ALCOHOL CONSUMPTION IN GREENLAND ‐AN EMPIRICAL ANALYSIS OF RATIONAL ADDICTION
Udarbejdet af Isabelle Mairey Hovedvejleder: Jørgen Trankjær Lauridsen Department of Business and Economics. Bi‐vejleder: Christian Kronborg Department of Business and Economics. Antal tegn: 208.489 Antal figurer: 8 Antal tabeller: 17 Antal tegn ialt: 215.505
Vejlederudtalelse vedr. speciale: Isabelle Mairey: Alcohol Consumption in Greenland – An Empirical Analysis of Rational Addiction. Isabelle Maireys speciale er udarbejdet og bedømt i foråret 2012. Emnet for afhandlingen er en teoretisk og empirisk belysning af om alkoholforbruget i Grønland kan beskrives ved hjælp af teorien om rationelt misbrug. Afhandlingens emne og konklusioner er meget relevante bidrag til såvel den fortsatte forskning i emnerne som den dagsaktuelle debat om samme. Afhandlingen offentliggøres hermed i den form hvori den blev indleveret til bedømmelse, med tilføjelse af et enkelt rettelsesblad. Center for Sundhedsøkonomisk Forskning har intet ansvar for eventuelle fejl eller misforståelser i afhandlingen eller for synspunkter der udtrykkes i samme. Odense, juli 2012 Christian Kronborg
Vejledere Center for Sundhedsøkonomisk Forskning (COHERE) Institut for Virksomhedsledelse og Økonomi
ACKNOWLEDGEMENTS First, I would like to thank Peter Bjerregaard for his great help in guiding me toward the research subject for my thesis and welcoming me at the center for Arctic public health research, National Institute of Public Health. Also his assistance in relation to my visit in Nuuk last fall when I attend Paarisa’s yearly conference on alcohol consumption in Greenland is much appreciated. In Nuuk Birgit Niclasen was also a great help to me. Her help in accommodating me during the first week of my stay and introducing me to everyone involved in alcohol politics in the Department for Health in Greenland was invaluable. Second I would like to thank all the helpful people working with Statistics Greenland for their great help in acquiring data especially Nielsine Joelsen‐Albrechtsen and Johanne Kleemann. Third I would like to thank Cecilia Petrine Petersen and Christina Viskum Lytken Larsen for numerous valuable insights about cultural norms in relation to alcohol consumption in a Greenlandic context. These insights have helped me to understand the Greenlandic culture and on the basis of this I have been able to form a framework in which the development in alcohol consumption in Greenland could be analyzed. Lastly I would like to thank everyone else among my friends and family with who I have discussed my thesis with.
ABSTRACT APRIL 2012 UNIVERSITY OF SOUTHERN DENMARK DEPARTMENT OF BUSINESS AND ECONOMICS MASTER THESIS ISABELLE MAIREY ALCOHOL CONSUMPTION IN GREENLAND – AN EMPIRICAL ANALYSIS OF THE RATIONAL ADDICTION THEORY
Alcohol is the most used intoxicant in Greenland and by far the majority of the adult population consumes alcohol. The social consequences of alcohol use in Greenland have received extensively attention, yet there has not been a single study on the relationship between alcohol use and the price of alcoholic beverages accounting for the addictive nature of alcohol. The purpose of this thesis is to test Becker and Murphy’s (1988) rational addiction model on 40 years of times series data on alcohol imports in Greenland. The empirical relevance for the rational addiction theory is assessed by examining the influence of past and future consumption and contemporaneous prices on current consumption. More specific the rational addiction theory maintains that past and future consumption of the addictive good should have a positive partial effect on current consumption while at the same time existing prices should have the conventional negative effect. Last but not least the short‐ and long‐run price elasticities from the rational addiction model are compared to the conventional price effects from the structural demand function which ignores addiction. The reason why this analysis is important is that ignoring addiction may provide misleading estimates of the price sensitivity of alcohol consumption. This may in turn lead to under estimations of the effect of changes in the price policy such as the hikes on alcohol taxes that have replaced the rationing system from 1979 to 1981 in Greenland. Variables included in the analysis were time series data on liters of pure alcohol imported in Greenland per person 15 years or older, index for development of prices and average per capita income from the period 1970‐2010. In the structural demand model for alcohol demand past and future consumption are endogenous variables and must be treated as such in estimating the model. Solving the endogenity problem amounts to
estimating the demand function by two‐stage least squares, with past and future prices and taxes serving as instrumental variables for past and future consumption. The statistical significance of the coefficient of future consumption provided a direct test of a rational model of addiction against an alternative model in which consumers are myopic. Looking at the magnitude of the parameters of past consumption, which measures the degree of reinforcement of addiction, the estimates in the myopic models of addition for beer, wine and spirituous all supported the mechanism of reinforcement whereby current consumption levels of the three alcoholic beverages depended on past consumption. On the basis of these findings, it can be concluded, that alcohol consumption in Greenland is consistent with addictive behavior. Further, estimation of the rational addiction models for wine quiet strongly reject the myopic consumer hypothesis by confirming that also future levels of wine consumption had a statistically significant partial effect on current consumption. The implication of this was that Greenlandic consumers of wine can be characterized as rational addicts. The estimated short‐ and long‐run price elasticities in the 2SLS model for rational addiction of wine on respectively ‐0.17 to ‐2.0 were considerably higher than the elasticity implied by the conventional model. According to these result, the role of price policy in limiting wine consumption seems to be quit important. Unfortunately no anticipation effect of the type predicted by the rational addiction model was found in the Greenlandic data on beer and spirituous consumption. This none‐finding has important policy implications, especially in relation to beer consumption, since the Greenlandic drinking culture indeed seems largely driven by beer consumption; people who consume beer and spirituous are not far‐sighted in their consumption behavior, and from the price estimates elasticities, they do not respond a great deal to price changes. Since many people consume alcohol in relatively small quantities the distribution of alcohol consumption is more continuous than the bimodal distribution likely to characterize consumption of other addictive goods such as cigarettes and drugs. The aggregated level in data when considering Greenland as one observation over time implies that data can be dominated by the consumer behavior of moderate and light drinkers. An optimal model testing rational addiction should include only heavy drinkers. However, I hope the findings from this study, along with the estimates of price elasticities of primarily wine prove a useful addition to the excising information on the relations between price, taxes and alcohol consumption in Greenland.
Retteark til speciale: Alcohol consumption in Greenland – An empirical analysis of rational addiction Indholdsfortegnelse. Afsnit 4.5 og 4.6 er registreret som hhv. afsnit 4.3 og 4.2 Discussion of results skulle have været afsnit 6.6 og Main findings skulle have været afsnit 6.7 s. 3: Skrivefejl i problemformuleringen Can Greenlandic consumers of alcohol be charaterized as rational addicits? And if so, are they then rational addicts Ordet rational optræder grundet en skrivefejl i øverste linje. I stedet skulle der have stået: Can Greenlandic consumers of alcohol be charaterized as addicits? And if so, are they then rational addicts s. 12: Figur 2.2.1 Her mangler en alternativ efterspørgselskurve D1i figur 2.2.1, Desuden er betegnelsen S for Udbudskurven forsvundet.
s. 60: Dummy variabel Dummie variablen antager værdien 1 i årene 1979 til 1982, og ikke i årene 1980 til 1982, som det står skrevet i afsnittet nederst på siden. s. 83: Skrivefejl Equation 10 = equation 6 Equation 12 = equation 7 Equation 13 = equation8
1.1. OBJECTIVE 1.2. STRUCTURE OF THE ANALYSIS
2. RATIONALE FOR PUBLIC INTERVENTION
2.1. THE COST OF ALCOHOL 2.1.1. NEGATIVE EXTERNALITIES 2.1.2. POSITIVE EXTERNALITIES 2.1.3. NET EXTERNALITIES 2.2. THE ECONOMICS OF TAXATION 2.2.1. IMPLICATION OF ALCOHOL TAXATION 2.3. OTHER REGULATIVE POLICIES 2.4. PRICE ELASTICITY OF ALCOHOL 2.5. MAIN FINDINGS IN RATIONALE FOR PUBLIC INTERVENTIONS
6 7 9 10 11 13 15 16 18
3. ADDICTION MODELS
3.1. THE BASIC MODEL 3.2. A MYOPIC MODEL OF ADDICTION 3.3. A RATIONAL MODEL OF ADDICTION 3.3.1. MODEL SPECIFICATIONS 3.3.1. ILLUSTRATION OF THE DYNAMICS OF RATIONAL ADDICTION 3.3.2. EFFECT OF PRICE CHANGES 3.3.3. ACQUISITION OF ADDICTIVE CAPITAL AND CATALYTIC CONSUMPTION 3.5 DISCUSSION OF THE RATIONAL ADDICTION THEORY 3.4. MAIN FINDINGS REGARDING ADDICTION MODELS
20 21 22 23 26 28 30 31 33
4. ALCOHOL CONSUMPTION IN GREENLAND
4.1. ALCOHOL CONSUMPTION IN GREENLAND, 1950‐2010 4.2. DISTRIBUTION AND HABITS OF CURRENT CONSUMPTION 4.3. EXTERNAL COST IN A GREENLANDIC CONTEXT 4.3.1 HEALTH SERVICES 4.3.2 SUICIDES 4.3.3 CRIME 4.4. DEVELOPMENT IN THE PASS‐THROUGH EFFECT AND THE REAL PRICE OF ALCOHOL 4.3. DISCUSSION OF ALCOHOL CONSUMPTION IN GREENLAND, 1970‐2010 4.2. MAIN FINDINGS CONCERNING ALCOHOL CONSUMPTION IN GREENLAND
36 40 42 42 43 44 45 49 50
5. METHODS AND DATA
5.1. ECONOMETRIC METHODOLOGY 5.2. DATA 5. 2.1. DEPENDENT VARIABLE 5.2.2. EXPLANATORY VARIABLES 5.3. SUMMERY STATISTICS OF THE VARIABLES 5.4. DISCUSSION OF METHODS AND DATA
52 54 56 58 61 62
6.1. PRELIMINARY TESTS 6.1.1. TEST OF ENDOGENITY 6.1.2. TEST OF INSTRUMENTS 6.1.3. TESTING OVER IDENTIFICATION RESTRICTIONS 6.2. MYOPIC MODELS OF ADDICTION 6.2.1. MYOPIC MODEL OF BEER ADDICTION 6.2.2. MYOPIC MODEL OF WINE ADDICTION 6.2.3. MYOPIC MODEL OF SPIRITUOUS ADDICTION 6.3. TEST OF THE MYOPIC ADDICTION MODELS 6.4. RATIONAL MODELS OF ADDICTION 6.4.1. RATIONAL MODELS OF BEER ADDICTION 6.4.2. RATIONAL MODEL OF ADDICTION FOR WINE 6.4.3. RATIONAL MODEL OF ADDICTION FOR SPIRITUOUS 6.5. PRICE ELASTICITIES DISCUSSION OF RESULTS 6.6. MAINE FINDINGS IN RESULTS
65 66 68 71 72 73 74 75 77 78 79 80 81 83 84 85
8.1. CONCLUSION REGARDING ADDICTIVE BEHAVIOR 8.2. CONCLUSION REGARDING RATIONAL ADDICTIVE BEHAVIOR 8.3. POLICY IMPLICATIONS OF CONCLUSIONS
89 89 89
APPENDIX 1: THE STRUCTURE OF ALCOHOL CONSUMPTION IN GREENLAND; 1960‐2010
APPENDIX 2: DEVELOPMENT IN GREENLANDIC ALCOHOL TAXES; 1971‐2010
APPENDIX 3. DEVELOPMENT IN BEVERAGES PRICES; 1971‐2010
APPENDIX 4. DEVELOPMENT IN AVERAGE INCOME; 1971‐2010
APPENDIX 5: DO‐FILES FROM STATA 10.1
FIGURE 2.2.1: Using the pigovian tax to intenalize externalities.
Figure 126.96.36.199: Implications of Becker and Murphy’s rational approach to addiction.
Figure 188.8.131.52: A change in full price of the addictive good.
Figure. 4.1.1.Composition and consumption of alcohol in Greenland, 1949‐1959
Figure 4.1.2: Composition and consumption of alcohol in Greenland, 1960‐1970
Figure 4.1.3: Composition and consumption of alcohol in Greenland, 1970‐2010
Figure 4.4.1. Pass‐through rate of alcohol taxes on real beverage prices
Figure 4.4.2. Development in the relative price of alcohol
Table 4.2.1: The proportion of harmful alcohol consumption patterns registered in the period from 1993 to 2007
Table 4.3.1: The percentage of domestic incidents involving people under influence of alcohol
Table 5.3.1 Summery statistics of the variables
Table 184.108.40.206: Hausman test comparing OLS and 2SLS estimations of the structural equation for alcohol demand
Table 220.127.116.11: Test for endogenity of the Ct 1 and Ct 1
Table 18.104.22.168: Test of instrument
Table 22.214.171.124: Test of P(t 1) , T (t 1) as IV for C (t 1) and P(t 1) , T (t 1) as IV for C(t 1)
Table 126.96.36.199: Estimates of the myopic model of addiction for beer
Table 188.8.131.52: Estimates of the myopic model of addiction for wine
Table 184.108.40.206: Estimates of the myopic model of addiction for spirituous
Table 6.3.1. Test of the myopic model for beer addiction
Table 6.3.2. Test of the myopic model for wine addiction
Table 6.3.3. Test of the myopic model for spirituous addiction
Table 220.127.116.11: Estimates of the Rational model of addiction for beer
Table 18.104.22.168: Estimates of the Rational model of addiction for wine
Table 22.214.171.124: Estimates of the Rational model of addiction for spirituous
Table 6.5.1: Price elasticities
1. INTRODUCTION The purpose of this thesis is to investigate the changes in price sensitivity for alcohol consumption in Greenland when the interdependency of past, current and future consumption is accounted for. This will be done by testing the rational addiction hypothesis on time series data for alcohol consumption in Greenland measured by the importation of alcohol per capita in the period 1970‐2010. Alcohol is the most used intoxicant in Greenland and by far the majority of the adult population consumes alcohol. The average yearly intake of pure alcohol has declined from 22 to 10 liters per person over the past 30 years (Statistics Greenland, 2011). For most consumers, drinking is habitual and strongly related to social interaction as well as celebration, but at the same time alcohol is one of the leading causes of health problems and social problems. Alcohol use causes accidents, diseases and loss of productivity, which all result in social costs. Thus, for this reason the government of Greenland has implemented several policy measures directed at minimizing the alcohol‐related impairments to society. The social consequences of alcohol use in Greenland have received extensively attention, yet there has not been a single study on the relationship between alcohol use and the price of alcoholic beverages accounting for the addictive nature of alcohol. One reason for this could be that consumption of alcohol is not considered to be nearly as addictive as smoking or consumption of drugs. Given the mounting evidence that lifestyle plays a more important role in health outcomes than medical care and treatment, it is relevant to investigate the empirical evidence of detrimental addictive behaviors with regard to the sensitivity to price. This is a much used policy tool in light of taxes and other Government regulations such as minimum legal drinking age for the purchase and consumption of alcoholic beverages. Greenland has a long history of prohibition, rationing and restricting in relation to alcohol consumption. Over time the alcohol legislation in Greenland has undergone many changes and has become less strict though there are still occasionally geographical prohibitions. Instead a progressive tax policy towards alcoholic beverages has been launched. Because of the wide range of different policy tools applied in order to diminish alcohol consumption it is interesting to analyze the impacts on the overall consumption of alcoholic beverages in Greenland over the past 40 years. Before considering how these restrictive means have affected the consumers it will be necessary to look into what characterize the Greenlandic consumers of alcohol.
The available scientific literature and public health reports on alcohol consumption in Greenland mainly focus on the medical or social consequences of heavy alcohol intake. None of these studies recognize the outline and character of the addictive alcohol consumption. This reveals a missing piece in the literature which attempts to account for addiction. Addiction would seem to be the antithesis of rational behavior and past behavioral models of addiction have often assumed that consumption of addictive goods was unresponsive to prices. The rational addiction model by Becker and Murphy (1988) is a relatively new direction that breaks with earlier perspectives of standard economic models in behavioral economics and develops a theory where rationality means a consistent plan to maximize utility over time. The theoretical model of addictive behavior emphasizes the interdependency of past, current and future consumption of an addictive good. Moreover, it is supposed that “…addictions, even strong ones, are usually rational in the sense of involving forward‐looking maximization and stable preferences…” (Becker & Murphy, 1988). The model assumes that consumers of an addictive good are rational in the sense that they can foresee the consequences of their current actions and consistently seek to maximize their utility over time. Becker and Murphy’s model yields several empirically testable predictions. The model maintains that past and future consumption should have a positive partial effect whereas current price should have the conventional negative effect on current consumption, as with normal goods. For aggregated data the two main testable predictions are inter‐temporal complementarities of consumption along with negative price effects and a long‐run price elasticity of demand exceeding the short‐run elasticity. Especially the last of these predictions is potentially highly relevant from a policy perspective. Ignoring the addictive behavior in alcohol consumption in Greenland may provide misleading estimates of the price sensitivity of alcohol and result in an underestimation of the effects of changes in price policy such as those that have been taking place in Greenland throughout the last 40 years.
1.1. OBJECTIVE The primary objective of this thesis is to examine the theoretical background of how alcohol use is affected by prices and to provide the first empirical research of the nature of alcohol addiction with Greenlandic data. In continuation of this, the study is also intended to add research to the small number of existing empirical studies of alcohol addiction. The main research questions are:
Can Greenlandic consumers of alcohol be characterized as rational addicts?
And if so, are they then rational addicts?
Here it is important to emphasize that the main object of this thesis is directed towards analyzing the second and highlighted of the subject above. This leads to more specified aspects, such as looking at differences in consumption patterns between different alcoholic beverages separately. A second aim is to see how, when the interdependency of past current and future consumption is accounted for, the results change with respect to price sensitivity. This thesis is the first study to test the rational addiction hypothesis on times series data on alcohol consumption in Greenland. This is an important analysis since it adds a missing piece to the existing Greenlandic alcohol‐related economic literature.
1.2. STRUCTURE OF THE ANALYSIS Theories of welfare and health economics form the backbone of the study, yet they are not sufficient for a careful examination of the research question. Results from earlier studies of optimal alcohol taxation and price sensitivity to alcohol in other countries provide an outset for comparison with the empirical analysis of alcohol consumption in Greenland. These studies and their results will be reviewed in section 3 on the rationale for public intervention in relation to alcohol consumption where the cost of alcohol will be treated initially. Several earlier studies of alcohol consumption have failed to explicitly recognize the addictive nature of alcohol consumption. Section 4 is still part of the theoretical frame work and here the focus will be on the differences between ordinary lifetime utility maximizing consumption models and models of addictive consumption. The key difference is that rational addicts, in contrast to myopic addicts, recognize that marginal utility of future consumption depends on current consumption. The focus of the thesis is solely on aggregated effects and issues such as individual monetary costs associated with alcohol abuse are not considered in the analysis. After a review of the development in alcohol consumption in Greenland during the period 1959‐2010 in section 4 follows a description of the applied econometrical methods in section 5. After this the examination proceeds to econometric analyses in section 6 in which the hypotheses regarding the models of rational addiction are tested on the empirical data of beer, wine and spirituous consumption in Greenland. The rational addiction hypothesis can be tested against the myopic addiction model, since the consumption in the myopic model will be a function of current and past prices whereas in the rational addiction model, future consumption should be significant as well. Predictions from the rational addiction model regarding inter‐temporal complementarities and negative cross‐price effects along with a long‐run price elasticity that exceeds the short‐run price elasticity will be tested when rational addiction can be statistically confirmed. Section 7 presents a discussion on the objectives, methods and results found in the analysis on rational addiction in Greenland and draws perspectives to results from studies on rational addiction and alcohol consumption in Scandinavia. Finally the thesis is summarized in section 8 where concluding remarks on the implication of the addictive consumption behavior in Greenland and future policy perspectives will be made.
2. RATIONALE FOR PUBLIC INTERVENTION Alcohol has a great magnitude of unintended negative consequences which affect the whole of society. The reason why people still continue to consume alcohol in large amounts could either be because they are not well informed about the risk of excessive alcohol intake, so they are unable to make rational choices about their consumption levels, they may have problems with their self‐control and are simply not able to abstain or maybe they are not directly affected by the external cost and therefore only consider the private cost of their consumption. In a markets economy, the consumer’s sovereignty is often regarded as given. The reason to intervene in private decisions could be paternalistic because there is reason to believe that the consumers of alcohol wish to drink less, but cannot do so on their own. However, public intervention could also appeal to economic efficiency. When drinkers underestimate the probabilities of ill health due to their consumption, the imperfect information provides an efficiency rationale for interventions such as taxes that tend to curb the consumption behavior. The next question is how to intervene effectively. Economics offer two key tools that may be effective in curbing consumption of a targeted bad causing the above mentioned negative external cost. These are taxes and restrictive regulations in the form of limited time span for servings of alcoholic beverages, age restriction in sales of alcoholic beverages and restrictions regarding advertisement. These are all tools that raise the full price of consuming alcohol with the intension of lowering the demand. The following subsection will initially review some of the societal costs associated with excessive alcohol consumption in a population. This will be followed by an introduction of a basic model for the economics of externalities introducing taxation followed by a discussion of the properties and implications of taxation and restrictive regulations as a tool for regulating self‐control mechanisms including a brief comparison of the overall expected welfare gains from taxation and administrative regulation, respectively. The last subsection presents various empirical efforts for identifying price elasticities for alcoholic beverages in order to establish whether alcohol consumption responds to price changes in the same manner as an ordinary good or if alcohol consumption is unresponsive to prices. Hereafter, the findings in this section will be briefly summarized and discussed in relation to the implication for testing the hypothesis regarding the rational addiction model on alcohol consumption in Greenland.
2.1. THE COST OF ALCOHOL
In standard economic theory the term externality refers to an unintended spill‐over effect that is not internalized through prices (Boardman, AE, 2006). External costs, or externalities, occur when a third party is unintentionally affected, regardless of the motivation of the individual causing the harm. In this section the term externality will be applied in relation to quantifying the unintended side effects that people consuming large amounts of alcohol impose on their families, friends, workplaces and to different institutions in society. Awareness of these externalities is important since they are a vital part of the full cost of abusive alcohol consumption and form the basis of legitimate societal concern about heavy alcohol abuse. Heavy alcohol abuse has both internal and external costs. Internal costs are private costs felt by the consuming part. These costs are assumed to have been considered when making the decision to drink alcohol and are therefore regarded as private costs. Calculating the cost of alcohol abuse, in which a corrective tax should be based, is very challenging. As will be seen, all though they might not be visible to the naked eye, the costs stretches deep into society and pose a significant expenditure on especially the health sector. Some costs are immediate while others, such as alcohol related illnesses, occur with a long lag from the initial abuse. This section will explain the nature of several of the negative unintended consequences that comes from heavy alcohol consumption and exemplify how the nature of the externality depends on the surrounding circumstances. Secondly, the subsection will look at some of the positive externalities that could come from a lesser degree of alcohol consumption and briefly describe in what way heavy alcohol consumption can be thought to have controversially positive external effects on society as a whole due to frequent premature death of heavy drinkers. Finally, the last subsection will comment on the overall net externalities when both the positive and negative externalities are added up.
2.1.1. NEGATIVE EXTERNALITIES There is no complete consensus on precisely what consequences from alcoholism should be included in the calculations for externalities, and even when a consensus exists, estimates of the magnitude vary widely depending on the measurement methods applied. The most frequently discussed externality from heavy alcohol consumption is the financial burden of the medical costs of treating alcohol related diseases and injuries. Of course the nature and size of this externality depends on the structure of the health sectors in different countries. The key to external costs is who is affected. If a drinker causes an accident and injures only himself, the costs of this action are private. He has engaged in risky behavior and had an adverse outcome. The nature of the externalities depends on the structure of his insurance. If the insurance coverage is high the magnitude of the externalities depends on the arrangement of the insurance policy and on the financer of the insurance. In countries with low levels of public health insurance the insurance coverage of alcohol related diseases will fall primarily on the persons suffering from alcohol related diseases or injuries themselves. In many cases the level of alcohol intake is private information and it can be difficult for insurance companies to screen for weekly levels of alcohol intake. Because of this, people with high alcohol consumption rarely pay more for their health insurance than people with less alcohol intake, which means that the latter unwillingly subsidize the former since people with a high alcohol intake indorse in more risky behavior. This is a negative externality to the people with insurance who do not have preferences for an excessive alcohol intake. In countries with national health insurance, where everyone is insured and the financing comes from tax revenues, a very large percentage of health care costs are identified as external cost due to excessive alcohol consumption. If the government pays the hospital and medical bills, then these costs may be external costs for other tax payers. But even in this case, it can be argued that costs are not necessarily fully external costs since e.g. an alcohol intoxicated driver has been paying tax and therefore indirectly has contributed to financing his insurance. If, following the example above, the alcohol intoxicated driver on the other hand causes a crash and injures someone else, this generates an external cost. In that case, the external cost exists irrespective of the nature of the insurance structure, since the injured person did not willingly engage in risky behavior.
