Political Parties and Representation of the Poor in - Rooney Center

Political Parties and Representation of the Poor in - Rooney Center

Political Parties and Representation of the Poor in the American States Elizabeth Rigby George Washington University [email protected] Gerald C. Wright...

591KB Sizes 0 Downloads 6 Views

Political Parties and Representation of the Poor in the American States

Elizabeth Rigby George Washington University [email protected]

Gerald C. Wright Indiana University [email protected]

Draft: April 6, 2011

Acknowledgements: The authors are grateful to the Russell Sage Foundation and the National Science Foundation for their financial support, as well as to Project Vote Smart for providing us data from their NPAT survey of candidates. Earlier versions of this paper received critical feedback and helpful guidance from Ernesto Calvo and Eric Plutzer, as well as participants at the 2008 State Politics and Policy Conference and the Summer 2010 Russell Sage Foundation Working Group on the Politics of Inequality. Yet, only the authors are responsible for any errors or for the interpretations presented.

1

Abstract

This paper examines the responsiveness of state parties to the policy preferences of different income groups. Motivated by recent work concluding that low-income citizens are virtually ignored in the American policymaking process, we ask whether a similar bias shapes the policy positions adopted by political parties. Using newly-developed measures of state party positions, we test three hypotheses regarding whether low-income preferences get incorporated at this early, formative stage of the policymaking process—finding no evidence that they do, except when low-income preferences overlap with those of higher-income constituents with whom parties are more likely to align. This differential responsiveness was most pronounced for economic issues and for Democratic parties in states with high levels of income inequality; it was less evident for Republicans’ social policy platforms, which reflected the preferences of middle-income groups. We discuss the implications of these findings for representation in this era of growing economic inequality.

2

Rates of income inequality in the United States have increased dramatically in recent decades, prompting concerns that the concentration of income and wealth compromises our political system’s ability to uphold its norm of political equality (APSA 2004; Jacobs and Skocpol 2005; McCarty, Poole, and Rosenthal 2006). This worry—that some citizens’ power and resources may overwhelm others’ right to equality of political voice—is not a new concern. It was 50 years ago that Schattschneider (1960) wrote his often-cited critique that “the flaw in the pluralist heaven is that the heavenly chorus sings with a strong upper-class accent” (pg. 34).Fueling these concerns, recent studies have concluded that policymakers are not equally responsiveness to different income groups (Bartels 2008; Gilens 2005, 2011; Hacker and Pierson 2010; Jacobs and Page 2005). Together, these studies suggest that American democracy is falling far short of its ideal of political equality due to policymakers’ heightened attention to more affluent constituents and political elites at the expense of the poor. These quite pessimistic accounts of unequal representation stand in sharp contrast to more than 40 years of accumulated work on representation in U.S. politics that concludes that by-and-large the democratic process works quite well, with a wide array of empirical evidence finding that elected officials and the policies they enact comport well with the policy preferences of the public as a whole (e.g., Erikson, Wright, and McIver 1993; Fiorina 1974; Miller and Stokes 1963). This substantial normative disjunction between the rosy picture of a healthy democracy painted by past work and the worrisome results of recent research calls for studies that illuminate the process by which differential responsiveness occurs, the conditions in which representation is likely to be more or less uneven, and the role of our major political institutions in exacerbating or minimizing the importance of economic resources in determining whose interests and preferences are represented. That goal motivates this analysis into the

3

responsiveness of the political parties to the preferences of different income groups at the early stage in which parties must aggregate diverse preferences into alternative platforms on which to run for election. We zero in on the parties because of their centrality in the health of democratic systems (Schattschneider 1960) and look specifically for the inclusion of societies disadvantaged into the political process (Key 1949). Piven (2006) acknowledged this centrality by describing parties as the “agencies that actualize—or fail to actualize—the ideal of reciprocity between voters and state elites on which the democratic idea rests” (pg. 4). In fact, the normative hope is that strong parties serve to equalize responsiveness across citizens and act as one of the primary avenues for representation of those with fewer resources to organize on their own. Yet, we also know that parties are constrained by strategic efforts to maximize votes and so likely focus their appeals on particular segments of society. In this analysis we ask whether the preferences of lower income citizens are included in state parties’ policy agendas. This is a crucial preliminary stage in the policymaking process because if a group’s interests are not represented by at least one political party, they hold little chance of securing policy to their liking. Thus, differential responsiveness at this early stage can set limits on the level of policy responsiveness possible in subsequent stages of the policy process. State parties provide a fruitful testing ground for examining this process since, although adopting the shared national brands of Democrats and Republicans, state parties are decentralized institutions that vary in their policy positions across states with different political, socio-economic, and demographic characteristics (Erikson, Wright, and McIver 1993; Brown 1995).We capitalize on this variation by comparing the degree to which Democratic and Republican parties take stances in alignment with the preferences of their low-, middle-, and

4

high-income constituencies, as well as assessing their responsiveness to each income group when its preferences diverge from those of the other income groups in the state. Drawing on empirical analyses of income-based political behavior and parties’ electoral strategies, we develop and then test a set of hypotheses based on our expectations (explained below) that the parties would pursue income-group specific appeals that differ across economic and policy issues—with Democrats appealing to lower income groups on economic issues and higher income groups on social issues, and with Republicans taking the opposite approach. Instead, we find that both parties tend to align their policy programs with the preferences of higher income groups so that the preferences of lower income groups are only incorporated when low-income preferences overlap with the preferences of the high-income group. The primary deviation from this pattern is for Republican parties’ social policy stances, which align most closely with those of the middle-income group. Finally, we find some support for fears that rising economic inequality is linked with political inequality, at least for Democratic parties, whose tendency to align with the preferences of higher income constituents is most pronounced in states with high levels of economic inequality. In contrast, Democratic parties in low-inequality states exhibit even responsiveness across income groups. Background A well-established line of research identifies a strong correlation between voters’ preferences and the decisions made by their elected representatives (Ansolabehere, Snyder, and Stewart 2001; Canes-Wrone, Brady, and Cogan 2002; Erikson and Wright 1980, 2005; Erikson, Wright, and McIver 1993; Fiorina 1974; Miller and Stokes 1963). However, these efforts typically treat the “mass public” as a single entity, masking potential differences in responsiveness to different groups of constituents. In fact, when researchers de-compose this

5

relationship to test for different levels of responsiveness to the opinions of Democrats, Republicans, and Independents, they find greater responsiveness to constituents identified with the representative’s political party (Clinton 2006; Miller 1964; Stone 1982; Wright 1989). These differences in responsiveness are particularly important to uncover if they align along income lines: a situation that would undermine the legitimacy of our representative democracy, which rests on its ability to insulate political decision-making from other sources of power in society (Walzer 1983). In fact, research examining differential responsiveness across income groups has found that the opinions of the wealthy matter much more than the opinions of the poor—a troubling (for the norm of equality in political representation) pattern of differential responsiveness. For example, among U.S. Senators, Bartels (2008) found substantially greater responsiveness to high-income constituents than to those in the middle-income bracket and virtually no responsiveness to the opinions of low-income constituents. Similarly, Gilens (2005; forthcoming) found little correlation between the public opinion preferences of low-income voters and the policies Congress enacts. Additionally, Jacobs and Page (2005) found foreign policy to be more congruent with elite rather than with mass opinion. Few studies have examined explanations for why this differential responsiveness occurs, the exception being Bartels (2008), who found that individual-level factors such as differences in citizens’ political knowledge, voting, or communication with elected officials had little, if any, impact on responsiveness. To understand the process by which differential responsiveness occurs, we turn our attention to the role of political parties. Parties play a central role in representation by aggregating individual candidates around coherent political alternatives as well as by providing on-going monitoring and discipline to ensure that elected officials pursue the party’s agenda

6

once in office (Aldrich 1995; APSA 1950; Piven 2006; Schattschneider 1942; Snyder and Ting 2002). Thus, the ability of parties to link constituents of different income groups with actors in the policy process becomes a critical ingredient in equal political representation. In looking at parties’ responsiveness to diverse income groups, it is clear, by definition, that the rich are advantaged. Resource theory (Brady, Verba, and Schlozman 1995; Verba, Schlozman, and Brady 2006) argues that because the rich have an advantage on wealth, and therefore on information, time, money, and civic skills (Winters and Page 2009), parties are likely to be most responsive to those with more resources. Our question is whether this resource difference translates into a political advantage in which those with more resources have a louder voice in shaping the policy positions pursued by parties. And if so, we ask whether parties’ differential responsiveness begins this early in the policymaking process, at the point when parties serve to aggregate the diverse preferences of their constituents. Party Responsiveness to Different Income Groups Based on the work of Bartels (2008) and Gilens (2005; forthcoming) we might expect that both parties respond to preferences of the affluent, with less attention given to those with middle incomes, and none to those with relatively meager incomes. However, the parties are quite different in the policies they advocate, and there is a long line of research that details the differences in the coalitions of the parties. On the basis of what we know about the parties and how they have evolved, we are drawn to a quite different set of expectations. We develop hypotheses for two common and widely perceived dimensions of policy: economic policy and the other social issues. Differences in economic policy came to define the difference between the parties as they developed through the New Deal (e.g., Hacker and Pierson 2010; Ladd and Hadley 1975;

