J, Marmot M. Do changes in effort-reward imbalance at work contribute to an explanation of the social gradient in angina? Occup Environ Med 2005;62:223–30. [6] Marmot MG, Bobak M, Davey Smith G. Explanations for social inequalities in health. In: Amick BC, Levine S, Tarlov AR, Chapman Walsh D, editors. Society & health. New York/Oxford: Oxford University Press; 1995. [7] Stansfeld SA, Head J, Marmot MG. Explaining social class differences in depression and well-being. Soc Psychiatry Psychiatr Epidemiol 1998;33:1–9. [8] Kouvonen A, Kivimaki M, Virtanen M, Heponiemi T, Elovainio M, Pentti J, et al. Effort-reward imbalance at work and the co-occurrence of lifestyle risk factors: cross-sectional survey in a sample of 36,127 public sector employees. BMC Public Health 2006;6:24. [9] Kuper H, Singh-Manoux A, Siegrist J, Marmot M. When reciprocity fails: effort-reward imbalance in relation to coronary heart disease and health functioning within the Whitehall II study. Occup Environ Med 2002;59:777–84. [10] Siegrist J. Reducing social inequalities in health: work-related strategies. Scand J Public Health Suppl 2002;59:49–53. [11] Bosma H, Marmot MG, Hemingway H, Nicholson AC, Brunner E, Stansfeld SA. Low job control and risk of coronary heart disease in Whitehall II (prospective cohort) study. BMJ 1997;314:558–65. [12] Siegrist J, Starke D, Chandola T, Godin I, Marmot M, Niedhammer I, et al. The measurement of effort-reward imbalance at work: European comparisons. Soc Sci Med 2004;58:1483–99. [13] Paoli P, Merile´ D. III European Survey on Working Conditions. Luxembourg: Office for Official Publications of the European Communities; 2001. [14] Parent-Thirion A, Ferna´ndez E, Hurley J, Vermeylen G. IV European survey on working conditions. Luxembourg: Office for Official Publications of the European Communities; 2007. [15] Moncada S, Llorens C, Gimeno X, Font A. Exposicio´n laboral a riesgos psicosociales en la poblacio´n asalariada

[16]

[17]

[18]

[19]

[20]

[21]

[22]

[23]

[24]

[25] [26] [27]

[28] [29]

[30]

[31]

147

espan˜ola. [Occupational exposure to psychosocial risk among employees in Spain.]. In: Moncada S, Llorens C, editors. Organizacio´n del Trabajo, Factores Psicosociales y Salud. [Work organization, psychosocial factors and health]. Madrid: ISTAS; 2007. Kristensen TS, Borg V, Hannerz H. Socioeconomic status and psychosocial work environment: results for a Danish national study. Scand J Public Health 2002;30 (Suppl 59):41–8. Kristensen TS, Moncada S, Llorens C, Navarro A. Psychosocial work environment in Spain and Denmark: a cross-cultural comparison. In: International Conference on Psychosocial Factors at Work; Okayama; 2005. Benach J, Muntaner C, Santana V, Chairs. Employment conditions and health inequalities. WHO Commission on Social Determinants of Health. Employment Conditions Knowledge Network (EMCONET). Final Report, 20 September 2007. MacDonald LA, Ha¨renstam A, Warren ND, Punnett L. Incorporating work organisation into occupational health research: an invitation for dialogue. Occup Environ Med 2008;65(1):1–3. Hofstede G. Culture’s consequences, comparing values, Behaviors, institutions, and organizations across nations. Thousand Oaks CA: Sage; 2001. Espelt A, Borrell C, Rodrı´guez-Sanz M, Muntaner C, Pasarı´n MI, Benach M, Kunst AE, Navarro V. Inequalities in health by social class dimensions in European countries of different political traditions. Int J Epidemiol. 2008;37(5):1095–105. Moncada S, Llorens C, Navarro A, Kristensen TS. ISTAS21 COPSOQ: versio´n en lengua castellana del cuestionario psicosocial de Copenhague [ISTAS21 COPSOQ: Spanish version of the Copenhagen Psychosocial Questionnaire]. Arch Preven Riesgos Laboral 2005;8(1):18–29. Pejtersen JH, Kristensen TS, Borg V, Bjorner JB. The second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II). Scand J Publ Health 2010;38(Suppl 3):8–24. Rose D, Pevalin DJ, Elias P, Martin J. Towards a European socio-economic classification: final report to Eurostat of the Expert Group. London and Colchester: ONS and ISER, University of Essex; 2001. Holland PW, Wainer H. Differential item functioning. Hillsdale, NJ: Lawrence Erlbaum Associates Inc; 1993. Nagelkerke NJD. A note on a general definition of the coefficient of determination. Biometrika 1991;78:691–2. Pejtersen JH, Kristensen TS. The development of the psychosocial work environment in Denmark from 1997 to 2005. Scand J Work Environ Health 2009;35:284–93. Greenacre MJ. Correspondence analysis in practice. London: Chapman & Hall/CRC; 2007. Lahera Sa´nchez A. La participacio´n de los trabajadores en la calidad total: nuevos dispositivos disciplinarios de organizacio´n del trabajo [Workers participation in total quality: new disciplinary tools of work organization]. REIS 2004; 106:63–102. Eironline. Collective agreements on changes in work organisation. Spain. Denmark. Comparative information. Thematic future. Dublin: European Foundation for the Improvement of Living and Working Conditions; 2005. Available on http://www.eurofound.europa.eu/eiro/thematicfeature9.htm (accessed 26 January 2009). Belkic K, Landsbergis P, Schnall P, Baker D. Is job strain a major source of cardiovascular disease risk? Scand J Work Environ Health 2004;30(2):85–128.