Another issue often discussed is the impact of alcohol abuse on the abuser’s family. They can be characterized as quasi external costs. When someone dies prematurely from alcohol abuse, family members suffer. The quasi expression is used because most of these costs can be defined as private in the sense that they are incurred by the primary decision making unit in our society, the family. While it is true that alcohol abuse may impose hardship on the family in terms of unemployment, spouse and child abuse, these costs are basically internalized within the family. Again, if government resources are used to address these abuse problems within the family, a case can be made in order for defining these costs as external. The cost of cirrhosis of the liver and other alcohol related diseases constitute both internal and external externality. The cost of pain and suffering are clearly borne by the abuser, while again the treatment cost can be private, social or both depending on the magnitude of government involvement in payment for indigent medical care. Due to the fact that individuals suffering from cirrhosis usually die before retirement but pay taxes and other forms of social security throughout their working lives the overall cost to the government depends on the net costs or benefits. Manning et al. (1989) found that the external costs of smoking with regard to the social security system were negative because heavy smokers died before they received their benefits. A similar set up could be the case regarding overuse of alcohol. There is also the subject of how to value the loss of life. In the human capital approach, value of life is computed as the present value of the discounted expected future earnings of the individual. This is the traditional measure of the value of life and has been used in past studies as well as being the legal norm (Boardman, AE 2006). It has the drawback that since it uses some subjectively determined interest rates to make the calculation, it does not reflect the individual’s attitudes toward risk. Further, it often lacks adjustments for investments made in social insurance funds. The other method for dealing with the cost of fatalities is the risk‐based “willingness to pay” method. If for example consumers are willing to pay a positive amount to reduce the risk of a car accident, then the value of lives lost in a car crash can be inferred from the response. Unfortunately, responses to this question are often based on hypothetical situations and not on actual behavior (Boardman, AE 2006).. As a result, the answers may not be accurate and empirical estimates using this approach show a great deal of variation. Regardless of which technique is used, it should be borne in mind that in the studies to date, most of the dollar value of the loss of life is a private cost. Loss of life is an external cost only for third‐party victims, such as those killed by drunken drivers. Similar considerations apply to morbidity costs.
2.1.2. POSITIVE EXTERNALITIES A light to moderate intake of alcohol has been suggested to reduce morbidity and mortality (Maclure M., 1993; Shaper AG., 1990; Klatsky et al., 1992). Moreover, the social side of drinking may have positive externalities in several ways. An Australian study found that those who drink up to three drinks a day are far happier than those who never drink (Robert A. Cummins, 2008). The same study found that people who did not drink at all had the lowest wellbeing of all drinking categories. Also labor market externalities both in terms of productivity and job opportunities can be positively affected since drinking as a societal and cultural norm might extend the network between individuals, making it easier to find employment. Socializing might also strengthen the bonds between co‐workers and thus have a positive effect on productivity. Many companies are aware of this and take the social side of drinking into consideration by serving alcohol to mark successful events and to promote the cohesion of staff in celebrations, while alcohol is prohibited in the normal operative functions. A recent study from Norway based on survey data found that alcohol consumption increases with a higher level of education and higher income (Strand BH., Steiro A., 2003). Even though this might be due to an income effect, it could also be due to a link between alcohol and extroversion. On a more societal level, people with high alcohol intake do not collect the full amount of their retirement benefits which they would otherwise have been entitled to. When comparing the average life expectancy for people in Denmark in the years 1997‐2001, a Danish study (2006) found that people who had an alcohol intake above the recommended level of alcohol died approximately 4‐5 years earlier compared to people with an average alcohol intake. This is a common finding. In addition, they do not impose the large nursing home and medical costs that usually occurs at old age. In this way they impose a positive externality onto society by subsidizing people with an alcohol intake within the recommended limits. Although it may seem inappropriate to calculate the societal benefits from premature death, the magnitude is offsetting when calculating the monetary value of this externality. The report on risk factors and public health in Denmark (2006), despite an overall net loss to society, estimate an external saving due to premature alcohol related deaths to a total of Dkk 501 million in the health sector.
2.1.3. NET EXTERNALITIES Adding up the positive and the negative external effects leads to a measure of the overall net effect of the action causing the unintended externalities. Efforts to calculate the net externality costs of alcohol are, as the previous sections show, not an easy task. Despite the apparent measurement problems, studies attempting to do so often find a negative net externality. Earlier studies have linked alcohol consumption to a considerable disease burden and large health care expenditures. Findings from studies based on individual level data from Copenhagen City Heart Study found a U‐shaped association between alcohol consumption and risk of admission for men, both regarding any admission and admissions due to alcohol‐related diseases. However, the results regarding women implied that the average weekly alcohol consumption was only vaguely associated with risk of admission and duration of hospitalization. The previously mentioned Danish study from 2006 reporting a net cost in the health sector of Dkk 597 million annually due to alcohol related treatments, also estimated an alcohol related production loss in Denmark amounting to Dkr 7.299 million per year based on the human capital method. The loss in production was mainly due to findings of the fact that people in the ages 25‐64 with high levels of alcohol consumption often leave the labor force early because of premature death. The economic cost to society is to some extend counterbalanced by a saving in future expenditures of Dkk 5.406 million due to earlier death among people with high alcohol abuse. Since the total net cost in production of 1.893 million Dkr is negative, heavy alcohol intake was evaluated as imposing a net externality to society. Studies investigating the relation between taxes on alcoholic beverages and the external cost of alcohol consumption, mainly in the United States, reported these external abuse costs as approximately USD 175 per gallon of pure alcohol in 1991 prices (Pogue & Sgontz, 1989. Manning, Keeler, Newhouse, Sloss & Wasserman, 1989). On the basis of this, they found that current taxes on alcoholic beverages was increasingly deemed too low to cover the external costs, despite the welfare losses they entail for drinkers who do not impose costs on others. Many of the studies estimating net externality cost omit monetary estimates of externality cost to the alcohol abuser’s family because the family is considered to be one economic unit. As discussed above, including these costs would have a significant effect on the magnitude of the cost, since families suffer through generations. The implication of this is an underestimation of the negative externality costs, and with the traditional externality‐based taxation policy the taxes set to internalize these external costs might not be optimal since all costs are not considered to the extent that they influence the externalities.
2.2. THE ECONOMICS OF TAXATION In a perfect competitive market where the price of a good adjusts to ensure that all trades which benefits both the seller and the buyer simultaneously will have effect, the introduction of a tax onto this good will distort the market equilibrium and create an inefficient market by introducing a deadweight loss. The term deadweight loss is used when the price received by the seller is less than the price paid by the buyer resulting in fewer trades and from this a welfare loss due to lesser market participation (Boardman, AE 2006). The value that is lost is defined as the deadweight loss from establishing the tax. From findings in the previous section regarding the cost of alcohol it is apparent that the market for alcohol is an imperfect market due to the relatively large amount of negative externalities associated with excessive alcohol intake, most importantly the public health care costs and costs that appear because of risky behavior causing accidents to others and implications not felt by the drinker. In the presence of negative externalities, the social cost of a market activity is not covered by the private cost of the activity. In such a case, the market outcome is not efficient and may lead to over‐consumption of the product. It is possible that the market can be corrected by introducing a tax on the commodity. While a revenue maximizing government would choose to impose high taxes on goods with relatively inelastic demand functions, since this would result in the highest level of revenue, a welfare maximizing tax revenue should only be adequate to compensate for the external costs of alcohol consumption by reflecting the full social costs of consumption and hereby increase economic efficiency and overall welfare. By that standard, it is not appropriate for the tax rate to take advantage of the inelastic demand or incorporate the drinkers’ heightened risk of injury, since this private cost is assumed to have already been taken into account when making the consumption decisions. On the other hand, to the extent that drinkers place others at risk of injury this should be reflected in the welfare maximizing tax rate, for at least two reasons. First, it is only fair that drinkers should compensate the public for the external costs of their choices. Second, if alcohol prices do not reflect the full social costs of consumption including the external costs, then consumers will drink too much, in the technical sense that the of margin their drinks will be worth less to them than the cost they pay. If the principle of consumer sovereignty is accepted, then it is the magnitude of the external costs that are relevant for the policy purposes on which a tax should be based.
FIGURE 2.2.1: USING THE PIGOVIAN TAX TO INTERNALIZE EXTERNALITIES
Figure 2.2.1 depicts in simple terms how the introduction of a tax internalizes the negative externality by raising the price paid by the consumer from P1 to P2 and hereby lower the quantity consumed from the private optimum level Q1 to the social optimal level Q2. The tax can equalize the marginal private cost and the marginal social cost if the government can accurately estimate the social cost. In more specific terms, the drinker would have to pay for the non‐pecuniary externality that is created. This would effectively reduce the quantity of alcohol consumed, moving the economy back to an efficient equilibrium. The alternative equilibrium at point A obtained from D1, where the reduction in consumption is less than the original reduction, illustrates that the attainable reduction in alcohol consumption imposed from a tax depends on true nature of the demand curve. Generally, the greater the demand elasticity in absolute terms, the greater the corresponding reduction effected by the excise tax. As the above illustration indicates, taxation can be an effective policy instrument to internalize external cost of an activity if an approximation to the true demand curve can be estimated.
2.2.1. IMPLICATION OF ALCOHOL TAXATION A recent review of 112 studies about the effect of alcohol tax confirmed that when alcohol taxes go up, drinking goes down, also among problem drinkers and youth (Wagenaar et al., 2009). Hence, excise taxes on alcoholic beverages can be effective alcohol‐control measures that can be used to promote public health but a great deal of issues need to be considered when deciding the actual level of a tax. To determine the optimal tax on alcohol, the reduction in external costs must be weighed against the welfare cost of the tax. The introduction of an alcohol excise tax or an increase in an already persisting tax rate results, as illustrated in figure 1, in a loss in consumer surplus. This loss is defined as the difference between the value of alcohol in a competitive market and the price the consumer ends up paying for the good after the tax has been introduced. Consumer surplus arises because they pay the same price for each unit of the good. This price equals the marginal value of the last unit purchased by the consumer since other units are more highly valued than the marginal unit. The welfare loss of the tax equals the loss in consumer surplus minus the revenue generated by the tax. On the basis of the external cost of USD 175 per gallon of pure alcohol found by Manning (1989), Pogue and Sgontz (1989), Manning et al. (1989) and Saffer and Chaloupka (1994) estimated the optimal tax on distilled in 1991 at respectively 73, 78 and 79 USD per gallon of pure alcohol as compared to the actual USD 35. This suggested that present time alcohol taxes could be more than doubled before the costs of the increased tax would exceed the benefits. The tax in itself is an important device in generating revenue for the government, but for the consumer a more important issue concerns the effect of tax increases on actual retail prices of alcoholic beverages. This was not considered in the above example. A comparison of the quantity of excise tax in the price of a liter of pure alcohol in 74 countries ranged from as much as 44.9 % percent in Norway to as little as 0.3 percent in Kyrgystan with an average of 17.3 percent (WHO 2011). From this it is likely that a consumer in Norway will be more sensitive to an increase in alcohol taxation than a consumer in Kyrgystan, all else equal. The industry producing and selling alcoholic beverages reacts to increased taxation be adjusting their pre‐ tax prices either upwards or downwards according to profit maximizing price levels. If drinkers are assumed to adjust their consumption according to the beverage prices then the design of an optimal tax based on economic efficiency needs to adjust for price changes that might exceed the change in taxes.
The extent to which changes in excise taxes can in fact be passed on to the final prices is an empirical question depending on the structure of the market. Studies by Cook (1981), Kenkel (2005), Young and BielinskaKwapisz, (2002) addressing this issue consistently found that in the U.S. market of alcoholic beverages, an over shifting of excise taxes occurs, meaning that prices for alcoholic beverages had gone up by more than the amount of tax increases in the same period. The rates by which tax increases were passed on to final retail prices, known as the pass through rates, ranged from 1.2 to 4.2, depending on specific brands and types of alcoholic beverages analyzed as well as on sales location. In this case, a tax rate designed to move consumption towards the Pareto‐efficient level may reduce consumption below intended level. Also, this can be an issue if the price increase causes incitements to induce smuggling, home production or substitution away from industrialized alcohol towards other intoxication products containing ethanol. These examples indicate that several concerns should be taken into account when calculating the optimal level of alcohol taxes. Consumers differ greatly in the quantity of alcohol they consume and several non monetary costs associated with alcohol intake do not increase in a linear manner. For example, liver cirrhosis incidences increase with the level of daily alcohol intake and with the duration of a steady intake above recommended levels (Risk factors and public health in Denmark. 2006). This means that consumers contribute with different levels of negative externalities depending on their consumption levels and habits. Unfortunately, it is difficult to account for this with a uniform excise tax. WHO notes that in some countries it is “official policy of the pricing system to steer people towards a particular type of low‐alcohol or non‐alcoholic beverage, in order to substantially reduce risky or high blood alcohol levels” (WHO 2004). Ideally, the marginal tax rate would reflect the total marginal cost of drinking. Since this way of taxing alcohol is not empirically possible, an alcohol tax will unavoidably lead to some level of individual over‐taxation of consumers who only consume alcohol occasionally and in small amounts, and to under‐taxation of heavy lifetime drinkers. From a health perspective alcohol taxes should also make alcohol beverages equally or progressively less affordable by offsetting the combined effect of inflation and increased consumer incomes and purchasing power. This however requires periodic increases in taxes to maintain the impact. Otherwise alcoholic beverages will become increasingly more affordable because taxes do not keep up with inflation and income. 14
2.3. OTHER REGULATIVE POLICIES In addition to tax‐related policies, several other regulations may also directly and indirectly affect the price of alcoholic beverages. These policies include, among others, restrictive regulation on wholesale and retail distribution, restricted advertisement or the imposing of minimum pricing policies. Research on the impact of these policies on the price drinkers pay for alcoholic beverages, drinking behavior and associated consequences are very limited. In practice, when restrictive controls have been implemented, results have been somewhat ambiguous. For instance, during prohibition in the U.S., alcohol‐related problems appeared to both decrease as measured by cirrhosis mortality rates and increase as measured by the illegal trade in alcohol and associated criminal activity (Gruenewald 1994). Drinking‐age hikes have been shown to have positive effects in reducing alcohol consumption, alcohol abuse, and motor vehicle accidents (O´Mally and Wagenaar, 1991). Minimum‐age laws are targeted at the population sub‐group that accounts for a disproportionate share of accidents and violence episodes and potentially would benefit most from changing their alcohol habits. Several factors are important here. First, one factor that favors the drinking‐age policy over tax hikes is the welfare loss that all segments in a population endure from a tax hike. Second, two factors against administrative regulation are, as mentioned by Grossman et al. (1994), the fact that enforcement and administrative costs associated with a uniform drinking age law are likely to be higher than those associated with a tax hike. And third, as mentioned by Kenkel (1993), the loss in consumer surplus due to a rise in legal drinking‐age cannot be partially offset by an increase in tax revenue. This leads to an important point regarding administrative regulation. The outcome of an administrative policy is relatively easy to predict, but it can be expensive to enforce and the loss in consumer surplus is in general not followed by an increase in government revenue. Advertising is thought to affect alcohol consumption through either information or persuasion. Bain (1956) found that advertisements differentiate one brand from another, creating increased brand loyalty. In contrast, if advertisements are viewed as information, advertisements can lower equilibrium prices and provide a better match of consumer’s preferences and feasible consumption bundles (Stigler and Becker 1977). This, however, does not imply an increased demand for the overall product category. Becker and Murphy (1993) proposed the assumption that if commercials increase the consumer’s marginal utility from consuming the good then advertisements will raise total consumption of the product. Empirically, studies have reported different results but Saffer (1993, 1997, 2003) found alcohol advertisements to be positively correlated with drinking levels, motor vehicle fatalities and to be especially effective on young drinkers. If this is true, restricting advertisement would reduce overall alcohol consumption.
2.4. PRICE ELASTICITY OF ALCOHOL The effect of taxation and other regulative policies depends on the price elasticity of demand on alcohol. This is a measure of the responsiveness of the quantity demanded of a good to a change in its price. It is defined as the percentage change in quantity demanded in response to a one percent change in price holding all other determinants of demand constant. The formula for the coefficient of price elasticity of demand for a good is given by1:
% change in quantity demanded Qd / Qd % change in price P / P
The above formula usually yields a negative value, due to the inverse nature of the relationship between price and quantity demanded for most goods. The review of 112 studies and 1003 price elasticity estimates of alcohol made by Wagenaar et al. (2009) found a total mean price elasticity of ‐0.51 on overall alcohol consumption when considering both aggregated and individual level studies. The total mean effect of price on beer consumption was ‐0.46, for wine it was ‐0.69 and for distilled spirits it was ‐0.80 also considering both aggregated and individual level studies. On the basis of 10 individual level studies concerning the effect of alcohol prices on heavy alcohol use a mean elasticity of ‐0.28 was found. Another extensive review of the economic literature on alcohol demand by Ornstein and Levy (1983) concluded that, based on studies using aggregate data, the price elasticity of demand for beer, wine, and distilled spirits are ‐0.3, ‐1.0, and ‐1.5, respectively. While the elasticity estimates for wine and spirits are somewhat smaller, the elasticity for beer is lager and suggests that beer consumption is relatively insensitive to price changes, whereas demand for wine and distilled spirits is more responsive to price. Although the findings from other studies are mixed in respect to the relative price sensitivity of abusive and non abusive drinkers, most studies have reported that heavy drinkers normally are less responsive to price changes than light drinkers2. Cook and Thauchen (1982) estimated that an increase in excise tax by one dollar (1967 prices) per gallon of pure alcohol would reduce the liver cirrhosis mortality rate by 5.4 percent indicating an influence from tax hike on to chronic excess consumption. Manning et al. (1995) found, using data from the 1983 National Health Interview Study, that moderate drinkers are the most responsive to changes in the price of alcohol, with an estimated price elasticity of –1.19, while both light and heavy drinkers have a price elasticity that is
Varian. H. (2006): Chaloupka and Laixuthai 1997; Chaloupka and Wechsler 1996; Cook and Moore 2000; French et al. 2006; Keng and Huffman 2007; Kenkel 1993, 1996; Laixuthai and Chaloupka 1993, Sloan et al. 1995; Stout et al. 2000
close to zero. Kenkel (1996) found somewhat different results using the 1985 wave of the same survey. He calculated separate elasticity estimates for heavy drinking by gender and found price elasticity of heavy drinking for men and women to be –0.52 and –1.29, respectively indicating that both men and women are responsive to changes in price, although women are much more so. Analyses using individual level data suggest that alcohol demand may be even more responsive to price than aggregate estimates indicate, possibly because this approach can obtain differential price responses among respondents of different age groups (Wagenaar 2005, Leung and Phelps 1993). There are reasons to argue that alcohol consumption by young people may be more responsive to prices than alcohol consumption by adults. Coate & Grossman (1987) found that price increases led to larger reductions in the fractions of heavy and fairly heavy adolescent drinkers than in the fraction of light drinkers. This was confirmed by subsequent studies by Cook & Moore 2001, Laixuthai & Chaloupka 1993 mand Young & BielinskaKwapisz 2006. Furthermore Laixuthai & Chaloupka (1993) reported that young adolescent drinkers were more price sensitive than older adolescent drinkers, which is a particularly important finding because Cook & Moore (2001) provided evidence supporting the notion that drinking habits are formed at young ages. Differences in gender and age groups also have significant implications. A study based on self‐reported data on drinking and driving obtained in the 1985 National Health Interview Survey estimated that a 10‐percent increase in the price of alcoholic beverages would reduce the probability of drinking and driving by about 7.4 percent for men and 8.1 percent for women (Kenkel, 1993). An even larger reduction of 12.6 percent among men and 21.1 percent among women would occur among people aged 21 years and younger. Chaloupka & Wechsler (1996) examined the effects of various factors on drinking and binge drinking among students in U.S. colleges and universities. The investigators estimated the potential results of a policy that would have adjusted alcohol taxes for the rate of inflation since 1951. The results implied that such a policy would have reduced the number of underage college women who drank in the past year by about 15 percent and the number of underage and older college women engaging in any binge drinking by roughly 20 percent. In contrast to these statistically significant negative effects of price on underage drinking and binge drinking by female students, no such effect was found for male students. These findings, on the whole, indicate that, in contrast to conventional addiction theory; the demand curves for alcoholic beverages can be negatively sloped. A number of the above studies do not take the nature and mechanism of addiction into consideration.
2.5. MAIN FINDINGS IN RATIONALE FOR PUBLIC INTERVENTIONS Alcohol consumption is perhaps the most common example of a legally addictive good. Unlike other addictive substances many persons regularly consume small quantities of alcohol without harming them self others. Moderate alcohol intake has even been shown to have positive outcomes in relation to socialization and has been shown to lower the risk of coronary heart disease in men. The adverse effects of alcohol come from overuse or misuse of alcohol. The overall cost of excessive alcohol consumption poses a negative net cost to society when monetary gains from premature deaths have been included. This net loss forms the legitimate reason for public interventions in order to reduce excessive consumption to a level where the private cost of consuming alcohol are equal to the social cost of enjoying alcohol. This can be done by the introduction of taxation or other regulative policies. Consequently, the responsiveness of alcohol to price changes is an important parameter in determining the optimal level of taxation or the actual impact of legal regulative policies. The conventional wisdom is that addictive goods are not sensitive to prices because small changes in such goods cause large changes in the marginal benefit of consumption. However, review of previous empirical studies on the relationship between prices and alcohol consumption seems to reveal a reversed correlation indicating negative price elasticity on the demand for alcoholic beverages. As mentioned in section 2.2.1, several studies have found that the prices of alcoholic beverages are strongly influenced by taxes, and on the basis of this, tax hikes have turned out to be effective in reducing the overall alcohol consumption. In relation to the object of this thesis findings in the above section are important since the insinuation of negative price relations, which is in accordance with the rational addiction theory, suggests that increasing the price of alcohol in Greenland through taxes will reduce consumption and hereby also the negative external cost, even if alcohol can be characterized as an addictive good. Because alcohol intake in large amount can lead to an addiction it is relevant to consider this when analyzing how prices influence consumption. The definition of addiction and the implication of how consumption of addictive products is fundamentally different from consumption of non‐addictive products is the topic of the following section.
3. ADDICTION MODELS “Whether a commodity conforms to the law of diminishing or increasing returns, the increase in consumption arising from a fall in prices is gradual; and further, habits which have once grown up around the use of a commodity while its price is low are not quickly abandoned when its price rises again” (Marshall, 1920). Alfred Marshall was the first economist to take an interest in addictive behavior. As he observed in 1920, consumption did not always respond to prices alone, but also to habits of consumption. By this observation he noted the differences between the short‐run and long‐run price responses, which play an important role in economic models of addiction. The psychological effects of harmful addiction are characterized by the mechanisms of reinforcement, tolerance and withdrawal. Reinforcement means that greater current consumption of a good raises its future consumption. Tolerance means that given levels of consumption are less satisfying when past consumption has been greater. Withdrawal refers to the negative physical and mental reactions that arise as consumption of an addictive substance is terminated. From this, consumption of an addictive good makes it progressively more costly in the short‐run to cease an addiction. Due to this blend of effects, a successful treatment of harmful addictions has proven difficult. These insights are incorporated into the economic approach to addiction. In the case of harmful alcohol addiction, an increase in past consumption lowers current utility through the effect of tolerance. At the same time, the marginal utility from current consumption is raised through the reinforcement effect. Lastly, the effect of withdrawal is present because a reduction in the addictive good lowers current utility. Reinforcement and withdrawal, but not tolerance, are also present in the case of a beneficial addiction. Early research considered addictive behaviors such as excessive alcohol abuse irrational and therefore unsuitable for conventional economic analysis. The proposition that price might be an important determinant of alcohol abuse has been treated with a good deal of skepticism. The following evaluation of alternative economic theories of addiction is based on their ability to contribute to an understanding of behavior, particularly price sensitivity, at the empirical level. Empirical research using a rational addiction framework is relatively new while similar research using myopic models is reasonably old.
