Race and Political Inequality in America: How Much and Why?

Race and Political Inequality in America: How Much and Why? John D. Griffin Assistant Professor Department of Political Science University of Notre Dame Griffin.58@nd.edu Brian Newman Assistant Professor Department of Political Science Pepperdine University bpn@duke.edu Paper prepared for the 2004 annual meeting of the Midwest Political Science Association, Chicago, IL, April 15-18.

Race and Political Inequality in America: How Much and Why? Abstract The U.S. prizes political equality, yet its citizens may not be represented equally. Concerned observers have long believed that racial minorities interests, in particular, are underrepresented in governmental decisions. We examine the degree to which the preferences of whites and African Americans correspond with the actions of their representatives in office. We find that across a wide array of policies, African Americans are severely underrepresented. However, on some issues salient to African Americans, they are marginally better represented than whites. We also discover that these disparities in representation are related to participation in elections. Senators tend to represent white voters better than white non-voters, but African American voters are not similarly rewarded, an asymmetry that accounts for part of the overall representation gap. 1

The U.S. prizes political equality, yet its citizens may not be represented equally. As Sidney Verba (2003, 663) put it, the equal consideration of the preferences and interests of all citizens is one of the bedrock principles in a democracy. However, concerned observers have long believed that racial minorities interests, in particular, are underrepresented in governmental decisions. Through much of American history, national, state, and local governments denied African Americans the full rights of citizenship, enshrining political inequality into law. Although the U.S. has taken steps to increase African Americans political power, and public officials have become more responsive to African Americans in recent decades (e.g., Keech 1968; Bullock 1981), "few disagree that blacks continue to be underrepresented in federal, state, and local government" (Guinier 1994, 8). This apparent failure on the bedrock democratic principle of political equality has prompted a variety of proposals for improving African Americans representation, ranging from mobilizing racial minorities in elections (e.g., Piven and Cloward 1988; Rosenstone and Hansen 1993), to the creation and maintenance of electoral districts where African Americans constitute a majority of constituents ("majority-minority" districts), to the adoption of cumulative voting (Guinier 1994), proportional representation (Lijphart 1998), and even compulsory voting (Lijphart 1997). Political debate over these reforms has largely guided recent studies of race and representation, with scholars focusing on the related questions of how representation changes as African Americans constitute greater portions of electoral constituencies, and the representational impact of majority-minority districts (Cameron, Epstein, and O Halloran 1996). However, the natural and healthy focus on these questions has left other important questions unexplored. In particular, while these studies show that African Americans are better represented as their numbers increase and when they are represented by other African 2

Americans, we do not know whether African Americans are equally represented compared to whites. In addition, the focus on the percentage of a constituency that is African American tends to treat African Americans as a monolithic political community rather than a relatively cohesive but still heterogeneous group. Finally, if indeed African Americans are underrepresented, prior work does not tell us whether the gains produced, for example, by the election of African Americans are adequate or insufficient to close the representation gap. Before we formulate or re-engineer public policies in an attempt to enhance representation of racial minorities, we must comprehend more fully both the extent and sources of political inequality. We address these issues by asking to what degree the opinions of whites and African Americans correspond with the actions of their representatives in office. Specifically, we evaluate whether the preferences of whites and African Americans are equally likely to correspond with the roll call behavior of their U.S. Senators in the 101 st to 107 th Senates (1989-2002). We find that across the full spectrum of policies, African Americans are severely underrepresented. However, their opinions are not underrepresented in every case. On some issues of special concern to African Americans, they are marginally better represented than whites. We also explore whether these disparities in representation are attributable to unequal rates of participation in elections, finding that turnout accounts for some but not all of the representation gap. These findings inform the race and representation literature and the consideration of public policies designed to promote political equality in the United States. Background In order for African Americans and whites to be represented unequally, the two groups must have different political preferences. There is no doubt that they do. In fact, Kinder and 3