7

Stonecash 2000). The Republicans had already become identified as the party of big business and the affluent through the 1920s, and the Great Depression provided the impetus for Democratic victories in the early 1930s. Roosevelt’s policies, presumably designed both to combat the effects of the Great Depression and to build a new national Democratic party, succeeded in then defining the Democrats as the party of common man. Thus, income has been a consistent and important factor in the two parties’ coalitions, with the Republicans favoring the interests of the rich including free-market policies, while the Democrats have been more concerned with the situation of the worker and have been more amenable to redistributive and regulatory politics to serve their interests. Interestingly, even with substantial changes in the party system that accompany increased dimensions of policy conflict, the economic differences between the parties have only gotten stronger (Stonecash 2000). Ample evidence confirms that economic issues remain central to the political identity of the political parties, with each maintaining their long-standing affinities for economically less fortunate (Democrats) versus the well-off Republicans (Ansolabehere, Rodden, and Snyder 2006; Bartels 2006). Economic issues have been and continue to be centrally important to voters in presidential voting. Thus, if we assume that politicians and parties listen to the core elements of their electoral coalitions, we are led to the unsurprising hypothesis that the Republican party will be relatively more responsive to the economic policy preferences of the rich and the state Democratic parties will be more responsive to those with lower incomes. The finding that economic issue preferences are better predictors of individual voting decisions in presidential elections does not mean that social issues have not played an important role in recent decades. Indeed, the evidence seems clear that race and social issues have been key

8

in turning southern whites to the Republican party (Abramowitz 1994; Bartels 2006; Carmines and Stimson 1989; Ladd and Hadley 1975) although clearly demographic changes and the growth of income-based issues in the region also have had a role (Brewer and Stonecash 2001; Layman and Carmines 1997; Shafer and Johnston 2006). One does not need to go as far as Thomas Frank’s (2004) portrait of duped poor rural Americans misguidedly voting for the Republicans, who are their economic ruin, to hypothesize that the Republicans’ adoption of more conservative racial and social issue positions was in part motivated by the desire to gain more votes than they could secure with their long-standing conservative positions on economic issues. In fact, parties’ polarization on social issues cuts across income groups in ways quite different than seen for economic issues. In almost all states, higher income is associated with more liberal positions on social issues, but more conservative issue stances on economic concerns (Gelman 2008, 90; Rigby and Wright 2011). Hence, we expect that Republicans attend to the concerns of the less affluent on social issues to augment the support among those with higher incomes— whose support they secure based on their economic issue positions. In contrast, we expect that Democrats seek support (and donations) from higher income voters by being responsive to this group’s more liberal inclinations on social issues, a pattern consistent with voting in recent presidential elections (Ansolabehere, Rodden, and Snyder 2006; Bartels 2008; Gelman 2008). Our third hypothesis is motivated by concerns over the recent growth in income inequality (Neckerman 2004). Specifically, we ask: does the heightened concentration of income and wealth at the top undermine the goal of political equality? If so, it would lend credibility to concerns that the “New Gilded Age,” with its massive economic inequalities, weakens the quality of American democracy in terms of the ideal of equal representation (e.g., Bartels 2008; Dahl 2006) as well as overall societal well-being (Frank 2007; Wilkinson and Pickett 2010).

9

Making this case, Winters and Page (2009) argue that the concentration of wealth provides both the material interest, the resources, and the motivation to exert potent political influence that allows the very wealthy to dominate the political process. This influence may stem from direct lobbying, campaign contributions, or other participation by those with the greatest resources; yet, it can also occur indirectly because the wealthy are more likely to control large organizations that are actively lobbying on their behalf (Winters and Page 2009). Other researchers focus on the process of opinion formation, another avenue by which income inequality may shape politics. The gap in income and related life circumstances work against a sense of common fate and concern (Hero 1998), resulting in a society divided into “haves” and “have nots,” with little social or political capital bridging Americans (Alesina and Glaeser 2004). And others examine inequality in political behavior, in which income inequality serves to depress the political interest, political discussion, and electoral participation for all but those in the richest quintile across countries (Solt 2008). Regardless of the mechanism, there is empirical evidence that income inequality has increased along with party polarization (McCarty, Poole, and Rosenthal2006) and, more generally, that income inequality is a fundamental input into the macro-political system in the United States (Hacker and Pierson 2010; Kelly 2009). For these reasons, we ask whether greater income inequalities are accompanied by larger differences in responsiveness to the affluent versus the less fortunate. Specifically, we hypothesize that party responsiveness to the best-off in society increases with levels of income inequality. Data and Measurement To test these hypotheses, we capitalize on the variation in party positioning and incomebased political characteristics across the states. We combine new data on state parties’ ideological positioning with estimates of income-group opinion for 47 states, excluding

10

Nebraska because of its non-partisan legislature and both Alaska and Hawaii due to their exclusion from the survey data used to estimate mass opinion and behavior. Estimating State-Level Opinion across Income Groups To estimate opinion across income groups, we use the National Annenberg Election Survey (see Romer et al. 2006); its large sample of the 2000 presidential electorate consists of 58,373 respondents, including a substantial number from each of the 48 continental states. A modest drawback to the Annenberg survey data for our purposes is that respondents were interviewed across a series of rolling, cross-sectional surveys in which different sets of (somewhat overlapping) questions were asked of each cross-section. To generate measures of policy preferences across these surveys, we pooled the cross-sectional surveys for 2000 into one dataset and identified 13 common policy items in the survey—each asking for respondents’ preferences on policy issues across a range of policy areas: economic issues, health and social welfare issues, social/moral issues, energy and environment, campaign finance, and legal and ethical issues. Three of these 13 items were asked on all the surveys,1 while the other 10 were asked in approximately half of the surveys. As a result, we used respondents’ answers on the questions they were asked to estimate their responses to other policy items. We imputed these missing data using imputation by multiple chained equations (see Royston 2005), which maximized similarities in responses across items. As a result, these data are best suited to capturing a latent construct (e.g., social policy liberalism) rather than precise views on a particular issue (i.e., the death penalty or abortion). In fact, simulations designed to assess the accuracy of these imputations (described in online Appendix C) indicate that the scales created using imputed

1

The observed N for all items is presented in Online Appendix A.

11

values correlate very highly with measures based on full information (r = .97 for economic issues and .96 for social issues). To generate the opinion scales used in these analyses, we factor analyzed two subsets of items.2 The first was a set of seven items most directly related to economic policy issues, such as tax policy, health care, and social welfare programs. Then, we factor analyzed six items tapping social policy issues such as abortion, the death penalty, gay rights, and school prayer. Additional information on these items and their factor loadings are presented in Appendix A. Most critical to this analysis, both factor models explained the vast majority (> 97 percent) of the variance among items, with eigenvalues well above one (and eigenvalues for both of the second factors below .50).3 Both scales are coded so that higher values indicate more liberal policy positions. [Figure 1 about here] To reliably estimate the policy preferences of different income groups across states, we then needed to collapse the nine-category income variable used by the Annenberg survey into fewer categories. Based on the distribution of cases across states, we divided the sample into strict income thirds within each state.4 Equally sized groups were an important component of our modeling strategy because we wanted to test a proportional notion of equality of representation in which each income group is influential in proportion to its size, rather than resources. This measurement strategy also allowed us to take a simple average of the policy preferences of each 2

Factor analysis is particularly efficient for large datasets such as these, especially due to our use of ordinal and categorical measures that would need to be dichotomized for use with NOMIATE or Heckman-Snyder approaches (although we note that the factor scores are almost identical to an alternative estimate from NOMINATE, r = .98). 3 These strong loadings likely represent a good deal of coherence among respondents’ economic or social policy views, though they are also biased upward due to our missing data imputation strategy that maximized the similarity among the items and also because of our focus on a small number of items for each sub-scale. 4 We adopted a method used by Bhatti and Erikson (2011) to estimate cut-points across the categorical income measure employed in the survey. Specifically, we recalculated the economic and social policy preference measures by taking state-specific averages for each of the nine income categories. The bottom 33 percent of respondents in each state were used to capture the preferences of the low-income group, the top 33 percent of income earners were used to capture the preference of the high-income group, with the middle-income group consisting of the respondents from 34th to 66th percentile in each state. We then sorted the data by income category and divided the sample into three relative thirds within each state.