3.1. THE BASIC MODEL Standard consumer theory treats consumption choices as temporally independent so that decisions at one point in time are independent of those at other times. This assumption about interdependence is built into models of optimal consumption over time by assuming a utility function (V) that is additively separable over time. This utility function has the form; V
U (C t )
Here Ct is consumption of a composite good at time t , U (Ct ) is the current period concave utility function at time t and is the time discount factor given by 1 /(1 r ) where r is the rate of time preference for the present. From this, lifetime utility (V ) is the discounted sum of utility in each period. If equation 1 is maximized subject to a lifetime wealth constraint and if the rate of interest is equal to the rate of time preference, the optimal quantity of Ct can be expressed as a function of the marginal utility of wealth (W ) and the price of C at time t given by Pt , as follows; C t C t (W , Pt )
Equation 2 specifies a demand function that holds the marginal utility of wealth constant. This is the standard demand function used in lifecycle utility maximization models (Heckman, 1974; Ghez & Becker, 1975). It is important to note, that both of the above equations rule out the effect of addiction because an increase in consumption in period t affects utility in period t alone and therefore has no effect on future utility. To allow for the possibility that a good is addictive, the utility function at time t should be rewritten as; U U (Yt , C t , S t )
Again U is considered a strongly concave function of Y , C and S . Here yt is consumption of the non‐ addictive good, ct is consumption of the addictive good and st is the stock of the addictive good. Past consumption of c affect current utility through a process of “learning by doing”, as summarized by the stock of “consumption capital” c . Both myopic and rational models of addiction stress that choices made today depend on choices made in the past. 20
3.2. A MYOPIC MODEL OF ADDICTION In myopic models of addiction, past consumption fuels current consumption, while at the same time individuals ignore the future consequences when making current consumption decisions. Most economists who have studied consumption behavior in relation to habits or addiction have assumed myopic or imperfect rational behavior. The focus here will be on two types of myopic models. In one, consumers ignore the effect of current consumption on future utility when they determine the optimal quantity of an addictive good. In this class of models, past consumption affect current consumption through an accumulated stock of habits, the term “reinforcement” from earlier. Since the stock measure in the current period utility function given by (3) depends on past decisions, this class of models is considered as one in which tastes are endogenous (for example, Houthakker and Taylor 1970; Pollak 1970, 1976; von Weizsacker 1971; El Safty 1976; Hammond 1976). In the second class of models, consumers have stable but inconsistent short‐run and long‐run preferences (for example, Schelling 1978, 1984a, 1984b; Elster 1979; Winston 1980; Thaler and Shefrin 1981). Thomas Schelling (1978) described a smoker who wants to quit the habit in this manner: "Everybody behaves like two people, one who wants clean lungs and long life and another who adores tobacco.... “. The two are in a continual contest for control. Or as Thaler and Shefrin (1981) characterized it "At any one point in time the individual is both a far‐sighted planner and a myopic doer with the planner and the doer in conflict”. In these imperfectly rational addiction models, the planner is often in command most of the time, but the rebellious side only needs to get occasional control to spoil the planners best laid plan. Those models in which consumers have spilt personalities have experienced limited application in empirical studies of the economics of addiction and the understanding of myopic behavior in this thesis is conceptually directed toward the first class of myopic addiction models. The myopic model does reflect the dependence of current consumption decisions on past behavior that characterizes the use of an addictive substance, and the long‐run effect of a permanent price change will exceed short‐run effect. However, the consumer in the myopic model of addiction ignores the future implications of addictive consumption when making current consumption decisions. An anticipated increase in future price will therefore not change current consumption since current consumption only is affected by consumption in the previous period.
3.3. A RATIONAL MODEL OF ADDICTION Contrary to conventional wisdom, Becker & Murphy (1988) stress that demand for an addictive goods is likely to be quite responsive to prices. So, even though the economic discussion of rational consumption of addictive goods had existed prior to the model developed by Becker and Murphy, it is now the most widely regarded addiction model in contemporary economic literature. It has been strongly supported empirically, and it will be the basis for the evaluation in this thesis on whether alcohol consumption in Greenland can be characterized as a rational addiction. The model follows prior literature from Ryder & Heal (1973), Stigler & Becker (1977), Spinnewyn (1981) and Iannaccone (1986) by considering the interaction of past and current consumption in a model with utility maximizing consumers. These authors focus on the optimal path of consumption when the utility function is not additively separable over time. That is, the lifetime utility function is given by the discounted sum of current period utility functions given by equation 3 rather than by equation 1. The model of rational addiction is constructed upon an interaction of standard demand theory, techniques ordinarily applied in capital theory and behavioral science in a “rational choice” framework where consumers of potentially addictive goods take account of future effects of current consumption in determining the optimal consumption quantity of an addictive good in the present period. More formally, consumers optimize intertemporally‐dependent preferences given time‐consistent exponential discounting of future utilities, perfect foresight and are subjected to an inter‐temporal budget constraint. The model avoids issues of time inconsistency and self‐control by assuming constant exponential discounting and assumes endogenous tastes where current consumption motivates future consumption given full prices reflecting market price and costs of use in terms of future addiction. The aim in any given period is the maximization of discounted lifetime utility. This utility is assumed to be a function of two goods, alcohol in this case and a composite good representing all other goods in the economy, as in equation (3). Rational addiction theory does not assume that consumers necessarily derive positive utility from being addicted. On the contrary, addiction can be caused by unhappiness due to negative life events whereby consumption of the addictive good is a form of self‐medication. The essence is that addicts would be less happy if prevented from consuming the addictive good. Lastly, it is important to mention that the rational addiction theory states that the initial unhappiness may be due to a temporary event, and the danger of starting an addictive consumption is that this temporary event may trigger a permanent addiction. 22
3.3.1. MODEL SPECIFICATIONS Alcohol is assumed to be addictive. The addiction is captured in the mathematical model allowing the utility derived from alcohol to depend not only on the level of current consumption but also on levels of past consumption. This process is expressed in a conduct of “learning by doing” or the term reinforcement from the previous section. The standard model of demand is modified by including the addictive capital stock detonated S as in (3). This addictive capital stock represents an economic measure of the level and strength of the addiction and is not necessarily a physical object.
The addictive capital S is assumed to increase with the consumption of alcohol and decrease over time. If the addiction is harmful an increase in the stock lowers utility U / S t U s 0 , while an increase in the stock raises utility if the addiction is beneficial U / S t U s 0 . The growth function is given by: S ct S t
The parameter δ reflects the rate of exogenous depreciation of the physical and mental effect of past consumption of the addictive good. From this it follows that the addictive capital will decrease in each period with the rate of δ unless alcohol is consumed to restock the addictive capital stock. With a length of life equal to T and a constant time preference σ, σ<0 Becker & Murphy assume that utility in the initial period will be given by;
U (O )
e t u y (t ), c(t ), S (t )dt
The utility function of the current period U (t ) is well behaved in the sense that an indifference curve between Ct the addictive good and Yt the non‐addictive good is convex to the origin. Also the lifetime utility function is separable over time in Yt , Ct and the stock of the addictive good S t . However, it is not separable in Ct and Yt alone. The rate of time preference is an important variable in the above lifetime utility function since this variable will determine how forward looking the consumer of the addictive good is. Present oriented individuals are potentially more addicted to harmful goods than future oriented individuals. 23
The reason for this is that they do not account for the future repercussion of their current behavior as much as an individual with a lower rate of time preference. An example, given by Becker & Mulligan (1997a), is that a young person with a high rate of time preference will be more willing to engage in risky behavior either because the future is not considered important or because of a lack of experience in perceiving the future correctly. With perfect capital markets and a constant rate of interest a rational consumer will, according to Becker and Murphy’s model, maximize the above utility function in subject to a budget constraint where the level of addiction is allowed to affect the level of income through its effect on the wage rate. When the utility function is quadratic and the rate of time preference for the present is equal to the market rate of interest, equation 3 generates a structural demand function for consumption of the addictive good c of the form:
Ct Ct 1 Ct 1 1 Pt 2 et 3 et 1
The mathematical derivation can be found in Becker & Murphy 1988. All variables except the error terms are defined in the first section of this paper. The term et contains unmeasured life cycle variables and et 1 measures future unmeasured life events.
measures the effect of an increase in past consumption on the marginal utility of current consumption. Greater value of is equivalent with a greater degree of reinforcement. Since is positive and 1 is negative, current consumption is positively related to past and future consumption and negatively related to current price. That is, cross‐price effects are negative, or stated differently, the quantity of an addictive good consumed in different periods are complements. These predictions pertain to a demand function that holds the marginal utility of wealth constant. As indicated in the previous section, this is the standard demand function employed in life cycle utility maximization models. Past prices are relevant because they affect past consumption of the addictive good, and a decrease in past consumption due to an increase in past prices reduces current consumption. The future price is relevant because its reduction raises future consumption which lowers the shadow price of current consumption. Stated in a different way, the demand function of a rational consumer exhibits the property of symmetry since an increase in past or future consumption cause current consumption to rise.
Another important implication of the rational addiction model implied by equation (6) is the differences between long‐ and short‐run responses to permanent price changes in the case of addiction. This can be seen more formally by solving the second‐order difference equation in (6) as has been done by Becker, Grossman and Murphy (1994, appendix A5). The effect on consumption in period t of a permanent reduction on price beginning in period t detonates the short term price elasticity. With t equal to 1, the expression in (7) gives the effect on current consumption of a completely unanticipated permanent reduction in price: dCt 1 dPt (1 )2
The effect of a permanent reduction in price in all periods on consumption in period t detonates the long term price elasticity. As t goes to infinity, the long run effect of a permanent reduction in prices can be approximated with the expression in (13). dC 1 dP (1 )(2 1)
Here 1 and 2 are roots of the second‐order difference equation in (6). The ratio of the long and short term effects is 2 /(2 1) which exceeds 1 since 2 1 for stability. A detailed derivation can be found in appendix 5A in Becker et al. (1994), and although a formal review of the proof of these roots are beyond the scoop of this thesis, an evaluation of appendix 5A would insinuate that the full solutions of the difference equation in (6) would require a rather complicated numeric optimization process in order to find the roots of the inhomogeneous difference equations. The solutions in appendix 5A are applied to the expressions presented in the results in section 6.5, although it is noted, that the solutions in 5A appears to be solutions to a homogeneous version of the actual second‐ order difference equation in (6). The main prediction in relation to these price elasticity expressions is that since ct 1 remains unchanged if a price change is not anticipated until period t, the long‐run price effect per definition exceeds the short‐run price effect.
3.3.1. ILLUSTRATION OF THE DYNAMICS OF RATIONAL ADDICTION The following illustration shows the demand for an addictive good c conditional on a fixed outline of current and future prices. The figure is a reproduction of the model presented in Becker, Grossmand and Murphy's paper from 1991. The consumption capital S (t ) is measured along the horizontal axis and consumption c (t ) is measured along the vertical axis. The positively sloped curve A1 provides the relation between c and S for an addicted consumer who has a particular utility function, who faces given prices of c and Y and who has a given level of wealth. Depending on the point of departure it can be determined whether the addictive capital stock is decreasing or increasing over time. If the demand curve given by A is above the line where c S then the addictive 1
capital will increase over time. When the consumption level of the addictive good exceeds the level of depreciation, then the level of capital stock will be positive. Opposite, if the consumption level of the addictive good is less than the depreciation rate, then the level of the capital stock will fall. When consumption is equal to the depreciation rate, then the addictive capital will not change. This constitutes the term steady state. The line c S gives all possible steady states where c and S are constant over time. FIGURE 126.96.36.199: IMPLICATIONS OF BECKER AND MURPHY’S RATIONAL APPROACH TO ADDICTION.
Since c is an addictive good the initial consumption stock S 0 depends on past consumption and past life cycle events. Acquisition of this initial addictive capital will be treated in a separate section to come. Both c 1
and S grow over time when S 0 is in the interval where A is above the steady‐state line. The degree of addiction is very sensitive to the initial level of addictive capital. If S 0 is below S '1 in the figure, a rational consumer would eventually lay off the addictive good. On the other hand, if S 0 is above
S '1 , then even a rational consumer will become addicted and end up consuming large quantities of the addictive good c. 1
From figure 188.8.131.52 it can be seen that there are three steady states. The intersection between the curve A
and the steady‐state line indicates two steady states at c'1 S '1 and c *1 S *1 and lastly all points where
c 0 and S S 1 will generate a steady state point where c 0 . Note that only the second and third sets of points are locally stable. If initially c 0 , S S 1 and a stressful event raises the stock of addictive capital to a level below S '1 then
c will become positive as a reaction to this event. As long as S initially remains lower than S '1 the consumer will eventually refrain from consuming c . Similarly, if initially c *1 S *1 and the consumer is the object of a positive event, maybe a new job, this can cause c to fall. If this event lowers S to a point still above S '1 then c will eventually begin to rise again and over time it will return to c c *1 . These two points thereby constitute locally stable steady states. If instead the initial level of consumption induces
c c'1 S '1 then even small changes in S will cause cumulative movements towards either c 0 or c c *1 . This steady state is then both locally and globally unstable. The reason why unstable steady states are an important part of the analysis of rational addiction is, that they provide information on why the same person sometimes is heavily addicted to alcohol or other goods and at other times lay off completely. To illustrate, this would be the case if alcohol is assumed to be a good consumed in tough times and the consumer is initially at c c *1 S *1 and experiences a favorable event that lowers the stock of addictive capital below the point of S S '1 , the unstable steady state with A . This would indicate that the 1
consumer would go from being strongly addicted to eventually giving up the addictive good c entirely. 1
In cases where A is very steep when S is below the unstable steady state, indicating powerful reinforcement in this interval, the consumer would go on a “cold turkey” and terminate the addiction. 27
3.3.2. EFFECT OF PRICE CHANGES According to conventional wisdom, the consumption of an addictive good is not responsive to prices as otherwise assumed in economic price theory inducing a downward‐sloping demand function. This conventional wisdom is contradicted by the model of rational addiction which, as introduced in the previous section, implies that addictive substances are likely to be quite responsive to price. To analyze rational addicts’ responses to changes in the cost of addictive goods, suppose the initial level is such that c c *2 S *2 along the curve A . A permanent decline in the price of c will raise the 2
consumption of the addictive good. If the full price permanently falls corresponding to a shift in the demand curve for c from A 2 to A1 in figure 2, then for each level of the addictive capital stock, this will cause an increase in the consumption of alcohol. This will cause the addict to acquire more addictive capital over time, moving the initial steady state to a new steady state with a higher consumption level. As a result, consumption increases from c * 2 to cˆ, and then c continues to grow over time since cˆ is above the steady state line. Consumption goes towards the new stable steady state at c c *1 S *1 . Long‐run responses to price changes exceed short‐run responses due to the fact that initial increases in consumption of addictive goods cause a subsequent growth in the stock of addictive capital, which then stimulates further growth in the consumption of c. The degree of addiction is expressed through A. The addiction is stronger when A is steeper, and since long‐run responses to price changes are also greater when A is steeper, strong addictions do not imply weak price elasticity. FIGURE 184.108.40.206: A CHANGE IN FULL PRICE OF THE ADDICTIVE GOOD.
If anything, rational addicts respond more to price changes than non‐addicts (Becker and Murphy, 1988). Even in the short‐run rational addicts respond to the anticipated growth in future consumption since future and current consumption of addictive goods are complements for them. However, the short‐run changes are smaller than the long‐run changes because the stock of addictive capital is fixed. The presence of unstable steady state for highly addictive goods means that the full effect of a price change on consumption could be much greater for these goods than the change between stable states from the previous example. A consumer with initial consumption capital between S ' 2 and S '1 in figure 220.127.116.11 would 2
be to the left of the stable state at c S ' 2 when price is p and demand curve A 2 is relevant. Equally the consumer would be to the right of stable state at c S '1 when price is p and demand curve A1 is 1
relevant. From this, a reduction in price from p 2 to p1 greatly raises the long‐run demand because the price decrease moves the consumption level from an initially low consumption to a new stable state at
c c *1 S *1 with a high level of consumption. This example illustrates that the total cost of an addictive good to consumers equals the sum of the monetary price of the good and the monetary value of any future adverse effects, such as the negative effects on future earnings and overall health from excessive alcohol intake. Both the short‐ and long‐run level of consumption can then be reduced by either a higher money price of the good or awareness about the future cost of health hazardous behavior. This is a very important implication of the theory of rational addiction that actively can introduce useful regulative mechanisms. It is intuitively plausible that as the monetary price becomes a bigger share of the total cost of c, long‐run changes in demand induced by a given percentage change in the money price get larger relative to the long‐run changes by an equal percentage change in future costs (Becker, Murphy and Grossman, 1991). This is based in the theoretical analysis from Becker and Murphy (1988) suggesting that younger and poorer consumers appear to discount the future more heavily and from this, besides being more sensitive to increased prices, they also place a smaller monetary value on health and other harmful consequences that will occur in the future. As shown from figure 18.104.22.168., the inclusion of an extra dimension of addictive capital into the traditional consumption model has caused a divergence away from the standard notion of a demand curve. In figure 2 prices are taken as exogenously determined and the level of alcohol consumption is a function of the level of capital stock of the addictive good.
3.3.3. ACQUISITION OF ADDICTIVE CAPITAL AND CATALYTIC CONSUMPTION When considering the model of rational addiction, all consumption must have an outset in the steady state origin where c = 0 and S = 0. This must be the dominating steady state. The reasoning is that there are no incentives to start consuming the addictive good c unless the consumer is initially endowed with addictive capital. To overcome this underlying principle it is necessary to assume that there are exogenous events that cause changes in the stock of addictive capital. Becker (1988) adds the term Z(t) to equation 4 to account for such life events. S ct S t Z (t )
If Z(t) is negative then these exogenous contemporary effects can induce a positive level of addictive capital and hereby cause a positive consumption of the addictive good. Assuming that Z(t) has a distribution where both the mean and standard deviation are small relative to the capital level of the lower steady state when c'1 S '1 then the majority of alcohol intake caused by these exogenous events will be temporary. This is consistent with the fact that most people enjoying alcohol from time to time do not develop an addiction. Becker and Murphy mention stressful events such as unemployment and divorce as examples of how a person might acquire addictive capital, as represented by the life event random variable Z(t) and hereby move from zero consumption levels to positive levels of consumption. Some could then afterwards progress into a steady state path and develop an addiction. Addictive capital is not freely disposable. Addictive capital will only forgo and depreciate over time. Some people will undergo unfavorable life events and recover relatively quickly while others will need more time. This blueprint can be transferred to the model of rational addiction as individually varying relatively high or low depreciation rates of the addictive capital. Alcohol intake in excessive amounts is considered an effective mean of blocking out the memories of unpleasant life events. This process is represented by the accumulation of addictive capital in the model. Excessive alcohol intake removes the imitate physical and mental the burden of the accumulated addictive capital while at the same time enforce the level of the future addictive capital.
3.5 DISCUSSION OF THE RATIONAL ADDICTION THEORY In the rational addiction theory, consumers are assumed to be rational, far‐sighted and to anticipate the future consequences of their current actions. The difference in this, compared to earlier models of addiction where addicts are myopic in the sense that they ignore the effects of current consumption on their future utility, is that the rational addict recognizes that the marginal utility of future consumption depends on current consumption. As noted by Chauloupka (1991), presuming rationality may appear to be a strong assumption. However, such a model is more consistent than the alternative myopic models in which consumers, when making consumption decisions, recognize the dependence of current consumption on past consumption but ignore the resulting dependency of future consumption on past and current consumption. Becker and Murphy point to a related issue in their 1988 paper. If people were myopic, spreading new information about health consequences of addictive goods would be of no use, since it would not affect the consumption behavior on whether or not begin consuming an addictive good. The theory of rational addiction is not without its critics. In this model addicts are utility maximizing which is inconsistent with Winston (1980) who observed regrets of initiating consumption of an addictive good among addicts. Akerlof (1991) argues that addicts in the rational addiction model choose to become consumer of an addictive good, well knowing the consequences. The implication of this is that there is no incentive to implement educational programs with the intent to educate current addicts about the negative effects of their addiction. This is incompatible with the intent of information campaigns as a public policy tool which in itself is an issue that Becker and Murphy criticize in the myopic addiction model. The distinction is that a rational addiction approach is partially affected by information prior to the initiation of an addictive consumption, but not when the consumption has been initiated. Orphanides and Zervos (1995) developed a revised rational addiction theory which incorporates learning and regrets and here by resolves some of the above criticisms. Here the basic idea is to allow for uncertainty rather than perfect foresight and a process of learning by doing through experimentation. This explains how individuals can voluntarily initiate a harmful addiction and the later regret it when they asses the real harm of the addiction. 31
Most recently Gruber and Köszegi (2001) question the assumption of time‐consistency in preferences in the rational addiction model. They find dropping the time consistent assumption still yields forward‐looking behavior but a strikingly different set of policy implication meaning that in their case regarding cigarette consumption, optimal government policy should depend not only on the externalities that smokers impose on others but also on the “internalities” imposed by smokers on themselves. The rational addiction model developed by Becker and Murphy has been applied to the several empirical studies of cigarette consumption, alcohol consumption, caffeine consumption, cocaine and other illicit drugs. The numerous empirical application of the rational addiction model compared to other recently developed addiction models is due to the fact that a key feature of the theory is that consumption of an addictive good will depends on future as well as pat consumption. This makes the theory relatively straightforward to test on empirical data since finding the coefficient for future consumption statistically significant in the structural demand function for the addictive good is a direct rejection of the myopic model of consumption behavior. This combined with the intuitively acceptable argument that consumer are forward looking to some extent, even though the rationality assumptions in the theoretical model might a strong assumption, makes this behavioral addiction model preferable.
3.4. MAIN FINDINGS REGARDING ADDICTION MODELS Addiction is a major challenge to the theory of rational behavior, and the theory of rational addiction does not claim that all behavior associated with addiction is consistent with rationality. However, the model does explain several features of addictive behavior and does appear to have a broader set of additional implications about addictive behavior than other more myopic approaches. The key empirical implication of the rational addiction model is that past and future expected prices influence current consumption and the assumption that addicts consume the addictive good with perfect foresight of the consequences of addiction. The reason why this model is applied to the analysis of alcohol consumption in Greenland is that the model yields several empirically testable predictions which contain valuable information on the nature of addiction. Individuals with a higher rate of preference for the present are shown to be likely to develop addictive behavior. Stressful events, such as divorces, unemployment and deaths of family members, which lower total utility whilst they raise the marginal benefit of addictive goods, have similar effects on consumption as price changes. Consequently individuals with the same preferences and endowments facing the same prices of the addictive good might differ with respect to the degree of addiction because of differences in the way they experience stressful events. For more aggregate level analyses the two main testable predictions are the presence of inter‐temporal complementarity of consumption and negative cross‐price effects, and the prophecy that the long‐run price elasticity of demand exceeds the short‐run elasticity. Especially the last of these predictions is potentially very important from a policy perspective since knowledge of the right demand elasticity is essential for calculations of effects on demand and tax revenue from higher taxes on alcohol. In this thesis it is highly relevant when assessing the consequences of the development in the Greenlandic alcohol policy during the past 40 years. As for the rational addiction hypothesis itself, it can be tested against the myopic addiction model since, according to the latter, consumption should be a function of current price and past consumption, whereas in the rational addiction model, the partial effect of future consumption should be significant as well. The next section will review the development of alcohol consumption in Greenland from 1970‐2010 in which a further analysis of the rational addiction model will be based.
4. ALCOHOL CONSUMPTION IN GREENLAND Considerable changes have taken place in the Greenlandic society primarily since the 1950s, but also long before. External influences, primarily from Denmark have resulted in various changes in patterns of production, material living conditions, lifestyle, consumption and health status. These distinct differences between Inuit and European cultures in social customs and attitudes do bring tensions and rapid socio‐ cultural changes are seen as causing detrimental effects on physical and mental health. There has been an increasing concern in public health perspectives regarding the social consequences of the growth in alcohol consumption and its implications on suicides, homicides, violence, accidents and to some extend alcohol‐ related illnesses (Young & Bjerregaard, 2008). An empirical study of the Greenlandic consumption of alcohol will be the subject of this thesis, where the center of attention will be the influence of price on alcohol consumption over time, with a recap from 1949‐1969 although mainly focusing on the period 1970‐2010. Existing literature on alcohol consumption in Greenland are modest and the following introduction to the history of alcohol in Greenland originates from a report by the Committee for social research in Greenland on the alcohol situation in West Greenland from 1961 (In Danish; Udvalget for samfundsforskning i Grønland 1961), a dissertation by Leif Sølling from 1974 on alcohol consumption in Greenland during 1960 to 1970 and chapter 2 on Greenland from Health Transitions in Arctic populations by Kue Young and Peter Bjerregaard. Alcoholic beverages were unknown in the traditional Inuit culture, but as early as in the Seventeenth Century intoxicating liquors were served and sold on various occasions, for instance when European whalers visited the settlements and trading stations. When the country was colonized in 1721 alcohol became more permanently available. This increased availability appears to have been creating problems from the beginning. Thus in 1782 the Danish Colonial Administration prohibited the distribution or sale of spirituous liquors to the Greenlandic population. Consumption of alcohol nevertheless continued to occur. Employees of the Royal Greenland Trade Department and the Christian Mission could obtain liquor, although not as much as the Danes. Liquor was also served as a reward for special jobs such as whale fleshing and coal mining. In the period up to 1929 various modifications were gradually introduced and as a result rationing was introduced in 1929 and continued, with various relaxations, until the sale of alcoholic beverages was derationized 25 years later. With the constitution of 1953, which incorporated Greenland as an integral province of the Danish Realm, the demands for an end to these prior discriminatory practices grew. On the December 15th 1954, alcohol rationing in Greenland was abolished. At that time alcohol consumption in Greenland was twice as high as in Denmark. 34
It is well documented that the Danish Administration until 1954 had combined an attitude of protective guardianship with that of issuing alcohol as a reward of remarkably good action, and thus projected status to the act of being able to consume alcohol (Schechter 1986). The nature of this has greatly influenced the Greenlandic population’s symbolic associations towards alcohol. The sale of alcohol was permitted as a symbol of the equal rights held by Greenlanders and Danes. At the same time as rationing was abandoned a great many changes were initiated in Greenland and alcohol consumption was especially high among the population living in the areas facing the greatest changes in the way of life (Sølling, L., 1974). These changes had their outset in a society with high disease prevalence, severe social problems and a decreasing economic capacity. The aim of expanding the institutional infrastructure was to build a modern society which would be self sufficient in a contemporary manner, based primarily on a modern fishing industry. A primary condition to achieve these societal changes was among other things a population concentration through urban migration from the smaller villages to the larger towns. The aim of integrating Greenland into the modern western society was in various ways interrupted, partly due to changing conditions outside Greenland such as an decrease in the worlds fishing resources and an increased interest in energy assets, and partly due to problems which emerged in relations to the relatively swift restructuring of the traditional Greenlandic society. Among the most important changes that have affected the health and wealth of many Greenlanders is the transition from subsistence hunting and fishing traditions to an economy based on labor and wage earning. This transition brought about profound alterations in the structure and tasks of the family as a unit. Many Greenlandic families found that their function had changed radically. The prior constitution where adults and children lived in a close relationship with a sense of balance between the upbringing and the society in which the children should later function had changed. Now families were split up and the adults worked outside the home meaning that the children to a greater extent would experience secondary socialization from people with different linguistic and cultural backgrounds. However, these transformations also brought about many possibilities for education and a horizontal expansion for alternative ways of living. The gap between the old and the new ways of life have been complicated to some and the changes in living conditions have had various social consequences, some resulting in strong use of mind altering substances such as alcohol. In the following section the focus will be directed towards the link between these changes and the development in the Greenlandic alcohol intake. 35
4.1. ALCOHOL CONSUMPTION IN GREENLAND, 1950‐2010 The first rationing scheme on alcohol in Greenland was abandoned in 1954. The result was that overall alcohol consumption rose from 7.2 liters of pure alcohol per person 15 years and older in 1949 to an average of 9.7 liters in 1955.