Sanders (1996, 27) conclude from years of opinion surveys that the racial divide in opinion is a divide without peer. African Americans and whites differ dramatically on explicitly racerelated policies like affirmative action, preferential hiring, college racial quotas, and equal employment policies, and disagree to a lesser but still substantial degree on implicitly racerelated policies like federal funding for education, health care, social security, law enforcement, and other social welfare policies (e.g., Kinder and Sanders 1996; Canon 1999; Kinder and Winter 2001). However, it is important to note that while African Americans are more politically cohesive than whites, there is significant variation among African Americans on a variety of political issues (e.g., Ellcessor and Leighley 2001; McClain and Stewart 2002; see Dawson 1994, 73ff for a review of literature dealing with the ways class shapes opinions among African Americans). Still, African Americans relatively cohesive and distinctive opinions open the possibility that their views are underrepresented. Scholars have long been concerned with this possibility, and have examined the ways elected officials respond to their African American constituents. Most studies of race and representation assess how the concentration of African Americans in electoral districts shapes various political outcomes. For instance, a raft of works have investigated how the racial composition of electoral districts affects public policies (e.g., Keech 1968; Radcliff and Saiz 1995), Democrats electoral success (Grofman, Griffin, and Glazer 1992), and legislators activities in office, typically their roll call behavior (Combs, Hibbing, and Welch 1984; Whitby 1985; Herring 1990; Grofman, Griffin, and Glazer 1992; Cameron, Epstein, and O'Halloran 1996; Lublin 1997; Canon 1999). These studies, which usually focus on the U.S. House, find that as the African American population in a district grows, Representatives vote first more 4

conservatively and then (with further growth) more liberally, although these patterns differ by region, district urbanicity, and legislators party affiliation and race (Combs, Hibbing, and Welch 1984; Grofman, Griffin, and Glazer 1992; Cameron, Epstein, and O'Halloran 1996; Canon 1999). In a similar vein, Bullock (1981) also found that the presence of African American voters altered Representatives behavior, showing that the advent of voting rights in the South produced a noticeable change in the tendency of Southern Representatives to support liberal positions and the roll call alternatives supported by the Leadership Conference on Civil Rights (LCCR). Even though concern with all officials representation of African Americans remains a central question, the introduction of racial redistricting and the consequent rise in the number of majority-minority districts sparked interest in related questions namely, whether African American officials represent African American constituents appreciably better than white representatives, and whether the benefits of descriptive representation come at significant costs in substantive representation. On one hand, majority-minority districts have boosted the number of African American representatives in the U.S. House (Davidson and Grofman 1994). On the other, concentrating African Americans in electoral districts dilutes their presence in surrounding districts, potentially leading to an increase in the number of more conservative Republicans elected and a decline in the number of more liberal Democrats, ultimately depressing the substantive representation of African Americans (Cameron, Epstein, and O Halloran 1996). As a result, the degree to which representation is enriched when African Americans represent African Americans has become a critical question. Some argue that white officials can represent African Americans just as well as an African American official can (Swain 1993), while others argue that descriptive representation enhances minority representation on a variety of fronts (e.g., Canon 1999; Mansbridge 1999; Tate 2003). 5

In summary, the race and representation literature has tended to focus on representatives reactions to the racial composition of their districts, and differences in the ways African American and white officials represent their African American constituents. In both approaches, analysts compare the representation of African Americans in some districts to the representation of African Americans in other districts. In contrast, we compare the relative representation of African Americans and whites across all districts. We ask whether Senators voting decisions reflect whites preferences better than African Americans preferences, an important question that has attracted little scholarly notice. One reason scholars have not compared the relative representation of white and African American preferences has to do with methods and data, a point we discuss below. Another may simply be that scholars generally assume whites are politically advantaged. After all, there are plenty of reasons to expect racial disparities in representation. First, whites are far better represented descriptively (e.g., Tate 2001). This descriptive representation may often translate into unequal substantive representation, since white legislators often represent their African American constituents differently than do African American legislators (e.g., Canon 1999). Second, there are simply more whites than African Americans in most electoral districts, and elected officials probably pay more attention to larger constituency subgroups. Third, whites tend to be more politically active on a variety of measures (e.g., Rosenstone and Hansen 1993; Verba, Schlozman, and Brady 1995), and political activity tends to reap representational rewards (e.g., Martin 2003; Griffin and Newman 2004). Despite these reasons to expect that whites are better represented, this may not always be the case. Conventional wisdom argues that politicians pay attention to constituents that are especially concerned with particular issues. We often refer to women s issues or racial 6

