P A R T I S A N R E S P O N S I V E N E S S IN T H E U S H O U S E OF R E P R E S E N T A T I V E S , 1997-2005 by J O H N R. M C A N D R E W S A THESIS S U B M I T T E D I N P A R T I A L F U L F I L L M E N T OF T H E R E Q U I R E M E N T S F O R T H E D E G R E E OF M A S T E R OF A R T S in T H E F A C U L T Y OF G R A D U A T E S T U D I E S (Political Science) T H E U N I V E R S I T Y OF B R I T I S H C O L U M B I A August 2007 © John R. McAndrews, 2007 Abstract In this paper, I develop and test a new theory of partisan responsiveness. The theory suggests that members of Congress w i l l be most responsive to the average same-party voter compared with the average voter on high polarization-low salience issues and most responsive to the average voter compared with the average same-party voter on low polarization-high salience issues. I test these predictions on four issues - free trade, the Patriot Act , taxes and abortion - using the 2000 and 2004 National Annenberg Election Surveys. The results offer only partial support for the theory. i i i Table of Contents Abstract i i Table of Contents i i i List o f Figures iv List of Tables '. v Acknowledgements v i i Dedication v i i i Introduction 1 A n Empirical Gap in the Representation Literature 5 A Theory of Partisan Responsiveness 13 Data Collection and Description 20 Two Models o f Congressional Responsiveness 27 Summary and Conclusion 41 Bibliography 43 Appendix 1 - List of N A E S policy questions 48 Appendix 2 - R o l l call votes used in the dependent variable indexes 52 Appendix 3 - Descriptive statistics 63 Appendix 4 - Correlations between opinion variables 66 Appendix 5 - Supplementary tables from regression analyses 69 Lis t of Figures Figure 1: Responsiveness predictions by issue type 19 Figure 2: N A E S issues by polarization and salience 24 Figure 3: Selected issues by polarization and salience 25 Figure 4: Conditional impact of partisans and independents on abortion, Democratic members only 40 Figure 5: Conditional impact of partisans and independents on abortion, Republican members only 41 Lis t of Tables Table 1: Descriptive statistics and Cronbach's alpha 26 Table 2: Model 1 - District mean and same-party mean responsiveness, all members .. 29 Table 3: Correlation between smaller-N district mean and variables specified below.... 30 Table 4: Model 1 - District mean and same-party mean responsiveness, all members (smaller-N district mean) 31 Table 5: Model 1 - District mean and same-party mean responsiveness, Democratic members only 34 Table 6: Model 1 - District mean and same-party mean responsiveness, Republican members only 35 Table 7: Model 2 - Conditional impact of partisans and independents on free trade, Democratic members only .38 Table 8: Model 2 - Conditional impact of partisans and independents on taxes, Republican members only 39 Table A 1: N A E S policy questions 48 Table A 2: House roll call votes used in free trade index 52 Table A 3: House roll call votes used in Patriot Act index 54 Table A 4: House roll call votes used in tax index 54 Table A 5: House roll call votes used in abortion index 60 Table A 6: Descriptive statistics for variables in free trade analysis 63 Table A 7: Descriptive statistics for variables in Patriot Ac t analysis 63 Table A 8: Descriptive statistics for variables in tax analysis 64 Table A 9: Descriptive statistics for variables in abortion analysis 65 Table A 10: Correlation matrix for opinion variables in free trade analysis 66 Table A l l : Correlation matrix for opinion variables in Patriot Act analysis 66 Table A 12: Correlation matrix for opinion variables in tax analysis 67 Table A 13: Correlation matrix for opinion variables in abortion analysis 67 Table A 14: Democratic mean residuals by selected pairs of similar districts 68 Table A 15: Republican mean residuals by selected pairs of similar districts 68 Table A 16: Model 1 - District and same-party mean responsiveness, all members without Texas (smaller-N district mean) 69 Table A 17: Change in adjusted R 2 (without Texas - with Texas) 69 Table A 18: Model 2 - Partisan and independent mean responsiveness with interactions, all representatives 69 Table A 19: Model 2 - Partisan and independent mean responsiveness with interactions, Democratic members only 70 Table A 20: Model 2 - Partisan and independent mean responsiveness with interactions, Republican members only '. 71 Table A 21: Model 2 - Conditional impact of partisan and independent means on Patriot Act , Democratic members only 71 Table A 22: Model 2 - Conditional impact of partisan and independent means on taxes, Democratic members only 72 Table A 23: Model 2 - Conditional impact of partisan and independent means on abortion, Democratic members only 72 v i Table A 24: Model 2 - Conditional impact of partisan and independent means on trade, Republican members only 72 Table A 25: Model 2 - Conditional impact of partisan and independent means on Patriot Act , Republican members only 73 Table A 26: Model 2 - Conditional impact of partisan and independent means on abortion, Republican members Only 73 Table A 27: Model 2 - Interaction terms only, Democrats only 73 Table A 28: Model 2 - Interaction terms only, Republicans only.. 74 Acknowledgements I am particularly grateful to Paul Quirk for his numerous helpful suggestions throughout the entire process. I would also like to thank Fred Cutler for his methodological advice and for his constructive criticism as second reader. Finally, I am grateful to Ben Bishin and Noah Kaplan for providing me with unpublished drafts of their forthcoming work. A n y errors are naturally my own. v i i i Dedication T o my wife Penny 1 Introduction Two essential questions in any representative democracy are: "who is represented?" and "on what issues?" In the U S , Christian evangelicals, for example, may be particularly influential on abortion but less so on international trade policy. For union members, it may be the reverse. Understandably, most American political science research to date attempts to answer only one question at a time, rarely both.. On the one hand, Clinton (2006), Bishin (2000), Wright (1989) and others carefully assess the political influence o f different segments of the electorate, but almost always in terms of a single aggregate measure (usually a liberal-conservative ideological dimension). This begins to answer the "who" but does not specify "on what." On the other hand, Egan (2007) compares congressional responsiveness on a variety of policy issues but only for the undifferentiated mass public. While this tackles the question of "on what," the "who" remains imprecise. Attempts to unite the two approaches are still limited. This paper seeks to help f i l l in this gap in the literature by simultaneously comparing the influence of different segments of the public on a wide variety of public policy issues. Beyond addressing this empirical gap, the paper makes three further contributions to the discipline. First, it begins to explain more fully the decision-making of representatives on an issue-by-issue basis. This paper, which focuses on constituency groups, is the first step in what should be a large process of systematically cataloguing who is and is not reflected in congressional actions on specific issues. Future research should, for example, examine - on the same issue-by-issue basis - elite-level influence involving think tanks, interest groups and lobbyists. Ultimately, understanding who is 2 influential on what issues provides the basis for a more productive future debate about who should be influential on what issues. Second, the theory developed here may help explain the polarization of Congress over the last thirty years (Aldrich 1995 and Fleisher and Bond 2004). The theory contends that, all else equal, as an issue becomes more polarized and less salient in the electorate, representatives have a greater incentive to respond, not to the average opinion of all voters in their districts, but rather to the average opinion of only those district voters who identify with their party. This increased incentive to respond to fellow partisans may translate into more extreme position-taking than would otherwise be the case i f representatives responded primarily to average voters. Although beyond the empirical scope of this project, the theory thus suggests by extension that polarization in . Congress may be due to increased responsiveness to partisan voters at the expense of average voters. Third, the paper focuses exclusively on the House of Representatives. Clinton (2006) notes that most of the work to date on congressional representation - and on subconstituency responsiveness in particular - focuses on the Senate. This is because state-level opinion data is historically more available and more reliable than congressional district-level opinion data. However, it is unreasonable to assume that conclusions drawn from an analysis of representation in the Senate can be generalized to the lower house: there are good theoretical reasons to expect that representatives exhibit a significantly different degree of responsiveness than senators. Representatives, for example, face voters at the polls three times more often than senators and are generally 3 elected from smaller and more homogenous constituencies. This paper helps correct the research imbalance. The approach adopted in this paper is possible in large part thanks to the newly available 2000 and 2004 National Annenberg Election Surveys ( N A E S ) . The N A E S offer huge random samples of the continental United States general population and ask dozens of specific public policy-related questions. These unprecedented sample sizes allow for reasonably accurate estimates of constituency and subconstituency preferences on several different policies. In effect, it should begin to tell us more about the "on what" and the "who" The particular electoral subconstituencies of interest in this paper are partisans, namely self-identified Democrats and Republicans. More specifically, I compare the influence of average voters to the influence o f representatives' fellow partisans, i.e. district Democratic voters i f the representative is a Democrat and district Republican voters i f the representative is a Republican (referred to here as "same-party" voters). Thus, the key research question of this work is: on what types df issues are members of Congress more responsive to same-party voters than average voters and on what types of issues are they more responsive to average voters than same-party voters? Put another way, this paper asks: when do representatives "play to their base?" Before proceeding further, it is worth briefly defining what is meant here by "responsive." The term responsiveness refers to the positive correlation across districts between members' voting records and the opinions of their constituents.1 A s noted , 1 Responsiveness can also refer to correlation across time. However, for the purposes of this paper, I limit the term to its cross-sectional sense: correlation across districts. 4 earlier, this paper is primarily interested in two types of responsiveness: to partisans and to the average voter. Partisan responsiveness is defined as a situation in which, all else equal, a unit change in the same-party mean opinion results - on average - in a bigger positive change in members' voting records than that prompted by a unit change in the district mean opinion. In other words, members respond more to same-party voters than to average voters. Conversely, mean voter responsiveness is defined as a situation in which, all else equal, a unit change in the district mean opinion results - on average - in a bigger positive change in members' voting records than that prompted by a unit change in the same-party mean opinion. In other words, members respond more to average voters than to same-party voters. It is also important to note here that responsiveness is not same as proximity. Proximity generally refers to how "close" representatives' issue positions are to those of their constituents in a unidimensional issue space. While proximity is likely the ideal method to analyze who is represented on what issues, representation scholars since Mi l le r and Stokes (1963) have confronted a common problem: the issue positions of elected officials and voters are rarely measured in the same way (Egan 2007). The policy positions of members o f Congress are usually derived from roll call voting; the mean positions of constituency groups are generally estimated by aggregating individual responses to public opinion surveys. Without common measures, evaluating proximity is impossible. 5 A s a substitute, scholars have historically turned to responsiveness. Clinton (2006), for example, refers to cross-district covariation as a necessary but insufficient condition for representativeness. This paper is in keeping with that tradition. Therefore, while parts of the theory proposed below are motivated by the concept of spatial proximity, the predictions derived from it and tested in this paper are strictly in terms of responsiveness. The paper is structured as follows. A short literature review situates the work in the larger body of research and explains why a mixed approach is needed to explore this under-tested variation in same-party voter and average voter influence on an issue-by-issue basis. The theory section makes specific predictions about this variation and derives the observable implications used in subsequent testing. The data and methods section then details how I compile the dependent variables (indexes of policy-specific House roll call votes) and measure the independent variables (district mean and same-party mean policy preferences). The results section presents and discusses ordinary least square (OLS) estimates of two models of congressional responsiveness. The paper concludes with a short summary of the analysis. An Empirical Gap in the Representation Literature The literature on representation is extensive and methodologically diverse. 3 Some works operationalize government output - the typical dependent variable - in terms of policy change (Page and Shapiro 1983), others in terms of congressional roll calls (Bartels 2005) or interest group ratings of individual representatives (Erikson and Wright 2 For a critical evaluation of responsiveness as a tool for assessing representation, see Achen 1977 and 1978. 3 For summaries on the state of scholarship in the field, see Wlezien and Soroka [2007], Burstein 2003, Jacobs and Shapiro 2000 and Glynn et al. 1999. 6 2000). Some measure public opinion - the key independent variable - at the national level (Page and Shapiro 1983 and Hurley 1991), others at the constituency-level (Miller and Stokes 1963 are the pioneer in this respect). Some studies tackle the question on an issue-by-issue basis (Quirk and Hinchliffe 1998 and Hurley and H i l l 2003), while others do so using broad issue indexes (Shapiro et al. 1990) or even broader measures of national "mood" (Stimson et al. 1995). Finally, some research is purely cross-sectional (Monroe 1979 and 1998), while others employ elaborate time-series analyses (Stimson et al. 1995). The findings are equally diverse. In this regard, Manza and Cook (2002) provide a useful typology. They classify the literature by three broad categories: 1) those works that find that public opinion has a large effect on public policy, 4 2) those, at the other end of the spectrum, that find only limited or minimal effects5 and 3) those who occupy the middle ground, arguing that public opinion does indeed affect public policy, but only in particular circumstances.6 For the purposes of this paper, however, the relevant literature can be divided into two main groups: 1) those works that analyze responsiveness on specific issues but often treat the electorate as an undifferentiated mass and 2) those that model subconstituency influence but are generally unconcerned with specific policies. The first group includes 4 This group includes: Stimson et al. (1995) who create a sophisticated domestic policy "mood" index along a single liberal-conservative ideological dimension based on hundreds of public opinion surveys; Monroe (1998) who analyzes over 500 specific policies from 1980 to 1993; and Quirk and Hinchliffe (1998) who conduct case studies of policy issues - such as the Clean Air Act and business regulation - on which elite and mass opinion have historically diverged. 