12

income third in each state rather than requiring weights to correct for varying sizes of each group. These aggregated preference measures (N=47 states) are presented in Figure 1, which displays the distribution of preferences of low-, middle-, and high-income constituents across the states. Consistent with previous work, we find different distributions of preferences for the economic (top panel) and social (bottom panel) policy preferences even though both measures are standardized on the same scale (M=0, sd=1) at the individual-level. In particular, we see a larger income gap in economic policy preferences—with the low-income group most liberal (M=.18, sd=.13) on these issues, followed by the middle-income group (M=-.02, sd=.13), and then the relatively more conservative higher income group (M=-.21, sd=.13). Yet, for social issues, we find the opposite pattern, with the high-income group being more liberal (M=.04, sd=.27), followed by the middle-income (M=-.04, sd=.23), and then the more socially conservative low-income group (M=-.09, sd=.20). Further, the income gap is smaller for social issues than for economic issues; however, the variation across states is larger for social issues than economic issues. Estimating State Party Ideology We define state party ideology as the ideology that emerges from the collective issue stances of the states’ Democratic and Republican candidates. This approach captures the policy positions candidates present to the electorate during the campaigns, which likely carry more force than the seldom-read official platforms. Further, by using electoral positions—rather than legislative votes—we capture party positions on the same scale across states and chambers, while avoiding state differences in patterns of party control, agenda control, and legislative procedures that could distort ideology measured through roll call votes across 94 chambers.

13

We pooled data gathered by Project Vote Smart (PVS), a non-partisan voter education service that administers the National Political Awareness Test (NPATs) to candidates for federal and state office.5 We used responses from 18,467 surveys completed by Democratic or Republican candidates for the state legislature, Congress, or governor.6 By pooling across years, we achieved adequate state samples, which average 373 (sd=182) and range from 115 in Louisiana to 968 in New Hampshire. Yet, since PVS individualizes each state’s survey for each election cycle, we needed to extract the overlapping questions across the 255 distinct surveys fielded from 1996 to 2005.7 We identified 87 items asked to at least half (> 9,000) of the Democratic and Republican party candidates during this time frame. On average, the variables were missing approximately a quarter of responses, missing items that were imputed using a parallel process as that used for the public opinion data described above (multiple imputations by chained equations; see Royston 2004 and Appendix C for more information). As with the public opinion measures, a set of simulations found this imputation approach to yield measures that are highly correlated with scales developed from full information (r=.99 for both issue scales). To generate measures of candidates’ economic and social policy preferences, we followed the same procedure as undertaken for the public opinion data and factor analyzed subsets of items capturing economic policy issues (25 items related to redistributive tax and spending policies and the social safety net) and social policy issues (20 items related to abortion, gun control, civil rights, school prayer). The specific items are listed in Appendix B, along with

5

Beginning in 2008, the NPAT was renamed the Political Courage Test. PVS conducts these surveys as a public service to provide the voters unbiased information about candidates’ policy positions and provides the responses on their website (http://www.vote-smart.org) and via a toll-free phone line. 6 This sample represents 13,628 unique candidates, as well as an additional 4,839 cases of candidates completing additional surveys in later election cycles, which we use to help capture changes in candidates’ issue positions over the decade. The findings are not sensitive to alternative specifications in which responses by candidates for Congress or Governor are weighted more highly, or when these cases are excluded from the sample. 7 For each of the five election waves, PVS fielded a unique survey in every state as well as a Congressional version given to all federal candidates across states.

14

information on the raw (versus imputed) number of cases and factor loadings. Beginning with the economic policy items, we factor analyzed the 25 items and extracted the first factor, which explained 78 percent of the variance (eigenvalue = 7.03). Predicted values were generated for each candidate-year and rescaled so that higher values indicated greater liberalism on economic policy issues. Candidates varied from -3.42 to 1.92 on this scale; not surprisingly, Democrats had higher scores (M=.65, sd=.61) than did Republicans (M=-.66, sd=.81). We then factor analyzed the twenty social policy items and extracted the first factor, which explained 65 percent of the variance (eigenvalue=4.53), and coded the scores so that higher values indicated more liberal positions on social policy issues. The individual scores varied from -2.01 to 1.89, with Democratic candidates more liberal (M=.55, sd=.77) than were Republicans (M=-.57, sd=.74). From these individual-level scores, we estimated each state party’s liberalism by taking the mean response of candidates for each party in the state. However, we also generated weighted scores that adjusted individual-level responses for patterns of non-response in the Project Vote Smart data. Their response rate of 32 percent (sd= 7 percent, ranging from 20 percent in Virginia to 47 percent in South Dakota) is about average for an elite survey such as this but nevertheless has the potential to introduce selection bias. We know of no theoretical reasons to expect that the NPAT selection process would yield a systematic re-ordering of the states in terms of their actual party ideologies. However, to attempt to account for such selection, we generated individual-level post-stratification weights that adjust the state-specific estimates for observed patterns of non-response. Based on the observed factors we included (office seeking, district safety, incumbency, election year), selection bias in these data appears to be quite modest.8In fact, the state-specific R2s explaining participation in the NPATs averaged only

8

These state-specific weights were calculated as the inverse of the predicted probability of completing the NPAT, estimated from a logistic regression in which the decision to respond was regressed on the following predictors:

15

.08 (sd=.05). Therefore, it is not surprising that correcting for possible selection bias at the individual-level had little effect on the aggregated scores for the state parties, which correlate highly (r=.96 and .98) with the simple state-party means. Empirical Strategy and Results Average Responsiveness Before moving into our primary aim—testing for differential responsiveness of state parties to constituents of different income groups—we first examine the positioning of state parties in terms of the average opinion in each state (the simple mean of all respondents in the state). These basic relationships are presented in Figure 2. The top scatterplot shows the relationship between parties’ economic policy liberalism and the average public opinion on economic issues. There is little relationship between the two parties on these issues (r=.24), which is primarily due to the near absence of a relationship between Democratic party liberalism and public opinion (r=.11). In contrast, Republican parties are more liberal on economic issues when they represent a public more liberal on economic issues (r=.48). The bottom panel presents the same relationships for our measures of social policy liberalism. Here we see a positive relationship between public preferences and both Democratic (r=.83) and Republican (.65) parties’ social policy liberalism. The two parties are correlated at .60 as well. [Figure 2 about here] Differential Responsiveness Our modeling approach combined two distinct notions of responsiveness as commonly operationalized in the literature. First, we considered alignment, which captures the bivariate

office seeking (governor, U.S. House, U.S. Senate, state house, state senate), electoral safety of district (district ideology folded so that high values represent more liberal districts for Democrats and more conservative districts for Republicans), incumbency, year dummies, and interactions between incumbency and safety as well as incumbency and each year dummy.

16

relationship between each group’s opinion and each party’s position. This is the more-basic notion of responsiveness adopted by Soroka and Wlezien (2008) and others who tend to find less evidence of differential responsiveness across income groups. In contrast, we also operationalized responsiveness in a second way, which we call influence, represented by the partial association between each income group’s opinion and each party program, controlling for the preferences of the other two income groups in the state. It is this second modeling approach that was used by Bartels (2008) and most others who have identified unequal responsiveness to various groups. Rather than adopting only one of these two common modeling strategies, we follow the lead of Gilens (2005; forthcoming)9 and tested for both notions of responsiveness. This was more than a methodological choice, but one designed to distinguish between two levels of responsiveness: alignment, which only considers whether a group’s preferences are included in party positions, and influence, which raises the bar, asking whether the group’s preferences are included even when these preferences diverge from those of the other income groups in the state. Teasing apart these two notions of responsiveness is particularly important for contextualizing differential responsiveness within our current public opinion landscape in which the preference gap between the rich and poor varies in size across specific issues, as well as among states (Gelman 2008; Gilens 2009; Rigby and Wright 2011; Soroka and Wlezien 2008). Rather than attributing parties’ responsiveness to overlapping preferences as evidence of equality in responsiveness (as did Soroka and Wlezien 2008) or excluding this alignment from an assessment of differential responsiveness (as did Bartels 2008), we wish to explore and identify these alternative patterns of responsiveness to the rich versus the poor. [Table 1 about here]

9

Gilens (2005) acknowledged the same reality by restricting his analysis to only those policy issues on which the poor and rich disagree. We instead estimate bivariate and then partial associations.