issues and typically assume that politicians respond disproportionately to the preferences of women and racial minorities when acting in these issue domains. Thus, we might expect Senators votes to correspond more closely to African Americans opinions for issues of greater concern to them. Whites may be advantaged in many issue domains, but representation may be more balanced for issues especially important to African Americans. Conceptualizing and Measuring Representation Representation is a complex, multifaceted phenomenon. Studying such a complex thing forces scholars to examine different facets of representation. Following the lead of Miller and Stokes seminal empirical study of congressional representation, many works examined the extent to which representatives respond to some summary statistic of their district s characteristics, like its mean opinion (Miller and Stokes 1963; Achen 1979; Erikson, McIver, and Wright 1993) or some proxy for mean opinion (Erikson, Luttbeg, and Halloway 1975; McCrone and Kuklinski 1979; Bullock and Brady 1983). Representatives behavior, usually roll call votes, is modeled as a function of their districts mean opinion. This approach provides a measure of legislators' responsiveness to district opinion, estimating the extent to which liberals represent the most liberal constituencies and conservatives represent conservative constituencies. It is designed to determine how much control constituents have over legislators activities. This design could be adopted to study the control various groups within each district exert. For example, Bartels (2002) used a variant of this approach to determine that Senators respond disproportionately to constituents with higher incomes. As a practical matter, though, the data requirements for this framework are extremely difficult to meet when studying groups within geographical constituencies, like racial groups. It is difficult to obtain a reliable measure 7

of a district s mean African American opinion because there are relatively few African Americans in electoral districts, meaning most surveys do not draw representative samples of them in many districts. Even in our study of U.S. Senators, where electoral districts equal entire states, pooling the General Social Survey (GSS) from 1973 to 2000 provides only 32 states with samples of 20 or more African Americans. Predictably, these small samples generally provide unreliable estimates of state-level African American ideology and opinion. In fact, we employed the generalizability measure Jones and Norrander (1996) present and found that state-level mean African American ideology and mean African American preferences concerning federal spending on social welfare, crime, education, health care, and racial issues were unreliable. Only African Americans attitudes toward abortion provided a reliable state-level measure (r =.88). 1 Therefore, we perform a traditional (i.e., state-level) analysis for votes on abortion policy, modeling Senators votes as a function of the average opinion among their states African Americans and whites, but not for votes in other issue areas or aggregate measures of voting behavior like NOMINATE scores. Instead, we examine another aspect of representation, the correspondence between individual-level opinion and roll call votes by Senators. Rather than asking how Senators respond to state-level opinion, we ask: across the nation, whose opinions most closely correspond to their Senators votes? Assessing opinion-vote correspondence provides a descriptive measure of one aspect of representation. Furthermore, it is important in its own right because the extent of opinion correspondence between constituents and Senators influences constituents evaluations of Senators performance (Binder, Maltzman, and Sigelman 1998), and 1 The coefficient ranges from 0 to 1. Jones and Norrander (1996, 302) classify values over.7 as highly reliable, between.6 and.7 as moderately reliable, and below.6 as unreliable. Items other than the abortion opinions often yielded coefficients below.5. State means for whites abortion attitudes were highly reliable, r =.96. 8

these evaluations are one indicator of how well individuals feel they are being represented (Tate 2001). However, we recognize that we are taking a limited look at representation, focusing on but one of many facets of a complex phenomenon. Focusing on roll call votes means ignoring other activities like bill sponsorship, speeches on the floor, working to convince other members to vote a particular way, and many other important endeavors (Canon 1999). Furthermore, the up or down nature of roll call votes is a blunt measure of Senators activities. However, roll call votes are an important substantive activity Senators perform and assessing the extent to which these votes correspond to African Americans and whites provides one way of ascertaining whether these groups are equally represented. Methods and Data To assess the relative degree of correspondence between Senators votes on one hand and whites and African Americans opinions on the other, we model each Senator s vote (or aggregate measure of votes like W-NOMINATE scores) as a function of his or her white and African American constituents opinions, controlling for each respondent s race. The model takes the form (1) y jk = a + b 1 x 1ik + b 2 x 2ik + b 3 x 3ik, where y jk is the roll call decision(s) of Senator j (j = 1, 2) from state k, a is a constant, x 1ik is the opinion of white constituent i in state k, x 2ik is the opinion of African American constituent i in state k, x 3ik is the race (African American or white) of constituent i in state k, and b 1, b 2, and b 3 are parameters to be estimated. Note that since we exclude respondents who are not African 9