148

S. Moncada et al.

[32] Hasle P, Hvid H. Current work politics. In: Hvid H, Hasle P, editors. Human development and working life. Aldershot: Ashgate; 2003. pp. 1–10. [33] Parent-Thiron A, Ferna´ndez Macı´as E, Hurley J, Vermeylen G, et al. Fourth European Working Conditions Survey. Dublin: European Foundation for the Improvement of Living and Working Conditions; 2007. [34] Rogaczewska AP, Larsen HH, Nordhaug O, Døving E, Gjelsvik M. Denmark and Norway: siblings or cousins? In: Brewster, C, Mayrhofer W, Morley M, editors. Human resource management in Europe – Evidence of convergence? Amsterdam: Elsevier; 2004. pp. 231–78. [35] Campell JL, Hall JA. The state of Denmark. In: Campell JL, Hall JA, Pedersen OK, editors. National identity and the varieties of capitalism. Copenhagen: DJØF Publishing; 2006. [36] Due J, Jensen CS, Madsen JS, Petersen JK. The survival of the Danish model. Copenhagen: DJOF Forlagene; 1994. [37] Knudsen H. Employee participation in Europe. London: Sage; 1955. [38] Lind J. Denmark – still the century of trade unionism. In: Hoffmann R, Waddington J, editors. Trade unions in Europe – facing challenges and searching for solutions. Brussels: European Trade Union Institute; 2000. pp. 143–82. [39] Eironline (2002). Denmark and Spain industrial relations profiles. In: http://www.eurofound.europa.eu/eiro/country/ Denmark.htm, and http://www.eurofound.europa.eu/eiro/ country/Spain.htm (accessed 8 July 2009). [40] Hvid H, Lund H, Pejtersen JH. Control, flexibility and rhythms. Scand J Work Environ Health Suppl 2008; (6):83–90. [41] Madsen PK. How can it possibly fly? The paradox of a dynamic labour market in a Scandinavian welfare state. In: Campell J, Hall J, Pedersen O, editors. National identity and the varieties of capitalism. The Danish experience. Montreal: McGill University Press; 2006. pp. 321–55. [42] Andersen SK, Mailand M. The Danish flexicurity model – the role of the collective bargaining system. Copenhagen: FAOS; 2005.