5 Ansolabehere et al. (2001), for example, determine that, between 1874 and 1996, the preferences of US congressional candidates have generally reflected those of the party and not those of the local mean voter. 6 Such conditions include the salience (Sharp 1999, Burstein 1985, Kuklinski and McCrone 1980) and the novelty (Petry and Mendelsohn 2004) of an issue, the time remaining before the next election and the type of decision-making institution (Baumgartner and Jones 1993). 7 the analyses of single issues like defence spending (Bartels 1991) and immigration (Cohen and Fleisher 2007). Both find strong links between district opinion and legislative behaviour. The group also includes those works that explicitly compare responsiveness across issues and issue domains. Monroe (1998) and Page and Shapiro (1983) engage in such cross-issue comparisons although their dependent variables measure policy change rather than legislative behaviour. Both conclude that responsiveness is highest on salient issues.7 Perhaps the most novel and counterintuitive explanation of cross-issue variation in responsiveness is Egan (2007). Egan's argument is based on a party's "ownership" of an issue. " A political party," he observes, " 'owns' an issue to the extent that the public assesses it can better 'handle' the issue than its opponents" (Egan 2007, p. 3). Egan argues that this issue-specific trust provides more latitude on the issue for a party and its candidates. According to this theory, members of Congress affiliated with a party that "owns" a particular policy issue (e.g. Democrats on environment or Republicans on defence) can better afford to be unresponsive to constituency opinion. Egan's evidence lends support to his theory. A s noted earlier, this line of research begins to clarify the types of issues on which representatives are most responsive. The second group in the literature asks: responsive to whom? It includes Wright (1989), Clinton (2006), Bishin (2000) and Kaplan [2007] who each assess subconstituency influence on legislative behaviour but only in terms of a 7 Monroe (1998) operationalizes salience using the traditional Gallup poll question: "what do you think is the most important problem facing this country today?" Page and Shapiro (1983) operationalize the same concept in two different ways. The first counts the frequency with which pollsters asked a policy-specific question over time. The second counts the frequency of "don't know" responses to a single survey question. 8 single ideological dimension: Wright and Clinton use self-reported ideological placement (e.g. "would you consider yourself a liberal, moderate, conservative?") while Bishin and Kaplan aggregate responses to specific policy questions. Surprisingly, despite these similar research strategies, there is no clear consensus from these scholars about which subconstituencies are most influential. Wright (1989) finds that senators are particularly responsive to the independent voters in their home states. In contrast, Clinton's (2006) analysis suggests that House Republicans are very responsive to district Republicans, while House Democrats are most responsive to district non-Democrats. Kaplan [2007] subsequently revises this finding upon replication of Clinton's data, arguing members of Congress primarily respond to their districts as a whole and that House Republicans and Democrats are substantively similar in their responsiveness to fellow partisans. Thus, on one side, there is good evidence that legislators are especially responsive to the public on highly salient policy issues as well as some surprising findings about "owned" issues. On the other, there is inconclusive evidence - and in terms of a single ideological dimension only - about whom exactly these legislators are responding to. Some research has begun to address this gap in the literature, asking not only "who is represented" but also "on what issues." They make valuable contributions but are often constrained by data limitations. Hurley (1991) is an early example. 8 She 8 Miller and Stokes (1963) is an even earlier example. They assess congressional responsiveness on three broad policies: social welfare, US foreign involvement and civil rights. Most interesting, for the purposes of this paper, is their finding that while congressional roll call voting on social welfare policy and district mean opinion are correlated at 0.3, the correlation between the same roll calls and the mean opinion of those who voted for the incumbent is 0.4. However, I do not emphasize their work in this paper for two reasons. First, despite collecting data on all three issues, Miller and Stokes do not offer specific results on the influence of the mean incumbent voter for the other two issues. Thus no cross-issue comparisons can 9 compares changes in the issue-specific voting records of Republican and Democratic senators in the 1980s against changes in the policy positions, on the same issues, of self-identified Republicans, Democrats and independents in the national electorate.9 Her data suggest that Republican senators moved in the opposite direction of Republican national opinion on several issues, most especially on defence, U S involvement in Central America, aid to minorities and women's equality. Conversely, she finds that changes over time among Democratic senators and Democratic voters nationally are more congruent. While these results are interesting, there are few good theoretical reasons to expect reelection seeking senators to respond systematically to changes in national public opinion since it is their state electorates that control their political future. Further analysis at the local level is needed. Shapiro et al. (1990) offer this. Here the dependent variables are different measures of liberalism - derived by the National Journal from Senate voting records - in economic and foreign policy. The independent variables are measures of economic and foreign policy liberalism among self-identified Democrats, Republicans and independents in each state, which the authors estimate by aggregating responses to specific policy questions contained in the National Election Studies' 1988 Senate Election Study. Consistent with their theoretical expectations, Shapiro et al. find that, on both economic and foreign policy, senators respond positively to the same-party voters, negatively to be made. Second, and more importantly, the reliability of their data for this purpose is suspect. Erikson (1978), for example, points out that Miller and Stokes have an average of only 11 usable opinions in each district. Page and Shapiro (1983) go on to describe the data as "based on small and unrepresentative district samples and are susceptible to varying interpretations" (Page and Shapiro 1983, p. 176). 9 Hurley (1991) measures public opinion using policy questions contained in the National Election Studies between 1980 and 1986. 10 voters of the opposing party and are fairly unresponsive to independents.1 0 While useful in addressing the empirical gap, Shapiro et al.'s issue categories are still too broad. A n analysis of responsiveness on economic policy would be more informative i f it were further divided into policies on job creation, taxes, the federal budget and debt, etc. The same could be said of foreign policy: are same-party voters more influential on international trade policy, military interventions or foreign aid? Indeed, the authors themselves recommend further research into "more specific issue areas" in the future (Shapiro etal. 1990, p. 617). Recently Hurley and H i l l (2003) argue persuasively that congressional responsiveness to same-party voters should be greatest on policy issues that are both simple and party-defining. A simple issue, borrowing Carmines and Stimson's (1980) taxonomy, is a long-standing, symbolic issue, on which voters often have a "gut" response. A party-defining issue is one on which opposing positions coincide with the traditional "lines of cleavage of partisan elites" (Hurley and H i l l 2003, p. 305). To illustrate their theory, Hurley and H i l l examine congressional responsiveness on a single issue - domestic welfare policy - three times over a thirty-year time period. 1 1 They find that in the late 1950s and early 1960s, when welfare policy was both simple and party-defining, representatives were especially responsive to same-party voters. The rationale behind Hurley and H i l l ' s (2003) concept of party-defining issues is plausible. Intra-party division on non-party-defining issues allows challengers to commit 1 0 Shapiro et al. (1990) also find that senators, in general, are less responsive on foreign policy than on other policy domains. This is consistent with Miller and Stokes's (1963) early work but runs counter to Monroe's (1990) findings. " Hurley and Hill measure public opinion on domestic welfare policy using relevant policy questions from the 1958 Miller and Stokes study, the 1978 National Election Study and the 1988 Senate Election Study. 11 credibly to responding to the district mean preference. This credibility forces the incumbent to respond more to the district mean or risk defeat at the polls. Conversely, the lack of intra-party division on party-defining issues means that challengers' commitments are much less credible. This lack of challenger credibility allows incumbents to respond more to same-party voters without fear that rivals w i l l exploit the issue to their advantage by appealing more to average voters. However, the underlying logic of the issue complexity argument is more problematic. Hurley and H i l l (2003) argue that the simpler an issue the more incentive •members of Congress have to "pay attention to constituency preferences" (Hurley and H i l l 2003, p. 306). Their rationale, in part, is that the simpler the issue the easier it is for a voter to monitor outcomes. In contrast, I contend that representatives are likely to respond to voters on salient issues regardless of whether they are complex or not. The key question for representatives - who seek to maximize their chances of reelection - is not whether voters understand an issue but rather whether it w i l l figure prominently in their voting decisions. There is also room to improve on Hurley and H i l l (2003)'s method and data. First, their qualitative operationalization of party-defining and simple issues is time consuming and removed from the survey questions used to gauge public opinion in their analysis. Second, Hurley and H i l l ' s measures of constituency and subconstituency opinion may not be reliable. Their use of Mi l l e r and Stokes's 1958 data in particular underscores the need for larger and more representative samples. 1 2 Third, by virtue of the 1 2 The 1958 Miller and Stokes data has a total of only 2,000 respondents across the 116 congressional districts the authors selected for study. Subsequent scholars have since cast doubt on the reliability of their district opinion estimates. See earlier note for details. 12 historical focus of the work, the data are old. The final year they analyze is 1988. It is worth reassessing the matter in a more contemporary context. The N A E S helps address these latter two concerns in particular. The most recent and extensive attempt to address the empirical gap is Bishin [2007]. Bishin advances a new subconstituency politics theory that contends that "candidates build a coalition of supporters by developing platforms that activate intense groups of constituents." (Bishin 2007, p. 16) The difference in intensity between the passionate minority and the more indifferent majority, he argues, allows politicians to respond to the minority position without seriously alienating the majority. Bishin tests his new theory by, among other things, analyzing the responsiveness of senators on abortion. His dependent variable is roll call voting on abortion-related matters between 1999 and 2005. His key independent variable is a combination of the mean abortion opinions of college-educated women and Christian evangelicals - the two subconstituencies he considers particularly passionate about the issue - in each state.1 3 Bishin finds that while his subconstituency opinion measure is statistically significant, the state mean opinion is not. Bishin's approach is novel and promising. Cross-group comparisons of issue intensity, in particular, are worth further study. However, the work jumps ahead of the current state of the literature without answering some of its outstanding questions. For example, there is yet no convincing evidence of those issues on which partisans are more influential than the average voter. While Bishin's subconstituency groups are more nuanced than those examined here, this paper argues that it is still premature to divide 1 3 Bishin measures subconstituency opinion using the 2000 National Annenberg Election Study. 13 further the public into smaller groups without first resolving the question of responsiveness among larger groups. Moreover, there is a good institutional reason to focus on partisans. The declining number of competitive districts (Fiorina 1977, Abramowitz et al. 2006) often means most political competition occurs in the primary, not the general, election. Many states employ closed primaries in which only voters registered as Democrats may vote in the Democratic primary and only voters registered as Republicans in the Republican primary. In this arrangement, same-party voters exert a particularly strong influence on representatives because reelection first requires satisfying same-party voters in the primary. College-educated women and evangelicals do not have this institutional advantage (unless as part o f these larger partisan groups). This paper, though having its own limitations (discussed below), seeks to tackle the weaknesses of those who have already begun to address the gap. Unlike Hurley (1991), this paper focuses on constituency-level responsiveness. It offers more narrowly-defined issues than Shapiro et al.'s (1990) economic and foreign policy indexes. It also develops, unlike Hurley and H i l l (2003), a theory that can be easily operationalized on quantitative grounds and then tests it on more than one issue. Finally, unlike Bishin (2007), this paper addresses the still outstanding question of whether partisans or average voters are more effectively represented in Congress. A Theory of Part isan Responsiveness In this section, I address the empirical gap by developing a theory of partisan responsiveness. The theory explains the logic of why members of Congress respond to voters at all and, more importantly, why they respond differently to different groups of 14 voters. Furthermore, it specifically addresses the questions of "who" and "on what" by identifying the types of issues on which members of Congress are likely to exhibit partisan responsiveness and those on which they are likely to exhibit mean voter responsiveness. These predictions serve as the basis for testing the theory in subsequent sections. A t the outset, the theory makes three assumptions. First, it assumes that representatives are, at minimum, boundedly rational and have one basic goal: to win reelection. 