17

We tested our hypotheses in reduced-form, cross-sectional OLS models that regressed the policy liberalism of each party (separately) on the preferences of each income-third in the state. Table 1 presents the results of these OLS models for both economic policy positions (top) and social policy positions (bottom). Beginning with Democratic parties, we find little alignment between party positions and the opinion of either low-income or middle-income citizens in the state. Instead, the economic platforms of Democratic parties are only aligned with the preferences of those in the top-income third in the state (b=.49, se=.19). Not surprisingly, this result persists in Model 4 (b=.64, se=.25), which tests for the independent influence of each group’s preferences on the issues in which income groups disagree. For Republican parties’ economic positions, we see significant coefficients for each of the alignment models, suggesting some positive correspondence between differences in the income groups’ preferences in the states and those of the Republican parties’ economic positions. This can be seen in the alignment coefficients for the low-income group (b=.60, se=.27), middle-income group (b=.70, se=.25), and high-income group (b=.96, se=.24). Yet, as shown in Model 4, when we test for the independent influence of each group’s opinion over and above the preferences of the other two groups, we find the only significant (at p<.05) association to be for the preferences of the high-income group (b=.91, se=.32). This suggests that the preferences of the lower income groups have no independent impact of the positions of the Republican parties on this issue. Clearly across the states, our hypothesis that the Democratic party would be more sensitive to the preferences of the poor is wrong. Both parties heed the concerns of the highest income group on economic policy. For social policy issues, both parties exhibit alignment with the preferences of each income group when considered alone. However, in Model 4, which tests for the independent influence of each income group, we see patterns of differential responsiveness to

18

social policy preferences for both parties. When the income groups disagree on social issues, Democratic parties are most aligned with the distinct preferences of the high-income group (b=.93, se=.25), while Republican parties are most aligned with the distinct preferences of the middle-income group (b=1.07, se=.34). [Figure 3 about here] These findings are brought together in Figure 3, which presents the predicted first difference for a one-unit change in each group’s opinion. It illustrates our general conclusion that although parties align their policy preferences with each of the three income groups, it is typically the highest income group that experiences any independent influence, or association over and above the preferences of the other two groups. We identify two exceptions to this general pattern. First, Democratic parties’ economic platforms are not at all aligned with the economic policy preferences of low- or middle-income groups. Second, Republican parties’ social policy platforms exhibit the most responsiveness (in terms of independent influence) to the middleincome group. In terms of representation of the poor, these findings indicate that the preferences of low-income Americans are only incorporated when they overlap with the preferences of the higher income groups to whom parties are more likely to attend. Further, on Democratic parties’ economic stances—presumably those of greatest material consequence to low-income citizens— the preferences of the low-income group have no independent effect at all. This is not to say that the Democratic party ignores the interests of the poor, but our data do strongly indicate that it is the definition of these interests by higher income groups that shapes party policy stances. Role of Income Inequality We now turn to examining our third hypothesis: that income inequality would serve to further skew responsiveness toward those at the top with more concentrated wealth. To really

19

test this hypothesis, we would need to model these relationships over time, which we are unable to do due to the absence of longitudinal measures of party positions. So instead, to gain some insight into this question, we capitalize on the variation in income inequality across states to ask whether responsiveness to the wealthy is greater in states with more skewed income distributions. We use a cross-sectional measure of income inequality based on market-income, pre-tax, and transfer family income drawn from the 1997–2005 March Current Population Surveys conducted by the U.S. Census Bureau. This takes the form of a gini coefficient (M= .50, sd= .03) that captures how the entire distribution of income deviates from equality. Higher values indicate greater income inequality. [Table 2 & Figure 4 about here] Table 2 presents the estimated models, which add the measure of income inequality along with three interaction terms (income inequality * each group’s opinion). For economic policy issues, the responsiveness of Democratic parties did vary by the level of income inequality in the states, with greater income inequality associated with even more of a skew in responsiveness toward the upper income third. But, the responsiveness of Republican parties did not vary by income inequality for economic issues nor for social issues. Republican parties remain most responsive to upper income preferences on economic issues and middle-income preferences on social issues, regardless of levels of economic inequality in the states. To illustrate the role of income inequality in moderating the responsiveness of Democratic parties, we used the models in Table 2 to estimate Democratic party liberalism on both economic and social policy issues in a state with low (one standard deviation below the mean) versus high (one standard deviation above) income inequality. These predicted values are presented in Figure 4. The pattern there indicates there may well be a basis for concerns that

20

heightened economic inequality exacerbates representational inequalities among the state Democratic parties. Importantly, in states with more equal distributions of income, we find no evidence of differential responsiveness across income groups. However, in states with higher levels of income inequality, only the preferences of the high-income groups matter—reflecting the same pattern for economic policy that holds across the board for Republican state parties. Further, variation in income inequality did not alter our earlier finding that on neither type of issue do parties attend to the distinct preferences of the lowest income third. Sensitivity Analysis: Does Participation Explain Responsiveness? Before concluding, we consider the possibility that parties’ differential responsiveness to the higher income groups may simply be a function of their heightened attention to voters, rather than non-voters in their state. Bartels (2008) also considers this possibility and tests for voting, as well as other forms of political participation (i.e., contributions) as potential explanations for the heightened responsiveness to the rich among U.S. Senators that he identified. However, he finds no evidence that this is the case. Despite this finding, we wish to examine this possibility using our data based on the extensive evidence that voters are not a random subset of the population but instead are wealthier and better educated than non-voters (Abramson, Aldrich, and Rohde 2007; Leighley 1995; Leighley and Nagler 1992; Rosenstone and Hansen 1993). Additionally, of particular relevance to our study, this income-based inequality in political participation has been linked to important consequences, such as the generosity of welfare policy (Allen and Campbell 1994; Avery and Peffley 2005; Campbell 2003; Fellowes and Rowe 2004; Hill and Leighley 1994; Husted and Kenney 1997; Ringquist, Hill, Leighley, and Hinton-Anderson 1997), which may begin in this early stage when parties adopt their policy priorities.

21

To conduct this sensitivity test, we use data from the Current Population Survey, which conducts a Voter Supplement in November every other year (for more information, see U.S. Census Bureau 2002). We drew from three surveys: 1996 (N=136,279), 2000 (N=134,723), and 2004 (N=156,519). For each dataset, we used the same method as described earlier to divide the sample into strict income thirds within each state. We derived state-level estimates for the percentage of each third that reported voting in the most recent presidential election and then averaged these values across the three datasets to estimate that 50 percent of the lower income, 64 percent of the middle-income third, and 75 percent of the high-income third reported voting in the most recent election. These estimates seem a little high, which is not surprising since we know that CPS respondents tend to over-report voting (Bernstein, Chada, and Montjoy 2003). Yet, we see no reason to expect any difference in over-reporting between wealthy and poor respondents in a way that would compromise our estimates. We then calculated the contribution of each income group to their total state electorate, determining that on average the low-income third contributed 26 percent of state voters (sd=2, range= 21 to 29 percent), the middle third contributed 34 percent of voters (sd=1, range= 31 to 35 percent), and the high-income third contributed 40 percent of voters (sd=2, range= 37 to 45percent). Each income group’s opinion measure was weighted by the contribution of that group to the state electorate. Additionally, we included two other variables to allow for groupspecific intercepts (which were needed due to the uneven weights across groups, see Bhatti and Erikson2011). Online Appendix D presents the results of these models, which are consistent with the earlier results using unweighted data, leading us to conclude that income bias in voting does not explain the differential responsiveness we identify.

22

Conclusion Motivated by recent research that has found skewed responsiveness among the policy outcomes of our political process, this paper asked whether a similar bias shapes the policy positions taken by political parties much earlier in the policymaking process. Since parties serve to aggregate diverse interests and function as mechanisms linking public opinion and policy outputs, any income-based bias in this process would raise both empirical and normative concerns, as well as identify a likely source of political inequality in the later, and more consequential, policy formation and outcome stage. To assess the equality of this process, we distinguished between two forms of responsiveness: alignment and influence, with alignment simply capturing whether parties align with the preferences of each income group (considered alone) and influence requiring a higher bar in which parties align with the unique preference of each income group controlling for the preferences of the other two groups. In general, we find alignment between the preferences of each income group and the positions adopted by both parties. This is encouraging in light of the take-away from much recent research concluding that the interests of the poor are completely ignored in our policymaking process. Instead, we find a healthy correlation between the preferences of the poor and parties’ positions. The exception is for Democratic parties on economic issues, in which party positions align only with those of the highest income-third. Clearly, this is not a minor caveat since it was Democratic parties that were expected to carry the interests of the poor, and economic issues are likely to have the greatest material impact on the well-being of the most disadvantaged citizens. Even more troubling from a normative ideal of equal representation is our examination of the independent influence of each income group’s preferences. We find both parties to be most responsive to the preferences of middle- and high-income constituents, with the lowest income

23

groups having no discernable independent influence on what either of the parties stand for on either set of issues. As a result, the representation of low-income citizens only occurs when their preferences happen to concur with the preferences of their economic betters. The lack of responsiveness to the poor does not necessarily mean that both parties cater to the narrow self-interests of high-income voters and contributors. Indeed, when we examine differences between the state parties, it is clear that the Democrats are a good deal more liberal and many of their policies advocated favor redistribution and regulations that would improve the material circumstances of those with lower incomes. So while the Democratic parties in the states do tend to stand for economic policies benefiting the “have nots,” the extent of that stance tends to be sharply circumscribed by preferences of the affluent. Does this make any difference? Normatively, it is less than satisfying to find that the least fortunate in society have no independent voice in the formulation of party platforms. Materially, while there is overlap across the states in the preferences of income groups, it is also clear (see Figure 1) that the poor (and even the middle-income group) prefer more liberal economic policies than do those in the upper income third. Therefore, we suspect that if the preferences of each income group were accorded equal weight, parties’ stances on economic issues would shift to the left. (And accordingly, although to a smaller degree, the Democratic party’s stance on social issues would likely shift somewhat to the right.) Finally, we note the variation in these patterns across states with higher versus lower levels of income inequality. Although drawn from a cross-sectional model that cannot establish whether rising income inequality has led to a greater bias in responsiveness, we can conclude that responsiveness to those in the high income group is greater in states with more income inequality. In those states both parties are aligned with the high-income group on economic