American or white, and by definition x 1 = 0 for African Americans and x 2 = 0 for whites, this setup is equivalent to using a standard multiplicative term as in the following equation: (2) y jk = a + b 1 x 1ik + b 2 x 2ik + b 3 x 1ik x 2ik, where x 1 is constituent opinion regardless of race, x 2 is the constituent s race and x 1 x 2 is the product of opinion and race. We present results in the former setup because it provides a direct view of Senator correspondence with whites and African Americans. A hypothesis test that b 1 = b 2 in the first equation is equivalent to the hypothesis test that b 3 = 0 in the second equation. This model is not intended to explain Senators vote decisions, but to measure the correspondence between Senators votes and constituents opinions/ideologies. In additional analyses for abortion votes, where aggregation of data provides reliable measures of state-level African American and white opinion, we also model Senators votes as a function of their state s mean African American opinion and mean white opinion. In both frameworks, if the ideologies and opinions of whites correspond better with the behavior of their Senators than do those of African Americans, we can conclude that African Americans are underrepresented. Both the traditional approach and our individual-level approach require measures of Senators roll call voting and African Americans and whites ideologies and opinions in the policy domain of the roll call vote. As a measure of Senators aggregate voting behavior for each of the 101 st to 107 th Congresses we use their W-NOMINATE coordinates, which range continuously from 1 to 1. This version of NOMINATE is appropriate given that it permits variation in legislators' scores across terms served. Studies of congressional roll call voting have frequently employed these and alternative versions of NOMINATE scores as dependent variables (e.g., Ansolabehere, Snyder, and Stewart 2001; Bartels 2002). As a practical matter, W-NOMINATE coordinates are highly correlated with alternative, summary measures of 10

legislator roll call behavior such as interest group ratings and Heckman-Snyder scores (Burden, Caldeira, and Groseclose 2000). We also model Senator roll call behavior on specific votes in different issue domains. First, we examine abortion votes because abortion is an issue that likely is salient for both racial groups and because African Americans opinions on abortion tend to be more conservative than those of whites. Some observers attribute African Americans underrepresentation to the fact that "one of the two major parties - the Republicans - has all but explicitly stated that it is willing to have itself regarded a white party (Hacker 1992, 201; see also Frymer 1999). Because Republicans generally advocate pro-life positions, this is one issue on which many African- Americans and many Republicans agree, so perhaps African Americans will not be as severely underrepresented on abortion votes. Second, we analyze votes on issues generally deemed especially salient to African Americans racial inequality, crime, health care, education, and welfare to examine whether votes correspond more to African Americans opinions on these issues. To pinpoint roll calls on abortion, we relied on the National Right to Life Committee (NRLC), which has identified 26 "Key Votes" since the organization began this practice in 1996. These include 10 roll calls in the 105 th Senate, 8 in the 106 th, 1 in the 107 th, and 7 in the 108 th. 2 To identify roll calls in domains salient to African Americans, we relied on the LCCR, which designated 15 Key Votes during the 105 th to the 107 th Congresses. A summary of the LCCR votes we included appears in Appendix B. For measures of constituent opinion, we use the GSS from 1973-2000, which samples from 44 states. 3 Although the GSS does not draw state-level samples, Brace, et al. (2002) found that it produces highly reliable measures of state opinion. Using the GSS abortion items, it was 2 A summary of the NRLC Key Votes can be provided by the authors on request. 3 The GSS does not survey citizens in Hawaii, Idaho, Nebraska, Nevada, New Mexico, and South Dakota. 11