Liter per person 15 years or older
FIG. 4.1.1.COMPOSITION AND CONSUMPTION OF ALCOHOL IN GREENLAND, 1949‐1959 12 10 8 6 4 2 0 1949
Industrially produced spirituous liquors
Data source: Statistics are from Alkohol situationen i Vest Grønland, 1961 and only considers West Greenland.
In the years after 1955 the alcohol intake decreased again. By 1959 it had reached a level of 7.3 liters and the Greenlandic level of alcohol consumption was now about the same level as before the rationing ended. This volume of consumption was on a level with the average consumption in the rest of Denmark fifty years earlier and 50 percent higher than the average consumption of alcohol in the rest of the Danish population in 1958 (Alkoholsituationen i Vestgrønland, 1961). As fig 4.1.1 illustrate a redistribution took place in the population’s consumption of alcohol during the years after the rationing was abandoned. There was a shift in consumption patterns towards the more expensive industrially produced liquor which rose by almost 40 percent from 1954 to 1955. In subsequent periods the average consumption of homebrew steadily declined and in 1959 it was less than half the consumption level in 1954. The new accessibility combined with the symbolic status linked to drinking alcohol, which was partly induced by the Danish Colonial Government, could very well have been a motivating factor causing a substitution effect away from the locally produced Imiaq leading to a continuing rise in consumption of industrially produced alcohol. Also, the import taxes introduced on industrially produced spirits in 1950 stayed unregulated throughout the decade and were not regulated again until 1961. From this, the relative price of alcohol can be assumed to have been decreasing over the period 1951‐1961.
The derationing of alcoholic beverages in 1954 and the rise in consumption in the first years here after did start a wide debate both publicly and privately regarding the social consequences of heavy alcohol intake. This debate fostered a report by the Danish Community Research Committee in 1961. This report stated that there was no need to worry about the level of alcohol consumption amongst the Greenlandic population. It concluded that the consumption of alcohol only appeared to be a pressing issue to a small group within the population. Chronic alcoholism, however, did not seem to occur to any substantial extent (Alkohol situationen i Vest Grønland, 1961)
Liter per person 15 years or older
FIGURE 4.1.2: COMPOSITION AND CONSUMPTION OF ALCOHOL IN GREENLAND, 1960‐1970 20 15 10 5 0 1960
Industrially produced spirits in total Wine Imiak
Beer Spirituous Total consumption
Data source: Statistics are from Alkoholforbruget I Grønland, 1974 and only considers West Greenland.
Even though the import taxes on alcohol were raised in 1961 the import of industrially produced liquor continued to rise in the years from 1960‐1965. In 1965, taxes on beer, wine and liquor were increased once more. The duties on wine and spirituous were increased yet again in 1967 and 1969 to motivate less harmful drinking pattern, and the duties for beer followed in 1968. From the development in consumption illustrated in fig. 4.1.2 it can be concluded that the products most subjected to large tax increases in the period 1961‐1968, showed the greatest decline in imports. The tax policies can then be assumed to have had the effect of approximately restoring past consumption composition with respect to the total distribution of alcohol intake, even though the transition to a much lower level that the Sobriety Commission of 1960 had hoped for was not realized. On the contrary the overall absolute level of both soft and strong alcoholic beverages had risen to more than the double compared to 1954 levels. 37
Liter per person 15 years or older
FIGURE 4.1.3: COMPOSITION AND CONSUMPTION OF ALCOHOL IN GREENLAND, 1970‐2010 25.00 20.00 15.00 10.00 5.00 0.00
Industrially produced spirits in total
Data source: Statistics Greenland http://bank.stat.gl/dialog/Saveshow.asp
From 1970 to 1974 alcohol imports rose again by 24% from 15 liters to about 19 liters if pure alcohol per. person 15 years and older. Now the alcohol consumption in Greenland was 70 percent higher than the registered consumption of 11 liter per. capita over 14 years in Denmark (Sølling, 1974, Danish statistical year book 1974). In 1975 import taxes on all alcoholic beverages were increased by up to 50 % (see appendix 2). This caused an immediate but very moderate drop in overall alcohol imports as seen in figure 4.1.3 indicating that something else was needed in order to decrease alcohol consumption. In 1978, following a public plebiscite, the Greenland Council passed a rationing ordinance (Landstingsforordning nr. 2 af 19. maj 1979 om salg og servering af stærke drikke). This attempt to control alcohol abuse by limiting the individual access and supply went into effect on August 1st 1979 just three months after the new Home Rule Government took office. This meant that every person aged 18 year and older was entitled to a sheet of 72 rationing points per month. One point equaled one beer, luxury‐strength beer required 1½ points, a bottle of wine first required 6 points but was later reduced to 3 point, 75 cl. bottle of fortified wine required 12 points, the same was the case for 35 cl. hard liquor and lastly a full bottle of hard liquor required 24 points. As a result the overall import and consumption were reduced markedly. As the accompanying figures shows, the alcohol importation decreased extensively to a level of 13.6 ‐ 14.6 liters per. person over 14 years in the year from 1979 to 1981.
The rationing system had several practical disadvantages which soon resulted in a much active black market for rationing points. After two and a half years, and without any public referendum, rationing was abruptly abandoned by the Home Rule Parliament except for Qaanaaq Municipality (Schechter 1986). During the entire period 1975‐1991, apart from three years of rationing in 1979 ‐ 1982, the average import was more than 15 liters of pure alcohol per. person over 14 years with a peak in the years following the abandoning of the rationing scheme, where imports in 1982 reached an all time maximum of 22 liters of pure alcohol per person. The rationing system was replaced with taxes hikes on all industrially produced alcoholic beverages and after 1983 the alcohol intake decreased. Even so, the Home Rule Government passed new restrictions in 1988 in order to further reduce the alcohol intake. Apparently the reduction did not reach target and in 1992 the tax legislation was reformed and even higher alcohol taxes were introduced (See appendix 2). The new progressive tax scale was based on pure alcohol volume percent. The tax per. quantity of pure alcohol increased from tax group to tax group, ie. a bottle of wine over 13.10 percent volume was more expensive than a bottle of wine under 13.10 percent volume despite quality. The charge resulted in the introduction of the 3.6 beer which was relatively cheaper, while strong drinks were significantly more expensive. The amendments were specifically intended to affect the import and consumption pattern to substitute away from the more heavy alcohol beverages. These progressive taxes were increased again in 2007 and have remained unchanged since. The extensive effort to use taxation as a tool for regulating the overall alcohol import may well explain much of the decline in total alcohol imports in the years 1983 to 2010, where an all time minimum of 10.5 liters per person were reached (Statistics Greenland 2011). In comparison the alcohol intake per person in Denmark for the same year was 10.1 liters. Also, from the 1st of January 2011 it is no longer legal to bring any kind of beer, wine, spirituous or other alcoholic beverages when traveling to Greenland. Whether this will contribute to a fall in consumption for the relatively large non‐Greenlandic population living, working and visiting Greenland or an increase in tax‐ revenues will be interesting to follow. The section to come will comment on the distribution and habits of current alcohol consumption and review the social cost of excessive alcohol consumption to the Greenlandic society. Following this, the last sub section will look closer at the development in the real alcohol prices and see how this is comparable to the development in overall alcohol consumption in Greenland. 39
4.2. DISTRIBUTION AND HABITS OF CURRENT CONSUMPTION The Inuit Health in Transition Study is a general health study of the Greenlandic population. It has been carried out in all municipalities in Greenland in 1993‐1994, in Denmark during 1997‐1999, in the Center of West Greenland in 1999‐2001 and again in the West of Greenland in 2005‐2007 including North Western towns and finally in the eastern part of Greenland in 2008‐2009 and Northern Greenland (Avanersuaq) in 2010. The data used in this section contains information on the development in alcohol consumption from the Study in 1993‐1994, 1999‐2001 and from 2005‐2007. From this, data on alcohol consumption in East Greenland from 2008‐2009 and Avanersuaq 2010 will not be included in the following review. Since problems of heavy alcohol abuse are generally larger in Avanersuaq compared to the rest of Greenland (Sundhedsundersøgelsen i Avanersuaq 2010), the implication of this exclusion can be an underestimation of the overall social issues and costs of alcohol abuse in Greenland. The Study defined heavy users of alcohol as consumers who have a frequent and simultaneous high level of alcohol consumption. Specifically it is defined as someone who had more than 10 drinks on the same day within the last week, or 2‐5 drinks the day before the study (Jeppesen & Krogh Larsen 2008). This definition of heavy drinking can be used both for data from 1993 and 2005‐2007 and is therefore used when the two population studies are compared. Binge‐drinking is defined as a pattern of alcohol consumption that brings the blood alcohol concentration (BAC) level to 0.08 percent or more. This pattern of drinking usually corresponds to 5 or more drinks on a single occasion (National Board of Health, Denmark 2010). In 1993 the definition of binge‐drinking was based on the number of drinks the respondent had on his/her last occasion of drinking. If this number was more than 5 drinks the respondent was categorized as being a binge‐drinker. In the Survey from 2005‐2007 intake of 5 or more drinks on the same occasion was used to define binge‐ drinking for both men and women. Finally a person with harmful alcohol consumption patterns was defined by questions constituting the extended CAGE test, a test that was developed to identify people with an alarming alcohol problem (MacKenzie et al. 1996). The majority of the respondents from the 2005‐2007 Health Survey have had experiences with alcohol. Only about 2 percent reported to drink alcohol on a daily basis and almost 68 percent reported to drink alcohol 1‐3 times per month or less. When considering the import statistics the level of alcohol imported per person 15 years and older have not changed considerably in the period 1993‐2007. The change in the proportion of harmful consumption patterns between the Population Surveys is depicted in table 4.2.1.
TABLE 4.2.1: THE PROPORTION OF HARMFUL ALCOHOL CONSUMPTION PATTERNS REGISTERED IN THE PERIOD FROM 1993 TO 2007 Harmful level of alcohol intake Heavy users Binge‐drinkers (CAGE‐test) 1999 2005‐2007 1993 2005‐2007 1993 2005‐2007 N = 1564 N = 1484 p value N = 1089 N = 1765 p value N = 1089 N = 1765 p value Men 31.0 35.9 0.005 24.8 16.5 <0.001 56.7 50.1 0.02 Women 16.1 24.8 <0.001 8.5 7.1 0.32 28.0 32.7 0.06 Total 22.7 30.0 <0.001 16.2 11.6 <0.001 41.6 40.9 0.72 Data source: The Inuit Health in Transition Study Report 2005‐2007
About 10% of the respondents had an alcohol intake above the recommended limit. A high average alcohol intake showed signs of correlation to social gradients and was twice as common among those respondents outside the labor market compared to those currently holding a job requiring an academic education (Jeppesen & Krogh Larsen 2008). Most participants had an average weekly alcohol intake below the limit of 21 for men and 14 for women drinks consisting of 12 grams of pure alcohol recommended by the Danish Board of Health in 2007. These recommendations were altered to respectively 7 and 14 drinks for women and men in 2010, but this section will use the prior standards as they have been applied in the report on the 2005‐2007 Health Survey. The percentage of men who fits the category of being heavy users, due to the given definitions, has gone down from 1993 to 2005‐2007. This tendency is not significant for women resulting in a growing ration of women to men defined as heavy users, although the overall percentage of heavy users has gone down. Among women there has also been an absolute increase in the percentage of binge‐drinkers whereas the percentage of men defined as binge‐drinkers have gone down. The proportion defined from the CAGE‐test, as having a harmful level of alcohol intake, has gone up between 1999 and 2007 for both men and women. The abuse patterns have changed and it is therefore interesting to look at how often alcohol is being consumed in Greenland. The average alcohol intake reported in the Survey is 7 drinks per week. The alcohol intake reported by the respondents however seems to be highly unrepresentative for the Greenlandic population in general since the reported average alcohol intake among the respondents per year only sums up to about 50 % of the average 11.9 liters of pure alcohol imported in 2005‐2007 per person 15 years or older in Greenland. This could indicate a not unfamiliar bias among the participant compared to the none‐participants where by people with an alcohol intake greatly exceeding the recommended limit abstain from participation in optional surveys since they can be exposed in revealing sensitive and undesirable information (Pernanen, 1974). Due to this issue the percentage consuming more than the recommended alcohol per week can be assumed to be a minimum estimate. 41
4.3. EXTERNAL COST IN A GREENLANDIC CONTEXT It is well known and well documented that alcohol abuse has significant economic, social, health and psychological consequences for both communities and individuals in Greenland (Young & Bjerregaard 2008). Drinking culture in Greenland is extensively based on so‐called intoxication drinking, where the achievement of drunkenness is a goal in itself and not merely a mean of amusement (Quote from slide, Paarisa’a seminar on alcohol consumption, 2011). This section will focus on three areas in which overuse of alcohol impose tremendous costs both in monetary term and in human capital aspects.
4.3.1 HEALTH SERVICES There seems to be a lower prevalence of liver diseases in Greenlanders with a high alcohol intake, compared to Danes with similar alcohol patterns (Lavik, Holmegaard & Becker 2006). Instead alcohol related accidents are a frequent source to loss of life years in Greenland, and probably also to loss of labor market participation and sickness in general. Alcohol is often a contributing factor in accidents resulting in deaths and injuries. A study of avoidable deaths in Greenland in the period 1968 to 1985 reported that alcohol was on average a contributing factor in at least 23 percent of the avoidable deaths during the entire period (Bjerregaard P. & Juel K. 1990). The proportion of alcohol related deaths in fatal accidents during the same period varied by region and type of accident ranging from 0 percent in shooting accidents to 51 percent when considering exposure to cold (Bjerregaard 1990). Data from the Inuit Health in Transition Survey also shows that a large percentage of both men and women in Greenland at some point have been exposed to violence or have experienced serious threats of violence from an alcohol intoxicated perpetrator. These studies indicate that the damage burden imposed on the health of the Greenlandic population from alcohol is extensive. It has been assumed that the health care system in Greenland use a substantial amount of resources on alcohol‐related injuries and accidents. To get a quantitative expression on the extent of this, the Department of Health has in Collaboration with Auditcenter Odense and Accident Analysis Group, made an audit study to clarify the extent. The results of this study has provided information on how much alcohol‐ related injuries and accidents take up capacity in the health service and resources spend on this group. 2403 acute inquiries were registered in the trial period and health personal assess in each case if clinical alcohol influence was prevalent. About 10 percent of all patients in the study were clinically evaluated to be under excessive influence of alcohol. 42
The proportion of alcohol‐intoxicated patients was significantly higher in cases concerning psychological or social problems, suicide, accidents or violence. Among the treated patients involved in accident or victims of violence respectively 16 ‐ and 62 percent were assessed to be under the influence of alcohol (Nexøe, Wilche et al. 2010). There was a correlation between alcohol intoxication and necessity of using very expensive services such as hospitalization and transfer to another hospital. Especially the latter is very costly because of the special geographical conditions in Greenland, where transportation of patients must be done by plane or helicopter. Due to the limited time period and the quantitative nature of this study, it was however not possible to establish any definite causal relationship between alcohol intake and acute medical emergencies. However, one can have a suspicion that a number of inquiries could have been avoided, if alcohol had not been involved.
4.3.2 SUICIDES Since the 1950s, the rapid social development in Greenland has been accompanied by increased suicide rates. The suicide rate has increased dramatically to one of the highest in the world. The current pattern of suicides among Greenland Inuit is radically different from that of industrialized countries in the Western Europe. Rates are considerably higher, especially among young men. It is not apparent from the epidemiological data which specific components that are most influential but there seems to be a correlation between alcohol abuse and suicidal thoughts, suicide attempts and actual incidences of suicide. Especially among 15 to 24 year old men the peak is 450‐500 per 100.000 person‐year (Bjerregaard & Lynge, 2006). The methods of suicide are generally drastic. Bjerregaard and Lynge reported in their paper from 2006 that shooting and hanging accounted for 91 and 70 percent among men and women respectively. In 30 percent of all incidents alcohol intoxication was mentioned in the post mortem police report. Two studies have been published about suicide attempts in Greenland (Grove & Lynge, 1979; Thorslund, 1992). These studies illustrate that those who attempted suicide, significantly more often than a control‐ group had experienced a disharmonious childhood with alcohol problems in the home. 86 percent had alcohol problems themselves compared to 5 percent in the control group. The Inuit Health in Transition Survey in Greenland from 1993‐1994 and 1997‐1998 reported alcohol problems in childhood home and being the victim of sexual violence in childhood as independently increasing the frequency of reported suicidal thoughts. Among those who reported frequent alcohol 43
problems in childhood home and also experiences of sexual abuse in their childhood, 82 percent had experienced having suicidal thoughts compared to 10 percent among those who had not experienced either. In 2005‐2007 the correlation was still present. Of those who had attempted suicide a minimum of 70 percent had experienced problems with alcohol and sexual violence in their childhood. This indicates a problematic relation.
4.3.3 CRIME There is a common tendency to assume that a very direct causal relationship exists between extensive alcohol intake and criminal behavior. Greenland's main problems in crime and domestic disturbances are generally linked to drinking, which has led to outlawing of alcohol in some towns and villages. It is not easy to estimate the role of alcohol in the total volume of crimes committed in Greenland. Police reports, which represent the most complete source of information on crimes, often lack accurate information on the perpetrator. Based on the Greenland Police Annual Statistics the following table depicts the percentages of domestic incidents where alcohol was involved. This table was presented at the 2011 Alcohol and Drug Seminar in Nuuk. T Brackets indicate incidents involving children. There is no indication on how alcohol influence was measured but it is likely that a clinical measure was used. Table 4.3.1 indicates that there is a strong association between the use of alcohol and domestic incidents involving the police. However, no comprehensive Greenlandic study has examined the nature of this likely link. More over analysis of all cases of violence in Nuuk during 2011 showed that 230 out of 270 cases of reported violence involved clearly intoxicated parties. In addition to this the Yearly Report in 2010 from the Greenlandic Police Districts states the experiences that the vast majority of both domestic violence and sexual offenses involve at least one drunk person. TABLE 4.3.1: THE PERCENTAGE OF DOMESTIC INCIDENTS INVOLVING PEOPLE UNDER INFLUENCE OF ALCOHOL Year Domestic incidences People under influence of alcohol involved 2005 3379 (561) 1570 2006 3695 (610) 1953 2007 3712 (505) 2017 2008 2788 (413) 1711 2009 3685 (551) 1423 2010 2635 (442) 1162 Data source: Presentation from Paarisa Alcohol and Drug Seminar, Nuuk 2011
Percent 46.5 52.9 54.3 61.4 38.6 44.1
4.4. DEVELOPMENT IN THE PASS‐THROUGH EFFECT AND THE REAL PRICE OF ALCOHOL A fundamental law of economics is the downward sloping demand curve, which states that as the price of a normal product rises, the quantity demanded of that product falls. From section 2.4 it appeared as if alcohol, although suspected of being inelastic, did respond to changes in price. Since the price of alcohol can be manipulated through excise tax policies, findings regarding the relationship between excise taxes, alcohol price and alcohol consumption clearly are relevant for policy‐ makers interested in reducing alcohol consumption and its adverse consequences. This section will briefly analyze the development in the real beverage prices for beer, wine and spirituous in relation to the development in excise taxes, known as the pass‐through rate from section 2.2.1, and the development in real average income. A total of 13 episodes of increases in excise taxes on all alcoholic beverages has taken place in Greenland during the period 1971‐2010. From this it is possible to study the symmetry of increases of excise taxes on prices of alcoholic beverages. Appendix 2 shows the amount of excise tax on beer, wine and spirituous per year in the entire period. On the basis of these taxes an average yearly excise taxes for each beverage category has been calculated and adjusted for the consumer price index. Appendix 3 contains information on the development in average beer, wine and spirituous prices in the years 1971‐2010 for beer and 1980‐2010 for wine and spirituous. To calculate the development in real prices over the period, these prices have also been adjusted for the consumer price index. Figure 4.4.1 on page 47 shows the development of both real excise tax and the real price for the three beverage categories. From all three tables in the figure it is apparent that inflation from 1971 to 1991 had eroded the effects of excise taxes to a greater extent than the development in real beverage prices. The decrease in prices was less than the average beverage excise taxes prior to 1992. This would indicate a pass‐through effect greater than one meaning that hikes in excise taxes were shifted to the consumers. Since 1992, when beverage taxes rose by roughly 50 percent, the growth in excise taxes has kept pace with the development in prices for wine and spirituous. For beer the tax increase in 1992 shifted the excise taxes without having an increasing effect on average real prices. From this it would seem, that excise taxes were not pass through to the consumers, as predicted from an assumption of an inelastic demand function. Instead they were undershifted and possibly resulted in a decrease in producer’s surplus. 45
Another important finding from the tables in figure 4.4.1 is that the real prices of beer (i.e., the prices after accounting for the effects of inflation) have declined over time whereas the real prices for wine and spirituous have fluctuated a bit more over the years. The 2010 real wine price appears to be over the 1981 level while the real price of spirituous is close to the status quo level from 1981. If alcohol use and abuse are sensitive to price, as the studies review in section 2.4 found, a decrease in the real prices will exacerbate alcohol consumption and from this, also the problems associated with alcohol use and abuse. In order to study the consequences of the changes in real beverages price it is necessary to consider their development in association to the development in the overall wealth of the Greenlandic population. Appendix 4 shows the development in the average real income in the years 1971 to 2010. On the basis of this average and the development in real beverage prices in appendix 3, figure 4.4.2 page 48 depicts the relative development of real alcohol prices in relation to the real average income in the years 1971 to 2010 for beer and 1981 to 2010 for wine and spirituous. Here it is very clear, that the relative price of beer has declined significantly over time. The relative wine price is almost unchanged and the relative spirituous price has also declined although not as substantially as the relative beer price. The implications of these finding are, that on average mainly beer and spirituous consumption has become progressively more affordable. From this it is apparent that alcoholic beverages have become more affordable because periodic tax increases, although they are partly passed through to the consumers of wine and spirituous, the prices do not keep up with inflation and hereby fails to adjust for changes in the real income. The structure of the alcohol consumption in Greenland in the period 1960 to 2010 is presented in appendix 1 where the percentage consumed of each beverage as part of the total consumption has been calculated. The figures show, that beer has been the dominating beverage amounting to roughly 70‐ to 80 percent throughout the entire period. This is in accordance with the findings in figure 4.4.2 since relative beer prices have been declining during the observed period making beer the more affordable beverage of the three. One problematic feature of the relatively large beer consumption is, that studies of beverage preferences suggest that compared to consumers preferring wine those who prefer beer tend to have increased risk of becoming heavy and excessive drinker (Jensen et al 2002). Maybe this can contribute to explaining why problematic alcohol consumption in table 4.2.1 seems to have increased, while overall consumption of alcohol has remain constant. 46
FIGURE 4.4.1. PASS‐THROUGH RATE OF ALCOHOL TAXES ON REAL BEVERAGE PRICES
PASS‐THROUGH EFFECT OF TAXES ON REAL BEER PRICE, 1971‐2010 INDEX (1971=100)
150 100 50 0 1970
Real beer price
Adjusted average beer tax
PASS‐THROUGH EFFECT OF TAXES ON REAL WINE PRICE, 1971‐2010 INDEX (1981=100)
200 150 100 50 0 1980
Adjusted wine price
Adjusted average Wine tax
PASS‐THROUGH EFFECT OF TAXES ON REAL SPIRITUOUS PRICE, 1971‐2010 INDEX (1981=100)
150 100 50 0 1980
Adjusted Spirituous price
Adjusted average spirituous tax
Data source: Data are provided from Statistics Greenland and statistical yearbooks from 1970‐2009
FIGURE 4.4.2. DEVELOPMENT IN THE RELATIVE PRICE OF ALCOHOL
DEVELOPMENT IN THE REAL PRICE OF BEER AND THE AVERAGE INCOME, 1970‐2010 INDEX (1970=100)
400 300 200 100 0 1970
Adjusted average income
Adjusted beer price
DEVELOPMENT IN THE REAL PRICE OF WINE AND THE AVERAGE INCOME, 1970‐2010 INDEX (1981=100)
200 150 100 50 0 1980
Adjusted average income
Adjusted wine price
DEVELOPMENT IN THE REAL PRICE OF SPIRITUOUS AND THE AVERAGE INCOME, 1970‐2010
200 150 100 50 0 1980
Adjusted average income
Adjusted Spirituous price
Data source: Data are provided from Statistics Greenland and Statistical Yearbooks from 1970‐2009.