[43] Philips K, Eamets R. Approaches to flexicurity: EU models. Dublin: European Foundation for the Improvement of Living and Working Conditions; 2007. [44] Scha¨fer G, Baryn M, Fritz M, Johansson A, Wieland U, editors. Europe in figures. Eurostat Yearbook 2006–7. Luxembourg: Office for Official Publications of the European Communities; 2008. [45] Eironline. Unskilled workers. Spain. Denmark. Comparative information. Thematic future. Dublin: European Foundation for the Improvement of Living and Working Conditions; 2005. Available On http://www.eurofound.europa.eu/eiro/thematicfeature10.htm (accessed 26 January 2009). [46] Ko¨hler HD. Las relaciones laborales espan˜olas en el marco europeo [Spanish industrial relations in the European context]. In: Migue´lez F y Prieto C (dir y coord): Las relaciones de empleo en Espan˜a. Madrid: Siglo XXI; 1999. [47] Prieto C. La degradacio´n del empleo o la norma social del empleo flexibilizado. [The degration of employment or the social norm of flexible employment.]. Sistema 2002;168–9. [48] Feveile H, Olsen O, Hogh A. A randomized trial of mailed questionnaires versus telephone interviews: response patterns in a survey. BMC Med Res Methodol 2007;7(27): doi: 10.1186/1471-2288-7-27. [49] Perkins JJ, Sanson-Fisher RW. An examination of self- and telephone-administered modes of administration for the Australian SF-36. J Clin Epidemiol 1998;51(11):969–73. [50] de Leeuw ED. Mode effects in survey research: a comparison of mail, telephone, and face to face surveys. Bull Methodol Sociol 1993;43:3–19. [51] Gimeno X, Llorens C, Moncada S, Navarro. Incidencias durante el trabajo de campo con encuestas personales en estudios epidemiolo´gicos. [Events during field work with personal questionnaires in epidemiologic studies.] Metodologı´a de encuestas, 2006; Monogra´fico incidencias en el trabajo de campo: 46–59. [52] Kunst A, Roskam A-J, van Agt H. The European Socioeconomic Classification (ESeC): exploring its potential to describe class differences in health among middle-aged men and women in 11 European countries. Anonymous Paper presented to the ESeC Validation Conferencesee. Lisbon; 2006.

Scandinavian Journal of Public Health, 2010; 38(Suppl 3): 149–155

COMMENTARY

A questionnaire is more than a questionnaire

TAGE S. KRISTENSEN Task-Consult, Østre Alle´ 35E, Gilleleje, Denmark

This special issue of the Scandinavian Journal of Public Health is devoted to articles on the Copenhagen Psychosocial Questionnaire (COPSOQ). I have been asked to participate with a personal contribution and have been given the freedom to write whatever I find appropriate. My main message in this article is that a questionnaire is not just a questionnaire. A questionnaire, such as COPSOQ, is a tool for creating theoretical insight, an eye opener for employees and employers, a way to create a new language, a bridge for building long lasting ties between researchers and workplaces, a way to give legitimacy to the field of psychosocial factors at work, an instrument for creating new personal and professional friendships, and – last but not least – a tool for improvement of the working conditions for thousands of employees and for increasing the productivity of the companies. I cannot cover all these aspects in detail but I will touch upon most of them in the following.

The aims of COPSOQ About 15 years ago Vilhelm Borg and I founded the psychosocial research group at the National Institute of Occupational Health in Denmark. Before this time the institute had given priority to physical, chemical, and ergonomic factors at work. After a short while we were approached by several work environment professionals who had serious problems in connection with the assessment of psychosocial factors at the workplace. A large number of private consultants and firms offered many different tools and questionnaires but the quality seemed to be low, there were no

national data, and – above all – there were no criteria for choosing one instrument over another. The professionals asked for national standards and guidelines, and – if possible – a national instrument. We decided to take up this challenge. During our discussions the idea of the ‘‘Three-level concept’’ quickly emerged: A long questionnaire for the researchers, a medium size for the professional work environment experts (occupational health services, labour inspection, private consultants, organizations, big companies), and a short version for the smaller companies and workplaces. As a natural consequence of this, the following goals of the Three-level concepts were formulated [1]:  To develop valid and relevant instruments for the assessment of psychosocial factors at work  To make national and international comparisons possible  To improve evaluations of interventions  To facilitate surveillance and benchmarking  To improve the communication between workplaces, work environment professionals, and researchers  To make it easier for the users to understand difficult concepts and theories

It is of course up to others to evaluate to what degree we have reached these goals, but I can honestly say that for the founding group the success of the COPSOQ concept has been much greater than we had ever imagined. I believe that the articles in this special issue demonstrate this point. Going beyond ‘‘the two models’’ From the beginning it was clear to us that the questionnaire should be theory based but not

Correspondence: Tage S. Kristensen, Task-Consult, Østre Alle´ 35E, DK.3250 Gilleleje, Denmark. Tel: +45 29 70 40 12. E-mail: [email protected] (Accepted 19 October 2009) ß 2009 the Nordic Societies of Public Health DOI: 10.1177/1403494809354437