possible to create reliable state-level measures of white and African American abortion opinion by pooling over time. We use these pooled opinions to model Senators NRLC roll call votes (see Appendix A for GSS question wording). 4 For our individual-level models of correspondence, we used the GSS ideological selfplacement item in models of Senators' W-NOMINATE scores, GSS abortion items in models of the same NRLC abortion votes, and GSS items related to civil rights and social welfare policies (attitudes regarding increasing or decreasing federal spending to improve conditions for African Americans, education, health care, and welfare assistance, and to reduce crime) in models of LCCR votes. We excluded NRLC or LCCR votes for which there was no related opinion item. For example, we excluded NRLC votes not explicitly relating to abortion (e.g., votes on euthanasia or lobbying restrictions). To determine whether GSS items were sufficiently related to LCCR roll call votes, we independently rated the nexus for the roll calls and the GSS federal spending items from zero to five. Only those roll calls receiving an average nexus rating of at least four were included in the analysis (see Appendix B for a list of LCCR votes and related GSS items). Some bills related to more than one GSS item. In these cases, we estimate separate models for each relevant opinion item. W-NOMINATE scores are continuous, so they were estimated using OLS; the NRLC and LCCR votes are binary, so these roll calls were estimated using probit. In the individuallevel models, for each GSS respondent we map the roll call behavior of each of the respondent's Senators onto the citizen's preferences. This results in models that contain two, and sometimes three, observations per GSS respondent. To account for this relatively minor non-independence of observations, we report robust standard errors. We also weight these estimations to account 4 Average state sample sizes were 751 and 138 for whites and African Americans respectively. States with fewer than 20 African Americans in the sample were excluded from the state-level analysis. All respondents were included in the individual-level analyses. 12

for disparities between the GSS sample and actual state populations, using Census data. In all estimations, we only include GSS responses that were acquired prior to the roll call or W- NOMINATE score that we model to limit the possibility that citizens opinions are affected by their Senators roll call decisions. All roll calls, NOMINATE scores, and GSS items are coded so that conservative alternatives are higher, so positive estimates for white and African American opinion indicate positive correspondence. Results Before we report estimates of the correspondence between constituents and their Senators votes, we note that, as expected, African American and white GSS respondents held significantly different opinions. Table 1 shows that, as a group, African Americans are significantly more liberal in their general orientation toward politics, more pro-life on abortion policy, and more supportive of increased federal spending on education, fighting crime, health care, welfare, and addressing racial inequality. For many of these items, African Americans opinions were more cohesive than whites, as the smaller standard deviations indicate. However, this is not always the case, and substantial standard deviations on all items remind us of important variation among African Americans, which makes the use of actual opinion data rather than blunt indicators like the percent of a district that is African American more appealing when available. 13

Table 1 Racial Differences in Opinion Whites African Americans Mean S.D. Mean S.D. Range Ideology 4.16 1.33 3.80 1.42 1-7 Abortion 1.34 0.33 1.42 0.34 1-2 Increase federal spending on: Education 1.49 0.65 1.28 0.49 1-3 Fighting Crime 1.37 0.59 1.29 0.57 1-3 Health Care 1.43 0.61 1.27 0.52 1-3 Race 1.99 0.71 1.20 0.44 1-3 Welfare 2.39 0.74 1.81 0.82 1-3 Data: GSS 1973-2000 All differences significant at the.0001 level. All variables coded so conservative (pro-life, less spending) values are higher. We begin our analysis of correspondence with the broad array of issues, modeling W- NOMINATE scores using constituents general ideology. As shown in Table 2, these models indicate that, across the full spectrum of issues, African Americans' ideologies correspond much less often with the behavior of their Senators compared to the ideologies of whites. In fact, in the first four Congresses, there is no relationship at all between African Americans ideologies and Senators' aggregate roll call decisions. P values of tests for parameter equality (tests that b 1 = b 2 in equation 1) show that whites were significantly better represented in all seven congresses. To compare the magnitude of the relationships between constituents ideologies and their Senators votes, we simulated the difference between the expected W-NOMINATE score of the most liberal whites Senators and the most conservative whites Senators. We then compare this to the difference in the expected score of the most liberal African Americans Senators and the most conservative African Americans Senators. The difference between these marginal effects provides an indication of the degree to which Senators votes correspond better with whites 14