4.3. DISCUSSION OF ALCOHOL CONSUMPTION IN GREENLAND, 1970‐2010 Greenland has a long history of prohibition, rationing and restricting in relation to alcohol consumption. Over time the alcohol legislation in Greenland has undergone many changes and become less strict in relation to rationing though there are still occasionally geographical prohibitions. A surprising finding when analyzing the symmetry between the development of the real price on alcoholic beverages in Greenland and the development in total consumption more closely is, that even though alcohol prices remain relatively high compared to for instance Denmark and large excise tax hikes have taken place frequently, the real price of beer has actually decreased in the period whereas the price of spirituous remained relatively constant. Only the real price of wine has ultimately increased by about 30 percent after experiencing a significant price drop during the rationing in 1979 to 1981. Further, from an economical view this interaction between development in prices and total consumption is surprising since, all else being equal; a decrease in the relative price of a normal good would lead to a rise in over all consumption. One explanation for this development would be that the full price of alcohol has increased over the period. This will not necessarily be reflected in the development in the monetary price, but could be the result of an increased awareness of the negative consequences of excessive alcohol abuse communicated through preventive information campaigns. When considering the development in the Greenlandic society in recent times, not many will counter argue that alcohol abuse in Greenland is highly influenced by the modernization process that has occurred during the past 60 years. In relation to the acquisition of addictive capital in the rational addiction model, alcohol abuse in Greenland can thus be a symptom of other phenomena’s, namely negative life events, and not merely a physical disease than can be treated with anta bus medication. Alcohol abuse can then be considered as a catalytic consumption since it is a consequence of a series of earlier negative life events. The implication of this is that alcohol abuse can be considered as a rational reaction, rather than a rational action (Bridges F, 2001). Without the catalyst the excessive alcohol consumption would not have occurred. Of course catalysts are individual and treating Greenland as one observation over time is not optimal but it will give some indication of a cultural “life event” that in many different ways have influenced generations of the Greenlandic people in a way that could be suspected of inducing heavy alcohol consumption as a coping strategy. 49
4.2. MAIN FINDINGS CONCERNING ALCOHOL CONSUMPTION IN GREENLAND This section has primarily described the aggregated development in the Greenlandic consumption pattern of alcohol throughout the last 60 year with a main focus on the period from 1970‐2010. The total consumption of industrially produced alcohol has declined from 14.4 liters of pure alcohol in 1970 to 10.5 liters in 2010 with an all time maximum of 22 liter of pure alcohol consumed in 1982 and 1987. Over the years alcohol consumption has changed greatly in relation to consumption structure and size, although the importance of beer in the Greenlandic consumption pattern is borne out by the fact, that total alcohol consumption in all years is largely driven by beer consumption. Several regulative policy tools have been applied in order to combat the excessive levels of yearly alcohol intake and throughout the period from 1970 to 2010 a total of 13 episodes of increases in excise taxes on all alcoholic beverages have taken place in Greenland and after 1992, the pass‐through rates of taxes on to prices have been close to one. A surprising finding, when alcohol prices adjusted for the consumer price index over the entire period is, that the price increase on beer and spirituous have only been of a temporary manner while the increase in real wine price has been long‐term. Overall beer prices have declined by 25 percent relative to the 1981 price. The lack of inverse symmetry between the development in prices and consumption of beer and spirituous is surprising since, all else being equal; a decrease in the relative price of a normal good would lead to a rise in over all consumption. This has not been the result in Greenland. Despite the decrease in consumption of alcoholic beverages, the costs of alcohol abuse are still significant in a Greenlandic context and stretches far beyond the consuming parties as was seen in section 4.2 and 4.3. Though the focus of the thesis is solely on aggregated effects of prices on alcohol consumption where issues such as individual costs and harms related to alcohol abuse are not considered in the analysis in section 6, it has been thought important to review the associating consequences of alcohol abuse in Greenland to comprehend the depth and width of the interconnected problems.
5. METHODS AND DATA In the former studies, researchers have used a variety of econometric models to investigate the effect of price on alcohol use. The most common method in the majority of the studies testing the rational addiction model is the use of instrumental variable (IV) techniques, such as two‐stage least squares (2SLS). These can be identified as structural methods and they are used mainly to cope with problems related to endogenity in one or more exogenous variables. Some studies have used longitudinal survey data to handle the statistical problem and to get more information on the individual fixed effects of drinking while others have used cross sectional data sampled from the same population in several periods and still others, as is the case with the current study, have considered an entire population as one observational unit and have used time series observation of relevant variables in several periods The following section initially presents the econometric methodology applied to test the model of rational addiction on alcohol consumption in Greenland in this study. After this introduction follows a description of first the dependent variable and hereafter the explanatory variables. Strength and weaknesses of the instruments will be mentioned forthcoming but will be summarized at the end of this section. Then summery statistic for the variables will be presented in table 5.3.1. Finally the last part of this section discuss of the quality and relevance of the data in testing a rational addiction model for alcohol consumption in Greenland.
5.1. ECONOMETRIC METHODOLOGY The econometric analysis of the nature alcohol consumption in Greenland during the period 1970‐2010 is based on the empirical specification of equation (6) given in section 2.2.3. C t C t 1 C t 1 1 Pt 2 et 3 e(t 1)
The first analysis will test whether alcohol consumption in Greenland can be characterized as an addictive behavior. The hypothesis of addiction is tested from the partial effect of prior consumption and present prices on current alcohol consumption in the myopic version of equation (6), where future consumption is not included in the analysis. H1 : beer , wine , Spirituous 0
Afterwards the rational addition hypothesis is tested by adding the component of future consumption to the structural equation, as in equation (6). H 2 : beer , wine , Spirituous 0
Ordinary least square estimation of equation (6) would lead to biased estimates of the structural parameters of interest. As noted by Becker et al. (1994) the unobserved variable that affect utility in each period given by et and et 1 are likely to be serially correlated. Even if these variables are uncorrelated ct 1 and ct 1 depends on et through the optimizing behavior. This will give rise to past and future consumption having a positive effect on current consumption even in the absence of addictive behavior when 0 . Equation (6) implies that current consumption is independent of past and future prices when past and future consumption is held constant. Therefore any effect of past and future prices must come through their effect on past and future consumption. Provided that the unobservables are uncorrelated with prices in these periods then past and future prices will serve as logical instruments for past and future consumption, since past prices expressions directly affect past consumption in the same way as future prices expressions affect future consumption. To sum up instrumental variables should be uncorrelated with the error term in equation (6) and be partially correlated with the deterrent variables (Woolridge chapter 15, 3th edition).
However IVs should be used with caution. Estimates are generally inconsistent if the instruments are correlated with the error term in the equation of interest. Also the selection of weak instruments, which are poor predictors of the endogenous instrumented variable in the first‐stage equation, can be an issue. In this case, the forecast of the endogen predictor by the instrument will be poor and the predicted values will have very little variation. Consequently, weak instruments are unlikely to have much success in predicting the ultimate outcome when they are used to replace the instrumented variable in the second‐stage equation. In the context of the alcohol consumption estimation, alcohol prices or taxes are weak instruments for alcohol consumption if consumption levels are largely unresponsive to changes in prices expressions. If higher prices do not induce people to reduce consumption then variation in prices reveals nothing about the effect of prior or future consumption on present consumption. If prices affect current consumption through channels other than through their effect on prior and future consumption then the instruments are invalid and the instrumental variables approach may yield misleading results. The following analysis has, as in prior literature (Fx, Becker, Grosmann &Murphy. 1994) attempted to correct for the possible endogenity problem by estimating equation (6) by a 2 stage least square regression using past and future expressions for the monetary cost of alcohol as instruments for past and future consumption. The statistical significance of the coefficients for past and future consumption measured by the instruments past and future costs, provides a direct test of the rational addiction model against an alternative model in which consumers are myopic and entirely backward looking indicating that current consumption only depends on current prices, lagged consumption, marginal utility of wealth and current life events as describes in section 2.3.2. Also the predictions from the rational addiction model regarding inter‐temporal complementarities and negative cross‐price effects along with a long‐run price elasticity that exceeds the short run price elasticity will be tested when rational addiction can be statistically confirmed.
All statistical models are estimated in STATA 10.1, a statistical program for regression analysis of cross‐ section and panel data. The commands applied have been found in STATA reference A‐Z version 9, 2005 and all do files can be found in appendix 5.
5.2. DATA The dataset applied in this analysis is constructed on the basis of information about the total consumption of alcohol measured by the amount of legally imported alcohol in Greenland during a period of 40 years from 1970‐2010, an index for the development in prices and absolute changes in import taxes for the different beverage categories beer, wine and spirituous and finally an estimate of the average income per year per person of Greenlandic heritage living in Greenland. No complete information exists on all the relevant indicators and information have been gathered from various sources, as will be described in relation to the specific variables. In general the data availability on factors such as income measures and beverage prices has been absent prior to respectively 1981 and 1977 and predictions of estimates for the missing observation have been calculted when necessary. The procedures for obtaining these estimates may vary according to relevant circumstances and will be thoroughly explained in the respective subsections describing the different included variables. Data for the empirical analysis of alcohol consumption consists of time series cross section of imports covering the period from 1971 through 2010 and has been provided by Statistic Greenland. The following section will present the variables chosen for the empirical analysis and account for their relevance, quality and origin. As a consequence of the strict regulation of trade in alcoholic beverages and the fact that Greenland is an geographically isolated limiting cross border trade, the statistics for alcohol imports when available is of very high quality as there is no problems of underreporting or difficulties on obtaining information from heavy‐drinking consumers, which can be the case when applying survey‐based data in analysis of alcohol consumption. Another advantage is the availability of statistics regarding imports according to type of alcoholic beverage, measured in liters of 100 per cent pure alcohol. Also, in contrast to previous studies on rational addiction
and alcohol (Grossman et al., 1998; Waters and Sloane, 1995), the past and future consumption levels used in this analysis is based on actual imports, and not constructed from data3. Finally it should be mentioned that the data in this analysis has one significant and very common draw back due to the aggregated nature. Data on average yearly alcohol imports and income measures can hide regional as well as individual variations. Average consumption may very well be dominated by moderate and light drinkers, which makes it difficult to investigate whether the rational addiction hypothesis also apply to heavy drinkers, who inflict the greatest social cost in relation to their alcohol consumption. Imports are reported on the basis of the calendar year running from 1st of January through 31st of December. Average per capita income is based on the fiscal year, and the price index for different alcoholic beverages pertains to January of the year at issue. Since changes in import taxes on alcoholic beverages have taken effect on different dates throughout the years, they will be attributed to the respective year when the amendment was put into effect. This of course can give cause or inexact measures for the effect of these tax increased if they are not effective in more than a few month in the year when they were issued. All prices, taxes and income measures were deflated to either 1971 or 1981 by the consumer price index for all goods. Further all variables, except the included dummy variable and a control variable indicating year, are given in logarithms following the main approach in the previous literature in considering the percentage change in consumption induced by one percent change in the included explanatory variables.
Waters and Sloan (1995) used data from the 1983 Health Interview Survey in 31 standard metropolitan statistical areas. Their measure of current consumption was the number of drinks in the most recent two weeks. The measure of past consumption was based on information on the history of drinking and the future consumption is generated from projections on current and past drinking. Grossman et al. (1998) used data from a representative panel of high school seniors in the 1976–85 period and the consumption measure was based on survey questions about the
number of drinks per month and the number of drinking occasions per month.
5. 2.1. DEPENDENT VARIABLE The dependent variable in all estimations is an index for the per capita consumption of litres of alcohol derived from beer, wine and spirits respectively measured in terms of 100 per cent alcohol equivalent imported to Greenland per year. The structure of Alcohol consumption in Greenland during the years 1960‐ 2010 as the relative share of each beverage type in the total consumption is depicted in appendix 1. All figures on alcohol imports are given in of liters of pure alcohol per capita older than 15 years by time and specification according to beer, wine and spirits per year. ALCOHOL CONSUMPTION Figures on alcohol importation in the years 1975‐2010 has been provided from Statistics Greenland. The statistics for yearly alcohol imports are based on tax group statistics for the different alcoholic beverages. The amount of pure alcohol is calculated on the basis of the total volume listed in the tax group statistics (Statistics Greenland, 2011). The data is provided to Treasury Department in Nuuk from the Customs department in Aalborg. In calculating the annual alcohol imports per. person over 15 years, the population average for the past two years has been applied. The assumption for the following analysis is that per capita consumption of alcohol based on the import statistics reflects the behavior of a representative consumer. In 1974 a report was published on the alcohol consumption in Greenland during the period 1961 to 1971. This report contains the data on specification on alcohol imports in the years 1960 to 1971 used in this analysis. For the remaining years from 1972‐1974 the alcohol import for the three categories have been estimated from a linear regression function in Excel based on an average of the growth rate in consumption three year before and three years after the relevant year. Alcohol consumption and specification in the years 1949‐1959 is from the publication “The alcohol situation in west Greenland” from 1961. This data only includes import and consumption in west Greenland, which accounts for approximately all consumption of industrial produced spirits and was inhabited by the majority of the population at the time. While the consumption of alcohol has been briefly described for the entire period 1950 to 2010, lack of adequate data on prices and income measures in several years prior to 1975 has resulted in a decision to limit the analysis of the rational addiction of alcohol in Greenland to the period 1970‐2010.
DIFFERENCES BETWEEN IMPORTS AND CONSUMPTION The statistics on alcohol provided by Statistics Greenland is limited to the amount of alcohol imported to Greenland. As far as imports are used as an indicator of consumption, it must be kept in mind that several factors may complicate the relationship between imports and consumption4: Household’s own‐production of alcoholic beverages is not included in the data. The implication on this is that actual consumption might be higher than the imports reported here. The magnitude of the household production of alcohol in Greenland is not familiar. Build up of stocks are also not considered. Since the population in Greenland is rather small this may have some importance, since uncertain and difficult transport conditions are a reality in many parts of the country. Due to this mechanism, imports could vary from year to year, even though there is no change in actual per capita consumption. Non‐declared imports (net‐border trade) can also cause a bias in consumption estimates. Because of high taxes and the large sea way traffic, the factor has some significance in Greenland even though the more traditional cross border shopping is limited by the geographical situation in Greenland in which Greenland has few entry opportunities for commercial travelers. Duty‐free purchases in international airports are not included in the import accounting methodology used in the Nordic countries. As of the 01 January 2011 it is no longer legal to bring any kind of beer, wine, liquor or other alcoholic beverages when entering Greenland. This should further limit the bias between actual consumption‐ and import measures. On a final note regarding this matter, it should be mentioned that Greenland has many tourists and seasonal workers visiting every year. They are often financially well established and have an economic freedom to enjoy alcohol in greater amounts. This has the implication that the total measure of liters of alcohol imported to Greenland can deviate from the actual amount consumed by the Greenlandic population, which is the population of interest in this analysis. Given that the magnitude of these factors, however, have not been calculated, it cannot be estimated exactly how big the difference is between imports and actual consumption. Even though it is important to mention the above factors they are not sought to influence the consumption of alcohol by the Greenlandic significantly. In summary, this means that the registered imports of alcohol to Greenland probably represent a very vital part of the total consumption by the Greenlander and will form the basis of a thorough analysis of the addiction behavior concerning alcohol in Greenland
These factors are mentioned in the report on alcohol import to Greenland during the period 2006 to 2010. This noted originated from March 7, 2012. The report can be found at; http://www.stat.gl/dialog/main.asp?lang=da&version=2010&link=AL&subthemecode=p2&colcode=p
5.2.2. EXPLANATORY VARIABLES The basic specification of the structural demand function in equation 2 is augmented with some exogenous variables. The information contained and the origin of these variables will be explained in the following section. PRICE MEASURES The price of three categories of alcoholic beverages is a vital variable in the empirical analysis. The two following subsections will describe the two different measures of the monetary prices felt by the consumers. These measures will be applied in two separate models to capture if differences in the effect on total consumption exist. BEVERAGE PRICES INDEXES The optimal measure capturing the development in the prices for beer, wine and alcohol over time is the price index on each beverage. The price indices per year for each type of alcoholic beverage in the period 1995 to 2005 were provided from Statistics Greenland. Indexes for additional years were found in statistical year books and have been deflated with the consumer price index for all goods to 1971 prices for beer and 1981 prices in the analysis of wine and spirits. In the years from 1970 to 1980 the indexes were not spilt into beer, wine and spirits. Instead the categories were Bayesian beer, pale/light beer and soft drinks in one index and wine and spirits in another index. The index for Bayesian has been used as an estimate for the beer price index in the period. The analysis for wine and spirits only considers the period from 1981 to 2010 where price index information is available. Alternatively as an experiment the indexes for wine and spirits for the entire period was estimated by adjusting the available price indexes according to their respective growth rates in the subsequent period from 1981‐1985 and the amount of alcohol consumed in respective years. The danger of this experiment was that these estimates could deviate from the actual price index during these particular years, especially since The Government of Greenland imposed restriction to the availability of alcoholic beverages per person in the year 1980. The predictions showed no effect on the price variables in the estimated models and for this reason the models presented in the result section is based on the original.
IMPORT TAXES Taxes are used as a proxy for prices since it is assumed, that increases in import taxes on alcoholic beverages is followed by an increase in the monetary price. Also the Greenlandic consumers may very well have more knowledge about a change in exercise taxes on alcohol than a change in actual prices since tax changes are publicly announced well before they are put into effect. For this reason import taxes will be tested as instruments for fast and future consumption in the same manner as actual prices in the empirical analysis, since it appears relevant to analyze how consumption changes when a tax is modified. Another advantage on applying taxes as a price proxy is, that data on changes in exercise taxes on alcoholic beverages are available in complete form for the entire period 1970‐2012 and therefore requires no additional adjustments besides the deflation applied to all price variables. The variables BeerTax, WineTax and SpiritsTax represents the respective deflated variables used in the analysis. INCOME MEASURES Average yearly real disposable income is the optimal measure for the amount of wealth that the Greenlandic population has available for spending and saving after income taxes have been accounted for. Unfortunately availability on this indicator only dates back to 2002 making it inadequate for the analysis. This has made it necessary to find alternative income measures for the empirical analysis. These proxy variables will be described in the following subsections and models have been estimated using both alternative income measures. AVERAGE TAXABLE INCOME Fortunately price trends dating back to 1971 are available, while the longest time series for income development goes back to 1988. This makes it possible to calculations an expression for the development in incomes from 1988 to 2010. From an earlier publication average income for the years 1960, 1970, 1971 and 1972 can be found (Sølling, L., 1974). The estimates used for the average income during the years 1961‐1969 an again from 1973‐1988 are based on an estimated average growth rate calculated on the basic of the available observation. Of course this method excludes all in between year variations in the growth rate for the average income, but it does however imply a trend for the development over the relevant intervals. 59
Since the price variables for beer is available for the period 1971‐2010 one average income variable income1 has been constructed by deflating the average income according to the consumer price index using 1971 as base year. Another income variable income2 has been adjusted according to the index with 1981 serving as base year, in order to fit the price data on wine and spirituous. DISPOSABLE GROSS NATIONAL INCOME Another way to control for in income is to use disposable gross national income for the period as a proxy variable. Even though gross national income per capita is not a measure of personal income but a measure of national income, a separate concept, it is often used as such an indicator, on the rationale that all citizens would benefit from their country's increased economic income. This reasoning will justify the application of the measure in this analysis. Data on GNI is available online at StatBank.gl for the period 1979‐2007. Data on GNI in additional years from 1971‐1978 and 2008‐2010 have been estimated from the growth trends in the subsequent or prior years. These variables have also been deflated by the consumer price index with 1971 and 1981 serving as base years and renamed to Disp.GNI1 and Disp.GNI2. DUMMY VARIABLE A dummy for the years 1980‐1982 has been introduced in the analysis as an experiment in order to analyze the possibility of capturing major changes in the consumption of alcohol induced by the restrictive regulation during these years. The value of this dummy variable will equal 1 in the years 1980‐1982 and will be null in all other years.
5.3. SUMMERY STATISTICS OF THE VARIABLES Table below summarizes all the variables included in the analysis. For each variable, the table presents the variables name and definition as well as the number of observations, range of values, mean and standard deviation. TABLE 5.3.1 SUMMERY STATISTICS OF THE VARIABLES
Dependent variables Cbeer
Liters of pure alcohol from beer
Liters of pure alcohol from wine
Liters of pure alcohol from spirits
Explanatory price variables Pbeer
Adjusted for CPI, 1971=100
Adjusted for CPI, 1981=100
Adjusted for CPI, 1981=100
Adjusted for CPI, 1971=100
Adjusted for CPI, 1971=101
Adjusted for CPI, 1971=102
Explanatory income variables Disp.GNI1
Adjusted, CPI, 1971=100
Adjusted, CPI, 1981=100
Adjusted, CPI, 1971=100
Adjusted, CPI, 1981=100
Additional variables D_rationing
1 if rationing, 0 otherwise
5.4. DISCUSSION OF METHODS AND DATA Overall testing a long‐run model of individual behavior on aggregated time series data is somewhat demanding due to the limited degrees of freedom from only one observation per period. Variables accounting for influential institutional changes might not have been exhausted in this analysis. Some changes like urbanization, increased awareness of health consequences of alcohol use and the age structure of the Greenlandic population are more gradual in their nature and therefore difficult to account for in an empirical analysis. These omitted variables are also likely to relate to the growth in income for the population as a whole and thus they assumed to be picked up by the income variables used in the analysis. The analysis includes the Greenlandic population aged 15 years and older, and hereby only considers the subgroup of the population who are legally permitted to buy alcoholic beverages. Over the past 10 years, drunkenness among young people in Greenland has been increasingly considered as a public health issue. (SIF’s Grønlandsskrifter nr. 22, 2011). This indicates that not all consumers of alcohol are included in the analysis. However the magnitude of alcohol consumed by under‐age drinkers it not thought to distort the average litres of alcohol consumed per person 15 years and older much compared to as for example the previous mentioned visitors and season workers visiting Greenland. Also average liters of alcohol consumed per year person 15 years or older has not been corrected for the 10 percent reporting to abstaining from consuming alcohol in the Public Health Survey from 2005‐2007. This could have some significance for the size of the dependent variable, but since the percentage of abstainers for all included years in unknown, no correction has been made. Moreover, if this percentage does not differ over time, the relevance of correction for it in this analysis insignificant, since it is the consumption as a function of income and prices in past, present and future periods which is relevant. It could be argued that in testing for rational addiction, distinguishing between different types of alcohol is irrelevant. Earlier studies have however shown that beer, wine and spirits are distinct kinds of goods in an economic sense and on the basis of this a model for each alcoholic beverage was chosen to be estimated separately. Also in contrast to arguing that all alcoholic beverages constitute one category of commodities, it could be argued that the rational addiction model may be more relevant for describing the intake of the harder alcoholic beverages such as spirits compared to beer and wine as the distribution of the spirits consumption is typically more bimodal indication that addictive behavior can be more easily captured by aggregated data on spirits consumption.
Since the focus of the thesis is solely on aggregated effects of prices on overall alcohol consumption it does not take specific drinking patterns into considerations. Especially in a Greenlandic context this will result in ignoring a very problematic relation, since drinking habits differ greatly among individuals in the population. About 10 percent of the total population in Greenland reports to abstain completely from consuming alcohol. On the other hand, a much larger percentage of those consuming alcohol regularly reports to consume a minimum of 5 drinks per occasion. This habit of binge‐drinking is fairly common in Greenland and it creates a problematic pattern where the achievement of drunkenness is a goal in itself and not merely a mean of amusement. The aggregated level in data when considering Greenland as one observation over time implies that data can be dominated by the consumer behavior of moderate and light drinkers and price elasticities estimated from aggregated data measures only the effect of price on the overall alcohol consumption in the population. From a public health policy point of view, Room (1984) contended that such a broad measure is not of central interest because what really matters is the effect of price on high‐risk drinking patterns. The availability and use of individual data could help in shedding light on debates that cannot be resolved using aggregated data. A very relevant subject is the issue of whether heavy drinkers in Greenland are responsive to price changes. On the basis of this, an optimal model testing rational addiction should include only heavy drinkers. Overall testing a long‐run model of individual behavior on aggregated time series data is somewhat demanding due to the limited degrees of freedom from only one observation per period. To gain efficiency a parameter restriction implied by the theoretical model could be imposed to constraint the coefficient for past and future consumption in a manner whereby it is only allowed to differ by a predetermined discount factor. An addition possibility would to impose the restriction of negative long‐run price elasticity. Neither of these possibilities has been tested in this analysis and hereby remains an option for future analyses.