1 4 This is a common assumption in the literature and is consistent with Mayhew's (1974) argument. Other goals - including those identified by Fenno (1973) and Hal l (1996) such as legislative influence and the implementation of personal policy preferences - are less important and indeed possible in the long term only through reelection. Second, the theory assumes that the proximity of candidates' policy positions is one of several important factors in voter decision-making (others factors, for example, may include constituency service and incumbency). This is a relaxed version of the classic Downsian condition that all voters "vote for the candidate closest to their own policy location" (Grofman 2004, p. 26). 1 5 Third, it assumes that same-party mean opinion and district mean opinion are not perfectly correlated across congressional districts. This is consistent with the findings of Hurley and H i l l (2003) and Clinton (2006). I also offer further evidence in support of this assumption in the results section. The theory begins from first principles: why do members of Congress bother responding to constituents at all? The answer flows directly from these assumptions. A l l 1 4 Bounded rationality relaxes the classic assumptions of rational choice theory by acknowledging that actors are often limited, among other things, by imperfect information and cognitive capacity (Jones 1999). 1 5 In contrast, this theory requires only that voters consider the proximity of candidate positions and accepts that some may not choose the closest candidate. 15 else equal, being responsive increases the likelihood that members' w i l l achieve their primary goal: reelection. B y adjusting their issue positions to reflect differences in constituency issue positions across congressional districts, members stay electorally competitive. In contrast, "when incumbent roll call behavior deviates from public preferences," observe Cohen and Fleisher, "challengers possess strong incentives to publicize these unpopular votes, which in turn may persuade constituents to vote against the incumbent" (Cohen and Fleisher 2007, p. 5). This is not to suggest, however, that members' and constituents' issue positions are "close" in the absolute sense. A s noted at the outset, without common measures such a statement cannot be tested. Instead, the argument here is that across congressional districts members change their issue positions in order to maintain what they consider an acceptable - but, for the purposes of this paper, unspecified - distance between their views and those of some or all of their constituents. To do otherwise is to increase the risk of defeat. If constituency responsiveness is a strategy by which members seek reelection, the next question is: who in the constituency do they respond to? If the same-party mean and the district mean vary independently across districts, then responding more to one means responding less to the another. Who then do boundedly rational reelection seeking members choose? The theory suggests they choose whichever group maximizes their chances of reelection. Furthermore, it contends that a group's reelection-maximizing potential is partly conditional on the characteristics of the specific issue in question. On some issues, responding more to average voters than same-party voters may maximize members' likelihood of reelection; on other issues, it may be the reverse. 16 Before examining these issue characteristics, it is useful to briefly reiterate why both average voters and same-party voters are important to members' reelection campaigns. Downs (1957), among many others, notes the pivotal role of the average voter. Responding to the issue positions of the mean voter should, by definition, maximize members' appeal to all voters on average. Why then are same-party voters also important and worth responding to? First, same-party voters control who is nominated by virtue of the closed list primary election system. Second, same-party voters constitute a more significant bloc of voters in the general election as the number of noncompetitive districts grows. Third, same-party voters are more likely to volunteer and to vote for their candidates in the general election and contribute financially to their campaigns than average voters. For these reasons, I argue partisan responsiveness is the optimal reelection strategy for most but not all issues. This leads to the final question: on what issue characteristics does the reelection-maximizing potential of same-party voters and average voters depend? In other words, on what types of issues does partisan responsiveness maximize members' chances of reelection and, conversely, on what types of issues does mean voter responsiveness do the same? To address this question, the theory combines Bishin's [2007] differential intensity argument with the near-consensus in the literature that salience promotes responsiveness in general (see Kukl inski and McCrone 1980, Page and Shapiro 1983, Monroe 1998, Sharp 1999 and Burstein 2003). The theory argues that as the difference in issue intensity between the same-party mean voter and district mean voter grows (the former becoming more intense relative to the latter), members should exhibit a greater degree of partisan responsiveness on the 17 issue. However, because measuring differences in issue intensity can be difficult, I use a more readily observable proxy: party polarization. 1 6 I argue that same-party voters care more about issues on which their respective ideal points are far from those of the opposing party and care less about issues on which the two parties are generally in agreement. Therefore, to restate the argument: the more polarized the issue, the greater the degree of partisan responsiveness. On a polarized issue, the electoral benefit to members of responding more to the more passionate same-party voters outweighs the cost of responding less to less passionate average voters. Same-party voters react to the greater responsiveness by increasing the likelihood of the incumbents' renomination in the primary, by providing a more reliable bloc of voters in the general election and by contributing more generously to the incumbents' campaigns. The reaction of average voters to the lesser responsiveness, while negative, is muted because they care much less about the issue. In other words, by responding primarily to same-party voters on issues only they care intensely about, members can secure their base without seriously alienating the district as 1 7 a whole. However, the theory also argues that the electoral risk of partisan responsiveness increases with an issue's salience, defined here as the importance o f the issue in the electoral decision-making of all voters on average. On a low salience issue, the risk of 1 6 Unlike issue intensity, party polarization can be easily measured by calculating the distance between the mean opinion of Democratic voters and the mean opinion of Republican voters in a one-dimensional issue space. 1 7 The concept of party polarization used here is very similar - at least superficially - to Hurley and Hill's (2003) notion of a party-defining issue. However, the underlying logic of the two concepts is different. My theory employs polarization to measure the difference in issue intensity between same-party voters and average voters. In contrast, Hurley and Hill's justification of party-defining issues, as noted earlier, lies in the ability of rival candidates to make credible commitments to the average voter. 18 alienating average voters is minimal. The issue is not an important part of voter decision-making. However, on a high salience issue, the risk of partisan responsiveness is much higher because the issue now figures prominently in the decision-making of same-party voters and average voters. If members' continue to pursue such a strategy on high salience issues, average voters may consider voting for rival candidates who are more responsive to them. In other words, as an issue becomes more salient, the optimum reelection strategy changes from one of partisan responsiveness to one of mean voter responsiveness. These arguments yield four predictions about whom members of Congress w i l l respond to on what types of issues. They are summarized in Figure 1. In Quadrant B (high polarization-low salience issues), the theory predicts a high degree of partisan responsiveness. Here, members have a strong incentive to respond to the same-party mean because same-party voters care more intensely about such issues than average voters. A t the same time, members have only a weak incentive to respond to the district mean because such issues still do not figure prominently in the electoral decisions of voters on average. 19 Figure 1: Responsiveness predictions by issue type PARTY POLARIZATION s A: L o w partisan — ^ A B : High partisan L responsiveness responsiveness I E N C E C: Mean voter D : Moderate partisan responsiveness responsiveness In Quadrant D (high polarization-high salience issues), the theory predicts a moderate degree of partisan responsiveness. Here, members' strong incentive to respond to same-party voters is tempered by the fact that the issue is now also important to the voting decisions of all voters on average. In Quadrant A (low polarization-low salience issues), I expect a low degree of partisan responsiveness. O n issues like these, members have little incentive to respond to either same-party voters or average voters because neither group cares very much about such issues. However, because of the importance of same-party voters generally (for the reasons given earlier), the theory predicts that members of Congress wi l l still be more responsive to same-party voters than to average voters (although the difference wi l l be small). 20 Finally, in Quadrant C (low polarization-high salience issues), the theory predicts mean voter responsiveness. In this last scenario, members have a strong incentive to respond to average voters because such issues figure prominently in their decision-making. At the same time, members have only a weak incentive to respond to same-party voters because such voters do not feel intensely about such issues. The remainder of the paper is devoted to testing these predictions. Data Collection and Description The National Annenberg Election Studies are some of the largest representative sample cross-sectional surveys of the American public ever.18 Each study contains dozens of policy-related questions designed to tap public sentiment on issues from taxes to gun control to abortion to Iraq. The 2000 NAES rolling cross-section, conducted between December 1999 and January 2001, contains 58,373 respondents. The 2004 edition, in the field between October 2003 and November 2004, contains 81,422 respondents. I collect and code the data in four main steps. First, I identify those policy questions contained in either NAES for which there was a sufficiently large response.19 The minimum threshold for a question to be included is roughly no more than 10% of 1 8 However, despite the studies' random digit dialing procedure, their authors acknowledge that the data slightly underrepresents certain demographic groups (Romer et al. 2006). As a result, Egan (2007) applies weights to each congressional district - based on US Census data - to correct for the underrepresentation of young people and visible minorities. In contrast, Clinton (2006), Kaplan [2007] and Bishin [2007] do not. I too have decided not to apply weights on the assumption that politicians are likely responsive to opinion weighted by the likelihood of answering a telephone survey, which - given this NAES method - requires no correction. 1 9 The NAES do not ask every respondent every policy question. Some questions, for example, are asked for only a few weeks while others are used only in split samples. 21 districts with fewer than 15 same-party respondents. 34 policy questions satisfy this criterion (see Appendix 1). Second, I match each of these policy questions with House roll call votes. This step is complicated by the fact that the 2002 election used new congressional district boundaries redrawn following the decennial census. The 2000 N A E S uses the 2000 districts, while the 2004 N A E S employs the post-census 2002 districts in all states but Texas. 2 1 Because it is crucial is a study of this kind to map accurately the congressional districts used in the recorded votes onto those used in the public opinion data, I therefore consider only those votes that took place between January 1997 and December 2002 for the 2000 N A E S and only those votes held between January 2003 and December 2005 for the 2004 N A E S . 2 2 2 0 1 set this threshold because the marginal increase in precision for every additional respondent declines dramatically beginning with approximately 15 respondents. Furthermore, I argue that 10% of districts below 15 same-party respondents is still an acceptable level of nonsystematic measurement error in an independent variable. As it turns out, three of the four issues 1 select for analysis have less than 1% of districts below 15 same-party respondents. The fourth - the Patriot Act - has 14% but is included because it is the best example of a high polarization-low salience issue available. 2 1 Three states - Maine, Pennsylvania and Texas - not only changed their districts for the 2002 election along with the rest of the country but did so again in time for the 2004 election (US Census Bureau 2004). Of these, the changes in Maine and Pennsylvania were trivial and are ignored in this paper. However, the Texas redistricting was extensive, not only redrawing boundaries but also renumbering districts (Texas 2007). An analysis at the county level reveals that some Texas districts in the 2004 N A E S do not conform to either the 2002 or the 2004 boundaries but are instead a mix of the two. For example, Angelina County was entirely in TX-2 under the official 2002 boundaries but entirely in TX-1 under the official 2004 boundaries. In contrast, the 2004 N A E S records 14 Angelina County respondents living in TX-2 and another 10 living in TX-1. Further analysis tentatively suggests that the N A E S rolling cross-section switched from the 2002 boundaries to the 2004 boundaries around July 2004. I address this potential problem in the results section. It is also worth noting that, while Georgia also conducted this type of extensive "mid-decade" redistricting, its new districts were used beginning only in the 2006 election (US Census Bureau 2007). Consequently, the 2004 NAES accurately records all of Georgia's districts according to the post-census 2002 boundaries. 2 2 The 2000 N A E S date range is the same as that employed by Egan (2007). I exclude votes prior to January 1997 on the grounds that they are too old to be a reasonable reflection of congressional opinion at the time of the 2000 NAES. I exclude votes after 2005 because Congressional Quarterly's 2006 Congressional Roll Call publication - my primary resource for evaluating roll calls - was unavailable at the time of data collection. 