opinions. 5 The results of these simulations reveal that at best, in a particular Congress the correspondence of African Americans ideologies with their Senators roll call decisions is half that of whites. On average, the correspondence of African Americans ideologies is only about one-third that of whites ideologies. Table 2 Correspondence Between Ideologies and Senator W-NOMINATE Scores Weighted Congress 101 st 102 nd 103 rd 104 th 105 th 106 th 107 th Average White Ideology 0.019 ** 0.018 ** 0.022 ** 0.035 ** 0.035 ** 0.037 ** 0.036 ** 0.030 [0.002] [0.002] [0.002] [0.003] [0.002] [0.003] [0.002] Black Ideology -0.001 0.002 0.005 0.006 0.015 ** 0.015 * 0.015 ** 0.009 [0.005] [0.005] [0.005] [0.006] [0.006] [0.006] [0.005] Black 0.072 ** 0.067 ** 0.109 ** 0.128 ** 0.098 ** 0.080 ** 0.076 ** 0.090 [0.024] [0.023] [0.023] [0.028] [0.025] [0.026] [0.023] Constant -0.282 ** -0.363 ** -0.434 ** -0.244 ** -0.226 ** -0.269 ** -0.309 ** -0.298 [0.011] [0.010] [0.010] [0.012] [0.010] [0.011] [0.010] Marginal Effects 0.115 0.109 0.133 0.210 0.210 0.224 0.216 0.183-0.008 0.013 0.030 0.036 0.090 0.089 0.090 0.056 p-value of Coef..000 0.005 0.004.000 0.001.000.000 Equality Test N 36,198 42,772 45,384 54,174 58,008 63,956 68,211 52,672 Robust standard errors in brackets + denotes p <.10; * p <.05; ** p <.01 5 Note that we are interested in the relative correspondence of whites and African Americans attitudes more than the level of either marginal effect on its own. We could calculate this just as easily using a smaller increment than minimum (most liberal) to maximum (most conservative) shift in opinion. Most of the marginal effects we report were simulated using CLARIFY (Tomz, Wittenberg, and King 2003). 15

Figure 1 illustrates these results, plotting predicted Senator W-NOMINATE scores for constituents with different ideologies and races. The predicted W-NOMINATE scores change relatively little as African Americans ideologies become more conservative, moving only.06 points. Scores change much more dramatically for whites, moving about.19 points. We can also conclude from Figure 1 that the predicted Senator W-NOMINATE score is roughly equivalent for whites who are only slightly conservative and for African Americans who are extremely conservative. To provide some tangible examples of the size of a (near).19 difference, in the 107 th Senate, the W-NOMINATE score of Sen. Max Baucus (D-MT) was -.27, while that of Sen. James Jeffords (I-VT) was -.054; the W-NOMINATE score of Edward Kennedy (D-MA) was -.94, while that of Earnest Hollings (D-SC) was -.73. Senators roll call votes across a range of issues correspond much more to whites general ideological perspectives than they do to African Americans general views. Figure 1: Correspondence of Ideology and Senator W-Nominate Score 0 Extremely Liberal Liberal Slightly Liberal Moderate Slightly Conservative Conservative Extremely Conservative -0.05-0.1 Pr(W-Nominate) -0.15-0.2-0.25-0.3 Whites African Americans 16

We also examine the correspondence between Senators roll calls on specific bills and citizens opinions on a variety of issues. In addition to the research design benefits to doing so, outlined above, a methodological reason to examine specific domains is that the ideological scale s meaning is not always clear for African Americans, who tend to be liberal on economic issues and conservative on social issues (McClain and Stewart 2002, 66-70). Therefore, the ideology measure may be more noisy for African Americans, depressing the amount of estimated correspondence. The items on specific policies may be more meaningful and clear, presumably making them more comparable for whites and African Americans. We begin with models of Senators abortion votes in the 105 th to 108 th Congresses as a function of their constituents' opinions at the individual level. As Table 3 shows, in every case, both whites and African Americans abortion opinions were significantly related to Senators votes. Tests of whether the coefficients for whites and African Americans differ show that Senators votes corresponded significantly more with whites than African Americans in 10 of the 25 votes (see column second from right). For these probit models, the marginal effects report the difference in the probability that the most pro-life whites (or African Americans ) Senators voted in a pro-life fashion as compared to the most pro-choice whites (African Americans ) Senators. For the 10 votes where Senators corresponded with whites opinions more, Senator correspondence with African Americans opinions was on average about 75 percent that of whites. At best on these votes, African Americans correspondence was 83 percent of whites and at worst, just over 60 percent. In the specific domain of abortion policy, Senators frequently vote more in line with the preferences of whites than those of African Americans. 17