6. RESULTS This section presents the empirically estimated myopic‐ and rational models of addiction based on data regarding alcohol consumption in Greenland in the years 1970‐2010. All analyses were initially estimated with both the income and Disp. GNI variables as measures of wealth. For all model estimations the implication of using Income was more significant than when the Disp. GNI variable was applied. Hence, only models using Income is presented in the result tables. Although data on average income is not complete in all years, one advantage from applying this variable is that observations are available from as early as 1960. Thus, it is possible to estimate an approximate income variable for the entire period 1970‐2010 with the actual income in 1960 as a baseline. Both the myopic and the rational models of addiction have been estimated separately for beer, wine and spirituous beverages. The reason for this is first of all the importance of investigating whether consumption of any the three alcoholic beverages can be characterized as an addictive consumption. This can be established from a positive impact of past consumption on current consumption in the myopic model. Consequently, if beer, wine and spirituous consumption can be defined as addictive on the basis of a positive impact of lagged consumption, it will then be interesting to observe how adding on period of lead prices to the model will affect current consumption. If inclusion of future prices results in a negative impact on current consumption, and hence adds to rejecting the myopic models of addiction, the next step will be to impose a constrain to insure, that the effect of future prices works solely through their effect on future consumption, and estimate the rational model of addiction for each alcoholic beverage. On the basis on the rational addiction model, it will be possible to test the two main prediction of (i) inter‐ temporal complementarities and negative cross‐price effects and (ii) the long‐run price elasticity of demand exceeding the short run price elasticity. Overall, the results from the analyses in this sections will not only reveal if consumption of beer, wine and spirituous beverages in Greenlandic can be characterized as an addictive behavior; it will also analyze the nature of this addictive behavior and be able to distinguish which alcoholic beverage generates the strongest addiction and which are the most sensitive to price changes.
6.1. PRELIMINARY TESTS Before testing the rational addiction hypothesis against the myopic addiction hypothesis, it is necessary to look into possible concerns with the data that can cause problems in the statistical model estimations. The 2SLS estimator is less efficient than the OLS estimator when the explanatory variables are exogenous (Wooldrige 2006). This step has been overlooked in several of the previous empirical studies of rational addiction possible due to the theoretical definition of the model for rational addiction whereby the optimization of equation (6) implies that the same error et directly affects consumption at all time periods. A positive serial correlation in the unobserved effect would incorrectly imply that past and future consumption positively affects current consumption, even if endogenity is not an issue. Still, it is useful to test for endogenity of the explanatory variables suspected not to be exogenous before applying the 2SLS estimator, in order to avoid using an IV estimator which is more computationally intensive and less efficient than the OLS estimator, if it is not a necessity. Testing for endogenity is rather simple and the first sub section to come will do so by performing a Hausman test that compares the OLS and 2SLS estimators. This is followed by an endogenity test of the variables of interest by obtaining the reduced form residuals and regressing them on the structural demand equation with an OLS regression. Without revealing too much, the next sections treat another overlooked issue in the literature: testing the significance of the instrumental variables. This step has also often been left out in the literature on empirical analyses of rational addiction. As mentioned earlier, estimates are generally inconsistent if the applied instruments are correlated with the error term in the equation of interest or uncorrelated with the instrumented variable. For this reason, it is particular important to perform a statistical test of the instruments and not settle with a logical argument for their applications. This will strengthen the findings in the final estimations of the rational addiction models. After these initial, but most important exercises, this section continues with the model estimations of first the myopic addiction model and here after the rational addiction models on Greenlandic alcohol consumption followed by estimations of the long‐ and short run price elasticity.
6.1.1. TEST OF ENDOGENITY The 2SLS estimator is less efficient than the OLS estimator when the explanatory variables are exogenous due to the large standard errors produced by the 2SLS estimator. Therefore, it is useful to test for endogenity of the explanatory variables suspected not to be exogenous before applying the 2SLS estimator. The structural equation (6) from section 3.3 is given by; C t C t 1 C t 1 1 Pt 2 et 3 e(t 1)
Where C (t 1) and C (t 1) are suspected of being partially endogenous explanatory variables by being correlated with the unobservable error term et that affects utility in each period. Fortunately the exogenous variables P(t 1) , P(t 1) , T (t 1) and T (t 1) do not appear in (4). If C (t 1) and C (t 1) are uncorrelated with et an OLS estimate should be used to estimate equation (6). One approach for testing if the OLS and 2SLS model estimations differ is the Hausman test. The test evaluates the significance of an estimator versus an alternative estimator and in this case the test directly compares the OLS and 2SLS estimates. The null hypothesis is here that the OLS estimator is consistent. If this hypothesis is accepted the OLS estimates will be preferred over of the 2SLS. If the hypothesis is rejected a visible differences between the coefficient estimates in the two models will be expected. The larger the difference, the bigger the Hausman statistic and the less similar are the coefficients. A large and significant Hausman statistic means a large and significant difference, and so the null hypothesis that the OLS methods are efficient will be rejected. The Hausman test has been performed and the results are in the table below. As it can be seen, the Hausman statistics are very small and the null hypothesis cannot be rejected. The implication of this is that the OLS estimates cannot be regarded as inconsistent. TABLE 22.214.171.124: HAUSMAN TEST COMPARING OLS AND 2SLS ESTIMATIONS OF THE STRUCTURAL EQUATION FOR ALCOHOL DEMAND Output values Beer Wine Spirituous chi2(5)
Prob. > chi2
Another way to check for endogenity of the parameter of interest is the reduced form approach. In order to test for endogenity of Ct 1 and Ct 1 the reduced form equations are obtained by regressing the two suspected endogenous variables on all endogenous variables including those in the structural equation and the additional IVs. Using past and future prices and taxes as instruments these reduced form equations will be; Ct 1 0 1 Pt 2Yt 3 d 4 Pt 1 5Tt 1 5 Pt 1 7 Tt 1 t 1
Ct 1 0 1 Pt 2Yt 3 d 4 Pt 1 5Tt 1 5 Pt 1 7 Tt 1 t 1
Here t 1 and t 1 are the error terms for the reduced form equations. Ct 1 and Ct 1 are uncorrelated with et if ‐ and only if ‐ respectively t 1 and t 1 are uncorrelated with et . Because it is possible to estimate the
reduced form of for Ct 1 and Ct 1 by OLS, it is possible to obtain the reduced form residuals of the error terms t 1 and t 1 and regress them on the dependent variable Ct using an OLS regression. This new structural equation is then given by: Ct 0 1 Pt 2Yt 3 d 4 vˆt 1 5 vˆt 1
Then, it is possible to test for joint significance of the residuals in the new structural equation by an F test. Joint significance will indicate that at least one suspected explanatory variable is endogenous. The results are presented in table 126.96.36.199. Since all computed F statistics exceed the 5 percent critical values the results indicate that one period lag and lead consumption are endogenous for consumption of both beer, wine and spirituous beverages. It does, however, not reveal if one or both variables have a partial effect on current consumption. The implication for this is, that OLS estimation of the structural demand equations beer, wine and spirituous beverages for Ct will lead to inconsistent estimates of the coefficients for lagged and lead consumption. After establishing that Ct 1 and Ct 1 are endogenous the analyses of the rational addiction model for alcohol consumption in Greenland can proceed using the 2SLS an instrumental variable approach although the results from the Hausman test established that it was not possible to reject the OLS estimates. One reason for the divergence in the results from the Hausman test and the reduced form approach is that the methods used are defined in terms of asymptotic theory and are therefore viable for large data sets, while application to relatively small data sets may well result in differences in indications. On these grounds both approaches will be applied in estimating the rational addiction models for beer, wine and spirituous beverages. TABLE 188.8.131.52: TEST FOR ENDOGENITY OF THE Ct 1 AND Ct 1 Output values Beer F‐statistics 29.12 5 % Critical value 3.36
Wine 16.87 3.40
Spirituous 7.33 3.40
6.1.2. TEST OF INSTRUMENTS When more than one endogenous explanatory variable is present in the structural equation, at least as many excluded exogenous variables are necessary in order to fulfill the order condition. The next steps in the analyses are to establish which exogenous variables that make good instruments for one period lagged and lead consumption. As assumed in section 5, the effect of past and future cost of alcohol on current consumption must appear through their effect on past and future consumption. Based on this hypothesis past and future prices serve as logical instruments for past and future consumption, since past prices directly affect past consumption in the same way as future prices affect future consumption. But also past and future excise taxes on alcoholic beverages could be good instruments since they also have an effect on current consumption through past and future consumption. The reason for this is, as earlier mentioned, that taxes changes often are announced and discussed in the public media well before the come into effect. Through this mechanism, the consumers are made aware of the price changes to come. It is not possible to test whether the covariance between the instruments and the unobservable error term in equation (6) is null, but since it is intuitively appealing and necessary for the further analysis is will be assumed that there is no other relation besides the effect that might appear through the instrumented variables. By contrast, the condition that P(t 1) and P(t 1) or T (t 1) and T (t 1) should be partially correlated with C (t 1) and C (t 1) can be tested by a simple OLS regression of the reduced form equations between the instruments, the instrumented variables and the other remaining exogenous variables from the structural equation. The results are presented in the table 184.108.40.206 below both the relevance of prices and taxes as instrument for past and future consumption. The coefficients of interest are the coefficients representing the partial effects of respectively past and future prices and taxes of beer, wine and spirituous on past and future consumption of each of three categories of alcoholic beverages. The significance of the remaining coefficient is not relevant in testing the validity of the instruments. For this reason, the coefficients for all other exogenous variables in the reduced form equations are not reported here. The t‐statistics for the coefficient estimate for past beer price and past wine price are both significant on a 5‐percent level. This implies, other things being held constant, that a 1 percent increase in past beer prices results in almost 2 percent decrease in past consumption. For wine a 1 percent decrease in past prices results in almost 1 percent decrease in past consumption. Past prices on these two goods have, as 68
expected, a partially negative effect on past consumption. Therefore, if P(t 1) is assumed to be uncorrelated with unobserved factors in the error terms, both Pb(t 1) and Pw(t 1) can be used for IVs for Pb(t 1) and Cw(t 1) . Besides being insignificant, the coefficient of Ps(t 1) also has an unexpected
direction. It is intuitively not logical that an increase in past price has a positive effect on past spirituous consumption. This strongly indicates, that Ps(t 1) is not a valid instrument for Ps(t 1) in the current demand function for spirituous. When looking at the results from the reduced form equation analyzing the effect of future prices on future consumption a similar effect as was found in the regressions of past prices on past consumption is present for wine. Conversely, the partial price effect on both consumption of beer and spirituous beverages appear to be positive. The beer coefficient is even statistically different from null, whereas the coefficient for spirituous remains insignificant. The results reported by the reduced form equations estimating the partial effect of import duties in period t 1 on consumption in the same period, are all significantly negative indicating an inverse effect of
respectively 0.28, 0.28 and 0.16 percent decrease in consumption in period t 1 from a 1‐percent increase in real import excise taxes on beer, wine and spirituous beverages in that same period. These results indicate that Tb(t 1) , Tw(t 1) and Ts(t 1) all are valid IVs for Cb(t 1) , Cw(t 1) and Cs(t 1) in that particular order. The increased level of significance further indicates that these previous import duty’s make stronger IV estimators for previous consumption than past prices. The reduced form equations estimating the partially correlation between T (t 1) and C (t 1) reports that all coefficient estimates for the tax variable are significant at either a 1‐percen or 5‐percent level. Further, all coefficients have the intuitively expected partially negative effect on consumption in period t 1 . The effect is strongest for spirituous beverages where a 1 percent increase in taxes on spirituous beverages in period t 1 , ceteris paribus, is followed by a 0.8 percent decrease consumption, while the effect for beer and wine
both are roughly 0.25 percent. From the analyses in table 220.127.116.11 it would seem that taxes on alcoholic beverages are the better of the two considered instruments to be used as an IV for past consumption were as there is no significant difference regarding future consumption. A final opportunity is to use both lagged and lead taxes and prices as instruments for lagged and lead consumption.
TABLE 18.104.22.168: TEST OF INSTRUMENT
Notes: * Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level
6.1.3. TESTING OVER IDENTIFICATION RESTRICTIONS In the previous section it was defined that an IV must satisfy two requirements: 1. Instruments must be uncorrelated with the error term in the structural equation and 2. it must be at least partially correlated with the expected endogenous explanatory variable. This first requirement could not be tested because it involves a correlation between the IV and an unobserved error. However, if more than one possible instrumental variable is present it is possible to effectively test whether the combination of them are uncorrelated with the error term from the structural equation assuming that at least one IV is exogenous. This is called testing the overidentified restrictions. The number of overidentified restrictions is simply the number of extra instrumental variables. In this case, both lagged and lead prices and taxes can be possible instruments for consumption. Hence, the analyses have one overidentifying restriction for each instrumented variable. Testing these overidentifying restrictions is rather simple. The approach is to obtain the 2SLS residuals of the structural demand equation and the run an auxiliary regression (Wooldridge 2006). The table beneath contains the results for testing the overidentified restriction for beer, wine and spirituous beverages when applying both lagged and lead prices and taxes as instruments for on period lagged and lead consumption. Since the F‐statistics do not exceed the critical value for any of the models, it is not possible to reject the null hypothesis stating that at least some of the IV are exogenous, therefore the inclusion of both taxes and prices pass the overidentification test, and a model applying both instruments will be estimated for comparison. TABLE 22.214.171.124: TEST OF P(t 1) , T (t 1) AS IV FOR C (t 1) AND P(t 1) , T (t 1) AS IV FOR C (t 1) Output values Beer Wine
5 % Critical value
6.2. MYOPIC MODELS OF ADDICTION After testing the instruments for C (t 1) the myopic models of addiction can be estimated for beer, wine and spirituous beverages separately. The estimations are presented in table 6.2.1 to 6.2.3. For each type of alcohol, four different models versions are reported. The first estimation (a) of the myopic model for addiction for each category of beverages is a 2SLS estimation where T (t 1) is applied as an instrument for C (t 1) while the second estimation (b) applies P(t 1) as an instrument for C (t 1) . The third estimation (c) for each beverage utilizes both T (t 1) and P(t 1) together as an instrument for C (t 1) . Finally the last estimation (d) of the myopic model for addiction for addiction is based on an ordinary least‐ square estimation. The implication is, that past consumption is treated as an endogenous variable in (a) –(c) because of the likelihood that the unobserved variables that affect current utility in t are serially correlated, as indicated by the parameter specific endogenity test in section 6.1.1. In (d) however, past consumption is treated as exogenous because the small results from the Hausman test implied that the OLS estimate of the structural demand equation for beer, wine and spirituous could not all together be regarded as inconsistent.
6.2.1. MYOPIC MODEL OF BEER ADDICTION When considering the variations in the coefficient estimates for C (t 1) in (a) ‐ (d), no striking differences appear for the beer models in table 126.96.36.199. Especially the three 2SLS models yield very similar coefficient estimates. However, they do not differ from the estimates in model (d), thus indicating that the predictions made by the OLS estimation are consistent and problems of endogenity are modest in this model. For all models R‐square are alike and about 83 percent of the variation in current consumption is explained by variation in the included explanatory variables. Excluding the dummy variable for rationing years, since it is insignificant in all estimations, might improve the overall model for beer demand. TABLE 188.8.131.52: ESTIMATES OF THE MYOPIC MODEL OF ADDICTION FOR BEER Beerb Independent variable Beera .9442126* .9266655* C (t 1) (.1077037) (.2389819) ‐.2633877 ‐.2515276 Pt (.2131844) (.2560825) ‐.1551539** ‐.1502761 Yt (.0596985) (.0861097) ‐.1348874 ‐.1339515 D1979 1981 (.064042) (.0645283) AdjustedR r n
Beerc .943125* (.1066952) ‐.2626526 (.2128496) ‐.155462** (.0595198) ‐.1348294 (.0640025)
Beerd .8179397* (.0818462) ‐.17804 (.2016233) ‐.1820762* (.0545975) ‐.1281515 (.0618738) 0.8397 40 a
Notes: Intercepts are not shown. * Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. 2SLS estimate with C (t 1) treated as endogenous and T (t 1) as instrument. b2SLS estimate with C (t 1) treated as endogenous and P(t 1) as instrument. c2SLS d estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous.
The coefficient C (t 1) accounting for addiction is statistically significant at the 1‐percent level in all models. This indicates that current consumption of beer is highly affected, positively by consumption in the previous year for the overall Greenlandic population. A 1‐percent increase in past consumption is followed by an average of around 0.9 percent increase in current consumption, ceteris paribus. An unexpected finding from model all models are that current wealth seems to be inversely related to current beer consumption. For model (a), (c) and (d) the negative coefficient is even statistically significant. This negative income elasticity is opposite of what is expected from economic theory where an increase in wealth, all else being equal, would result in an overall increased consumption. The short‐term price elasticities detonated by the coefficient for Pt vary ‐0.18 in (d) to ‐0.26 in (a) and (d) but are statistically insignificant for all model estimation. However, they all indicate a negative impact on current consumption, as expected by economic consumption theory but as opposite to the conventional assumption about inelastic demand function for addictive goods.
6.2.2. MYOPIC MODEL OF WINE ADDICTION In the myopic model of addiction considering wine, the differences between the coefficients in the 2SLS and the OLS estimation are slightly more obvious by introspection but there are still modest problems of endogenity in the models and all coefficient estimations points toward similar effects of the exogenous variables on wine consumption. The relatively large R‐square values indicate that about 80 percent of the variation in wine consumption can be explained by the exogenous variables in the model. TABLE 184.108.40.206: ESTIMATES OF THE MYOPIC MODEL OF ADDICTION FOR WINE Wineb Independent variable Winea .8211207* .7515469** C (t 1) (.165266) (.278471) ‐.3993104 ‐.4918153 Pt (.2739424) (.4024649) .2662023 .3081755 Yt (.2299215) (.2617545) .8820769* .8322689* D1979 1981 (.1853261) (.2424729) AdjustedR r n
Winec .8139575* (.1624287) ‐.4088209 (.2705692) .2705176 (.2284965) .8769561* (.183632)
Wined .6228302* (.1166238) ‐.6625779 (.2192853) .3856574 (.2084126) .7403241* (.1588247) 0.8128 30 a
Notes: Intercepts are not shown. * Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. 2SLS estimate with C (t 1) treated as endogenous and T (t 1) as instrument. b2SLS estimate with C (t 1) treated as endogenous and P(t 1) as instrument. c2SLS d estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous.
The coefficient estimate for C (t 1) in (a)‐(c) are slightly larger than in the OLS model, but so are the standard deviations and the coefficient in (b) is only significant on a 5‐percent level instead of at the 1‐ percent level as in the other model estimations. However, they all point to a positive addictive behavior. As was the case in the myopic model for beer, the short‐term price elasticities are negative though statistically insignificant for all estimations of the myopic demand function for wine. Even though not much interpretation can be put into insignificant coefficient estimates, it is worth noticing that they indicate a slightly larger numerical elasticity than was found for beer. This matches finding from the previously mentioned studies in section 2.4 that point to the findings of beer being less sensitive to changes in price than both wine and spirituous beverages. The significance level of income elasticity is also below an acceptable in the myopic demand estimations for wine. However the direction of the potential partial effect is positive in the wine models as opposed to the beer models, indicating that an increase in income would lead to an increase in consumption. The coefficient for the dummy variable for rationing in the years 1979‐1981 is positive and statistically significant at the 1‐percent level in all models, this is consistent with the finding a in section 4.1 where it was seen that wine consumption increased relative to other alcoholic beverages during these years. 74
6.2.3. MYOPIC MODEL OF SPIRITUOUS ADDICTION A considerably larger difference between model estimations is detected in the spirituous models, where the OLS estimation for C (t 1) appears somewhat smaller indication problems of endogenity in the explanatory variables. The implication of this is, that OLS estimation of the structural demand equations for spirituous for will lead to inconsistent coefficient estimates. The fact that the R‐squared are especially low in the 2SLS models is, in itself, not important, since OLS minimizes the sum of squared residuals. Nevertheless, the indication is, when including the OLS model, that very little variation in the consumption of spirituous beverages can be explained by variation in the included explanatory variables. TABLE 220.127.116.11: ESTIMATES OF THE MYOPIC MODEL OF ADDICTION FOR SPIRITUOUS Independent variable Spirituousa Spirituousb 1.315886* 1.383062 C (t 1) .4328549 2.062227 ‐.5540792 ‐.6174696 Pt .707288 2.011962 ‐.2398581 ‐.2249973 Yt .3725253 .5870392 ‐.050395 ‐.027941 D1979 1981 .3321891 .7484191 AdjustedR r n
Spirituousc 1.321044* .3897399 ‐.5598052 .6873196 ‐.2385157 .3717126 ‐.0483668 .3274924
Spirituousd .8000363* .1516776 ‐.0753686 .4885734 ‐.3520835 .2931592 ‐.2199633 .2471556 0.5436 30 a
Notes: Intercepts are not shown. * Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. 2SLS estimate with C (t 1) treated as endogenous and T (t 1) as instrument. b2SLS estimate with C (t d 1) treated as endogenous and P(t 1) as instrument. c2SLS estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous.
All coefficient estimates for the impact for lagged consumption indicates an addictive behavior. Coefficient estimates in (a), (c) and (d) are all statistically significant at the 1‐percent level, and it is worth noticing that especially the 2SLS coefficients indicate a large effect of lagged consumption. A 1‐percent increase in one period lagged consumption leads to a 1.3‐percent increase in current consumption. The direction of the shot‐term price elasticities, income elasticities and the consumption in rationing years has the same implication of a negative partial effect as in the myopic demand for beer, although neither these effects are significant here. The size of the standard deviations in (b) indicates that P(t 1) is a weak instrumental for C (t 1) in the myopic model of spirituous demand. Thus, the size of the coefficients in all 2SLS models are alike, the use of weak instrumental variables can cause inconsistency and imprecision in the IV/2SLS estimators. This is consistent with the finding in table 18.104.22.168 where P(t 1) was insignificant in the reduced for equation when testing the validity of the instruments.