150

T. S. Kristensen

attached to one specific theory (such as the Job Content Questionnaire) [1, p. 439]. Thus, the COPSOQ I, which was developed around the year 2000 and presented in English in 2005, included most of the dimensions of the seven influential psychosocial theories reviewed by Kompier in his paper on models of psychosocial factors at work [2]. At this time it was very common in psychosocial research to ‘‘cover’’ the psychosocial work environment by including the two or three dimensions of the job strain model and perhaps also the two dimensions of the effort–reward imbalance model. (Later a third model was added: The justice at work model [3]). It was one of our main points that the two or three well-known models did not paint the whole picture. Many relevant and important psychosocial factors at work were left out, and this had serious implications for research as well as prevention. The papers of the present issue of this journal illustrate the point perfectly well. In the presented studies of psychological well-being, intention to leave, and long term absence the following ‘‘new’’ risk factors appear to be independent predictors: Emotional demands [4–6], meaning of work [4,6,7], predictability [6,8], role clarity or conflicts [5,8], and commitment to the workplace [6,7]. In comparison, influence at work (decision authority) is only a significant predictor in one of the studies [8]. In the paper by Burr et al. [4] this point is illustrated in a particularly clear way since neither the job strain model nor the effort– reward imbalance model predict ill psychological health, while high emotional demands and low meaning of work turn out to be significant predictors. The paper by Aust et al. [9] on an unsuccessful intervention study is particularly interesting. This study shows a number of differences between the intervention and control group with regard to psychosocial factors at work. (Three of these differences were statistically significant and in the ‘‘wrong’’ direction). These differences were only found because of the use of a comprehensive questionnaire such as COPSOQ. The additional insight gained by using a broad questionnaire has also been illustrated by a number of other studies from our psychosocial research group published since 2005. In studies of absence, burnout, exclusion from the labour market, and return to work, the following significant independent risk factors have been identified: Meaning of work [10–13], predictability [10,11,13,14], quality of leadership [10,11,15], role conflicts [10,11,15], and emotional demands [10,15]. Seen across these studies [10–15] and the results reported in this issue [4–9], it is remarkable to see the high number of studies showing significant associations with emotional demands,

meaning of work, and predictability. This is not the place to elaborate further on this strong pattern, but our results clearly deserve theoretical consideration and discussion.

From job factors via relational factors to social capital In the international psychosocial literature it is common practice to review and evaluate the importance of psychosocial factors by summing up the number of ‘‘positive’’ and ‘‘negative’’ studies. The role model seems to be natural science where a given risk factor (such as smoking) is assumed to have the same effects regardless of time and place. It seems to me that such an approach is naı¨ve and unfruitful. Experience, as well as a vast amount of studies, indicate that the importance of a specific risk factor depends on context. Context could be gender, time period, culture, labour market conditions, social class and many other contextual factors. Instead of asking if, for example, job strain leads to heart disease, we should ask: Under what conditions do what psychosocial factors lead to what outcomes? Two examples will illustrate this point. First, it well known that cardiovascular diseases used to be diseases of the upper classes in western Europe, but that the trend reversed so that they are today diseases of the lower classes. Two English studies published 50 years apart clearly illustrate this point [16,17]. And second, a number of studies of psychosocial factors at work clearly show that risk factors for men and women differ markedly with very little overlap [13,15,18]. Indeed, it is very rare to see analyses stratified for gender finding the same patterns for the two sexes. During my time as a researcher in psychosocial factors at work, I have experienced a definite change with regard to the context of the labour market in my country. My first study was on work and health of women, published in 1978 [19]. (Those were the days when we wrote thick books but not international articles!). The main focus was on the typical features of the jobs of unskilled women in industry: high work pace, monotony, low influence, etc. Therefore, Karasek’s job strain model came as sent from heaven [20], and I was fortunate enough to have a typewritten copy months before this famous article was published in 1979. In my next study on slaughterhouse workers I was still focusing very much on the job characteristics of these workers, and I discovered that high job strain was not only associated with absence from work [21], but also with use of

A questionnaire is more than a questionnaire medicine [22], accidents, stress, fatigue and many other endpoints [23]. This research and the research of many of my colleagues in Sweden and Denmark had a decisive impact on Danish society. Psychosocial factors went from being controversial and almost taboo to being accepted and established. One of the manifestations of this was the high priority given to female industrial workers in connection with the establishment of the Danish occupational health services. Another clear sign of social and political acceptance was the national action plan against one-sided repetitive work launched in 1993. This action plan was supported by the minister of labour as well as the main organizations of employers and employees. Since the year 2000 a new picture has emerged. In brief there has been a transition from ‘‘job factors’’ (such as quantitative demands, decision authority and skill discretion) to ‘‘relational factors’’. Relational factors are characteristics of relations between colleagues, between employees and supervisors, or between employees and customers/clients. Typical examples are predictability, support, quality of leadership, recognition, social community, emotional demands, demands for hiding emotions, work– family conflicts, role conflicts, interpersonal conflicts, bullying, violence, and sexual harassment. The new situation is briefly described as: ‘‘We are each other’s work environment’’. This shift has not only been noticeable in our research (as already described above) but is also confirmed by the labour inspection, the labour unions, work environment professionals, workers’ compensation system, etc. As far as I can see, there are three main explanations: First, many of the industrial jobs have moved out of the country to countries with lower pay and poorer working conditions. Second, autonomous groups, customer focus, project organization etc. has expanded in all sectors. And third, the public sector with a great emphasis on ‘‘people work’’ has expanded. And while I write these lines a third paradigm of psychosocial factors is manifesting itself in Denmark: The social capital of the workplaces has come into focus. In 2008 a so-called white book on the topic was published by the National Work Environment Council in collaboration with the National Research Center for Work Environment [24]. In this book we have defined the social capital of the workplace and reviewed the international literature on work and social capital. Social capital in the workplace is defined as collaborative capabilities of the company based on trust and justice. Along with this, a number of ongoing studies have suggested that the two scales of trust (so-called ‘‘vertical’’ trust) and justice of the