22 I select all votes using the subject indexes contained in Congressional Quarterly's annual Congressional Roll Call publications. To be included for analysis, a vote must conform to the date ranges specified above and meet six further criteria. The goal is to identify, as best as possible, the true preference of each representative on each specific issue. First, a vote must be justifiably related to the text of the N A E S survey question. This is essential for an issue-by-issue comparison of responsiveness. Second, by extension, a vote must not be on a combination of issues (as is often the case, for example, with supplemental and emergency appropriations that contain funding for multiple policy areas). A member's vote must to be on a specific issue and cannot not be influenced by unrelated provisions contained the bi l l or motion. Third, it must be non-unanimous. B y definition, using non-varying data would be futile. Fourth, a vote must not be of a procedural nature (this requirement excludes votes on rules, previous questions and appeals of rulings by the chair). A member's position has to address the substance of the issue, not the procedural matters involved in its passage. Fifth, a vote must not be on a compromise where members at both extremes of the issue vote against the measure in the hopes of securing a more favourable b i l l later. Such a vote does not reveal the true preference of at least one extreme. Six, all votes included in an issue index must be correlated with each other greater than .2. This is designed both as a check on the vote selection process as well as to ensure that the issue-specific additive indexes generated from these votes are reasonably internally consistent. O f the 34 policy issues I identify in Step 1, only two do not have matching roll call votes that satisfied these 23 requirements.^ Appendix 2 contains the list of roll call votes that meet these criteria for the four issues eventually selected for analysis in this paper. A full list o f matching roll calls for all 32 N A E S policy issues is available from the author. In the third step, I select four of the remaining 32 policy issues in manner that allows me to test the theory's predictions. 2 4 Party polarization is operationalized by calculating the absolute value of the difference between the mean Democratic opinion and the mean Republican opinion nationally for each issue. 2 5 The bigger the difference, the more polarized the issue (and thus, according to the theory, the more intensely partisans should feel about it). To operationalize issue salience I borrow from Page and Shapiro (1983) who measure the concept by recording the percentage of respondents who answer "don't know." This is on the assumption that the less important the issue in voter decision-making, the higher the frequency of non-response. 2 6 Therefore, the more "don't knows" as a percentage of all responses on the issue, the less salient the issue (and thus, according to the theory, the less important the issue to all voters on average). Figure 2 plots all 32 N A E S policy issues according to these measures of polarization and salience. l i They are 1) the role of gays in the military (2000 NAES) and 2) the investment of Social Security contributions in the stock market (2004 NAES). 2 4 Not all of the 32 cases are unique policy issues however. Similar questions appear in both surveys (e.g. abortion, military spending, school vouchers, etc.). Nevertheless, because the questions were asked four years apart (and are matched against different sets of roll calls) I consider them separate cases. 2 5 All survey questions are rescaled 0 to 1. 2 6 One might argue that, given the theory, issue polarization and salience should be measured separately in each congressional district and the results aggregated into a single polarization score and a single salience score for each issue. 1 argue, however, that this alternative operationalization would not yield substantively different results from simply measuring polarization and salience at the national level. 24 Figure 2: N A E S issues by polarization and salience 25 20 c o 15 10 0.000 • • • * • • • • • • • 0.100 0.200 0.300 0.400 Absolute difference between mean Democrat and mean Republican It is worth noting at this point that these issues may not represent the full range of issue polarization and salience. In particular, Burstein (1985) observes that the use of survey data risks automatically selecting for issues with at least a modicum of salience. In other words, pollsters are unlikely to ask respondents about issues that are totally unimportant to all voters on average. Therefore, political scientists reliant on these surveys w i l l not have public opinion data on completely non-salient issues. While I agree with Burstein, I argue,that the alternatives are either a) less desirable (i.e. inferring issue-specific opinions from non-opinion data such as demographics or election results) or b) impractical at this time (such as fielding my own national survey of 30,000 to 60,000 respondents). O f the 32 issues, I select four - one for each quadrant in the theory - for further analysis. They are chosen because 1) they are suitable examples of their respective categories (i.e. reasonably close the extreme values) and 2) they each have several highly 25 relevant roll cal ls . 2 7 Figure 3 plots them along the same scale as Figure 2. The four issues are: a) free trade (low polarization-low salience), b) the Patriot Act (high polarization-low salience), c) taxes (low polarization-high salience) and d) abortion (high polarization-high salience). Trade, the Patriot Act and abortion are drawn from the 2004 N A E S and taxes from the 2000 N A E S . Figure 3: Selected issues by polarization and salience 25 20 ° 15 c o "° 10 • Trade Patriot Act 0 0.000 • Taxes • Abortion 0.100 0.200 0.300 0.400 Absolute difference between mean Democrat and mean Republican With the issues selected, the fourth and final step is to generate the dependent and independent variables. I create four separate datasets, one for each issue. For each issue, the dependent variable is an additive index of those roll calls that match the policy 1 1 1 use this second selection mechanism to decide between multiple suitable issues in a single category. While all 32 issues have relevant matching roll calls, some are more directly relevant than others. For example, in the high polarization-low salience category, both the issue of an income gap and the Patriot Act satisfy the first selection mechanism. The income gap question asks respondents whether they favour or oppose the federal government working to reduce the income gap between the rich and the poor. Matching roll calls include public housing, welfare and homelessness funding. The Patriot Act question asks respondents whether they think the Patriot Act is a good or a bad thing. The matching roll calls in this case deal with the renewal of Patriot Act itself in 2005. I consider the Patriot Act to have the more relevant roll calls because the operative term appears in both the question and the roll calls. 26 question. Consistent with Cohen and Fleisher's (2007) approach, each index counts the number of times a member voted for the liberal position (i.e. anti-free trade, anti-Patriot Act , pro-taxes or pro-abortion) and then, in order to account for absences, divides this figure by the number of votes cast by the member. 2 9 The resulting score ranges from 0 to 1 and can be interpreted as the percentage of times a member voted the liberal position (e.g. i f the member has an trade index score of .75, he voted against free trade agreements 75% of the time). Table 1 presents summary statistics for each issue index. The high Cronbach's alpha suggests that the individual roll calls form a suitably reliable scale for each issue. Table 1: Descriptive statistics and Cronbach's alpha Number of votes included Number of Values Number of members (N) Mean Standard Deviation Alpha Free trade 10 19 479 .313 .365 .934 Patriot Act 4 5 433 .452 .448 .923 Taxes 36 81 534 .432 .401 .986 Abortion 10 11 479 .391 .446 .971 I derive the independent variables from the N A E S by calculating, for each congressional district and on each of the four issues, the mean opinion of self-identified Democrats, Republicans and independents as well as the mean opinion of all voters. 3 1 For descriptive statistics of the independent variables, see Appendix 3. 2 8 1 obtain the roll calls by congress from Poole and Rosenthal's website (http://voteview.com). 2 9 1 exclude Alaska's at-large congressional district and Hawaii's two congressional districts because the N A E S does not poll in these states. 3 0 Cohen and Fleisher (2007) use this simple additive method as well as Poole and Rosenthal's (2007) more commonly-used W - N O M I N A T E computer program to each generate an index of 32 roll call votes on immigration policy. They find that the simple additive technique and the W - N O M I N A T E technique yield highly correlated results. 3 1 I exclude non-responses (i.e. "don't know" and "refused") from these calculations. 27 Two Models of Congressional Responsiveness Before presenting the analysis, it is useful to reiterate briefly what the theory does and does not address. The theory proposed in this paper deals with responsiveness, not proximity. It does not attempt to predict how close members are, on average, to the district mean opinion or to the mean opinion o f same-party voters. Instead, the theory makes issue-specific predictions about the degree of partisan and mean voter responsiveness. These predictions can be tested by examining across districts the extent to which differences in district mean and same-party mean opinions covary with differences in voting behaviour. I test the theory's predictions by estimating the following model: (Model 1) Issue-specific roll call index = intercept + Bi(district mean) + B2(same-party mean) + B3(Democrat) + e, where • Issue-specific roll call index is the proportion of times a House member voted for the liberal position (e.g. anti-trade, pro-abortion etc.), • District mean is the average opinion of all respondents in the member's district on approximately the same issue, • Same-party mean is the average opinion - on approximately the same issue - of all respondents in the member's district who self-identify with the party of the member, and • Democrat is a dichotomous variable that records the party affiliation of the member (Democrat=l, Republican=0). Bi represents the average change in the vote index for a unit change in the district mean opinion. B2 represents the average change in the roll call index for a unit change in 28 the same-party mean opinion. With this, the theory's four quadrant predictions can be condensed and expressed in mathematical terms as three readily observable predictions. They are as follows. First, B2 should be greater than Bi on the Patriot Act, abortion and free trade (i.e. members should exhibit partisan responsiveness on all three issues). Second, the difference between B2 and Bj should be largest on the Patriot Act , followed by abortion and free trade in that order (i.e. members should exhibit high partisan responsiveness on the Patriot Act, moderate partisan responsiveness on abortion and low partisan responsiveness on free trade). Third, Bi should be greater than B 2 on taxes (i.e. members should exhibit mean voter responsiveness). Table 2 presents the results of an O L S estimation of Model 1. Each column contains the estimated coefficients, and their standard errors, for each issue. The dependent variables, coded 0 to 1, record the proportion of times a member voted for the liberal position. The district mean and same-party mean are both scaled 0 to 1 with 1 being the most liberal position. This permits a straightforward interpretation of the coefficients. For example, a 1 percentage point increase in a district's mean pro-abortion position results, on average, in a 1.377 percentage point increase in the pro-abortion position of the corresponding member. In other words, as a district becomes, on average, more favourable to abortion, its representative w i l l vote the pro-abortion position more frequently. 29 Table 2: Model 1 - District mean and same-party mean responsiveness, all members Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 District mean ( b i ) .754** .452** .880*** 1 377*** (.291) (.147) (.210) (.247) Same-party mean (D2) -.235 .128 -.085 -.050 (.245) (.110) (.148) (.223) Democrat .549*** y23*** 731*** .644*** (.026) (.038) (.018) (.055) Constant -.214** -.107** _ 2\9*** - 772*** (.084) (.045) (.056) (.085) N 478 432 532 478 Adj . R 2 .532 .799 .850 .684 Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 However, Table 2 generally runs counter to the theory's predictions. None of the four same-party coefficients are statistically significant. In fact, on trade, taxes and abortion, the data suggest that members are, on average, negatively responsive to the views of same-party voters (i.e. the more liberal the average same-party voter, the less liberal the voting record of the corresponding member). These seemingly anomalous findings could, however, be the result o f 1) higher measurement error in the same-party mean variable compared with the district mean variable because of the smaller Ns used to construct the former and 2) high correlation between the district mean variable and the same-party mean variable. To address this, I remeasure the district mean variable for all four issues so that it contains roughly the same amount of measurement error as the same-party mean variable. To do so, I randomly drop respondents from the 2000 and 2004 N A E S such that the new Ns are equivalent to the respective number of Republicans in the original datasets (the smaller party in both the 2000 and 2004 samples is the Republican Party). 3 2 This reduces the 2000 and 2004 N A E S to 16,266 and 25,022 respondents respectively. 30 A t the same time, I also assess whether the same-party mean variable and the new smaller-N district mean variable meaningfully deviate from perfect correlation for all four issues. A s noted earlier, this is a key assumption of the theory. If the two variables are near-perfectly correlated, then members have no incentive to respond differently to same-party voters and average voters. Table 3 presents the correlation coefficients - for each issue - between the smaller-N district mean variable and the same-party, Democratic and Republican mean variables. The mean opinions of same-party voters and average voters are most highly correlated across districts on free trade and least so on the Patriot Act . Moreover, the two variables do not come close to perfect correlation on any issue. This suggests that the opinion variables are sufficiently independent to proceed with the analysis. 3 4 Table 3: Correlation between smaller-N district mean and variables specified below Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 Same-party mean .643 .370 .527 .584 Democratic mean .658 .482 .558 .717 Republican mean .602 .211 .385 .669 Table 4 presents the results of the O L S estimation of Model 1 substituting the new smaller-N district mean variable for the previous larger-N district mean variable. 3 5 The 3 3 For correlation matrices of alt public opinion variables, see Appendix 4. 3 4 In order to be more confident that these deviations from near-perfect correlation are substantively real (and not the result of noise due to measurement error), I also separately regress - for three of the four issues - the Democratic mean and the Republican mean on the smaller-N district mean and inspect the residuals. The results are presented in Appendix 4. I conclude from this that the independent variation in the measures of public opinion is mostly real. 3 5 Due to potential errors in Texas's district boundaries in the 2004 NAES, I also estimate Model 1 without Texas. The results are presented Appendix 5 (Tables A16 and A17). The difference in adjusted R 2 between the two estimates is negligible. Moreover, Texas's mean residuals from Model 1 (not shown) are close to the national mean. Consequently, 1 include Texas in all subsequent estimates. 