According to the parameter estimates of the myopic models presented in table 22.214.171.124 to 126.96.36.199, consumption of beer, wine and spirituous beverages is positively and significantly related to past consumption. This is consistent with the hypothesis that alcohol consumption is an addictive behavior. According to the parameter estimates in each model, spirituous is the most addictive substance follow beer and finally wine. The result regarding spirituous beverages as the most addictive substance is intuitively expected since this is the strongest of the three beverages and thereby also the most habit forming. The range of the estimates for beer and wine would by this logic be reversed, since wine is the stronger of the two beverages. But beer is also the cheaper. Especially in a Greenlandic context, where wine is fairly expensive relatively to beer and spirituous beverages when considering the strength measured in percent of pure alcohol, wine can be considered as a luxury good that is primarily enjoy be the higher social classes. This could explain why, in an aggregated context, wine does not appear to be as addictive as beer although it has a stronger alcoholic content. A further confirmation of this hypothesis can be found when considering the coefficient for the effect of income in the three models. Wine is the only beverage where current consumption is positively related to income, even though this is not significant. Directions of the effect of current income on beer and spirituous consumption, were only the coefficient for beer is significant, are unexpected in the sense that the coefficients are negatively related to current consumption of both beverages. This tendency is not compatible with economic theory of the demand for a normal good. Even though the coefficients for the myopic models indicate that consumption of all three categories of alcohol is inversely related to current prices these associations are not significant on an acceptable level of . When considering the dummy variable for rationing in the period 1979‐1981, it is only significant in the myopic models for wine, where the direction of the partial effect is opposite to the insignificant effect in the beer and spirituous models. This indicates that wine consumption has increased in those years. This tendency is not surprising when considering the construction of the point system. As mentioned a bottle of wine ultimately required 3 point were as a beer required one point and a bottle 35 cl. of hard spirituous beverage required 12 points. This made a bottle of wine relatively more affordable compared to both beer and spirits when considering the alcohol percentage. To check if the models op myopic consumption of beer, wine and spirituous beverages are correct a one period, lead of prices are added to the models in table 188.8.131.52 to 184.108.40.206. Since only the effect of adding this variable to the model is relevant in this context, all other coefficients have been left out of the result in tables
6.3. TEST OF THE MYOPIC ADDICTION MODELS The results of adding one period lead price to the myopic addiction functions are presented in table 6.3.1 to 6.3.2 below. When one period lead of taxes is added to the models of myopic addiction the coefficients are negative, as predicted by the rational addiction theory, for almost all model estimations except the OLS estimations of beer and spirituous beverages. Even though the coefficients are insignificant in all model estimations for beer and spirituous beverages, the significant coefficients on all the wine models suggest that decisions about current consumption of this beverages depend on future prices. These findings are inconsistent with a myopic model of addiction, but consistent with a rational model of addictive behavior in which a raise in expected future prices reduces expected future consumption, which in turn reduces current consumption. Even though these results reject the myopic model of addiction, they do not provide definitive evidence in support of the rational addiction model, because they do not impose the constraint that the effect of future prices works solely through future consumption. However, they do suggest that consumers do consider the future prices of wine in their current consumption decisions and therefore it is worth trying to obtain the structural estimates of the rational addiction demand functions for wine in particular. TABLE 6.3.1. TEST OF THE MYOPIC MODEL FOR BEER ADDICTION a Independent variable Beer
P (t 1)
Notes:*Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. 2SLS estimate with C (t 1) treated as b c endogenous and T (t 1) as instrument. 2SLS estimate with C (t 1d) treated as endogenous and P(t 1) as instrument. 2SLS estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous. a
TABLE 6.3.2. TEST OF THE MYOPIC MODEL FOR WINE ADDICTION Independent variable Winea ‐.67441** P (t 1) (.2508855)
Wineb ‐.6939092* (.2361229)
Winec ‐.6765217* (.2470456)
Wined ‐.6905471* (.234698)
Notes:*Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. 2SLS estimate with C (t 1) treated as b c endogenous and T (t 1) as instrument. 2SLS estimate with C (t 1d) treated as endogenous and P(t 1) as instrument. 2SLS estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous. a
TABLE 6.3.3. TEST OF THE MYOPIC MODEL FOR SPIRITUOUS ADDICTION Independent variable Spirituousa ‐.3546984 P (t 1) (.7065704)
Spirituousc ‐.3725435 (.7084261)
Spirituousd .0448694 (.5360125)
Notes:*Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. 2SLS estimate with C (t d1) treated as c endogenous and T (t 1) as instrument. 2SLS estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous. a
6.4. RATIONAL MODELS OF ADDICTION Even though the inclusion of one period lead prices in the myopic models of addiction for beer and spirituous beverages did not turn out to be statistically significant, the rational addiction models for both of these beverages will be estimated in any case for. This is done because the indication of the effect of on period of lead prices in the myopic demand function was negative, as predicted by the rational addiction model, although coefficient was not statistically different from null. The results from the myopic model estimations in section 6.3 suggested that very little internal difference occurred between the 2SLS model estimations for all three types of beverages. Several strategies can be pursued in fitting the rational addiction model. In section 4.1 both prices and taxes were tested as valid instruments. Here no significant difference was found between the two as instruments for future consumption. This, combined with the positive test of the overidentifying restriction models which concluded that inclusion of both taxes and prices was a valid instrument for one period lagged and lead consumption, only the overidentifying restriction model and the OLS estimates of the rational addiction models for beer, wine and spirituous beverages will be presented in the following section.
6.4.1. RATIONAL MODELS OF BEER ADDICTION When considering the results of the estimated rational addiction models for beer in table 220.127.116.11, a considerably large difference exist between the two model estimations when it comes to the parameters of primary interest C (t 1) and C (t 1) . TABLE 18.104.22.168: ESTIMATES OF THE RATIONAL MODEL OF ADDICTION FOR BEER Independent variable C (t 1)
Beerc 1.051086 (1.084035)
Beerd .5047075* (.1016827)
C (t 1)
‐.1247131* (.0512297) 0.8902 40
D19791981 AdjustedR r n
Notes: Intercepts are not shown. * Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. c d 2SLS estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous.
The OLS model for beer demand is broadly consistent with the rational addiction hypothesis. Past and future consumption variables enter with significantly positive coefficients representing inter‐temporal complementarities of consumption, whereas the price term, though not significant, has a negative direction. Also, the income elasticity has a negative, insignificant direction while the dummy variable for the rationing years once more indicate a significant fall in beer consumption in the years from 1979 to 1981. The discount factor ( ) , given by the ratio of the coefficient of future consumption to the coefficient for past consumption, yields a plausible interest rate of 6 percent. The results from the instrumental variables estimation in (c) are presented as a mean of checking the sensitivity of the OLS estimates to a potential endogenity bias. The information that comes from the somewhat larger and insignificant coefficients in the 2SLS estimation indicate problems of endogenity in the explanatory variables in the OLS model, as expected by the theoretical frame work for the rational addiction model. Most importantly, the coefficient estimate for C (t 1) is reversed in the 2SLS model compared to the OLS estimate. The implication of this is that there is no statistically significant proof for the anticipation effect of future price on current beer consumption in Greenland, as otherwise predicted in the OLS model.
6.4.2. RATIONAL MODEL OF ADDICTION FOR WINE From table 22.214.171.124, which contains the rational addiction models for wine, it can be seen that the coefficient for future wine consumption enters with a positive and significant sign in both model. More over the size of the coefficient estimates for C (t 1) and C (t 1) and the standard deviations for all coefficients in both models are fairly similar. This is reassuring as the results in (d) otherwise could be contaminated by endogenity bias. TABLE 126.96.36.199: ESTIMATES OF THE RATIONAL MODEL OF ADDICTION FOR WINE Independent variable
Winec .5831872* (.1788965)
Wined .4683799* (.1233865)
C (t 1) C (t 1)
D19791981 AdjustedR r n
Notes: Intercepts are not shown. * Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. c d 2SLS estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous.
The positive and significant future consumption coefficient is consistent with the hypothesis of rational addiction and inconsistent with the hypothesis of myopic addiction. Evidently, the estimates indicate that wine consumption is a rational addiction, in the sense that changes in both past and future consumption significantly affects current consumption of wine. Furthermore, both the price and income terms have the directions predicted by the rational addiction theory, although they are both statistically insignificant at an acceptable level. Although the results indicate that there is evidence of the expected anticipation effect from the rational addiction theory in the structural model for wine consumption, there is still the issue of the discount factor. The discount factor will need to be evaluated to further decide whether the estimates are fully consistent with rational addiction. The point estimates of the discount factor of 0.78 in (c) and 0.82 in (d) are in the low end of a plausible estimate, since they corresponds to interests rates of respectively 27.9 and 21.2 percent which are in the high end of realistic interest rates measures.
6.4.3. RATIONAL MODEL OF ADDICTION FOR SPIRITUOUS In the rational addiction models for spirituous in table 188.8.131.52 the coefficients in the two models differ only slightly although all coefficients have the same effect on the spirituous consumption. TABLE 184.108.40.206: ESTIMATES OF THE RATIONAL MODEL OF ADDICTION FOR SPIRITUOUS Independent variable Spirituousc .5409992 C (t 1) (.9792222) .4874423 C (t 1) (.6610222) ‐.2313701 Pt (.5341985) ‐.1558719 Yt (.2680103) ‐.3606317 D19791981 (.43631389 AdjustedR r n
Spirituousd .3852234* (.1353915) .5515523* (.1157666) ‐.138643 (.3577535) ‐.1553334 (.2188549) ‐.4192061** (.1856265)
Notes: Intercepts are not shown. * Statistically significant at the 1‐percent level. **Statistically significant at the 5‐percent level. c d 2SLS estimate with C (t 1) treated as endogenous and T (t 1) , P(t 1) as instrument. OLS estimate with C (t 1) treated as exogenous.
Even though only the coefficients for C (t 1) and C (t 1) are significant in the OLS model, the direction of the estimates for C (t 1) and C (t 1) in both models for spirituous demand is consistent with the rational addiction hypothesis. Past and future consumption variables both enter with positive coefficients, as in the rational demand model for wine, representing inter‐temporal complementarities of consumption. The price term for spirituous beverages also has a negative direction as expected though it is not significant in either of the models. Like in the myopic demand model, the income elasticity in the rational demand model for spirituous also has an unexpected negative direction. The dummy variable in both models indicates that also consumption of spirituous beverages fell in rationing years. The coefficient is significant in the OLS model, and it has doubled in size compared to the myopic demand model for spirituous, where it was in significant. A noteworthy difference between the two models in 220.127.116.11 is that past consumption seems to have a larger impact than future consumption in (c) while it is opposite in (d), yielding discount factor larger than 1 and on the basis of this, a negative interest rate in the OLS model. This is not consistent with the rational addiction hypothesis and on the basis of this there is no statistically significant evidence of rational addiction in the structural model for spirituous consumption, despite the insinuation from the positive coefficients for one period lagged and lead consumption.
To sum up the findings from this sub section on rational addiction models, all estimated effects of past and future consumption on current consumption are significantly positive in the three OLS model. This positive and significant past consumption effects are consistent with the hypothesis that alcohol consumption is addictive for both consumption of beer, wine and spirituous beverages, as was found in the myopic models. The positive and significant future consumption coefficient is consistent with the hypothesis of rational addiction and inconsistent with the hypothesis of myopic addiction. Clearly, the estimates indicate the alcohol consumption is addictive in the sense that inter‐temporal complementarities between past and future consumption significantly affect current consumption. Besides the negative interest rate in the spirituous model the two other interest rates are plausible, especially the 6 percent rate in the beer model. Unfortunately the estimations of the rational addiction models, in all but the wine models, fail in providing statistically significant proof of the rational addiction model for alcohol consumption in both the 2SLS and the OLS estimations simultaneously. Due to the possible endogenity bias in the C (t 1) and C (t 1) estimates in the OLS models for beer and spirituous beverages, consumption of these alcoholic beverages cannot be determined as a rational addiction. For all models, a negative cross‐price tendency is presented. Even if it is not statistically significant different from null, it does indicate a decreasing tendency in consumption when prices go up. The price effects vary between ‐0.1 for beer to ‐0.3 for wine in the OLS models and ‐0.02 for wine and ‐0.3 for beer in the 2SLS. The result from this is, that the suspected rank of price sensitivity is inconclusive. When considering the differences in the magnitude of the parameters of past and future consumption, which measures the degree of reinforcement of addiction, they range between 0.4 to 0.6 for C (t 1) and the same for C (t 1) when excluding the negative coefficient estimate from the 2SLS model for beer. In comparison to the myopic models, there are no signs of consumption of spirituous beverages being more addicting than beer and wine. In the OLS models beer would seem to be the most addictive beverage with a coefficient estimate of C (t 1) on 0.5 followed by 0.46 from wine and lastly 0.39 from spirituous beverages. This rank is not changes in the 2SLS models where beer however seems to be almost twice as habit forming as both wine and spirituous.
6.5. PRICE ELASTICITIES As mentioned in section 3.3, a very important implication of Equation 10 in the rational addiction model is that the response to permanent price changes differs between the short and the long term. The short term price effect is the response to the change in price in period t and all future periods that is not anticipated until t, whereas the long term effect refers to the price change in all periods. This section will use the expressions in equations (12) and (13) combined with the estimates of the statistically significant OLS models for all beverages and the statistically significant 2SLS model for wine to compute the short‐run and long‐run price elasticities for beer, wine and spirituous consumption with respect to various price changes as defined in section 3.3.2. TABLE 6.5.1. PRICE ELASTICITIES
Ordinary price effects
Short‐term price effect
Long‐term price effect
From the price effects in table 6.5.1 it can be seen, that a 1‐percent increase in for example beer prices will have very different effect on the consumption of beer depending on whether the actual price effect is given by the ordinary price effect measure or the effects from the rational addiction model. If consumers can be characterized as rational addicts, as was the case with wine in table 18.104.22.168, an unanticipated increase in the wine price in period t will cause a decrease in short term consumption of 0.2 to 0.7 percent in the rational addiction model and around 0,3 in the myopic addiction model. The estimated price effect is somewhat larger in the rational addiction model, but the difference is not overwhelming. If, on the other hand, the price change in all periods the myopic model of addiction will still predict a price effect of 0.3 whereas the rational addiction model will predict a much larger decrease of approximately 2‐percent decrease in wine consumption. Even though the ordinary price effect are not tested to be statistically different from null in any of the model estimated in section 6.4.3 and the inter‐temporal complementarities of consumption in the OLS models for beer and spirituous cannot be supported by finding in the complimentary 2SLS models for these beverages, the table above is nevertheless presented to confirm that there is a divergence between the ordinary price effect and the price effects predicted by the rational addiction model. This is important since ignoring addiction might give rise to misleading estimates of the actual price sensitivity on alcohol consumption. 83
DISCUSSION OF RESULTS No anticipation effect of the type predicted by the rational addiction model was found in the Greenlandic data on beer and spirituous consumption. Of more importance is the fact, that no evidence for a negative lead price effect on consumption can be established for beer and spirituous. The above findings are clearly incompatible with the habit formation mechanism outlined in the rational addiction model. One possible explanation for this negative result could be that Greenlandic consumers of these alcoholic beverages are typically fairly short sighted in their consumption behavior. Another explanation for the insignificant results might also be that Greenlandic consumers of beer and spirituous did not anticipate the price changes, or perceived them as being temporary. In the Rational addiction model people will not adjust their consumption as much, when the price change is considered to be temporary as if they are considered permanent. This explanation is especially relevant in relation to the present case of alcohol consumption in Greenland, where price changes for beer and spirituous turned out to be temporary. In the long‐run, the real price of beer has actually declined. One argument for the lack of rational behavior in general mentioned by Skog & Melberg (2006) is, that a large instantaneous increase in prices as when excise taxes increase, bring about a substantial reduction in current consumption because consumers respond on the basis of a fairly inflexible cognitive behavior. However, after a while most consumers will become used to the new price level, and old habits (addictive capital) could bring about a regression towards the old consumption level. In sum, there are several possible mechanisms that can explain why consumption of beer and spirituous in Greenland could not be established as a rational addiction. Besides the above qualitative arguments it is important to mention, that the lack of significant results also could be a consequence of methodological problems or misspecifications of the model. Estimation the model for rational addiction in sub‐periods could potentially yield different results. From this the effect of the policy implication in specific periods could be analyzed more thoroughly. Also estimations of the rational addiction model for geographically limited areas could be interesting. Unfortunately such disaggregated data is not available for requisitioning from Statistics Greenland currently.
6.6. MAINE FINDINGS IN RESULTS Initially this section has performed several important preliminary tests that have been greatly overlooked in the previous literature on empirical analyses of the rational addiction hypothesis. These tests were inconclusive regarding problems of endogenity of the variables measuring lagged and lead consumption resulting in both OLS and 2SLS model estimations. Also the significance of the instruments where tested before continuing with the model estimations. The findings in section 6.2 established that consumption of beer, wine and spirituous beverages in Greenland can be defined as a myopic addiction since current consumption of these substances depends on current and past factors. The results in table 22.214.171.124 to 126.96.36.199 established that an increase in past consumption had a significant positive partial effect on current consumption. From this, an increase in past price, by reducing past consumption, would also reduce current consumption. Although the estimated price effects could not be established to be statistically significant different from null, the indication was that an increase in current price for all beverages would reduce current consumption. The significance of the myopic model were tested by adding one period lead prices to the structural demand equations for all three beverages since an anticipated increase in future price was expected to have no effect on current consumption. For the wine model the coefficient for lead prices was statistically significant and hereby wine consumption was rejected as a myopic addiction. Next, the rational addiction models were estimates for all models, even though only wine consumption showed clear indications of not being a myopic addiction. The results from this were as expected; positive inter‐temporal complementarity, as predicted by the rational addiction model, could only be significantly established for wine consumption. However, the point estimates of the discount factor for wine corresponded to an interest rate of roughly 20 percent, which is in the high end of a realistic interest rate. This does induce some concerns regarding the reliability of the rational addiction model for wine. Finally, this section tested a second key prediction of the rational addiction theory; the difference between the long‐ and short‐term price effects. The finding s in table 6.5.1 confirms that there is a divergence between the ordinary price effect and the price effects predicted by the rational addiction model and that the long term price effects exceed the short term effects. This has one potentially important policy implication since ignoring a rational addiction behavior might give rise to misleading estimates of the price sensitivity of alcohol consumption. The findings from this section and their implication for Greenlandic alcohol policy measures will be discussed in the next section of this thesis. 85
7. DISCUSSION The following section will briefly discuss the main on the theory, methods and results found in the analysis on rational addiction in Greenland. Price elasticities of the magnitudes found in this study may be regarded as surprising, particularly since alcohol consumption and drinking has previously been found less responsive to price. However, the results reported here are not inconsistent with earlier studies cited in section 2.4.. Greenland has a long history of prohibition, rationing and restricting in relation to alcohol consumption. Over time the alcohol legislation in Greenland has undergone many changes and become less strict in relation to rationing though there are still occasionally geographical prohibitions. A surprising finding when analyzing the symmetry between the development of the real price on alcoholic beverages in Greenland and the development in total consumption more closely is, that even though alcohol prices remain relatively high compared to for instance Denmark and large excise tax hikes have taken place frequently, the real price of beer has actually decreased in the period whereas the price of spirituous remained relatively constant. Only the real price of wine has ultimately increased by about 30 percent after experiencing a significant price drop during the rationing in 1979 to 1981. From an economical view this interaction between development in prices and total consumption is surprising since, all else being equal; a decrease in the relative price of a normal good would lead to a rise in over all consumption. In the rational addiction theory, consumers are assumed to be rational, far‐sighted and to anticipate the future consequences of their current actions. Because of the controversial assumption that additive behavior can be treated in a rational choice framework, the theory of rational addiction is not without its critics The difference in this, compared to earlier models of addiction where addicts are myopic in the sense that they ignore the effects of current consumption on their future utility, is that the rational addict recognizes that the marginal utility of future consumption depends on current consumption. All though the rationality assumptions in the theoretical model might be a strong assumption, the empirically testable assumptions regarding inter‐temporal complementarities and the notion that long‐run price elasticities exceeds short‐run price elasticities, combined with the intuitively acceptable argument
that consumer are forward looking to some extent, makes this behavioral addiction model preferable fro empirical analyses compared to other behavioral models of addiction. The variables describing the amount of alcohol imported to Greenland is an index for the per capita consumption of liters of alcohol derived from beer, wine and spirits respectively measured in terms of 100 per cent alcohol equivalent imported to Greenland per year. Because imports are used as an indicator of consumption, it must be kept in mind that several factors may complicate the relationship between imports and consumption. This study has a number of other limitations. Variables accounting for influential institutional changes might not have been exhausted in this analysis and in general testing a long‐run model of individual behavior on aggregated time series data is somewhat demanding due to the limited degrees of freedom from only one observation per period. An advantage in this data is, that as a consequence of the strict regulation of trade in alcoholic beverages and the fact that Greenland is an geographically isolated limiting cross border trade, the statistics for alcohol imports when available is of very high quality as there is no problems of underreporting or difficulties on obtaining information from heavy‐drinking consumers, which can be the case when applying survey‐based data in analysis of alcohol consumption. Also, in contrast to previous studies on rational addiction and alcohol (Grossman et al., 1998; Waters and Sloane, 1995), the past and future consumption levels used in this analysis is based on actual imports, and not constructed from data. To gain efficiency a parameter restriction implied by the theoretical model could be imposed to constraint the coefficient for past and future consumption in a manner whereby it is only allowed to differ by a predetermined discount factor. An addition possibility would to impose the restriction of negative long‐run price elasticity. Ideally, estimation of this model would be carried out with a pooled cross‐sectional or panel data set to control for stable individual factors that may be correlated with both alcohol prices and alcohol consumption. Previous analyses using individual level data suggest that alcohol demand may be even more responsive to price than aggregate estimates indicate, possibly because this approach can obtain differential price responses among respondents of different age groups.
8. CONCLUSION This study provides new information bearing directly on public health concerns about the adverse consequences of excessive alcohol consumption, a problem often thought beyond the reach of policy interventions. The study successfully for accounted addiction and provide an analysis of the price sensitivity and the interdependency of past, current and future consumption in relation to alcohol consumption in Greenland. The evidence presented in this thesis suggests that higher prices for wine would reduce consumption. This finding is consistent with earlier studies on the price elasticity on wine indicating that higher prices would reduce drinking. The primary object of this thesis was to analyze if alcohol consumption in Greenland could be characterized as a rational addictive behavior. To investigate this, it was initially necessary to assess whether reinforcement of past consumption fuelled current consumption of alcohol and whether current prices had the conventional negative effect on current consumption in Greenland. This was done by estimating the myopic demand functions for beer, wine and spirituous, respectively. After this, the analysis proceeded to estimate the rational addiction model by adding the effect of future consumption to the structural demand functions for the three categories of alcoholic beverages. Addiction is a major challenge to the theory of rational behavior, and the theory of rational addiction does not claim that all behavior associated with addiction is consistent with rationality. The key empirical implication of the rational addiction model is that present, past and future expected prices influence current consumption and the assumption that addicts consume the addictive good with perfect foresight of the consequences of addiction. Review of previous empirical studies on the relationship between prices and alcohol consumption seems to reveal a reversed correlation indicating negative price elasticity on the demand for alcoholic beverages. Several studies have found that the prices of alcoholic beverages are strongly influenced by taxes, and on the basis of this, tax hikes have turned out to be effective in reducing the overall alcohol consumption. Throughout the period from 1970 to 2010 a total of 13 episodes of increases in excise taxes on all alcoholic beverages have taken place in Greenland, and after 1992 the pass‐through rates of taxes on to prices have been close to one.
A surprising finding, when adjusting the alcohol prices for the consumer price index over the entire period, is that price increases on beer and spirituous have only been of a temporary manner while the increase in real wine price has been long‐term.
8.1. CONCLUSION REGARDING ADDICTIVE BEHAVIOR Looking at the magnitude of the parameters of past consumption, which measures the degree of reinforcement of addiction, the estimates in the myopic models of addiction for beer, wine and spirituous all support the mechanism of reinforcement whereby current consumption levels of the three alcoholic beverages depend on past consumption. On the basis of these findings it can be concluded that alcohol consumption in Greenland is consistent with addictive behavior.
8.2. CONCLUSION REGARDING RATIONAL ADDICTIVE BEHAVIOR Moving to the rational addiction models, both the OLS and 2SLS estimation of the rational addiction model for wine quite strongly reject the myopic consumer hypothesis by confirming that also future levels of wine consumption have a statistically significant partial effect on current consumption. The consistency between the two model estimations is reassuring, since the OLS estimates may be contaminated by endogenity bias. The implication of this is that Greenlandic consumers of wine can be characterized as rational addicts. Though not statistically significant in the 2SLS models, there are also positive indices of rational addiction of beer and spirituous in the OLS models. However, for all beverages in the analysis price and income elasticities were insignificant, and for beer and spirituous the income elasticities are opposite normal economic intuition.
8.3. POLICY IMPLICATIONS OF CONCLUSIONS The support of the key prediction from the rational theory, i. e. that past and future consumption of wine causes current consumption to rise, has one important policy implication: Ignoring addictive behavior in wine consumption might result in misleading estimates of the price sensitivity. The estimated short‐ and long‐run price elasticities in the 2SLS model for rational addiction of wine on respectively ‐0.17 to ‐2.0 are considerably higher than the elasticity implied by the conventional model of ‐ 0.02 and other earlier estimates derived from models not recognizing addiction such as ‐0.69 found by Wagenaar et al. (2009) and ‐1.0 found by Ornstein and Levy (1993). According to these result, the role of price policy in limiting wine consumption seems to be quite important. 89
Unfortunately, no anticipation effect of the type predicted by the rational addiction model was found in the Greenlandic data on beer and spirituous consumption. This finding has important policy implications, especially in relation to beer consumption, since the Greenlandic drinking culture indeed seems largely driven by beer consumption; people who consume beer and spirituous are not far‐sighted in their consumption behavior, and from the price elasticities in table 6.5.1, they do not respond a great deal to price changes. Overall I have found that wine addiction in Greenland corresponds to rational addiction behavior, whereas other instruments than monetary prices will be needed to influence the consumption of beer and spirituous. Among these could be efforts to raise the full price through campaigns on the negative effects of excessive alcohol consumption. I hope these findings along with the estimates of price elasticities of primarily wine prove a useful addition to the excising information on the relations between price, taxes and alcohol consumption in Greenland.