24

issues and Democrats also reflect this group’s preferences on social issues. A different pattern, across parties and issues, was found in states in which income inequality is lower. In these states, we found both parties aligned with the middle-income group on social issues; Republicans remained responsive to the upper income group on economic issues, and the economic positions adopted by Democratic parties exhibited no bias in favor of one income group or the other. This low-inequality pattern of responsiveness is consistent with a party constituency notion of representation in which the two parties draw from and respond to distinct constituencies on economic issues, while using social issues to make appeals to a broader constituency. Adopting this view, it is not surprising that the Republican party pays less attention to the poor on economic issues: to do otherwise would violate their general stance since before the New Deal. More surprising, is the degree to which Democratic parties’ appeals to lower income constituencies depends on the level of income inequality in the state. In states with low inequality, Democratic parties may indeed be serving to equalize responsiveness. Under conditions of higher income inequality, Democratic parties may still draw votes from lower income groups, but seem to respond only to higher income constituencies when taking policy stances—serving to exacerbate any income bias in representation at this early, formative stage of the policymaking process.

25

References Abramowitz, Alan I. 1994. "Issue Evolution Reconsidered: Racial Attitudes and Partisanship in the U.S. Electorate." American Journal of Political Science 38 (1):1–24. Abramson, Paul R., John H. Aldrich, and David W. Rohde. 2007. Change and Continuity in the 2004 and 2006 Elections. Washington, DC: CQ Press. Aldrich, John A. 1995. Why Parties? The Origin and Transformation of Political Parties in America. Chicago: University of Chicago Press. Alesina, Alberto, and Edward L. Glaeser. 2004. Fighting Poverty in the US and Europe: A World of Difference. Oxford: Oxford University Press. Allen, Michael Patrick, and John L. Campbell. 1994. “State Revenue Extraction from Different Income Groups: Variations in Tax Progressivity in the United States, 1916 to 1986.” American Sociological Review 59:16996. Ansolabehere, Stephen, Jonathan Rodden, and James M. Snyder, Jr. 2006. "Purple America." Journal of Economic Perspectives 20 (2):97–118. Ansolabehere, Stephen, James M. Snyder, and Charles Stewart. 2001. "Candidate positioning in US house elections." American Journal of Political Science 45 (1):136–59. APSA. 1950. "Toward a More Responsible Two-Party System: A Report of the Committee on Political Parties of the American Political Science Association." American Political Science Review 44:1–99. APSA. 2004. American Democracy in an Era of Inequality. Task Force on Inequality. Washington, DC: APSA. Avery, James M., and Mark Peffley. 2005. “Voter Registration Requirements, Voter Turnout, and Welfare Eligibility Policy: Class Bias Matters.” State Politics and Policy Quarterly 5: 47–67. Bartels, Larry M. 2006. "What's the Matter with What's the Matter with Kansas?" Quarterly Journal of Political Science 1 (2):201–26. Bartels, Larry M. 2008. Unequal Democracy: The Political Economy of the New Gilded Age. Princeton, NJ: Princeton University Press. Bernstein, Robert A., Anita Chada, and Robert Montjoy. 2003. “Cross-State Bias in Voting and Registration Over-reporting in the Current Population Surveys.” State Politics and Policy Quarterly 3(4): 367–68.

26

Bhatti, Yosef, and Robert S. Erikson. 2011. “How Poorly are the Poor Represented?” in Who is Represented, eds.Peter Enns and Christopher Wlezien. New York: Russell Sage Foundation. Brady, Henry E., Sidney Verba, and Kay Lehman Schlozman. 1995. “Beyond SES: A Resource Model of Political Participation.”American Political Science Review 89(2):271–94. Brewer, Mark D., and Jeffrey M. Stonecash. 2001. "Class, Race Issues, and Declining White Support for the Democratic Party in the South." Political Behavior 23 (2):131–55. Brown, Robert D. 1995. Party Cleavages and Welfare Effort in the American States. American Political Science Review 89 (1): 23-33. Campbell, Andrea. 2003. How Policies Make Citizens: Senior Political Activism and the American Welfare State. Princeton, NJ: Princeton University Press. Canes-Wrone, Brandice, David W. Brady, and John F. Cogan. 2002. "Out of Step, Out of Office: Electoral Accountability and House Members' Voting." American Political Science Review 96 (1):127–40. Carmines, Edward G., and James A. Stimson. 1989. Issue Evolution: Race and the Transformation of American Politics. Princeton, N.J.: Princeton University Press. Clinton, Joshua D. 2006. “Representation in Congress: Constituents and Roll Calls in the 106th House.”The Journal of Politics 68(2):397–409. Dahl, Robert A. 2006. On Political Equality. New Haven, CT: Yale University Press. Erikson, Robert S., and Gerald C. Wright. 1980. "Policy Representation of Constituency Interests." Political Behavior 2:91–106. Erikson, Robert S., and Gerald C. Wright. 2005. "Voters, Candidates, and Issues in Congressional Elections." In Congress Reconsidered, eds. Lawrence C. Dodd and Bruce I. Oppenheimer. Washington, D.C.: CQ Press, 134–77. Erikson, Robert S., Gerald C. Wright, and John P. McIver. 1993. Statehouse Democracy: Public Opinion and Policy in the American States. New York: Cambridge University Press. Fellowes, Matthew C., and Gretchen Rowe. 2004. “Politics and the New American Welfare States.”American Journal of Political Science 48(2):362–73. Fiorina, Morris P. 1974. Representatives, Roll Calls, and Constituencies. Lexington, MA: Lexington Books. Frank, Robert H. 2007. Falling Behind: How Rising Inequality Harms The Middle Class. Berkeley: University of California Press. 27

Frank, Thomas. 2004. What's the Matter with Kansas? How Conservatives Won the Heart of America. New York: Metropolitan Books. Gelman, Andrew. 2008. Red State, Blue State, Rich State, Poor State: Why Americans Vote the Way They Do. Princeton: Princeton University Press. Gilens, Martin. 2005. "Inequality and Democratic Responsiveness." Public Opinion Quarterly 69:778–96. Gilens, Martin. 2009. “Preference Gaps and Inequality in Representation.” PS: Political Science and Politics 70(2): 335–41. Gilens, Martin. forthcoming. Paying the Piper: Economic Inequality and Democratic Responsiveness in the United States. New York: Russell Sage Foundation. Hacker, Jacob S., and Paul Pierson. 2010. Winner-take-all Politics: How Washington Made the Rich Richer—and Turned its Back on the Middle Class. New York: Simon & Schustser. Hero, Rodney E. 1998. Faces of Inequality: Social Diversity in American Politics. New York: Oxford University Press. Hill, Kim Quaile, and Jan E. Leighley. 1994. "Mobilizing Institutions and Class Representation in U.S. State Electorates." Political Research Quarterly 47:137–50. Hill, Kim Quaile, Jan E. Leighley, and Angela Hinton-Andersson. 1995. "Lower-Class Mobilization and Policy Linkages in the United States." American Journal of Political Science 39: 75–86. Husted, Thomas A., and Lawrence W. Kenney. 1997. "The Effect of the Expansion of the Voting Franchise on the Size of Government." Journal of Political Economy 105:54–82. Jacobs, Lawrence R., and Benjamin I. Page. 2005. "Who Influences U.S. Foreign Policy?" American Political Science Review 99:107–24. Jacobs, Lawrence R., and Theda Skocpol, eds. 2005. Inequality and American Democracy: What We Know and What We Need to Learn. New York: Russell Sage Foundation. Kelly, Nathan. 2009. The Politics of Income Inequality in the United States. New York: Cambridge University Press. Key, Jr., V.O. 1949. Southern Politics in States and Nation.New York: Knopf. Ladd, Everett Carl, and Charles D. Hadley. 1975. Transformations of the American Party System: Political Coalitions from the New Deal to the 1970s. New York: Norton.