31 data lend support to my explanation of Table 2's unexpected results. The same-party mean now has a statistically significant effect as well as the expected sign in three of the four issues (trade, Patriot Act and abortion). The district mean continues to have a statistically significant effect in three of the four issues (Patriot Act , taxes and abortion). Given this improvement, I use the new smaller-N district mean variable in all subsequent estimates.3 6 Table 4: Model 1 - District mean and same-party mean responsiveness, all members (smaller-N district mean) Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 District mean (bi) -.036 .226*** .450*** .509*** (.163) (.080) (.127) (.163) Same-party mean (b 2) .324* .223** .130 .618*** (.178) (.095) (.122) (.173) Democrat 519*** 711*** 717*** .511*** (.024) (.037) (.017) (.048) Constant -.074 -.047 - 127*** -.561*** (.072) (.030) (.043) (.074) N 478 432 532 478 Adj . R 2 .526 .798 .849 .670 b 2 - bi (p-value) .360 -.003 -.320 .109 .248 .984 .148 .726 Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 To test the theory's predictions, I subtract the estimated district mean coefficient (bi) from the estimated same-party mean coefficient (b 2) for each of the four issues. The last row of Table 4 presents these differences. 3 7 A positive term suggests partisan 3 6 An alternative to this reduced-N random sampling method is Stata's eivreg ("error-in-variables regression") command. Clinton (2006) partly adopts this approach. This estimation technique allows the user to specify the reliability of all independent variables used in a regression. In this case, because the independent variables are aggregates of individual-level responses, the reliabilities can be obtained using calculations developed by either Wright, Erikson and Mclver (1985) or Jones and Norrander (1996). Unfortunately, this approach is not possible here due to deadline constraints. The results, however, should be substantively similar to those presented here, apart from a gain in the efficiency of the estimates with the errors-in-variables technique. 3 7 Egan (2007) is skeptical about cross-issue coefficient comparisons because the strength of a coefficient may be strongly affected by the wording of the question. He argues that some questions may better 32 responsiveness (i.e. that, on average, members respond more to the average same-party voter than the average district voter). A negative term suggests mean voter responsiveness (i.e. that, on average, members respond more to the average district voter than the average same-party voter). The p-value is the probability that there is no difference between the two coefficients in the population. The results are generally inconsistent with the theory. There is no statistically significant difference between b 2 and bi on any of the four issues. However, because of the modest N in each dataset (i.e. between 432 and 532 members of Congress) and the fairly high correlation between the same-party mean variable and the district mean variable, it would take a large substantive difference to be statistically significant. For example, on abortion, the difference between b 2 and bi would need to be roughly .624 - a very large difference - in order to approach statistical significance at the .05 level . 3 8 Nevertheless, we can still test the predictions by cautiously interpreting the difference between b 2 and bi where appropriate and by comparing the statistical significance of the b 2 and bi coefficients themselves. Table 4 suggests partial confirmation of the first prediction. The coefficient b 2 is indeed greater than bi for abortion and trade, though not for the Patriot Act . However, the results are highly inconsistent with the second prediction. The data strongly suggest that on the Patriot Act - the issue predicted to have the highest degree of partisan responsiveness - members of Congress are equally responsive to same-party voters and measure the public's true opinion than others. I contend that by instead comparing across issues the difference between two same-issue coefficients (i.e. b2-b,) I can avoid the danger posed by variation in question wording. 3 8 1 estimate this figure by calculating a linear combination of b2-bi and doubling the standard error of the estimate (2 * .312). 33 average voters. A 1 percentage point increase in a district's same-party mean anti-Patriot Act position results, on average, in a .223 percentage point increase in the frequency with which the corresponding member votes the anti-Patriot Ac t position. The same change in the district mean results, on average, in a .226 percentage point increase. Furthermore, on free trade - the issue predicted to have the lowest degree of partisan responsiveness -members appear to be considerably more responsive to the same-party mean than the district mean. A 1 percentage point increase in a district's same-party mean anti-free trade position results, on average, in just under one-third of a percentage point increase in the frequency with which the corresponding member votes the anti-free trade position. Average voters appear to have no statistically significant effect on trade policy. Table 4 does, however, provide strong support for the third prediction - namely, mean voter responsiveness on taxes. The coefficient bi is greater than b 2 and the difference approaches statistical significance. Moreover, the district mean has a large positive effect while the effect o f the same-party mean is statistically insignificant. A 1 percentage point increase in a district's mean pro-tax position results, on average, in nearly half a percentage point increase in the frequency with which the corresponding member votes the pro-tax position. It is also worth noting here that even when controlling for the district mean and the same-party mean, members' party affiliations continues to exert a substantial effect. 3 9 For example, the average Democratic member votes the pro-tax position 71.7 percentage points more often than the average Republican member. The data strongly suggest that 3 9 Given the lack of control variables, it is also possible to interpret this as other factors correlated with congressional party but uncorrelated with local opinion having a substantial effect. This is consistent with Clinton's (2006) findings. 34 variation in district and same-party opinion does not entirely account for variation in members' voting records. A s noted earlier, Clinton (2006) finds important differences in how members of the two congressional parties respond to district subconstituencies. To address this, I reestimate Model 1 for Democratic and Republican members separately. The results are presented in Tables 5 and 6. Table 5: Model 1 - District mean and same-party mean responsiveness, Democratic members only Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 District mean (bi) -.047 .237** .632*** .724** (.311) (.116) (.233) (.282) Same-party mean (b 2) .169 .159 .071 .828*** (.356) (.121) (.211) (.307) Constant .530*** 543*** -.355** (.141) (.059) (.077) (.156) N 223 200 247 223 Adj . R 2 -.008 .048 .053 .195 b 2 - b i (p-value) .216 -.078 -.560 .104 .724 .707 .169 .850 Standard errors in parentheses. * p.< 0.10, ** p < 0.05, ***p < 0.01 The evidence for Democratic members is very inconsistent with the theoretical predictions. The data suggest partisan responsiveness on only a single issue: abortion. On average, Democratic members appear to respond slightly more to Democratic voters on the issue than they do to average voters (although there is no statistically significant difference between the two effects). However, as predicted, Democratic members exhibit a high degree of mean voter responsiveness on taxes. A 1 percentage point increase in a district's mean pro-tax position results, on average, in nearly two-thirds of a percentage point increase in the frequency with which the corresponding Democratic member votes the pro-tax position. 35 Table 6: Model 1 - District mean and same-party mean responsiveness, Republican members only Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 District mean (bi) -.008 .217* .211* .207 (.125) ( •HI ) • (.108) (.166) Same-party mean ( b 2 ) .446*** .359** .096 .404** (.132) (.163) (.115) (.171) Constant -.148*** -.070 -.024 -.265*** (.054) (.044) (.046) (.077) N 255 232 285 255 , Adj . R 2 .065 .048 .0162 .072 b 2 - bi (p-value) .454 .142 -.115 .196 .054 .534 .535 .527 Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 In contrast, the evidence for Republican members is more consistent with theoretical predictions, though not perfectly so. A s predicted, Republican members exhibit partisan responsiveness on trade, the Patriot Ac t and abortion and exhibit mean voter responsiveness on taxes. However, the degree of partisan responsiveness among the four issues again contradicts the theory's arguments. On the highly polarized but low salience Patriot Act, Republicans members appear to respond fairly equally to Republican voters and average voters. Furthermore, there is a large and statistically significant difference between the effects of the Republican mean and the district mean on low polarization-low salience trade policy. A 1 percentage point increase in a district's Republican mean anti-free trade position results, on average, in a .446 percentage point increase in the frequency with which the corresponding Republican member votes the anti-free trade position. To reiterate, Republican members appear to exhibit partisan responsiveness on more issues than their Democratic colleagues. 4 0 Controlling for the district mean, Democratic voters have a statistically significant effect on Democratic members on only This is consistent with Clinton's (2006) finding in terms of a single ideological dimension. 36 one issue (abortion). Republican voters have such an effect on three policies. However, when Democratic members do display partisan responsive behaviour on abortion, they are more than twice as responsive to their base as Republicans are to theirs. Controlling for the district mean, a 1 percentage point increase in a district's Democratic mean pro-abortion position results, on average, in a .828 percentage point increase in the frequency with which the corresponding Democratic member votes the pro-abortion position. In contrast, the same change among Republican voters produces only a .404 percentage point increase among Republican members. In sum, the data provide only limited support for the theory. The effect of the same-party mean is rarely statistically different from the effect of the district mean. When it is statistically different, as is the case for Republicans on free trade, the theory does not predict the high degree o f partisan responsiveness suggested by the data. Where the theory does appear to have some traction, however, is on taxation. It correctly predicted that members would exhibit mean voter responsiveness on such an issue. In no estimation of Model 1 - whether for all members or Democrats and Republicans separately - does the same-party mean opinion on taxes ever exert a statistically significant effect on congressional voting behaviour. Overall, though, the theory fairs poorly. While this may be attributed to methodological problems — partisan intensity and salience may be inadequately operationalized or the smaller-N means may permit too much error - the data strongly suggests the theory is incorrect or, at best, incomplete. To assist in its revision, I set aside the theory's emphasis on comparing same-party mean and district mean effects and instead model the effects of Democratic, Republican and independent mean opinion -37 conditional on the size of each group in the district electorate - on members' voting behaviour. This second model is follows: (Model 2): Issue-specific roll call index = intercept + B t (Dem mean) + B 2 ( G O P mean) + B 3 ( I n d mean) + B^percent Dem) + B5(percent GOP) + B6(percent Ind) + B 7 (Dem mean)(percent Dem) + B 8 (GOP mean)(percent GOP) + B 9(Ind mean)(percent Ind) + B] 0(Democrat) + e, where • . Dem mean is the average opinion all respondents in the district who self-identify as Democrats • GOP mean is the average opinion of all respondents in the district who self-identify as Republicans • Ind mean is the average opinion of all respondents in the district who self-identify as independents • percent Dem is the number of self-identified Democratic respondents as a percentage of all respondents in the district • percent GOP is the number of self-identified Republican respondents as a percentage of all respondents in the district • percent Ind is the number of self-identified independent respondents as a percentage of all respondents in the district, and • the remaining terms are interactions of the above variables. 4 1 However, this model - which I estimate separately for all four issues and for both congressional parties - generates a large number of coefficients which cannot be 4 1 This is a slight modification of a model in Shapiro et al. (1990). Shapiro et al. model same-party mean, opposing party mean and independent mean conditional on the size of each group in the district electorate. By presenting my results for Democratic and Republican members separately, my model is the same as theirs in all but variable names. 38 realistically presented here. 4 2 Instead, I select two tables and two figures for inclusion on the basis of substantive or unusual findings. The remainder are in Appendix 5 (Tables A l 8-A26). A l l findings - whether formally presented here or not - are briefly summarized in the text. On free trade, Democrats appear highly responsive to Republican voters (Table 7). Cells contain the OLS-estimated coefficients and their standard errors for the respective mean opinion variables at the group size specified. For example, when Republicans constitute 30% of the district electorate, a 1 percentage point increase in the mean Republican position results, on average, in a 1.066 percentage point increase in the frequency with which the corresponding Democratic member votes against free trade. A t no point do Democratic voters have a statistically significant effect. In contrast, the data suggest that Republican members are virtually exclusively responsive to Republican voters on the issue. Table 7: Model 2 - Conditional impact of partisans and independents on free trade, Democratic members only Size of group as % of district total Democrats ' Republicans Independents 20% -1.090 .179 .252 (.679) (.246) (.367) 3.0% -.705 1.066*** .630* (.427) (.361) (.345) 40% -:319 1 953*** 1.007 (.331) (.625) (.716) 50% .067 2.841*** 1.385 (.495) (.918) (1.147) Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 4 2 The econometrics literature suggests that interaction terms should be modeled with both their original variables (Kaplan [2007]). However, I also estimate a revised version of Model 2 containing only the interaction terms and present the results in Appendix 5 (Tables A27 and A28). 39 On the Patriot Act, the data suggest that when Democratic members are responsive to local opinion at all , they are attentive to independents. Republican members, once again, are heavily influence by district Republican voters. On taxes, Democratic members are responsive to Democratic and Republican voters when either group constitutes approximately 30% of the district electorate. Quite surprisingly, Republican members appear particularly influenced by Democratic voters (Table 8). Even when Democrats constitute as little as 30% of the electorate, a 1 percentage point increase in the Democratic mean results, on average, in a . 