151

COPSOQ II [25] could be a very good operationalization of company social capital. At company level the correlation of trust and justice is very high, so the average score of these two scales is an obvious measure of social capital. In order to illustrate the type of findings in the current projects, two simple figures are presented below. Both figures are based on a study of the social capital of all the elementary schools (n ¼ 12) in a provincial municipality of Denmark. Figure 1 shows the association between the dimension of leadership quality and social capital. The unit of analysis is the workplace (school), and the figure shows a perfect correlation between the two dimensions. This finding corresponds to the results of several other ongoing studies. We still have not found a workplace with discrepancy between social capital and leadership quality. Figure 2 shows the clear association between the average level of recognition and the social capital of the schools. It should be emphasized that not all associations look like this. There are very weak or no association between social capital and average levels of, e.g., quantitative demands and emotional demands. The results illustrated by the two figures are typical for the findings in the ongoing studies of company social capital: (1) We have found strong correlations at workplace level with factors such as leadership, sense of community, social support, recognition, and predictability. (2) We have also found clear correlations with job satisfaction, commitment to the workplace, intention to stay, sickness absence, and a number of measures of psychological well-being. (3) We now have suggestive, but encouraging, results on performance indicators such as productivity, customer satisfaction, quality of services etc. The two figures illustrate another striking feature of these studies: the large differences in social capital between workplaces within the same structural and economic framework. The schools illustrated in Figures 1 and 2 function under the same municipal authority, the same laws and regulations, the same economic budget, the same general agreement, and have (approximately) the same kind of children. And in spite of this the score for social capital varies from 50 points to almost 80! (As pointed out by Pejtersen et al [26] a difference of 5–10 points makes a difference for the employees). In other words, the social capital differences between these schools cannot be explained by outside structural or economic factors, which indicates that ‘‘any school can do it’’! So far, our research strongly suggests that a factor of principal importance for developing high social capital is leadership quality.

152

T. S. Kristensen Quality of leadership Ly

No

Sp

Sv

70

Te So

60

Hy Ba

Ka 50 Ro Mø

40

Gr Social capital 50

60

70

80

Figure 1. Social capital and quality of leadership at elementary schools in a Danish provincial community. Average scores on the two COPSOQ scales. (School names are abbreviated to keep anonymity. Average scores for Danish employees are indicated with dotted lines).

Recognition

Ly 80

Te

Sp No

So

Sv

70 Hy

Ba DK 60

Ka Mø

Ro

Gr 50 Social capital 50

60

70

80

Figure 2. Social capital and recognition at elementary schools in a Danish provincial community. Average scores on the two COPSOQ scales. (School names are abbreviated to keep anonymity. Average scores for Danish employees are indicated with dotted lines).

In Denmark the interest in company social capital has been overwhelming during the past couple of years. A considerable number of research projects have been launched, meetings and seminars are arranged every week, workplaces initiate development plans for increasing company social capital, and private consultants – as always – offer their assistance. During my time as a researcher I have not experienced a similar wave of enthusiasm before. In summary, the focus of Danish psychosocial work environment research during the years 1980–2010 has developed through three stages. (1) Job factors. (2) Relational factors. (3) Company factors.