28

Layman, Geoffrey C., and Edward G. Carmines. 1997. "Cultural Conflict in American Politics: Religious Traditionalism, Postmaterialism, and U.S. Political Behavior." The Journal of Politics 59:751–77. Leighley, Jan E. 1995. “Attitudes, Opportunities and Incentives: A Review Essay on Political Participation.” Political Research Quarterly 48:181–209. Leighley, Jan E., and Jonathan Nagler. 1992. “Individual and Systemic Influences on Turnout – Who Votes, 1984.” Journal of Politics 54:718–40. McCarty, Nolan M., Keith T. Poole, and Howard Rosenthal. 2006. Polarized America: The Dance of Ideology and Unequal Riches. Cambridge, MA: MIT Press. Miller, Warren E. 1964. "Majority Rule and the Representative System of Government." In Cleavages, Ideologies and Party Systems: Contributions to Comparative Political Sociology, eds. Erik Allardt and Yrjo Littunen. Helsinki: Transaction of the Westermarck Society. Miller, Warren E., and Donald E. Stokes. 1963. "Constituency Influence in Congress." American Political Science Review 57:45–56. Neckerman, Katheryn, ed. 2004. Social Inequality. New York: Russell Sage Foundation. Piven, Frances Fox. 2006. Challenging Authority: How Ordinary People Change America. Lanham, MD: Rowman & Littlefield Publishers. Rigby, Elizabeth, and Gerald C. Wright. 2011. "Whose Statehouse Democracy? State Policy Responsiveness to Rich vs. Poor Constituents in Rich vs. Poor States." In Peter Enns and Christopher Wlezien (Eds.) Who Is Represented? New York: Russell Sage Foundation. Ringquist, Evan, Kim Hill, Jan Leighley, and Angela Hinton-Anderson. 1997. “Lower-Class Mobilization and Policy Linkage in the U.S. States: A Correction.” American Journal of Political Science 41(1):339–44. Romer, Daniel et al. 2006. Capturing Campaign Dynamics 2000 & 2004.Philadelphia: University of Pennsylvania Press. Rosenstone, Steven J., and John Mark Hansen. 1993. Mobilization, Participation, and Democracy in America. New York: Macmillan. Royston, Patrick. 2005. "Multiple Imputation of Missing Values." Stata Journal4:227– 41. Shafer, Byron, and Richard Johnston. 2006. The End of Southern Exceptionalism: Class, Race, and Partisan Change in the Postwar South. Cambridge, MA: Harvard University Press. Schattschneider, E. E. 1942. Party Government. New York: Rinehart. 29

Schattschneider, E.E. 1960. The Semi-Sovereign People. New York: Holt, Rinehart and Winston. Snyder, James M., and Michael M. Ting. 2002. “An Informational Rationale for Political Parties.” American Journal of Political Science 46: 90-110. Solt, Frederick. 2008. “Economic Inequality and Democratic Political Engagement.” American Journal of Political Science 52 (1): 48-60. Soroka, Stuart N., and Christopher Wlezien. 2008. “On the Limits to Inequality in Representation.” PS: Political Science and Politics 69(4): 1032–46. Stone, Walter J. 1982. "Electoral Change and Policy Representation in Congress: Domestic Welfare Issues from 1956–1972." British Journal of Political Science 12:95–115. Stonecash, Jeffrey M. 2000. Class and Party in American Politics. Boulder, CO: Westview Press. Ura, Joseph Daniel, and Christopher R. Ellis. 2008. “Income, Preferences, and the Dynamics of Policy Responsiveness.” PS: Political Science & Politics 41(4): 785–94. U.S. Census Bureau. 2002. Voting and Registration in the Election of November 2000. Report 1902–2002. Available at: http://www.census.gov/prod/2002pubs/p20-542.pdf Verba, Sidney, Kay Lehman Schlozman, and Henry Brady. 2006. Voice and Equality: Civic Voluntarism in American Politics. Cambridge, MA: Harvard University Press. Walzer, Michael. 1983. Spheres of Justice. New York: Basic Books. Wilkinson, Richard G., and Kate Pickett. 2010. The Spirit Level: Why Greater Equality Makes Societies Stronger. New York: Bloomsbury Press. Winters, Jeffrey A., and Benjamin I. Page. 2009. “Oligarchy in the United States?” Perspectives on Politics 7(4): 731–51. Wright, Gerald C. 1989. "Policy Voting in the U.S. Senate: Who Is Represented?" Legislative Studies Quarterly 14:465–86.

30

Table 1. Parties’ Responsiveness to Policy Preferences of Each Income Group

(1)

Democratic Parties (2) (3)

(4)

(1)

-0.31 (0.29) -0.09 (0.33) 0.64 * (0.25) 0.83 * (0.07)

0.60 * (0.27)

Republican Partie s (2) (3) (4)

Economic Issues Low-Income

-0.18 (0.21)

Middle-Income High-Income Constant R-squared

0.07 (0.20) 0.49 * (0.19) 0.68 * 0.64 * 0.74 * (0.05) (0.03) (0.05) 0.02

0.00

0.12

0.40 (0.37) 0.70 * -0.12 (0.25) (0.43) 0.96 * 0.91 * (0.24) (0.32) -0.72 * -0.60 * -0.41 * -0.50 * (0.06) (0.03) (0.06) (0.09)

0.19

0.10

-0.27 (0.32) 0.59 + (0.31) 0.93 * (0.25) 0.43 * (0.04)

0.86 * (0.19)

0.15

0.26

0.29

Social Issue s Low-Income

1.36 * (0.20)

Middle-Income High-Income Constant R-squared

1.33 * (0.15) 1.20 * (0.12) 0.59 * 0.52 * 0.42 * (0.04) (0.03) (0.03) 0.50

0.64

0.70

0.73

-0.19 (0.35) 0.91 * 1.07 * (0.14) (0.34) 0.68 * -0.02 (0.14) (0.28) -0.52 * -0.55 * -0.62 * -0.56 * (0.04) (0.03) (0.04) (0.05) 0.31

0.48

0.35

0.48

Note: N=47, excludes AK, HI, NE. Coefficients from OLS regression models in which measures of party policy positions are regressed on the mean opinion of each group. Models 1-3 examine alignment: the bivariate relationship between each group’s opinion and the party position (regardless of preferences of other two groups. Model 4 tests for influence: the partial association after controlling for the opinion of the other two groups. Standard errors are presented in parentheses beneath each coefficient. *=p<.05, +=p<.10.

31

Table 2. Differential Responsiveness, by State Income Inequality Economic Issues Democratic Parties

(1) Low-Income Mid-Income High-Income Income Inequality Low X Ineq. Mid X Ineq. High X Ineq. Constant

R-squared

(2)

-0.30 (0.29) -0.01 (0.35) 0.62 * (0.25) -0.85 (0.99)

0.54 (5.25) 7.53 (5.41) -10.51 * (4.14) 4.55 (2.97) -1.94 (10.67) -15.03 (10.76) 22.45 * (8.32) 1.25 * -1.40 (0.49) (1.45) 0.20

0.34

Social Issue s

Republican Parties

(1)

(2)

0.38 (0.37) -0.29 (0.44) 0.96 * (0.32) 1.64 (1.27)

-6.87 (7.11) -1.91 (7.34) 1.53 (5.61) -1.94 (4.03) 14.96 (14.96) 3.11 (14.58) -1.10 (11.28) -1.30 * 0.42 (0.63) (1.97) 0.31

0.35

Democratic Parties

(1)

(2)

-0.46 (0.34) 0.69 * (0.32) 0.94 * (0.25) -1.68 (1.18)

3.33 (6.57) 10.07 (5.19) -15.23 (5.43) -4.11 (2.03) -7.43 (13.47) -19.65 (10.55) 32.82 (11.08) 1.26 * 2.46 (0.58) (0.98) 0.74

Republican Parties

0.80

(1)

+ * *

+ * *

-0.03 (0.38) 0.98 * (0.35) -0.02 (0.28) 1.42 (1.31)

-3.14 (7.90) -0.30 (6.25) -2.37 (6.53) 2.50 (2.44) 6.85 (16.20) 2.13 (12.69) 4.51 (13.32) -1.26 + -1.75 (0.64) (1.18) 0.49

Note: N=47, excludes AK, HI, NE. Coefficients from OLS regression models in which measures of party policy positions are regressed on the mean opinion of each group. Model 1 adds the measure of income inequality (market gini) to the basic models presented in Table 1; Model 2 includes three interaction terms to test for a conditional effect of preferences across states with different levels of income inequality. Standard errors are presented in parentheses beneath each coefficient. *=p<.05, +=p<.10.

32

(2)

0.56

Figure 1. Public Policy Preferences, by Income Group

A. Economic Issues

LowIncome

MiddleIncome

HighIncome

-.5

0

.5

B. Social Issues

LowIncome

MiddleIncome

HighIncome

-.5

0

.5

Note: N =47 states. Policy preferences estimated from 2000 Annenberg Survey, aggregated to the state-level.