167 percentage point increase in the frequency with which the corresponding Republican member votes the pro-tax position. A t no point does the Republican mean appear to have a statistically significant effect. Table 8: Model 2 - Conditional impact of partisans and independents on taxes, Republican members only Size of group as % of district total Democrats Republicans Independents 20% -.076 .267 -.192 (.120) (.188) (.140) 30% .167** .092 -.050 (.084) (.108) (.115) 40% .410*** -.084 .092 (.149) (.207) (.237) 50% .653*** -.260 .233 (.245) (.359) (.385) Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 Abortion is particularly unusual. The data suggest that Democratic members are highly responsive to Democratic voters but to a decreasing extent as they become more numerous in the district (Figure 4). For example, at 20% of the electorate, a 1 percentage point increase in the mean pro-abortion position among Democratic voters results, on 40 average, in a 1.15 percentage point increase in the frequency with which the corresponding Democratic members votes the pro-abortion position. However, at 50% of the electorate, the same change in mean Democratic opinion prompts only a .87 percentage point change in voting behaviour. Conversely, Democratic members become increasingly responsive to Republican voters as they become more numerous in the district. Figure 4: Conditional impact of partisans and independents on abortion, Democratic members only — • — Democrats - - • — Republicans - - - A- - - Independents 20% 30% 40% 50% Size of group as % of district total '. A similar phenomenon occurs among Republican members (Figure 5). A s district Republicans grow in number, their influence on abortion wanes. Conversely, as Democrats and independents become more numerous, their influence appears to grow. Future theories of partisan responsiveness should ideally be able to account for these unusual patterns in representation. 41 Figure 5: Conditional impact of partisans and independents on abortion, Republican members only — - • — Democrats - - • - - Republicans —A- - - Independents 20% 30% 40% 50% Size of group as % of district total In concluding this section, it is important to identify briefly a potential weakness of both models: omitted variable bias. Kaplan [2007] faults Clinton (2006), for example, for failing to control for voter income and the proportions of African Americans and individuals over 65 years in each district. I too omit such controls for two reasons. First, there is a risk of considerable collinearity between these types of control variables and the measures o f constituency and subconstituency opinion used here. Parsing out the effects of each would be very difficult. Second, this potential collinearity is likely to vary by issue (i.e. income might be highly correlated with public opinion on taxes but not on abortion), rendering cross-issue comparisons impossible. Summary and Conclusion In this paper, I identified an under-developed area in the U S representation literature: cross-issue variation in responsiveness to partisan groups. To address this gap, I developed a new. theory suggesting that partisan responsiveness is partly a function 42 of the characteristics of the issue, namely an interaction between its polarization and salience. I argued that members of Congress exhibit partisan responsiveness on high polarization-low salience issues and mean voter responsiveness on low polarization-high salience issues. I tested these propositions using the unprecedented accuracy offered by the massive sample sizes of the 2000 and 2004 N A E S . The results, however, provided only limited support for the theory. While the data suggest members exhibit mean voter responsiveness on the low polarization-high salience issue of taxation, there is no statistically significant difference between the effects of the same-party mean and the district mean. In fact, statistically significant differences between the two effects occur only on trade and then only among Republican members of Congress. This latter finding is inconsistent with the theory. I then conducted further analysis of the influence of Republican, Democratic and independent voters contingent on the size of each group in the district electorate. These results suggest some unexpected patterns in responsiveness and pose interesting questions for further empirical work. 43 Bibliography Achen, Christopher H . 1977. "Measuring Representation: Perils of the Correlation Coefficient." American Journal ofPolitical Science 21 (4): 805-815. Achen, Christopher H . 1978. "Measuring Representation." American Journal of Political Science 22 (3): 475-510. Aldr ich, John H . 1995. Why Parties? The Origins and Transformation of Political Parties in America. Chicago: University of Chicago Press. Abramowitz, A lan I., Brad Alexander and Matthew Gunning. 2006. "Incumbency, Redistricting, and the Decline of Competition in U .S . House Elections." Journal of Politics 68 (1): 75-88. Ansolabehere, Stephen, James M . Snyder, and Charles Stewart III. 2001. "Candidate Positioning in U .S . House Elections." American Journal of Political Science 45 (1): 136-59. Bartels, Larry M . 1991. "Constituency Opinion and Congressional Policy Making: the Reagan Defence Buildup." American Political Science Review 85 (2): 457-74. Bartels, Larry M . 2005. Economic Inequality and Political Representation. Paper presented at the 2002 annual meeting of the American Political Science Association, Boston, M A . Baumgartner, Frank R. and Bryan D . Jones, B . D . 1993. Agendas and Instability in American Politics. Chicago: University of Chicago Press. Bishin, Benjamin G . 2000. "Constituency Influence in Congress: Does Subconstituency Matter?" Legislative Studies Quarterly 25 (3): 389-415. Bishin, Benjamin G . Forthcoming. Tyranny of the Minority: The Subconstituency Politics Theory of Representation. Draft manuscript provided by author. Burstein, Paul. 1985. Discrimination, Jobs, and Politics: The Struggle for Equal Employment Opportunity in the U.S. since the New Deal. Chicago: University of Chicago Press. Burstein, Paul. 2003. "The Impact of Public Opinion on Public Policy: A Review and an Agenda." Political Research Quarterly 56 (1): 29-40. Carmines, Edward G . and James A . Stimson. 1980. "The Two Faces o f Issue Vot ing." American Political Science Review 74 (1): 78-91. 44 Clinton, Joshua D . 2006. "Representation in Congress: Constituents and Ro l l Calls in the 106th House." Journal ojPolitics 68 (2): 397-409. Cohen, Jeffrey E . and Richard Fleisher. 2007. "Global Versus Specific Opinion and Senator Rol l Cal l Voting: The Case of Immigration Policy, 2006." Paper presented at the 2007 annual meeting of the Midwest Political Science Association, Chicago, IL. Congressional Roll Call. 1997-2005. Washington, D C : C Q Press. Downs, Anthony. 1957. An Economic Theory of Democracy. N e w York: Harper and Row. Egan, Patrick J. 2006. "Issue Ownership and Representation in the United States: A theory of legislative response to constituency opinion." Except from doctoral dissertation. Available online at http://socs.berkeley.edu/~pjegan/. Erikson, Robert S. 1978. "Constituency Opinion and Congressional Behavior: A Reexamination of the Miller-Stokes Representation Data." American Journal of Political Science 22 (3): 511-535. Erikson, Robert S. and Gerald C. Wright. 2000. "Representation of Constituency Ideology in Congress." In D . W . Brady, J.F. Cogan and M . P . Fiorina (Eds.), Continuity and Change in House Elections (pp. 149-177). Stanford, C A : Stanford University Press. Fenno, Richard F. 1973. Congressmen in Committees. Boston: Little, Brown. Fiorina, Morris P. 1977. "The Case of the Vanishing Marginals: The Bureaucracy Did It." American Political Science Review 71 (1): 177-181. Fleisher, Richard and Jon R. Bond. 2004. "The Shrinking Middle in Congress." British Journal of Politics 34: 429-451. Glynn, Carroll J., Susan Herbst, Garrett J. O'Keefe and Robert Y . Shapiro. 1999. Public Opinion. Boulder, C O : Westview Press. Grofman, B . 2004. "Downs and Two-Party Convergence." Annual Review of Political Science 7: 25-46. Hal l , Richard L . 1996. Participation in Congress. New Haven: Yale University Press. House of Representatives R o l l Calls [data]. Available online at http://www.voteview.com. 45 Hurley, Patricia A . 1991. "Partisan Representation, Realignment, and the Senate in the 1980s." Journal of Politics 53 (1): 3-33. Hurley, Patricia A . and K i m Q. H i l l . 2003. "Beyond the Demand-Input Model : A Theory of Representational Linkages." Journal of Politics 65 (2): 304-326. Jacobs, Lawrence R. and Robert Y . Shapiro. 2000. Politicians Don't Pander: Political Manipulation and the Loss of Democratic Responsiveness. Chicago: University of Chicago Press. Jones, Bradford S. and Barbara Norrander. 1996. "The Reliability of Aggregated Public Opinion Measures." American Journal of Political Science 40(1): 295-309. Jones, Bryan D . 1999. "Bounded Rationality." Annual Review of Political Science 2: 297-321. Kaplan, Noah. Forthcoming. " O f Models and Measures: Constituency Influence in the 106th House." Draft manuscript provided by author. Kukl inski , James H . and Donald J. McCrone. 1980. "Policy Salience and the Causal Structure of Representation." American Politics Research 8 (2): 139-164. Mayhew, David R. 1974. Congress: the Electoral Connection. New Haven: Yale University Press. Manza, Jeff and Fay Lomax Cook. 2002. "The Impact of Public Opinion on Public Policy." In Jeff Manza, Fay Lomax Cook and Benjamin .1. Page (Eds.), Navigating Public Opinion: Polls, Policy, and the Future of American' Democracy (pp. 17'-32). Oxford: Oxford University Press. Mil ler , Warren E . and Donald E . Stokes. 1963. "Constituency Influence in Congress." American Political Science Review, 57 (1): 45-56. Monroe, Alan D . 1979. "Consistency between Public Preferences and National Policy Decisions." American Politics Quarterly 7 (1): 3-19. Monroe, Alan D . 1998. "Public Opinion and Public Policy, 1980-1993." Public Opinion Quarterly, 62 (6): 6-28. National Annenberg Election Surveys, 2004 and 2006 [data]. Available in Daniel Romer, Kate Kenski, Kenneth Winneg, Christopher Adasiewicz and Kathleen Hal l Jamieson. 2006. Capturing Campaign Dynamics, 2000 and 2004: the National Annenberg Election Survey. Philadelphia: University o f Pennsylvania Press. 46 Page, Benjamin I. and Robert Y . Shapiro. 1983. "Effects of Public Opinion on Policy." American Political Science Review, 11 (1): 175-190. Petry, Francois and Matthew Mendelsohn. 2004. "Public Opinion and Policy Making in Canada 1994-2001." Canadian Journal of Political Science 37 (3): 505-529. Poole, Keith T. and Howard Rosenthal. 1997. Congress: A Political-Economic History of Roll Call Voting. New York: Oxford University Press. Proximity One. 2007. 110th Congress Congressional District Demographics. Available online at proximityone.com/cdl 1 Ortl .htm. Quirk, Paul J. and Joseph Hinchliffe. 1998. "The Rising Hegemony of Mass Opinion." Journal of Policy History 10 (1): 19-50. Romer, Daniel, Kate Kenski, Kenneth Winneg, Christopher Adasiewicz and Kathleen Hal l Jamieson. 2006. Capturing Campaign Dynamics, 2000 and 2004: the National Annenberg Election Survey. Philadelphia: University of Pennsylvania Press. Shapiro, Catherine R., David W . Brady, Richard A . Brody and John A . Ferejohn. 1990. "Linking Constituency Opinion and Senate Voting Scores: A Hybrid Explanation." Legislative Studies Quarterly, 15 (4): 599-621. Sharp, Elaine B . 1999. The Sometime Connection: Public Opinion and Social Policy. Albany: State University of N e w York Press. Stimson, James A . , Michael B . MacKuen, and Robert S. Erikson. (1995). "Dynamic Representation." American Political Science Review, 89 (3): 549-565. State of Texas. 2007. Texas Redistricting. Available online at www.tlc.state.tx.us/redist/redist.htm. U S Census Bureau. 2004. 109th Congressional District Summary Files. Available online at factfinder.census.gov/jsp/saff/SAFFInfo.jsp?_pageId=sp4_decennial_cdl09. U S Census Bureau. 2007. 110th Congressional District Summary Files. Available online at factfinder. census. go v/j sp/saff/S AFFInfo .j sp? _pageld=sp4_decennial_cd 110&_su bmenuld=&ds_name=&_ci_nbr=&qr_name=&_industry=. Wlezien, Christopher and Stuart N . Soroka. Forthcoming. "The Relationship between Public Opinion and Policy." In R. Dalton and H . Klingemann (Eds.), Oxford Handbook of Political Behaviour. Oxford: Oxford University Press. 47 Wright, Gerald C. 1989. "Policy Voting in the U . S . Senate: Who is Represented?" Legislative Studies Quarterly, 14 (4): 465-486. Wright, Gerald C. , Robert S. Erikson, and John P. Mclver . 1985. "Measuring State Partisanship and Ideology with Survey Data." Journal of Politics 47 (2): 469-89. 48 Appendix 1 - List of NAES policy questions The following is a list o f the 34 N A E S policy questions that meet the criterion of roughly no more than 10% of districts with fewer than 15 same-party respondents. The list is sorted in declining order of national sample size (N). „ Table A 1: N A E S policy questions Year Original Variable Name Question Wording N % of districts with same-party N <15 2004 cceOl The federal government banning all abortions—do you favor or oppose the federal government doing this? 63496 0 2004 ccc39 The federal government giving tax credits or vouchers to help parents send their children to private schools—do you favor or oppose the federal government doing this? 63062 0 2004 cce21 Would you favor or oppose an amendment to the U.S . Constitution saying that no state can allow two men to marry each other or two women to marry each other? 62175 0 2004 ccb82 The federal government negotiating more free trade agreements like N A F T A — d o you favor or oppose the federal government doing this? 55838 0 2000 cbd02 Give tax credits or vouchers to help parents send their children to private schools— should the federal government do this or not? . 54886 0 2000 cbbOl The amount of money Americans pay in taxes—is this an extremely serious problem, serious, not too serious or not a problem at all? 54822 0 2000 cbpOl The amount of poverty in the United States—is this an extremely serious problem, serious, not too serious or not a problem at all? 54795 0 2000 cbj07 Maintaining a strong military defense— should the federal government spend more money on this, the same as now, less or no money at all? 54780 0 49 Year Original Variable Name Question Wording N % of districts with same-party N <15 2000 cbg06 Restricting the kinds of guns that people can buy—should the federal government do more about this, the same as now, less or nothing at all? 54599 0 2000 cbmOl Trying to stop job discrimination against blacks—should the federal government do more about this, the same as now, less or nothing at all? 54192 0 2000 cbf02 Make it harder for a woman to get an abortion—should the federal government do this or not? 54177 0 2000 cbcOl Social Security benefits—should the federal government spend more money on this, the same as now, less or no money at all? 54089 0 2000 cbgl2 The number of criminals who are not punished enough—is this an extremely serious problem, serious, not too serious or not a problem at all? 54001 0 2000 cbe02 Providing health care for people who do not already have it—should the federal government spend more money on this, the same as now, less or no money at all? 53973 0 2000 cbhOl Limi t the amount of money that can be given to political parties—should the federal government do this or not? 53540 0 2000 cbl05 Trying to stop job discrimination against homosexuals—should the federal government do more about this, the same as now, less or nothing at all? 53136 0 2000 cbb05 Which do you personally think is more important, cutting taxes or strengthening the Social Security system? 