Today the focus is not so much on the association between psychosocial factors and health, because the importance of psychosocial factors is generally accepted. The focus is much more on the macro level: what factors shape the psychosocial work environment? In our paper on the development of psychosocial factors in Denmark during the period 1997 to 2005 we concluded: ‘‘It is noteworthy that almost all models and most research on psychosocial factors at work deal with the impact of work environment factors on health and health-related factors. If our goal is to understand the forces shaping and changing the work environment, this focus is clearly

A questionnaire is more than a questionnaire insufficient. We need theories and research regarding the factors that shape the psychosocial work environment of the future.’’ [27, p 291]. In the present issue Llorens et al. focus on the importance of labour management practices using the Spanish/Catalan version of COPSOQ [28]. Much more research on this important issue is necessary. My main point in this connection is that the two versions of COPSOQ [1,25] have been able not only to reflect and document the transition from job factors to company factors but also to contribute to this development of a more global understanding of psychosocial factors. If we had chosen a ‘‘modelspecific’’ instrument instead of an open instrument this could not have been possible. A tool is not only a passive instrument but also a means of gaining new insight and perspective. Furthermore, the development of COPSOQ has been stimulated by statistical analyses going ‘‘beyond Cronbach’s alpha’’ using modern psychometric methods of scale analysis [26,29–31]. In this connection the tremendous impact of Jakob Bjorner’s work cannot be overestimated. As an example, the use of analyses for differential item function (DIF) paved the road for the use of the two scales for job demands: one for amount of work (‘‘quantitative demands’’) and another for work tempo (‘‘work pace’’) [25,30].

A tool for interventions and improvements As I mentioned in the introduction, the COPSOQ was not only meant to be a research instrument, but also to be a tool for workplace surveys. The medium and short questionnaires have now been used by thousands of workplaces in Denmark as well as in other countries, mainly Spain and Germany [32–36]. Promising developments have taken place in countries such as Chile [37], China, Iran, France, Belgium and elsewhere. The workplace surveys have given workplaces a solid basis for making priorities in the continued work for the improvement of the working conditions. This has often been a difficult task since psychosocial factors are not well established risk factors in many countries. Among the many obstacles the ‘‘psychosocial pioneers’’ have not only met ignorance but also neglect and strong resistance. This resistance often comes from employers and organizations as well as government agencies and work environment professionals. The general tendency has, however, been positive. Psychosocial factors are slowly but steadily becoming recognized in more and more countries. Change takes time.

153

We have identified two distinct processes in many countries. First, the COPSOQ has given workers, employers and experts a common language and, above all, a much ‘‘richer’’ language. Suddenly they can communicate using terms such as role clarity, emotional demands and meaning of work. And secondly, many of the COPSOQ pioneers have used the prestige and recognition of the Scandinavian researchers as a lever in their own national context. The words of a professor with grey hair and a long CV from a distant country often carries more weight than the words of a well-known local person! The field of workplace surveys, risk assessment, interventions and improvement is an extremely important one that should be given much higher priority in the future. The article by Aust et al. in this issue [9] illustrates this point in an excellent way. After all, the ultimate goal of psychosocial work environment research is to improve the working conditions and to create better workplaces in the future. Our article on the Danish development [27] shows that the trend can be negative even in countries where psychosocial factors have been given high priority.

A tool for creating networks For me, as well as for many others, working with COPSOQ has had an unanticipated but very positive side effect: the creation of long lasting networks and friendships. At the national level the use of COPSOQ has often been the first step in a process leading to further collaboration with companies, organizations and other researchers. At present I participate in a number of research projects based on close collaboration with workplaces in which COPSOQ is an important part. Also, one of Denmark’s largest unions has recently used COPSOQ in a comprehensive study of the work environment of its members. Very often one kind of collaboration leads to the next. For instance, we have worked with a large financial company for some years in connection with the regular workplace surveys. This has now resulted in a large prospective research project looking into the associations between psychosocial work environment factors and Key Performance Indicators (KPIs). At the international level strong ties have developed between researchers and others from a large number of countries. Collaboration has often been two-sided but two more formal COPSOQ conferences have also taken place. The first was in Denmark in 2007 and the second in Germany in 2009. The next conference will be in Barcelona in 2011. An excellent group of three, Jakob Bjorner, Salvador