33

Figure 2. Party Platforms, by Average Public Opinion in the State

AZ

VA CAFL IL TX MN WA NM MIOR CO UT WI ID SCDE NC WY ND KS IN ALOH MT SD MO TN OK KY LA GA NV

.5

IA

CT PAMD NJ ME

VT RI

WV

VA NJ

DE

-.5

ND IA WY

CT

MIOR TN KY AZ OH MS AR NC WINM NVSC

VT MA

IN

SD

UT MO

ID

WA

MN

CO TX OK GA

-.3

RI

PA MD WV

AL

KS

NY

FLIL

LA

-1

Republican Parties

NH

NY MA

AR MS

0

Democratic Parties

1

A. Economic Issues

-.2

ME

NH

MT CA

-.1

0

.1

.2

.3

Average State Opinion: Economic Issues

1

CA

TX NC SC UT MO

.5

Democratic Parties

B. Social Issues

KY LA

-.5

MS SD

IN GA

ND

KY LA MS

NY DE

OK AR ND

MA

IL IA KS IN

AR TN

VT

NJRI CT

TN

AL

WV

MA

RI

WY AL

WV

-1

Republican Parties

0

SD

IL CO WA AZ FL OR WI MD NM MN IA ME DE MI VA OH KS PA MT ID NV

NY NJ NH CT

UT NC MO SC GA TX

ID OH

ME FL OR PA WY NV WI CA MIVA MD NM MT AZ COWA MN

VT NH

OK

-.3

-.2

-.1

0

.1

.2

.3

Average State Opinion: Social Issues

Note: N=47 states. Average Policy Liberalism Score for Candidates from Democratic and Republican Parties, estimated from Project Vote Smart data.

34

Figure 3. Parties’ Responsiveness to Preferences of Different Income Groups A. Economic Issues Democrats: Influence

Republicans: Alignment

Republicans: Influence

-1

0

1

2

-1

0

1

2

Democrats: Alignment

.

Low

Middle

High

.

.

Low

Middle

High

.

B. Social Issues Democrats: Influence

Republicans: Alignment

Republicans: Influence

-1

0

1

2

-1

0

1

2

Democrats: Alignment

.

Low

Middle

High

slope

.

.

Low

Middle

High

.

95% CI

Note: N=47 states. Predicted first differences for a one-unit change in each group’s opinion. Alignment estimates reflect bivariate relationship between the group’s opinion and party position. Influence estimates reflect the partial association after controlling for the opinion of the other two groups. 35

Figure 4. Democratic Party Responsiveness, by State Income Inequality A. Economic Issues High Income Inequality: E conomic Issu es

.

.

-1

0

1

2

Lo w Income Inequality: Economic Is sues

Low

Middle

High

.

slope

Low

Middle

High

.

95% CI

B. Social Issues High Income Inequality: Social Issues

-1

0

1

2

Lo w Income Inequality: Social Issues

.

Low

Middle

High

slope

.

.

Low

Middle

High

.

95% CI

Note: N=47 Democratic parties. Predicted first differences for a one-unit change in each group’s opinion, controlling for the preferences of the other two groups (influence models). Compare estimates under conditions of low (-1 sd) versus high (+1 sd) income inequality. Confidence intervals for these point estimates generated with CLARIFY. Parallel estimates are not presented for Republican parties since their responsiveness did not vary by income inequality but remained most responsive to high-income preferences on economic issues and middle-income preferences on social issues.

36

Online Appendices

Appendix A. Items Used to Estimate Income Group’s Preferences Appendix B. Items Used to Estimate Parties’ Programs Appendix C: Additional Information on Missing Data Imputation Appendix D. Models with Group Preferences Weighted by Political Participation

37

Appendix A. Items Used to Estimate Income Group’s Preferences Raw N

Range

Economic Policy Liberalism Should spend on health care for uninsured (Q111b) Favor universal health care for children (Q91d) Should spend on Medicare (Q111g) Should spend on Medicaid (Q111h) Should spend on aid to mothers with children (Q111e) Should reduce income differences (Q136e) Inheritance tax should be cut (Q113a & Q113b)

55,549 29,084 24,501 24,317 24,055 23,758 18,292

0 0 0 0 0 0 0

3 1 3 3 3 1 1

0.74 0.67 0.63 0.75 0.65 0.46 0.27

Social Policy Liberalism Favor restricting abortion (Q91b & Q38c) Should stop job discrimination against gays (Q110a) Favor death penalty (Q268b) Favor gays in the military (Q268c) Should allow school prayer (Q136d) Should ban abortion (Q136a)

54,876 54,767 29,496 27,955 24,139 24,010

0 0 0 0 0 0

1 3 1 1 1 1

-0.73 0.36 -0.11 0.52 -0.42 -0.71

Factor Loadings

Notes: Items from 2000 Annenberg National Election Survey used in analysis.

38

Appendix B. Items Used to Estimate Parties’ Programs Raw N Economic Policy Liberalism Reduce funding for Health Care Reduce funding for Education (K-12) Reduce funding for Education (Higher) Decrease Income Taxes (incomes greater than $75,000) Reduce funding for Welfare Decrease Capital Gains Taxes Decrease Cigarette taxes Guaranteeing medical care to all citizens is not a responsibility of government Reduce state government regulations on the private sector Eliminate government-funded welfare and advocate privately-funded assistance Decrease Income Taxes (incomes less than $75,000) Decrease Property Taxes Decrease Sales Taxes Provide tax incentives to businesses that hire welfare recipients Provide tax credits for businesses that provide child care for their employees Increase state funds for professional development and salaries of public school teachers Increase employment and job training programs for welfare recipients Ensure that citizens have access to basic health care, managed care Increase access to public transportation for welfare recipients who work Increase state funds for hiring of additional teachers Provide child care for welfare recipients who work Increase state funding for programs to retrain displaced workers Support state government funding of programs for at-risk youth Increase state funding for Head State in order to serve additional children Increase state funds to provide child care for children in low income working families

39

Factor Range Loading

14,627 17,401 11,760 13,369 16,638 15,434 13,648

0-6 0-6 0-6 0-6 0-6 0-6 0-6

-0.70 -0.65 -0.65 -0.63 -0.62 -0.62 -0.57

16,346 14,919

0-1 0-1

-0.52 -0.48

15,870 13,409 10,814 12,585 10,158

0-1 0-6 0-6 0-6 0-1

-0.40 -0.39 -0.21 -0.14 0.33

15,050

0-1

0.43

15,895 12,961 10,282 12,322 14,295 13,745 14,919 10,292

0-1 0-1 0-1 0-1 0-1 0-1 0-1 0-1

0.54 0.58 0.58 0.60 0.60 0.62 0.66 0.67

12,535

0-1

0.72

9,844

0-1

0.74

Items Used to Estimate Parties’ Programs, Continued Favor allowing citizens to carry concealed firearms Prohibit public funding for abortions and organizations that advocate or perform abortions Ease state restrictions on the purchase and possession of firearms Repeal state restrictions on the purchase and possession of firearms by lawabiding citizens Abortions should be limited by waiting periods and notification requirements for minors Limit welfare benefits given to recipients if they have additional children Support sex education programs which stress abstinence Abortions should be legal only when the life of the woman is endangered Abortions should be legal only when the pregnancy resulted from incest/rape Abortions should always be illegal Do you believe that the [state] government should recognize same-sex marriage Abortions should be legal only within the first trimester of pregnancy Do you believe that the [state] government should include sexual orientation in discrimination clause Increase state restrictions on the purchase and posession of firearms Abortions should always be legally available Maintain the restrictions on the purchase and possession of firearms Require a license for gun possession Ban the sale or transfer of all forms of semi-automatic weapons Require manufacturers to provide child-safety locks with firearms Require background checks of gun buyers at gun shows

14,706

0-1

-0.65

9,969 16,733

0-1 0-1

-0.65 -0.55

16,636

0-1

-0.51

10,717 13,238 9,793 12,578

0-1 0-1 0-1 0-1

-0.43 -0.40 -0.32 -0.30

15,982 14,849

0-1 0-1

-0.17 -0.16

9,613 13,695

0-1 0-1

-0.03 0.01

12,424 13,043 17,509 11,044 9,381 10,618 13,306 9,435

0-1 0-1 0-1 0-1 0-1 0-1 0-1 0-1

0.44 0.50 0.51 0.54 0.56 0.62 0.65 0.67

Notes: Items from Project Vote Smart’s NPAT Survey, Pooled 1996–2004 used in analysis.