46890 1 2004 ccd35 Do you think the U.S . should keep military troops in Iraq until a stable government is established there, or do you think the U .S . should bring its troops home as soon as possible? 46736 1 2004 ccc32 and ccc33 Do you favor or oppose allowing workers to invest some of their Social Security contributions in the stock market? O R Al lowing workers to invest some of their 46466 0 50 Year Original Variable Name Question Wording N % of districts with same-party N <15 Social Security contributions in the stock market—do you favor or oppose this? 2004 ccbl3 The federal government reducing federal taxes—do you favor or oppose the federal government doing this? 40451 3 2004 ccc41 The federal government trying to reduce the income differences between rich and poor Americans—do you favor or oppose the federal government doing this? 39139 3 2004 cce32, cce33 and cce34 3 similarly worded questions: one which asks whether Congress should extend assault weapons ban and two others which ask whether respondent favours assault weapons ban 39118 • 2 2000 cbg05 Do you personally favor or oppose requiring a license for a person to buy a handgun? 32924 7 2000 cbgOl Do you personally favor or oppose the death penalty for some crimes? 32877 7 2000 cbe08 Do you personally favor or oppose using government funds to make sure that every child in the U S is covered by health insurance? 32358 9 2004 ccc40 Providing financial assistance to public elementary and secondary schools—should the federal government spend more on it, the same as now, less, or no money at all? 31450 7 2000 cbel4 Do you personally favor or oppose giving patients the right to sue their health maintenance organization or H M O ? 31414 9 2004 ccd03 Mili tary defense—should the federal government spend more on it, the same as now, less, or no money at all? 31207 7 2000 cblOl Do you personally favor or oppose allowing homosexuals to serve openly in the United States military? 31048 10 2004 ccd57 Homeland security—should the federal government spend more on it, the same as now, less, or no money at all? 30863 8 51 Year Original Variable Name Question Wording N % of districts with same-party N <15 2004 ccd34 Rebuilding Iraq—should the federal government spend more on it, the same as now, less, or no money at all? 30644 8 2000 cbc05 Do you personally favor or oppose allowing workers to invest some of their Social Security contributions in the stock market? 28391 14 2000 cbh02 Do you personally favor or oppose a ban on all soft money campaign contributions? 27182 18 2004 ccd67 Which comes closer to your view—would you say the Patriot Act is a good thing for America or a bad thing for America? 26703 14 52 Appendix 2 - Roll call votes used in the dependent variable indexes The following lists are compiled using the subject indexes of Congressional Quaterly's annual Congressional Roll Call publications, 1997-2005. The titles and descriptions are adapted from C Q , the Library of Congress (http://thomas.loc.gov) and C - S P A N (http://www3.capwiz.com/c-span/dbq/issuesdbq/votesearch.dbq). The position recorded at the conclusion of each description is the liberal position on each vote (i.e. anti-free trade, anti-Patriot Act, pro-taxes and pro-abortion). Table A 2: House roll call votes used in free trade index Issue 1: Free trade (2004): The federal government negotiating more free trade agreements like N A F T A — d o you favor or oppose the federal government doing this?) Year Date Roll call number CQ Title Description Yea-Nay 2003 July 24 430 H R 2 7 3 9 . U S -Singapore Trade/ Question of Engrossment and Third Reading Third reading of bi l l to implement free trade agreement with Singapore. Nay = -anti-free trade. 309-114 2003 July 24 431 H R 2 7 3 9 . U S -Singapore Trade / Motion to reconsider Motion to k i l l another motion that would have the House reconsider its vote on third reading of the bi l l . Nay = anti-free trade. 269-153 2003 July 24 432 H R 2 7 3 9 . U S -Singapore Trade/ Passage Passage of b i l l to implement free trade agreement with Singapore. Nay = anti-free trade 272-155 2003 July 24 434 H R 2 7 3 8 . U S -Chile Trade/ Question of Engrossment and Third Reading Third reading of b i l l to implement free trade agreement with Chile. Nay = anti-free trade. 299-129 2003 July 24 435 H R 2 7 3 8 . U S -Chile Trade / Motion to reconsider Motion to k i l l another motion that would have the House reconsider its vote on third reading of the b i l l . Nay = anti-free trade. 276-152 53 Issue 1: Free trade (2004): The federal government negotiating more free trade agreements like N A F T A — d o you favor or oppose the federal government doing this?) Y e a r Date R o l l call number C Q Tit le Description Yea-Nay 2003 July 24 436 H R 2 7 3 8 . U S -Chile Trade/ Passage Passage of bi l l to implement free trade agreement with Chile. Nay = anti-free trade 270-156 2004 July 14 375 H R 4759. U S -Australia Trade / Passage Passage of b i l l to implement free trade agreement with Australia. Nay = anti-free trade 314-109 2004 July 22 413 H R 4 8 4 2 . U S -Morocco Trade / Passage Passage of b i l l to implement free trade agreement with Morocco. Nay = anti-free trade 323-99 2005 July 28 443 H R 3045. Dominican Republic -Central America Free Trade Agreement ( C A F T A ) / Passage Passage of bi l l to implement free trade agreement with Dominican Republic and 5 Central American countries. Nay = anti-free trade 217-215 2005 Dec. 7 616 H R 4 3 4 0 . U S -Bahrain Trade/ Passage Passage of bi l l to implement free trade agreement with Bahrain. Nay = anti-free trade 327-95 54 Table A 3: House roll call votes used in Patriot Act index Issue 2: Patriot Act (2004): Which comes closer to your view—would you say the Patriot Ac t is a good thing for America or a bad thing for America? Year Date Roll call number CQ Title Description Yea-Nay 2005 June 15 258 Fiscal 2006 Science, State, Justice, and Commerce Appropriations / Sanders Amendment Prohibits funding to search library and bookstore sales records. Yea = anti-Patriot Act . 238-187 2005 July 21 413 Patriot Act Reauthorization / Recommit Recommits with instructions to apply 4-year sunset clauses to all key provisions. Yea = anti-Patriot Act . 209-218 2005 July 21 414 Patriot Act Reauthorization / Passage Extends or makes permanent most of the original Patriot Act ' s provisions. Nay = anti-Patriot Act. 257-171 2005 Dec. 14 627 Patriot Ac t Reauthorization / Conference Report Extends or makes permanent most of the original Patriot Act ' s provisions. Nay = anti-Patriot Act. 251-174 Table A 4: House roll call votes used in tax index Issue 3: Tax (20C extremely serious ] 0): The amount of money Americans pay in taxes—is this an problem, serious, not too serious or not a problem at all? Year Date Roll call number CQ Title Description Yea-Nay 1997 A p r i l 15 78 H J Res. 62. Tax Limitation Constitutional Amendment / Passage Constitutional amendment requiring two-thirds vote in both houses to raise taxes. Nay = pro-taxes. 233-190 55 Issue 3: Tax (200 extremely serious 0): The amount of money Americans pay in taxes—is this an problem, serious, not too serious or not a problem at all? Year Date Roll call number CQ Title Description Yea-Nay 1997 May 21 144 H Con Res. 84. Fiscal 1998 Budget Resolution / Conservative Action Team Substitute Multiple provisions, one of which increases the net tax cut from $85 bil l ion to $192.5 bil l ion over 5 years. Nay = pro-taxes. 119-313 1997 June 26 245 H R 2 0 1 4 . Fiscal 1998 Budget Reconciliation -Revenue/ Passage Passage of b i l l that provides net tax cut of $85 bil l ion over 5 years. Nay = pro-taxes. 253-179 1998 Apr i l 22 102 H J Res. 111. Tax Limitation Constitutional Amendment / Passage. Constitutional amendment requiring two-thirds vote in both houses to raise taxes. Nay = pro-taxes. 238-186 1998 June 5 208 H Con Res. 284. Fiscal 1999 Budget Resolution / Conservative Action Team Substitute Multiple provisions, one of which calls for $150 bil l ion in tax cuts over 5 years. Nay = pro-taxes. 158-262 1998 Sept. 26 469 H R 4 5 7 9 . Tax Cuts / Passage Passage of b i l l that provides $80.1 bil l ion in tax cuts over 5 years. Nay = pro-taxes. 229-195 1999 March 25 77 H Con Res. 68. Fiscal 2000 Budget Resolution / Adoption Multiple provisions, one of which calls for $778.5 bil l ion in tax cuts over 10 years. Nay = pro-taxes. 221-208 56 Issue 3: Tax (20C extremely serious 0): The amount of money Americans pay in taxes^—is this an problem, serious, not too serious or not a problem at all? Year Date Rol l call number CQ Title Description Yea-Nay 1999 Apr i l 15 90 Ft J Res. 37. Tax Limitation Constitutional Amendment / Passage Constitutional amendment requiring two-thirds vote in both houses to raise taxes by more than an 'insignificant' amount. Nay = pro-taxes. 229-199 1999 July 22 333 H R 2 4 8 8 . Tax Reconciliation / Passage B i l l provides $792 bi l l ion in tax cuts over 10 years. Nay = pro-taxes. 223-208 1999 Aug. 5 379 H R 2 4 8 8 . Tax Reconciliation / Conference Report B i l l provides $792 bi l l ion in tax cuts over 10 years. Nay = pro-taxes. 221-203 2000 Feb.10 15 H R 6 . Alleviate "Marriage Penalty" Tax / Passage B i l l cuts taxes for married couples by approx. $182 bi l l ion over 10 years. Nay = pro-taxes. 268-158 2000 March 9 41 H R 3 0 8 1 . Tax Revisions / Passage Mult iple provisions, one of which provides $122.7 bi l l ion in tax cuts over 10 years. Nay = pro-taxes. 257-169 2000 March 23 73 H Con Res. 290. Fiscal 2001 Budget Resolut ion/ Sununu Substitute Mult iple provisions, one of which increases tax cuts above the level provided in the original resolution. Nay = pro-taxes. 78-339 57 Issue 3: Tax (200 extremely serious 0): The amount of money Americans pay in taxes—is this an problem, serious, not too serious or not a problem at all? Y e a r Date R o l l call number C Q Tit le Description Yea-Nay 2000 A p r i l 12 119 H J R e s . 94. Tax Limitation Constitutional Amendment / Passage Constitutional amendment requiring two-thirds vote in both houses to raise taxes by more than an 'insignificant' amount except in times of war. Nay = pro-taxes. 234-192 2000 May 10 155 H R 3709. Internet Tax Moratorium / Long-Term Extension Provides 99-extension on moratorium on state and local taxes on the Internet. Nay = pro-taxes. 90-336 2000 June 9 254 H R 8. Estate Tax Repeal / Passage B i l l completely phases out estate and gift taxes by 2010. Nay = pro-taxes. 279-136 2000 July 12 392 H R 4810. "Marriage Penalty" Tax Relief / Passage B i l l cuts taxes for married couples by approx. $182.3 bil l ion over 10 years. Nay = pro-taxes. 269-159 2000 July 20 418 H R 4 8 1 0 . Alleviate "Marriage Penalty" Tax / Conference Report B i l l cuts taxes for married couples by approx. $89.8 bil l ion over 5 years. Nay = pro-taxes. 271-156 2000 Sept. 7 458 H R 8. Estate Tax Repeal / Veto Override Overrides President Clinton's veto of the estate and gift taxes repeal. Nay = pro-taxes. 274-157 2000 Sept. 13 466 H R 4 8 1 0 . Alleviate "Marriage Penalty" Tax / Veto Override Overrides President Clinton's veto of "marriage penalty" tax relief. Nay = pro-taxes. 270-158. 58 Issue 3 : T a x (20C extremely serious 0): The amount o f money Amer i cans pay i n taxes—is this an prob lem, serious, not too serious or not a p rob lem at a l l? Y e a r D a t e R o l l c a l l n u m b e r C Q T i t l e D e s c r i p t i o n Y e a - N a y 2001 M a r c h 8 45 H R 3 . Income T a x Reduc t ion / Passage B i l l cuts income taxes by restructuring the tax brackets. N a y = pro-taxes. 230-198 2001 M a r c h 28 68 H C o n Res . 83. F i s c a l 2002 Budget Reso lu t ion / F lake Substitute M u l t i p l e prov is ions , one o f w h i c h increases the tax cut f rom $1.6 t r i l l ion to $2.2 t r i l l i on over ten years. N a y = pro-taxes. 81-341 2001 M a r c h 28 70 H C o n Res . 83. F i sca l 2002 Budget Reso lu t ion / A d o p t i o n M u l t i p l e prov is ions , one o f w h i c h cal ls for $1.6 t r i l l i on in tax cuts over 10 years. N a y = pro-taxes. 222-205 2001 M a r c h 29 75 H R 6 . Mar r iage tax Reduc t i on / Passage B i l l cuts taxes for marr ied couples by $399.2 b i l l i on over 10 years. N a y = pro-taxes. 282-144 . 2001 A p r i l 4 84 H R 8. Estate Tax R e l i e f / Passage B i l l complete ly phases out estate and gift taxes by 2011 . N a y = pro-taxes. 274-154 2001 A p r i l 25 87 H J R e s . 4 1 . T a x L im i ta t i on Const i tu t ional Amendmen t / Passage Const i tu t ional amendment requi r ing two-thirds vote i n both houses to raise taxes by more than an ' ins ign i f i cant ' amount except i n t imes o f war. N a y pro-taxes. 232-189 2001 M a y 16 118 H R 1836. T a x -Cu t Reconc i l i a t i on B i l l / Passage B i l l prov ides for $958.3 b i l l i on i n tax cuts over 11 years. N a y = pro-taxes. 230-197 59 Issue 3: Tax (20C extremely serious 0): The amount of money Americans pay in taxes—is this an problem, serious, not too serious or not a problem at all? Year Date Roll call number C Q Title Description Yea-Nay 2001 May 26 149 H R 1836. Tax Cuts Reconciliation B i l l / Conference Report B i l l provides for $1.35 tril l ion in tax cuts over 10 years. Nay = pro-taxes. 240-154 2002 Feb. 6 10 H Con Res. 312. Tax Support / Adoption Sense of House that 2001 tax cuts should not be suspended or repealed. Nay = pro-taxes. 235-181 2002 Apr i l 18 103 H R 586. Permanent Tax Cuts / Concur With Senate Amendments Mot ion to concur in House amendment to Senate amendments that makes the 2001 tax cuts permanent. Nay = pro-taxes. 229-198 2002 June 6 219 H R 2 1 4 3 . Estate Tax Repeal / Passage B i l l permanently extends repeal of estate and gift taxes. Nay = pro-taxes. 256-171 2002 June 12 225 H J Res. 96. Tax Limitation Constitutional Amendment / Passage Constitutional amendment requiring two-thirds vote in both houses to raise taxes by more than an 'insignificant' amount except in times of war. Nay = pro-taxes. 227-178 2002 June 13 229 H R 4019. Married Couples Tax Rel ief / Passage B i l l extends tax cuts for married couples beyond 2010. Nay = pro-taxes. 271-142 2002 Sept. 4 371 H R 5203. Education Tax Break / Passage B i l l makes permanent recent changes to education savings accounts. Nay = pro-taxes. 213-118 60 Issue 3: Tax (200 extremely serious 0): The amount of money Americans pay in taxes—is this an problem, serious, not too serious or not a problem at all? Year Date Roll call number CQ Title Description Yea-Nay 2002 Sept. 19 401 H Res. 524. House Support for Estate Tax Repeal / Adoption Sense of the House that Congress should pass the repeal of the estate and gift taxes before it adjourns. Nay = pro-taxes. 242-158 2002 Oct. 2 430 H R e s . 543. House Support for Married Couples Tax Relief / Adoption Sense of the House that Congress should pass legislation permanently extending tax cuts for married couples. Nay = pro-taxes. 285-130 Table A 5: House roll call votes used in abortion index Issue 4: Abortion (2004): The federal government banning all abortions—do you favor or oppose the federal government doing this? Year Date Roll call number Title Description Yea-Nay 2003 May 22 215 H R 1588. Fiscal 2004 Defense Authorization / Abortion as Mili tary Facilities Al lows privately-funded abortions at U S military facilities abroad. Yea = pro-abortion. 201-227 2003 June 4 241 "Partial-Birth" Abortion Ban / Recommit Recommits bi l l with instruction to allow "partial-birth" abortions to preserve life or health of mother. Yea = pro-abortion. 165-256 2003 June 4 242 "Partial-Birth" Abortion Ban / Passage Bans "partial-birth" abortions unless necessary to save mother's life. Nay = pro-abortion. 