154

T. S. Kristensen

Moncada and Matthias Nu¨bling, will be in charge of the future development of COPSOQ. For me it is time to step back, but I will certainly follow future developments with great interest and enthusiasm. Acknowledgement During the many years of work with the COPSOQ and related issues many colleagues from a large number of countries have contributed. I cannot mention everybody here, but only the most active and enthusiastic. My sincere thanks go to: Chantal Brisson, Rene´e Bourbonnais, Michel Ve´zina (Canada), Jian Li (China), Yuko Odagiri, Norito Kawakami, Akizumi Tsutsumi, Teruichi Shimomitsu (Japan), Hans Martin Hasselhorn, Matthias Nu¨bling (Germany), Clara Llorens, Salvador Moncada, ¨ stergren, Johan Ariadna Galte´s (Spain), Per-Olof O Hallqvist, To¨res Theorell, Kerstin Ekberg (Sweden), Yawen Cheng (Taiwan), Robert Karasek, Peter Schnall, and Paul Landsbergis (USA). References [1] Kristensen TS, Hannerz H, Høgh A, Borg V. The Copenhagen Psychosocial Questionnaire (COPSOQ). A tool for the assessment and improvement of the psychosocial work environment. Scand J Work Environ Health 2005;31:438–49. [2] Kompier M. Job design and well-being. In: Schabracq MJ, Winnbust JAM, Cooper CL, editors. The handbook of work and health psychology. Chichester (UK): John Wiley & Sons; 2003. pp. 429–54. [3] Kivima¨ki M, Elovainio M, Vahtera J, Ferrie JE. Organisational justice and health of employees: prospective cohort study. Occup Environ Med 2003;60:27–34. [4] Burr H, Albertsen K, Rugulies R, Hannerz H. Do dimensions from the Copenhagen Psychosocial Questionnaire predict vitality and mental health over and above the job-strain and effort-reward-imbalance models? Scand J Public Health 2010; 38(Suppl 3):59–68. [5] Rugulies R, Aust B, Pejtersen JH. Do psychosocial work environment factors measured with scales from the Copenhagen Psychosocial Questionnaire predict registerbased sickness absence of three weeks or more in Denmark? Scand J Public Health 2010;38(Suppl 3):42–50. [6] Li J, Fu H, Hu Y, Shang L, Wu Y, Kristensen TS, Mueller BM, Hasselhorn HM. Psychosocial work environment and intention to leave the nursing profession: results from the longitudinal Chinese NEXT study. Scand J Public Health 2010;38(Suppl 3):69–80. [7] Clausen T, Christensen KB, Borg V. Positive work-related states and long-term sickness absence: A study of register based outcomes. Scand J Public Health 2010;38(Suppl 3): 51–58. [8] Albertsen K, Rugulies R, Garde AH, Burr H. The effect of the work environment and performance-based self-esteem on cognitive stress symptoms among Danish knowledge workers. Scand J Public Health 2010;38(Suppl 3):81–89. [9] Aust B, Rugulies R, Finken A, Jensen C. When workplace interventions lead to negative effects: learning from failures. Scand J Public Health 2010;38(Suppl 3):106–119.

[10] Rugulies R, Christensen KB, Borritz M, Villadsen E, Bu¨ltmann U, Kristensen TS. The contribution of the psychosocial work environment to sickness absence in human service workers: results of a 3-year follow-up study. Work Stress 2007;21:293–311. [11] Borritz M, Bu¨ltmann U, Rugulies R, Christensen KB, Villadsen E, Kristensen TS. Psychosocial work characteristics as predictors for burnout: findings from a 3-year follow up of the PUMA study. J Occup Environ Med 2005; 47:1015–25. [12] Labriola M, Lund T, Christensen KB, Kristensen TS. Multilevel analysis of individual and contextual factors as predictors of return to work. J Occup Environ Med 2006;48:1181–8. [13] Nielsen ML, Rugulies R, Christensen KB, Smith-Hansen L, Kristensen TS. Psychosocial work environment predictors of short and long spells of registered sickness absence during a 2-year follow up. J Occup Environ Med 2006;48:591–8. [14] Nielsen ML, Rugulies R, Smith-Hansen L, Christensen KB, Kristensen TS. Psychosocial work environment and registered absence from work: estimating the etiologic fraction. Am J Industr Med 2006;49:187–96. [15] Lund T, Labriola M, Christensen KB, Bu¨ltmann U, Villadsen E, Burr H. Psychosocial work environment exposures as risk factors for long-term sickness absence among Danish employees: results from DWECS/DREAM. J Occup Environ Med 2005;47:1141–7. [16] Logan WPD. Mortality from coronary and myocardial disease in different social classes. Lancet 1952;I:758–9. [17] Marmot MG, Shipley MJ, Rose G. Inequalities in death – specific explanations of a general pattern? Lancet 1984; I:1003–6. [18] Albertsen K, Lund T, Christensen KB, Kristensen TS, Villadsen E. Predictors of disability pension over a 10-year period for men and women. Scand J Public Health 2007; 35:78–85. [19] Kristensen TS. Kvinders helbred og arbejde. Helbred, arbejds- og familieforhold for ansatte i 7 kvindebrancher. Publikation nr. 9. København: Institut for Social Medicin, Københavns Universitet; 1978 [Womens’ health and work. Health, working- and family-conditions for employees in 7 female industries. In Danish]. [20] Karasek RA. Job demands, job decision latitude, and mental strain: implications for job redesign. Adm Sci Q 1979; 24:285–308. [21] Kristensen TS. Sickness absence and work strain among Danish slaughterhouse workers: an analysis of absence from work regarded as coping behaviour. Soc Sci Med 1991; 32:15–27. [22] Kristensen TS. Use of medicine as a coping strategy among Danish slaughterhouse workers. J Soc Adm Pharm 1991; 8:53–64. [23] Kristensen TS. Arbejdsmiljø, stress og helbred i den danske slagteribranche. København: FADL’s Forlag; 1994. 185 s. (Thesis). [Work environment, stress and health in the Danish slaughterhouse industry. In Danish]. [24] Olesen KG, Thoft E, Hasle P, Kristensen TS. Virksomhedens sociale kapital. Hvidbog. København: Arbejdsmiljøra˚det og Det Nationale Forskningscenter for Arbejdsmiljø; 2008 [The social capital of the companies. A white book]. [25] Pejtersen JH, Kristensen TS, Borg V, Bjorner JB. The second version of the Copenhagen Psychosocial Questionnaire (COPSOQ II). Scand J Public Health 2010;38(Suppl 3): 8–24.