9. PERSPECTIVES In a study from 1999 Bentzen et al. analyze the rational addiction model on alcohol consumption in the Nordic countries in the period 1960 to 1994. The object of this study was to assess the consequences of a liberalized alcohol policy and lower alcohol prices which was expected in Sweden and Finland, as a result of their European Union membership. Greenland is not a member of the European Union or about to become so5. The reason why this study is interesting in relation to the object of this thesis is, that it is interesting to compare the finding regarding rational addiction in Scandinavia and the findings in section 6 regarding Greenland. From Scandinavian study show that there are considerable differences among the Nordic countries in the structure of alcohol demand. Consumption in Denmark is largely driven by beer consumption, as is the case with the Greenlandic consumption behavior, whereas the development in alcohol consumption in Sweden and Finland is highly influences by spirituous consumption and the both beer and spirituous take op equal shares of the total consumption in Norway. Also the differences between the developments in real prices differ between the countries. Here Denmark is the only country having experienced declining real alcohol prices in the period, as has been the development in Greenland. Overall, the Scandinavian analyses strongly reject the myopic consumer hypothesis and find implications for rational addiction behavior particular for spirituous consumption but also for beer and wine consumption. This is the opposite of the finding in this analysis regarding alcohol consumption in Greenland. Here the coefficient for past consumption was significant in the myopic models for both beer, wine and spirituous whereas only the coefficients for past and future consumption of wine had a partial effect on current wine consumption in the rational addiction models. According to this, the Greenlandic consumers appear to be less far‐sighted than their Scandinavian neighbors and the role of price policy in curtailing alcohol consumption seems to differ as a result of this. A future perspective in research of alcohol and addictive behavior could gain strength from analyzing what constitutes these differences in the discounting of time between Greenland and other neighboring countries in Scandinavia. Included in this analysis could be a study of the implication of a circular time perception, as it is known from the prehistorically Inuit culture. 5
As part of the Danish National Community, Greenland was a member of the EC until The Greenland Home Rule was introduced in 1979, and after a consultative referendum Greenland chose to leave the EC with effect from 1 February 1985.
REFERENCES Akerlof, George A (1991). Procratination and Obedience. A.E.R. Papers and Proc. 81; 1‐19. Udvalget for samfundsforskning i Grønland (1961). Alkoholsituationen i Vestgrønland. Danmark Bain JS (1956). Barriers to New Competition. Cambridge, MA: Harvard University press. Becker GS & Mulligan CB (1997). The Endogenous Determination of Time Preference.The Quarterly Journal of Economics. Oxford University Press. Becker GS & Murphy KM. (1993). “A Simple Theory of Advertising as a Good or Bad,” Quarterly Journal of Economics, 108, 941‐964. Becker GS & Murphy KM. (1988). A Theory of Rational Addiction. Journal of Political Economy, 96(4), 675– 700. Becker GS, Grossman M & Murphy KM. (1994). An empirical analysis of cigarette addiction. American Economic Review, 84:396‐418. Becker GS, Grossman M & Murphy KM (1991). Rational Addiction and the Effect of Price on Consumption. The American Economic Review, Vol. 81, No. 2, Papers and Proceedings of the Hundred and Third Annual Meeting of the American Economic Association. (May, 1991), pp. 237‐241. Befolkningsundersøgelsen i Grønland 2005‐2007 Levevilkår, livsstil og helbred. Bjerregaard P & Dahl‐ Pedersen Ik (2008). Statens institut for Folkesundhed, Syddansk Universitet. Bentzen J, et al (1999). Rational addiction and alcohol consumption: Evidence from the Nordic Countries. Journal of Consumer Policy 22: 257‐279 Bjerregaard P & Lynge I (2006). Suicide – A challenge in Modern Greenland. Archives of Suicide Research, 10;2:209‐220 Bjerregaard P & Juel K (1990). Avoidable accidents in Greenland 1968‐1985: Variation by region and period. Arctic Medical Research 1990;49:119‐127 Bjerregaard P (1990). Fatal accidents in Greenland. Arctic Medical Research 1990;49:132‐142 92
Boardman AE et al.(2006). Cost‐benefit Analysis – Concept and practice. 3th edition. Pearson, Prentice Hall Bridges JFP. Understanding Heroin as a ‘Rational Addiction’. European Addiction Research 2001, Vol. 7, No. 2: 78‐82 Chaloupka FJ & Andwechsler H (1996). Binge drinking in college: The impact of price, availability, and alcohol control policies. Contemporary Economic Policy 14(4):112–124. Chaloupka FJ (1988). An Economic Analysis of Addictive Behavior: The Case of Cigarette Smoking. Ph.D. dissertation, City University of New York Graduate School. Chaloupka F J (1991). Rational addictive behavior and cigarette smoking. Journal of Political. Economy, 99:722‐742. Coate D & Grossman M (1998). Effects of Alcoholic Beverage Prices and Legal Drinking Ages on Youth Alcohol Use, Journal of Law and Economics, Vol. 31, No. 1 (Apr., 1988), pp. 145‐171 Cook PJ & Moore MJ (2001). Environment and persistence in youthful drinking patterns. In: Gruber, J., Ed. Risky Behavior Among Youth: An Economic Perspective. Chicago: University of Chicago Press, pp. 375–437. Cook PJ & Tauchen G (1982). The effect of liquor taxes on heavy drinking. Bell Journal of Economics 13(2):379–390. Cook PJ (1981). The effect of liquor taxes on drinking, cirrhosis, and auto fatalities. In: M.H.Moore and D. Gerstein, eds., The Alcohol Problem: Beyond the Shadow of Prohibition (National Academy Press, Washington DC): 255‐285. El‐Safety AE (1976). Adaptive behavior, demand and preferences. Journal of Economic Theory 13:298‐318. Elster J (1979). Ulysses and the Sirens: Studies in Rationality and Irrationality. Cambridge, England:Cambridge University Press. Ghez G & Becker GS (1975). A Theory of the Allocation of Time and Goods Over the Life Cycle, National Bureau of Economic Research : distributed by Columbia University Press (New York) Grove O & Lynge I (1979). Suicide and attempted suicide in Greenland: A controlled study in Nuuk. Acta Psychiatrica Scandinavica, 60, 375‐391
Grossman M, Chaloupka FJ,Saffer H & Laixuthai A (1994). Alcohol price policy and youths: summary of economic research. Journal of Research on Adolescence, 4:347‐364. Gruber J & Koszegi B (2001): Is Addiction 'Rational?' Theory and Evidence, Quarterly Journal of Economics, 116(4), 1261–1305. Gruenewald Pauli ., Ph.D,. Economics and Prevention, Alcohol Problems and Availability. National Institute on Alcohol Abuse and Alcoholism. RESEARCH. Monograph‐25, 1994; 59‐90 Hammond PJ (1976). Endogenous tastes and stable long run choice. Journal of Economic Theory, 13:329‐ 340. Heckman, J. (1974): “Life Cycle Consumption and Labor Supply: An Explanation of the Relationship between Income and Consumption Over the Life Cycle,” The American Economic Review, 64, 188–194." Houthakker H S & Taylor LD (1970). Consumer Demand in the United States, 1929‐1970: Analyses and Projections (2nd Ed.). Cambridge, MA: Harvard University Press. Iannaccone L R (1984). Habit formation and consumption capital, with an application to religious participation. Ph.D. Dissertation, University of Chicago. Iannaccone LR (1986). Addiction and Satiation. Economics Letters 21 (1986):95‐99. ensen MK, Andersen AT, Sørensen TI, Becker U, Thorsen T, Grønbaek M. Alcoholic beverage preference and risk of becoming a heavy drinker. Epidemiology. 2002 Mar;13(2):127‐32. Jeppesen C & Krogh NL (2008). Alkohol og Hash. Befolkningsundersøgelsen i Grønland 2005‐2007 ‐ Levevilkår, livsstil og helbred, page 61‐74, København, Syddansk Universitet. Statens Institut for Folkesundhed. Kenkel DS (1993). Drinking, driving and deterrence: The effectiveness and social costs of alternative policies. Journal of Law and Economics 36(2):877–914. Kenkel DS (1996). New estimates of the optimal tax on alcohol. Economic Inquiry, 34:296‐319. Kenkel DS (2005). Are Alcohol Tax Hikes Fully Passed Through to Prices? Evidence from Alaska. American Economic Review Papers & Proceedings 95 (2): 273‐277.
Klatsky AL, Armstrong MA & Friedman GD (1992). Alcohol and mortality. Ann Intern Med Laixuthai A & Chaloupka FJ (1993). Youth alcohol use and public policy. Contemporary Policy Issues 11(4):70–81. Lavik B, Holmegaard C & Becker U (2006). Drinking patterns and biochemical signs of alcoholic liver disease in Danish and Greenlandic patients with alcohol addiction. Int J Circumpolar Health; 65: 219‐27. Leung SF & Phelps CE (1993). My Kingdom for a Drink ... ?: A Review of Estimates of the Price Sensitivity of Demand for Alcoholic Beverages in Hilton, M.E. and G. Bloss eds. Economics and the Prevention of Alcohol‐ Related Problems: Proceedings of a Workshop on Economic and Socioeconomic Issues in the Prevention of A1cohol‐Related Problems, October 10‐11, 1991 Bethesda, MD NIAAA Research Monograph No. 25 1993. MacKenzie DM. et al. (1996). Identifying hazardous or harmful alcohol use in medical admissions; a comparason of audit, Cage and Brief Mast. Alcohol and Alcoholism 1996;31:591‐599 Maclure M (1993). Demonstration of deductive meta analysis: ethanol intake and risk of myocardial infarction. Epidemiol Rev 1993;15:328 51 Manning WG, Keeler EB, Newhouse JP, Sloss EM & Wasserman J (1989). The Taxes of Sin: Do smokers and Drinkers pay their way? The National Center for Health Services Research and Health Care Technology Assessment. Manning WG, Blumberg L & Lawrence HM (1995). The demand for alcohol: The differential response to price. Journal of Health Economics, Volume 14, Issue 2, June 1995, Pages 123‐148 Marshall, Alfred. Principles of Economics. 1920. Library of Economics and Liberty. National Board of Health, Denmark 2010Sundhedsstyrelsens nye udmelding vedrørende alkohol, 31/8 2010 Nexøe J, Wilche JP, Færgemann C, Kjeldsen AB, Munck A & Lauritsen J (2010). Volds‐ og alkoholrelaterede skader behandlet i det grønlandske sundhedsvæsen. Rapport. Odense: Audit Projekt Odense. O'Malley PM & Wagenaar AC (1991). Effects of minimum drinking age laws on alcohol use, related behaviors and traffic crash involvement among American youth: 1976‐1987. J Stud Alcohol. 1991;52:478‐ 491.
Ornstein S I & Levy D (1983) Price and Income Elasticities and the Demand for Alcoholic Beverages. in Galanter, M., ed. Recent Developments in Alcoholism Vol. I New York: Plenum 303‐345 1983. Orphanides A & Zervos D (1995). Rational addiction with learning and regret. Journal of Political Economy, 103, 739‐758. Pernanen K ( 1974). Validity of Survey Data on Alcohol Use. In Don Gibbons et al. (eds.), Research Advances in Alcohol and Drug Problems. Vol. 1. John Wiley and sons, New Political Economy, 103, 739‐758. York Pollak R (1970). Habit formation and dynamic demand functions. Journal of Political Economy 78, 745–63. Pollak RA (1976). Interdependent Preferences. American Economic Review, 66:3 (June), 309‐320. Pogue TF & Sqontz LG (1989). Taxing to Control Social Costs: The Case of Alcohol. The American Economic Review (AER), 79(1), 235 ‐ 43. Robin Room, Alcohol Research Group, Institute of Epidemiology & Behavioral Medicine, Medical Research Institute of San Francisco, 1816 Scenic Ave., Berkeley CA 94709. From Annual Review of Public Health 5:293‐317, 1984. Cummins RA (2008) Commentary : preference‐weighted health‐related quality of life and addiction ‐ a critique, Addiction, vol. 103, no. 8, pp. 1330‐1332, Wiley‐Blackwell Publishing Ltd, Abingdon, England Ryder HE. Jr., Heal GM (1973). Optimal Growth with Intertemporally Dependent Preferences. Review of Economic Studies, vol. 40: pp. 1‐33. Saaffer H & Chaloupa FJ (1994). Alcohol tax equalization and social costs. Eastern Economic Journal 20(1):33–43, 1994. Saffer H (1997). Alcohol Advertising and Motor Vehicle Fatalities, The Review of Economics and Statistics, Vol. 79, No. 3 (Aug., 1997), pp. 431‐442 Schechter EJ (1986). Alcohol rationing and control systems in Greenland. Contemporary Drug problems. Winter 1986 Schelling TC (1978). Economics, or the Art of Self‐Management, The American Economic Review, Vol. 68, No. 2, Papers and Proceedings of the Ninetieth Annual Meeting of the American Economic Association (May, 1978), pp. 290‐294 96
Schelling TC (1984). Self‐Command in Practice, in Policy, and in a Theory of Rational Choice, The American Economic Review, Vol. 74, No. 2, Papers and Proceedings of the Ninety‐Sixth Annual Meeting of the American Economic Association (May, 1984), pp. 1‐11 Folkesundhed blandt skolebørn ‐resultater fra HBSC Greenland undersøgelsen 2010. Niclasen BV, Bjerregaard, P (2011). Statens institut for Folkesundhed, Syddansk Universitet. Shaper AG (1990). Alcohol and mortality: a review of prospective studies. British Journal of Addiction. Skog JO & Melberg HO (2006). Becker's rational addiction theory: An empirical test with price elasticities for distilled spirits in Denmark 1911‐1931.. Addiction. ISSN 0965‐2140. 101, s 1444‐ 1450 Spinnewyn, Frans 1981; Rational habit formation. European Economic Review, 15 (1981), pp. 91–109 Statistics Geenland. (2011). Greenland in figures 2011. Available from http://www.stat.gl/dialog/main.asp?lang=da&theme=Greenland%20in%20Figures&link=GF Stigler GJ & Becker GS (1977). De Gustibus Non Est Disputandum. American Economic Review. 67: 76‐90. Strand BH & Steiro A (2003). Alcohol use, income and education in Norway 1993 – 2000. Tidsskrift for Den norske lægeforening 2003, 123: 2849‐2053. Sundhedsundersøgelsen i Avanersuaq 2010.Bjerregaard P, Dahl‐Petersen IK (2011). SIF’s Grønlandsskrifter nr. 23 Sølling L (1974). Alkoholforbruget I Grønland 1960‐1970. Bilag 3. Grønlands Import af Alkoholiske Drikke 1960‐1970. [Alcohol consumption in Greenland 1960‐1970. Appendix 3; In Danish] Copenhagen: Socialforskningsinstituttet Thaler RH & Shefrin HM (1981). An economic theory of self‐control. Journal of Political Economy 89, 392– 406. Thorslund J ( 1992). Ungdomsselvmord og moderniseringsproblemer blandt Inuit I Grønland [Youth suicides and problems of modernization among the Inuit in Greenland; in Danish] Holte: SOCPOL Carl Christian von Weizsäcker, Steady state capital theory. Springer‐Verlag, 1971 ‐ 102 sider Varian, H. (2006). Intermediate Microeconomics – A modern approach. 7th addition. Norton.
Wagenaar, A. C., Toomey, T. L., & Erickson, D. J. (2005). “Preventing youth access to alcohol: Outcomes from a multi‐community time‐series trial.” Addiction, 100(3): 335‐345. Wagenaar AC, Matthew JS & Kelli AK (2009).Effects of beverage alcohol price and tax levels on drinking: a meta‐analysis of 1003 estimates from 112 studies. Addiction Volume Waters TM & Sloan FA (1995). Why do People Drink? Tests of the Rational Addiction Model, Applied Economics, 27, 727‐736. World Health Organization. Global Status Report on Alcohol and Health 2004 World Health Organization. Global Status Report on Alcohol and Health 2011 Winston G (1980). Addiction and backsliding: A theory of compulsive consumption. Journal of Economic Behavior and Organization, 1:295‐324. Wooldridge JM (2006). Introductory Econometrics, Third Edition. Chapter 15. Thomson South ‐western Young DJ & Bielinska‐Kwapisz A (2002). Alcohol taxes and beverage prices. National Tax Journal LV‐1: 57‐73. Young DJ & Bielinskakwapisz A (2006). Alcohol prices, consumption and traffic fatalities. Southern Economic Journal 72(3):690–703. Young TK & Bjerregaard P (2008). The Circumpolar Inuit: health of a population in transition. København: Munksgaard.
APPENDIX 1: THE STRUCTURE OF ALCOHOL CONSUMPTION IN GREENLAND; 1960‐2010 (PERCENTAGE SHARES OF THE TOTAL CONSUMPTION) Year 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Beer 40,1 46,7 47,6 52,2 50,5 55,9 62,9 73,4 75,1 68,1 74,2 76,0 74,6 75,7 75,9 77,0 76,2 74,9 72,8 78,6 81,6 69,0 74,4 74,3 73,7 76,1 73,1 71,3 66,2 65,8 63,7 59,8 68,1 70,2 70,8 75,1 75,4 72,7 73,4 72,6 72,4 70,1 67,2 68,5 65,8 65,3 66,7 66,0 73,1 69,8 71,5
Wine 28,1 19,5 16,0 13,0 12,7 11,4 7,8 6,6 6,3 11,0 10,2 11,2 11,0 11,1 11,2 9,7 12,2 11,4 12,0 9,6 9,6 23,1 16,3 16,0 17,3 17,5 18,1 18,2 21,1 20,4 21,9 25,1 19,6 16,4 16,0 13,1 13,2 15,7 13,7 15,1 15,4 17,6 19,7 20,2 22,2 22,7 22,4 23,9 20,1 22,7 20,7
Spirituous 31,8 33,9 36,4 34,8 36,8 32,6 29,3 19,9 18,6 20,9 15,5 12,9 15,4 13,3 13,0 13,6 11,6 13,7 15,2 11,7 8,8 8,0 9,4 9,7 9,0 6,4 8,8 10,6 12,6 13,7 15,4 15,1 12,3 13,5 13,2 11,8 11,4 12,5 12,9 12,4 12,2 12,3 13,1 11,3 12,1 12,0 10,9 10,1 6,8 7,5 7,8
Total 5,7 7,6 8,0 11,5 12,8 15,1 16,7 17,0 19,4 13,5 15,4 15,7 18,0 19,1 19,1 18,4 18,3 18,6 18,6 15,8 13,6 15,6 21,8 21,1 19,9 18,4 20,1 22,0 16,7 17,5 15,5 15,0 15,1 12,8 13,3 12,5 12,8 12,8 13,3 13,2 13,5 12,4 12,3 11,5 11,6 12,1 11,7 11,9 10,5 10,7 10,4
APPENDIX 2: DEVELOPMENT IN GREENLANDIC ALCOHOL TAXES; 1971‐2010
Development in excise tax 1971‐1991
300 250 200 150 100 50 0
Wine, not above 14 % per liter Spirituous, not above 60 % per liter Beer per 35 cl, ordinary pilsner and lager
Other wine, per liter Other Spirituous, per liter Export beer, per 35 cl
Dkr. Index: 1971=100
Development in real excise tax 1971‐1991 90 80 70 60 50 40 30 20 10 0
Wine, not above 14 % per liter Spirituous, not above 60 % per liter Beer per 35 cl, ordinary pilsner and lager
Other wine, per liter Other Spirituous, per liter Export beer, per 35 cl
Data source: Data are provided from Statistics Greenland and statistical yearbooks from 1970‐2009.
APPENDIX 3. DEVELOPMENT IN BEVERAGES PRICES; 1971‐2010
350 300 250 200 150 100 50 0 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Development in spirituous price, 1981‐2010
Adjusted spirituous price CPI; 1981=100
Data source: Data are provided from Statistics Greenland and statistical yearbooks from 1970‐2009.
APPENDIX 4. DEVELOPMENT IN AVERAGE INCOME; 1971‐2010 Development in adjusted average income, 1971‐2010
25000 20000 15000 10000 5000 0
Adjusted average income, 1971=100
Data source: Data are provided from Statistics Greenland and statistical yearbooks from 1970‐2009.
APPENDIX 5: DO‐FILES FROM STATA 10.1 use "C:\Documents and Settings\isma\Skrivebord\stata_data.dta", clear
**Hausman test in stata** ivregress 2sls lncbeer pbeer income1 d (lncbeer1 lncbeer2= hausman, save regress lncbeer pbeer lncbeer1 lncbeer2 income1 d hausman ivregress 2sls cwine pwine income2 d ( cwine1 cwine2= hausman, save regress cwine pwine cwine1 cwine2 income2 d hausman
beertax1 pbeer1 beertax2 pbeer2)
winetax1 pwine1 winetax2 pwine2)
ivregress 2sls cspirits lnpspirits income2 d ( cspirits1 cspirits2= spiritoustax2 pspirits2) hausman, save regress cspirits lnpspirits cspirits1 cspirits2 income2 d hausman
**Test of endogenity of C(t-1) and C(t+1) regress lncbeer1 pbeer income1 d pbeer1 beertax1 pbeer2 beertax2 predict res_beer1, res regress lncbeer2 pbeer income1 d pbeer1 beertax1 pbeer2 beertax2 predict res_beer2, res regress lncbeer pbeer1 income1 d res_beer1 res_beer2 test res_beer1 res_beer2 regress cwine1 pwine income2 d pwine1 winetax1 pwine2 predict res_wine1, res regress cwine2 pwine income2 d pwine1 winetax1 pwine2 predict res_wine2, res regress cwine pwine income2 d res_wine1 res_wine2 test res_wine1 res_wine2
regress cspirits1 lnpspirits income2 d pspirits1 spiritoustax1 pspirits2 spiritoustax2 predict res_sprit1, res regress cspirits2 lnpspirits income2 d pspirits1 spiritoustax1 pspirits2 spiritoustax2 predict res_sprit2, res regress cspirits lnpspirits income2 d res_sprit1 res_sprit2 test res_sprit1 res_sprit2
**Instrumental variable check in stata** regress regress regress regress
lncbeer1 pbeer pbeer1 income1 d lncbeer2 pbeer pbeer2 income1 d lncbeer1 pbeer beertax1 income1 d lncbeer2 pbeer beertax2 income1 d
regress regress regress regress
cwine1 cwine2 cwine1 cwine2
regress regress regress regress
cspirits1 cspirits2 cspirits1 cspirits2
pwine pwine pwine pwine
pwine1 income2 d pwine2 income2 d winetax1 income2 d pwine2 winetax2 d
lnpspirits pspirits1 income2 d lnpspirits pspirits2 income2 d lnpspirits winetax1 income2 d lnpspirits winetax2 income2 d
**Over identifying** regress lncbeer pbeer lncbeer1 income1 d( pbeer pbeer1 beertax1 income1 d) test lncbeer1 regress lncbeer pbeer lncbeer1 lncbeer2 income1 d ( pbeer pbeer1 beertax1 pbeer2 beertax2 income1 d) test lncbeer1 lncbeer2 regress cwine pwine cwine1 income2 d ( pwine pwine1 winetax1 income2 d) test cwine1 regress cwine pwine cwine1 cwine2 income2 d ( pwine pwine1 pwine2 winetax1 test cwine1 cwine2
cwine2 income2 d)
regress cspirits lnpspirits cspirits1 income2 d ( lnpspirits pspirits1 spiritoustax1 income2 d) test cspirits1 regress cspirits lnpspirits cspirits1 cspirits2 income2 d ( lnpspirits pspirits1 pspirits2 spiritoustax1 spiritoustax2 income2 d) test cspirits1 cspirits2
**Model estimations in stata** *myopic models for beer* regress lncbeer pbeer lncbeer1 regress lncbeer pbeer lncbeer1 regress lncbeer pbeer lncbeer1 regress lncbeer pbeer lncbeer1
*myopic regress regress regress regress
models for wine* cwine cwine1 pwine cwine cwine1 pwine cwine cwine1 pwine cwine cwine1 pwine
income1 income1 income1 income1
income2 income2 income2 income2
d ( pbeer pbeer1 income1 d) d ( pbeer beertax1 income1 d) d ( pbeer beertax1 pbeer1 income1 d) d
d ( winetax1 pwine income2 d) d ( pwine1 pwine income2 d) d ( winetax1 pwine1 pwine income2 d) d
*myopic models for spirituous* regress cspirits cspirits1 lnpspirits regress cspirits cspirits1 lnpspirits regress cspirits cspirits1 lnpspirits regress cspirits cspirits1 lnpspirits
income2 income2 income2 income2
d ( d ( d ( d
spiritoustax1 lnpspirits income2 d) pspirits1 lnpspirits income2 d) spiritoustax1 pspirits1 lnpspirits income2 d)
*rational models for beer* regress lncbeer pbeer lncbeer1 lncbeer2 income1 d ( pbeer pbeer1 pbeer2 income1 d) regress lncbeer pbeer lncbeer1 lncbeer2 income1 d ( pbeer beertax1 beertax2 income1 d) regress lncbeer pbeer lncbeer1 lncbeer2 income1 d ( pbeer beertax1 pbeer1 beertax2 income1 d) regress lncbeer pbeer lncbeer1 lncbeer2 income1 d
*rational models for regress cwine cwine1 regress cwine cwine1 regress cwine cwine1 regress cwine cwine1
wine* cwine2 cwine2 cwine2 cwine2
pwine pwine pwine pwine
income2 income2 income2 income2
d ( winetax1 winetax2 pwine income2 d) d ( pwine1 pwine2 pwine income2 d) d ( winetax1 pwine1 winetax2 pwine2 pwine income2 d) d
*rational models for spirituous* regress cspirits cspirits1 cspirits2 lnpspirits income2 d ( spiritoustax1 spiritoustax2 lnpspirits income2 d) regress cspirits cspirits1 cspirits2 lnpspirits income2 d ( pspirits1 pspirits2 lnpspirits income2 d) regress cspirits cspirits1 cspirits2 lnpspirits income2 d ( spiritoustax1 pspirits1 spiritoustax2 pspirits2 lnpspirits income2 d)