40

Appendix C: Additional Information on Missing Data Imputation To generate state-level estimates of public opinion liberalism for each income group, as well as for each political party, we pooled respondents across a set of cross-sectional surveys with different sets of (somewhat overlapping) questions. For our public opinion measures, we identified 13 items that were often asked across a series of rolling cross-sectional surveys fielded leading up to and right after the 2000 Presidential election. For the measure of parties’ liberalism, we identified 46 items repeatedly asked across a set of 225 state- and year-specific surveys administered to candidates for elected office between 1996 and 2004. Note: a listing of the specific variables, descriptive statistics, and the raw N (number of non-missing cases) are presented in Appendix A (public opinion variables) and Appendix B (party variables). Since all respondents were not asked all questions, we estimated the missing data using imputation by multiple chained equations (see Royston 2005), which predicts survey participants’ responses based on their preferences among items they were asked. Beginning with the item missing the least amount of data, missing cases were estimated based on the covariance matrix of all other non-missing responses, and for the dataset of policymaker positions a variable capturing a linear trend capturing the year/campaign in which positions were measured. This process was repeated for each item, ultimately using the full set of observed and imputed items to estimate missing data for the variable with the most instances of missing data. We estimated five imputed datasets and averaged values across these five datasets to generate the values used in later analyses. This imputation approach maximizes similarities in responses across items, weighing most heavily the cases with the most observed cases. As a result, these data are best suited to capturing a latent construct (e.g., social policy liberalism), as we use them, rather than precise views on a particular issue (i.e., the death penalty or abortion). Below we present information on the distribution of missing cases across all the variables used to construct our scales. In addition, we present results of simulations we conducted in order to assess the accuracy of our imputed values. A. Proportion of Cases Imputed

The histograms that follow present the proportion of cases imputed for each variable, first for the variables used in the public opinion scales (Figure 1) and then for the variables used to estimate parties’ liberalism (Figure 2). These histograms illustrate two different patterns of missingness. For the public opinion variables, each scale relies on one or two variables asked of all respondents, while the other variables are asked only of half the sample. Information on the shared question (serving as a bridge) was paired with information on respondents’ responses to the other variables in order to estimate the missing data for the approximately half of respondents not asked these questions. In contrast, the variables used in the scales capturing parties’ liberalism exhibit a more even pattern of missingness. These variables, drawn from the NPAT survey, are missing an average of 25 percent of cases (s = 14). No variables were missing more than half of respondents since we used that as a threshold for identifying “commonly-asked” questions from the NPAT survey.

41

Figure 1. Proportion of Cases Imputed, Public Opinion Liberalism

0

2

Density 4

6

8

a. Public Opinion on Economic Issues (M=.49, s=.22)

0

.2

.4

.6

Proportion of Public Opinion Cases Imputed

N= 7 variables.

0

2

4

Density 6

8

10

b. Public Opinion on Social Issues (M=.36, s=.27)

0

.2

.4

.6

Proportion of Public Opinion Cases Imputed

N= 6 variables.

42

Figure 2. Proportion of Cases Imputed, Parties’ Liberalism

0

1

2

Density 3

4

5

a. Parties’ Position on Economic Issues, (M=.23, s=.13)

0

.1

.2 .3 .4 Proportion of NPAT Cases Imputed

N= 26 variables.

0

2

Density

4

6

b. Parties’ Position on Social Issues (M=.27, s=.15)

0

.1

.2

.3

.4

Proportion of Imputed NPAT Cases

N= 20 variables.

43

B. Accuracy of Imputed Estimates To estimate the accuracy of our imputed estimates we undertook a simulation for a sub-set of cases in which we had complete data. For each measure of the four scales, we identified two component variables that exhibited a typical (median) proportion of missing cases. For the public opinion measures, these variables were: (1) favoring universal health care for children, (2) government should reduce income differences, (3) favor gays in the military, and (4) ban all abortions. We selected approximately 25 percent of the 7,940 cases with complete data across these four variables. For these 1,946 cases, we deleted the individual’s responses for each of these four variables, setting responses to missing. After imputing data for the full set of respondents and variables, we compared the imputed values for the deleted cases to the respondents’ actual responses, identifying significant differences for all four variables. Additionally, we rounded the imputed results to one of the two original responses (zero or one) and calculated the percent of cases in which the imputation matched the respondent’s true response. The imputations classified cases correctly, 89 percent, 63 percent, 69 percent and 84 percent of the cases, respectively. Additionally, to assess the degree to which differences between known and imputed values influenced the economic and social liberalism scales used in the paper, we re-calculated both scales using the imputed (rather than known) values for these four variables and 1,946 respondents for whom we deleted known values. Figure 3 presents the comparisons between the factors scores computed with and without the imputed values. These scores are highly correlated, .969 for economic issues and .963 for social issues. Figure 3. Public Opinion Liberalism Scores, Imputed versus Known Values

-6

Scores with Imputed Data -4 -2 0

2

a. Public Opinion on Economic Issues

-4

-2

0

2

Scores with Known Data

Note: N=1,946 randomly-selected cases in which known values on two variables were replaced with imputed values. Scatter plot compares economic liberalism scores when computed with known values versus imputed values. The two scales are highly-correlated (r=.969). 44

-2

Scores with Imputed Values -1 0

1

b. Public Opinion on Social Issues

-2

-1

0 Scores for Known Values

1

2

Note: N=1,946 randomly-selected cases in which known values on two variables were replaced with imputed values (mean across five imputations). Scatter plot compares social liberalism scores when computed with known values versus imputed values. The two scores are highlycorrelated (r=.963).

To estimate the accuracy of our imputed estimates for policymakers’ positions, we selected four variables with a median proportion of missing data in need of imputation. These variables were: (1) providing child care for welfare recipients who work, (2) supporting increased funding for hiring additional teachers, (3) increasing state restrictions on the possession of firearms, and (4) abortion should be legal only when the life of a woman is in danger. We selected 25 percent of the 5,111 policymakers with complete data across these four variables. For these 1,242 cases, we deleted the individual’s responses for these four variables, setting their responses to missing. After imputing data for the full set of respondents and variables, we compared the imputed values for the deleted cases to the respondents’ actual responses. Next, we rounded the imputed results to one of the two original responses (zero or one) and calculated the percent of cases in which the imputation matched the respondent’s true response, which we found to occur 77 percent, 78 percent, 84 percent and 72 percent respectively. Additionally, to assess the degree to which differences between known and imputed values influenced the economic and social liberalism scales used in the paper, we re-calculated both scales using the imputed (rather than known) values for these four variables and 1,252 respondents for whom we deleted known values. Figure 4 presents the comparisons between the factors scores computed with and without the imputed values. These scores are highly correlated, .995 for economic issues and .993 for social issues.

45

Figure 4. Policymakers’ Liberalism Scores, Imputed versus Known Values

0 -1 -3

-2

Scores (Imputed Values)

1

2

a. Policymakers on Economic Issues

-3

-2

-1

0

1

2

Scores (Known Values)

Note: N=1,242 randomly-selected cases in which known values on two variables were replaced with imputed values. Scatter plot compares economic liberalism scores when computed with known values versus imputed values. The two scores are highly-correlated (r=.995).

1 0 -1 -2

Scores with Imputed Data

2

b. Policymakers on Social Issues

-2

-1

0

1

2

Scores with Known Values

Note: N=1,242 randomly-selected cases in which known values on two variables were replaced with imputed values. Scatter plot compares social liberalism scores when computed with known values versus imputed values. The two scores are highly-correlated (r=.993). 46

Appendix D. Models with Group Preferences Weighted by Political Participation Economic Issues Democratic Parties

(1) Low-Income Mid-Income High-Income Low Vote High Vote

-0.08 (0.06) -0.01 (0.09) 0.17 * (0.08) -13.71 * (2.54) -10.91 * (2.47)

Income Inequality Low X Ineq. Mid X Ineq. High X Ineq. Constant

R-squared

8.72 * (1.60) 0.53

(2) -0.26 (1.15) 2.07 (1.51) -2.49 (1.42) -12.00 (2.67) -9.18 (2.66) 1.79 (2.72) 0.33 (2.33) -4.09 (3.00) 5.38 (2.87) 6.72 (2.19) 0.59

Social Issues

Republican Parties

(1)

+ * *

0.11 (0.10) -0.03 (0.15) 0.31 * (0.13) -1.58 (4.25) -4.06 (4.14)

1.50 (2.67)

-0.51 (1.92) -1.21 (2.52) 0.57 (2.38) -1.86 (4.47) -5.76 (4.45) 1.34 (4.56) 1.27 (3.89) 2.12 (5.01) -0.47 (4.81) 1.57 (3.55)

0.30

0.38

+ *

(2)

Democratic Parties

(1) -0.04 (0.07) 0.14 (0.19) 0.34 * (0.09) -13.98 * (3.75) -11.54 * (3.68)

8.69 * (2.39) 0.80

Republican Parties

(2) 1.41 (1.60) 1.85 (1.69) -4.66 (2.11) -11.00 (3.57) -9.18 (3.53) -4.07 (1.84) -3.04 (3.27) -3.53 (3.43) 10.18 (4.32) 8.96 (2.35) 0.83

(1)

*

-0.06 (0.09) 0.34 * (0.12) 0.02 (0.11) 0.69 (4.64) -2.57 (4.55)

0.27 (2.96)

1.68 (1.96) 0.35 (2.07) -4.20 (2.59) 2.11 (4.38) -3.87 (4.33) 0.96 (2.26) -3.30 (4.01) -0.39 (4.21) 8.58 (5.30) -0.05 (2.89)

0.85

0.64

*

* *

(2)

Note: N=47, excludes AK, HI, NE. Coefficients from OLS regression models in which measures of party policy positions are regressed on the mean opinion of each group. Each group’s preference measure was weighted for the contribution of each group to the state electorate (based on turnout estimates generated from the 1996, 2000, and 2004 Current Population Survey, November Voter Supplement). Standard errors are presented in parentheses beneath each coefficient. *=p<.05, +=p<.10.

47