A questionnaire is more than a questionnaire [26] Pejtersen JH, Bjorner JB, Hasle P. Determining minimally important score differences in scales of the Copenhagen Psychosocial Questionnaire. Scand J Public Health 2010;38(Suppl 3):33–41. [27] Pejtersen JH, Kristensen TS. The development of the psychosocial work environment in Denmark from 1997 to 2005. Scand J Work Environ Health 2009;35: 284–93. [28] Llorens C, Alo´s R, Cano E, Font A, Jo´dar P, Lo´pez V, et al. Psychosocial risk exposures and labour management practices. An exploratory approach. Scand J Public Health 2010; 38(Suppl 3):125–136. [29] Bjorner JB, Pejtersen JH. Evaluating construct validity of the second version of the Copenhagen Psychosocial Questionnaire through analysis of differential item functioning and differential item effect. Scand J Public Health 2010;38(Suppl 3):90–105. [30] Kristensen TS, Bjorner JB, Christensen KB, Borg V. The distinction between work pace and working hours in the measurement of quantitative demands at work. Work Stress 2004;18:305–22. [31] Thorsen SV, Bjorner JB. Reliability of the Copenhagen Psychosocial Questionnaire. Scand J Public Health 2010; 38(Suppl 3):25–32. [32] Moncada S, Llorens C, Navarro A, Kristensen TS. ISTAS21: Versio´n en lengua castellana del cuestionario psicosocial de Copenhague (COPSOQ) [Catalan version of

[33]

[34]

[35]

[36]

[37]

155

the Copenhagen Psychosocial Questionnaire (COPSOQ)]. Arch Prev Riesgos Labor 2005;8:18–29. Moncada S, Pejtersen JH, Navarro A, Llorens C, Burr H, Hasle P, Bjorner JB. Psychosocial work environment and its associations with socioeconomic status. A comparison of Spain and Denmark. Scand J Publich Health 2010; 38(Suppl 3):137–148. Nu¨bling M, Hasselhorn HM. The Copenhagen Psychosocial Questionnaire (COPSOQ) in Germany: from the validation of the instrument to the formation of a job-specific database of psychosocial factors at work. Scand J Public Health 2010; 38(Suppl 3):120–124. Nu¨bling M, Sto¨ssel U, Hasselhorn H-M, Michaelis M, Hofmann F. Methoden zur Erfassung psychischer Belastungen [Methods for assessment of psychosocial factors]. Dortmund, Berlin, Dresden: Schriftenreihe der Bundesanstalt fu¨r Arbeitsschutz und Arbeitsmedizin; 2005. Nu¨bling M, Sto¨ßel U, Hasselhorn HM, Michaelis M, Hofmann F. Measuring psychological stress and strain at work: evaluation of the COPSOQ Questionnaire in Germany. GMS Psychosoc Med 2006;3:Doc05. Available from: http://www.egms.de/en/journals/psm/20063/psm000025.shtml SUSESO-ISTAS 21. Manual de uso. Cuestionario de evaluacio´n de riesgos psicosociales en el trabajo [Questionnaire for assessment of psychosocial factors at work]. Santiago: Superintendencia de Seguridad Social; 2009.