282-139 61 Issue 4: Abortion (2004): The federal government banning all abortions—do you favor or oppose the federal government doing this? Year Date Roll call number Title Description Yea-Nay 2003 Oct. 2 530 "Partial-Birth" Abortion Ban / Conference Report Bans "partial-birth" abortions unless necessary to save mother's life. Nay = pro-abortion. 281-142 2004 May 19 197 H R 4 2 0 0 . Fiscal 2005 Defense Authorization / Abortion at Mili tary Facilities Al lows privately funded abortions at U S military facilities abroad for women in military or military dependents. Yea = pro-abortion. 202-221 2005 A p r i l 27 141 Abortion Notification / Professional, Transportation and Medical Provider Exemption Exempts taxi and bus drivers and medical professionals from criminal prosecution for transporting a minor across state lines to obtain an abortion. Yea = pro-abortion. 179-245 2005 Apr i l 27 142 Abortion Notification / Grandparent and Clergy Exemption Exempts minor's grandparents and members of the clergy from prosecution for transporting minor across state lines to obtain an abortion. Yea = pro-abortion. 177-252 2005 A p r i l 27 143 Abortion Notification/ Recommit Recommits b i l l with instructions to eliminate the right of the father to sue i f he caused the pregnancy through rape or incest. Yea = pro-abortion. 183-245 62 Issue 4 : A b o r t i o n (2004): The federal government banning all abortions—do you favor or oppose the federal government doing this? Y e a r D a t e R o l l c a l l n u m b e r T i t l e D e s c r i p t i o n Y e a - N a y 2005 Apr i l 27 144 Abortion Notification / Passage Criminalizes the transportation of abortion-seeking minors across state lines in circumvention of state law requiring parental notification. Nay = pro-abortion. 270-157 2005 May 25 216 Fiscal 2006 Defense Authorization / Abortion at Mili tary Facilities Al lows privately funded abortions at U S military facilities abroad. Yea = pro-abortion. 194-223 63 Appendix 3 - Descriptive statistics Table A 6: Descriptive statistics for variables in free trade analysis Cases Mean Standard Deviation Min Max R o l l call index 479 .313 .365 0 1 District mean (larger-N) 484 .512 .074 .333 .748 District mean (smaller-N) 484 .510 .096 .250 .797 Same-party mean 483 .499 .088 .213 .818 Democratic mean 484 .529 .086 .324 .781 Republican mean 484 .475 .097 .179 .818 Independent mean 484 .525 .096 .205 .833 % Democratic 484 32.895 9.302 15.116 68.447 % Republican 484 30.143 9.478 3.636 54.070 % Independent 484 26.612 6.203 10.833 55.102 Democratic Member 483 .464 .499 0 1 Table A 7: Descriptive statistics for variables in Patriot Act analysis Cases Mean Standard Deviation Min Max R o l l call index 433 .452 .448 0 1 District mean (larger-N) 437 .393 .092 .177 .730 District mean (smaller-N) 437 .392 .137 .000 .861 Same-party mean 436 .356 .207 .000 .875 64 Democratic mean 437 .531 ,138 .125 .882 Republican mean 437 .196 .111 .000 .800 Independent mean 437 .436 .147 .000 1.000 % Democratic 437 32.979 9.396 15.116 68.447 % Republican 437 29.934 9.597 3.636 54.070 % Independent 437 26.675 6.310 10.833 55.102 Democratic Member 436 .461 .499 0 1 Table A 8: Descriptive statistics for variables in tax analysis Cases Mean Standard Deviation Min Max R o l l call index 534 .432 .401 0 1 District mean (larger-N) 536 .392 .048 .271 .585 District mean (smaller-N) 536 .392 .065 .227 .598 Same-party . mean 534 .387 .083 .181 .672 Democratic mean 536 .438 .076 .252 .672 Republican mean 536 .344 .062 .083 .593 Independent mean 536 .400 .400 .167 .592 % Democratic 536 32.042 9.694 12.707 65.766 % Republican 536 27.138 8.527 3.333 50.000 % Independent 536 28.335 6.710 12.613 60.185 Democratic Member 534 .466 .499 0 1 65 Table A 9: Descriptive statistics for variables in abortion analysis Cases Mean Standard Deviation Min Max Rol l call index 479 .391 .446 0 1 District mean (larger-N) 484 .649 .092 .403 .956 District mean (smaller-N) 484 .654 .106 .383 .983 Same-party mean 483 .614 .166 .260 .964 Democratic mean 484 .750 .100 .452 .964 Republican mean 484 .509 .110 .083 .969 Independent mean 484 .694 .100 .365 .948 % Democratic 484 32.895 9.302 15.116 68.447 % Republican 484 30.143 9.478 3.636 54.070 % Independent 484 26.612 6.203 10.833 55.102 Democratic Member 483 .464 .499 0 1 66 Appendix 4 - Correlations between opinion variables Table A 10: Correlation matrix for opinion variables in free trade analysis District mean (larger-N) District mean (smaller-N) Same-party mean Democratic mean Republican mean Independent mean District mean (larger-N) 1 District mean (smaller-N) .809 1 Same-party mean .807 .643 1 Democratic mean .802 .658 .677 1 Republican mean .767 .602 .681 .472 1 Independent mean .803 .659 .502 .516 .480 1 Table A l l : Correlation matrix for opinion variables in Patriot Act analysis District mean (larger-N) District mean (smaller-N) Same-party mean Democratic mean Republican mean Independent mean District mean (larger-N) 1 District mean (smaller-N) .626 1 Same-party mean .625 .370 1 Democratic mean .650 .482 .402 1 Republican mean .318 .211 .176 .048 1 Independent mean .578 .300 .163 .250 -.055 1 67 Table A 12: Correlation matrix for opinion variables in tax analysis District mean (larger-N) District mean (smaller-N) Same-party mean Democratic mean Republican mean Independent mean District mean (larger-N) 1 District mean (smaller-N) .752 1 Same-party mean .691 .527 1 Democratic mean .792 .558 .580 1 Republican mean .493 .385 .387 .200 1 Independent mean .695 .531 .320 .385 .211 1 Table A 13: Correlation matrix for opinion variables in abortion analysis District mean (larger-N) District mean (smaller-N) Same-party mean Democratic mean Republican mean Independent mean District mean (larger-N) 1 District mean (smaller-N) .873 1 , Same-party mean .702 .584 1 Democratic mean .809 .717 .508 1 Republican mean .748 .669 .553 .483 1 Independent mean .778 .669 .391 .574 .488 1 To test whether the independent variation evident above is substantively real (and not the result of measurement error), I separately estimate and record the residuals of the following two models: Democratic mean = intercept + Rx(smaller-N district mean) + e, and 68 Republican mean = intercept + fli(smaller-N district mean) + e. I then identify four pairs of similar congressional districts on the basis of geographic proximity and comparable demographic characteristics. This is done using commercially repackaged 2000 Census data (Proximity One 2007). Mississippi 1 and 2 are rural Southern districts. N e w Jersey 7 and 11 are wealthy Northeastern districts. Minnesota 2 and 3 are highly educated Midwestern districts. California 36 and 40 are urban Western districts. If the residuals for these pairs of similar districts are significantly different, then I cannot rule out the possibility that deviations from near-perfect correlation are due to noise. . Tables A14 and A15 list the residuals for the selected districts. Unfortunately, because the Census data are not readily available for the 2000 districts, I am unable to compare the residuals for the issue of taxes. A majority of paired districts have residuals with the same sign, suggesting that the difference between the actual and predicted partisan means is substantively similar in similar districts (i.e. both above or both below the regression line). For example, the Republican mean in Mississippi 1 is slightly more anti-trade than expected i f predicted solely on the basis of the district mean (a residual of .029). This is also the case in Mississippi 3 (a residual of .044). Furthermore, several paired districts with differently signed residuals are fairly close to the regression line (i.e. both residuals are near zero). Few paired districts have very different residuals. I conclude from this that the independent variation in the opinion variables is mostly real. Table A 14: Democratic mean residuals by selected pairs of similar districts Trade Patriot Act Abortion MS-01 .002 -.064 -.112 MS-03 .069 .029 .034 NJ-07 -.035 -.168 -.003 NJ-11 -.019 -.202 .005 M N - 0 2 .013 .006 .098 MN-03 .088 -.067 .005 C A - 3 6 .023 .083 .024 CA-40 -.028 .028 .107 Table A 15: Republican mean residuals by selected pairs of similar districts Trade Patriot Act Abortion MS-01 .029 .025 -.045 MS-03 .044 -.063 .021 NJ-07 .109 .198 .089 NJ-11 .062 -.103 .083 M N - 0 2 -.030 -.104 -.067 MN-03 -.016 -.018 -.091 C A - 3 6 .001 .069 .079 C A - 4 0 .134 .194 -.062 A p p e n d i x 5 - S u p p l e m e n t a r y tab les f r o m reg ress ion ana lyses 69 Table A 16: Model 1 - District and same-party mean responsiveness, all members without Texas (smaller-N district mean) F r e e t r a d e 2004 P a t r i o t A c t 2004 T a x e s 2000 A b o r t i o n 2004 Dis t r ic t mean (bi) - .009 . 1 8 8 * * .428** - * . 4 8 5 * * * (.169) (.085) (.130) (.173) Same-party mean ( b 2 ) .280 . 2 1 3 * * .122 . 6 3 8 * * * (.182) (.100) (.123) (.183) Democra t 7 2 2 * * * 7 1 6 * * * . 5 1 0 * * * (.025) (.039) (.018) (.051) Constant - .064 - .030 - . 1 1 5 * * - . 5 5 6 * * * (.075) (.031) (.044) (.078) N 438 400 500 437 A d j . R 2 .55 .804 .850 .674 b 2 - b i (p-value) .289 .024 -.305 .153 .368 .878 .173 .645 Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 Table A 17: Change in adjusted R 2 (without Texas - with Texas) F r e e t r ade 2004 P a t r i o t A c t 2004 T a x e s 2000 A b o r t i o n 2004 A l l representatives .024 .005 .001 .004 Table A 18: Model 2 - Partisan and independent mean responsiveness with interactions, all representatives F r e e t r a d e 2004 P a t r i o t A c t 2004 T a x e s 2000 A b o r t i o n 2004 Democra t i c mean -.212 .070 .142 - .500 (.556) (.271) (.298) (.512) G O P mean -.543 -.114 .041 - .180 (.380) (.198) (.265) (.288) Independent mean .004 .166 .231 -1 .115* (.632) (.309) (.458) (.584) % Democra t i c ,001 -.002 .003 - . 0 3 1 * * * (.010) (.005) (.004) (.012) % Repub l i can - . 0 1 7 * * - . 0 0 8 * * - .005 - 0 2 3 * * * (.007) (.004) (.004) (.006) % Independent - .002 -.003 .002 - . 0 4 7 * * * (.013) (.006) (.007) (.016) D e m mean * % Democra t i c .005 .001 .003 . 0 3 5 * * (.016) (.008) (.008) (.015) 70 G O P mean .030** .012 .008 .015 * % Republican (.013) (.008) (.012) (.010) Ind mean * .008 .002 -.003 .051** % Independent (.024) (.012) (.017) (.022) Democrat .464*** 735*** .669*** 542*** (.031) (.027) (.017) . (.030) Constant .369 .283 -.140 2.309*** (.638) (.333) (.300) (.768) N 478 432 .532 . 478 Adj . R 2 .543 .800 .858 .700 Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 Table A 19: Model 2 - Partisan and independent mean responsiveness with interactions, Democratic members only Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 Democratic -1.862 .165 • .799 1.338 mean (1.282) (.500) (.597) (1.035) G O P mean -1.595** .087 -.353 -.956** (.702) (.238) (.426) (.406) ' Independent -.504 -.119 .747 -.631 mean (1.181) (.358) (.818) (.883) % -.012 -.0001 .011 -.007 Democratic (.018) (.008) (.007) (.021) % -.045*** -.008 -.014 -.053*** Republican (.016) (.005) (.009) (.011) % -.014 -.005 .010 -.048* Independent (.024) (.007) (.014) (.025) Dem mean .039 -.0009 ' -.013 -.009 * % Democratic (.032) (.0128) (.015) (.026) G O P mean 089* * * -.004 .033 .058*** * % Republican (.031) (.013) (.024) (.019) Ind mean * .038 .011 -.017 .037 % Independent (.045) (.014) (.031) (.033) Constant 2.170* 1.070** .095 2.410* (1.230) (-447) (.557) (1.296) N 223 200 247 223 • Adj . Rz .077 .121 .130 .315 Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 71 Table A 20: Model 2 - Partisan and independent mean responsiveness with interactions, Republican members only Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 Democratic mean .225 .678 -.561* -.641 (.570) (.469) (.312) (.624) G O P mean .724 -.081 .619 1.419* (.678) (.831) (.496) (.795) Independent mean .243 .485 -.476 -1.730** (.543) (.623) (.425) (.766) % Democratic -.002 .009 -.009* -.017 (.011) (.010) (.005) (.016) % Republican .003 -.003 .005 .017 (.010) (.007) (.006) (.011) % Independent .003 .004 -.004 -.038* (.012) (.0109) (.006) (.021) Dem mean * % Democratic -.003 -.024 .024** .033 (.020) (.017) (.011) (.022) G O P mean * % Republican -.010 .0166 -.018 -.034 (.018) (.024) (.017) (.022) Ind mean * % Independent -.008 -.010 .014 .0672** (.021) (.024) (.016) (.029) Constant -.345 -.373 .273 .582 (.755) (.586) (.329) (1.119) N 255 232 285 255 Adj . Rl .052 .047 .066 .221 .Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 Table A 21: Model 2 - Conditional impact of partisan and independent means on Patriot Act, Democratic members only Size of group as % of district total Democrats Republicans Independents 20% .147 .008 .096 (.257) (.110) (.116) 30% .139 -.031 .204* (.151) (.192) (.115) 40% .130 -.070 .311 (.111) (.305) (.225) 50% .122 -.110 .419 (.186) (.425) (.356) Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 72 Table A 22: Model 2 - Conditional impact of partisan and independent means on taxes, Democratic members only Size of group as % of district total Democrats Republicans Independents 20% .539 .299 .406 (.330) (.202) (.270) 30% .410* .625* .236 (.223) (.369) (.254) 40% .280 .951 .065 (.183) (.585) (.499) 50% .150 1.277 -.105 (.245) (.812) (.791) Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 Table A 23: Model 2 - Conditional impact of partisan and independent means on abortion, Democratic members only Size of group as % of district total Democrats Republicans Independents 20% 1.149** .196 .109 (.535) (.185) (.293) 30% 1.054*** 773*** .480* (.324) (.272) (.262) 40% .960*** 1.349*** .850 (.253) (.429) (.520) 50% .866** 1.926*** 1.220 (.404) (.603) (.832) Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 Table A 24: Model 2 - Conditional impact of partisan and independent means on trade, Republican members only Size of group as % of district total Democrats Republicans Independents 20% .174 .522 .086 (.204) (.327) (.161) 30% .149 .421** .008 (.131) (•174) (.154) 40% .123 .320** -.071 (.265) (.142) (.332) 50% .098 .219 -.150 (.449) (.276) (.533) Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 73 Table A 25: Model 2 - Conditional impact of partisan and independent means on Patriot Act, Republican members only Size of group as % of district total Democrats Republicans Independents 20% .191 .250 .294* (.156) (.376) (.166) 30% -.052 .416** .199 (.114) (.187) (.152) 40% -.296 .581*** .103 (.242) (.203) (.368) 50% -.539 .747* .008 (.402) (.399) (.604) Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 Table A 26: Model 2 - Conditional impact of partisan and independent means on abortion, Republican members only Size of group as % of district total Democrats Republicans Independents 20% .017 .730** -.386 (.229) (.369) (.236) 30% .346** .386** .285 (.150) (.190) (.203) 40% .675** .041 .957** (.292) (.185) (.441) 50% 1.003** -.303 1.629** (.490) (.361) (.717) Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 Table A 27: Model 2 - Interaction terms only, Democrats only Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 Dem mean * % Democratic .012*** 007*** .013*** 0 2 i * * * (.004) (.002) (.003) (.003) G O P mean *% Republican -.004 -.006 -.004 .012** (.005) (.005) (.005) (.006) Ind mean * % Independent .010* .007** .008** .009*** (.005) (.003) (.004) (.003) Constant .263* .688*** .556*** -.181 (.141) (.061) (.091) (.155) N 223 200 247 223 Adj . Rz .051 .113 .115 .208 Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01 74 Table A 28: Model 2 - Interaction terms only, Republicans only Free trade 2004 Patriot Act 2004 Taxes 2000 Abortion 2004 Dem mean * '% Democratic .003* -.0009 .006*** 009*** (.002) (.003) (.002) (.002) G O P mean * % Republican 007*** .015*** -.001 .0004 (.002) (.004) (.002) (.003) Ind mean * % Independent .005** .008** .002 012*** (.002) (.003) (.002) (.002) Constant .194*** -.105 .010 -.358*** (.063) (.067) (.048) (.081) N 255 232 285 255 Adj . Rz .057 .060 .050 .158 Standard errors in parentheses. * p < 0.10, ** p < 0.05, ***p < 0.01