Table 3 Preference-Vote Correspondence on Senate Abortion Votes Senate Marginal Effects Test RC # White Opinion Black Opinion Race Constant Whites Blacks p-value N 105 th Congress RC13 0.41 ** [0.02] 0.42 ** [0.04] 0.00 [0.07] -0.65 ** [0.02] 0.16 0.17 0.82 60,560 RC71 0.38 ** [0.02] 0.36 ** [0.04] 0.11+ [0.07] -0.32 ** [0.02] 0.15 0.14 0.56 60,560 RC129 0.42 ** [0.02] 0.48 ** [0.04] -0.01 [0.07] -0.45 ** [0.02] 0.16 0.18 0.20 60,560 RC167 0.41 ** [0.02] 0.37 ** [0.04] 0.09 [0.07] -0.59 ** [0.02] 0.16 0.15 0.39 60,109 RC190 0.41 ** [0.02] 0.38 ** [0.04] 0.06 [0.07] -0.57 ** [0.02] 0.16 0.15 0.56 60,085 RC215 0.33 ** [0.02] 0.22 ** [0.04] 0.12+ [0.07] -0.75 ** [0.02] 0.13 0.08 0.01 60,414 RC105 0.40 ** [0.02] 0.41 ** [0.04] -0.02 [0.07] -0.55 ** [0.02] 0.16 0.16 0.82 60,560 RC176 0.46 ** [0.02] 0.38 ** [0.04] 0.10 [0.07] -0.60 ** [0.02] 0.18 0.15 0.07 56,378 RC277 0.38 ** [0.02] 0.36 ** [0.04] 0.11+ [0.07] -0.32 ** [0.02] 0.15 0.14 0.56 60,560 RC282 0.36 ** [0.02] 0.33 ** [0.04] 0.05 [0.07] -0.49 ** [0.02] 0.14 0.13 0.45 58,987 106 th Congress RC148 0.44 ** [0.02] 0.39 ** [0.04] 0.06 [0.07] -0.68 ** [0.02] 0.18 0.16 0.25 60,560 RC197 0.44 ** [0.02] 0.39 ** [0.04] 0.08 [0.07] -0.68 ** [0.02] 0.18 0.15 0.22 60,117 RC336 0.43 ** [0.02] 0.41 ** [0.04] 0.01 [0.07] -0.70 ** [0.02] 0.17 0.16 0.64 60,161 RC337 0.44 ** [0.02] 0.41 ** [0.04] 0.02 [0.07] -0.71 ** [0.02] 0.17 0.16 0.61 60,033 RC338 0.45 ** [0.02] 0.41 ** [0.04] 0.00 [0.07] -0.74 ** [0.02] 0.18 0.16 0.31 59,011 RC340 0.46 ** [0.02] 0.37 ** [0.04] 0.17* [0.07] -0.45 ** [0.02] 0.18 0.15 0.05 59,011 RC134 0.44 ** [0.02] 0.39 ** [0.04] 0.06 [0.07] -0.68 ** [0.02] 0.18 0.15 0.25 60,201 107 th Congress RC160 0.42 ** [0.02] 0.28 ** [0.04] 0.17 * [0.07] -1.04 ** [0.03] 0.15 0.09 0.00 53,477 108 th Congress RC47 0.49 ** [0.02] 0.32 ** [0.04] 0.24 ** [0.07] -0.69 ** [0.02] 0.20 0.13 0.00 59,844 RC48 0.44 ** [0.02] 0.32 ** [0.04] 0.23 ** [0.07] -0.90 ** [0.02] 0.17 0.12 0.01 60,229 RC49 0.54 ** [0.02] 0.42 ** [0.04] 0.18 ** [0.07] -0.70 ** [0.02] 0.21 0.17 0.01 58,326 RC51 0.55 ** [0.02] 0.43 ** [0.04] 0.21 ** [0.07] -0.68 ** [0.02] 0.22 0.17 0.01 58,897 RC192 0.48 ** [0.02] 0.34 ** [0.04] 0.22 ** [0.07] -0.77 ** [0.02] 0.19 0.13 0.00 59,963 RC267 0.51 ** [0.02] 0.40 ** [0.04] 0.17 * [0.07] -0.87 ** [0.02] 0.20 0.15 0.02 56,925 RC402 0.55 ** [0.02] 0.42 ** [0.04] 0.26 ** [0.07] -0.75 ** [0.02] 0.22 0.17 0.01 57,971 Standard errors in brackets + denotes p <.10; * p <.05; ** p <.01 To examine the robustness of these findings, and generally to compare results obtained using the traditional, state-level approach to estimates derived using individual measures of 18

opinion, we modeled the same abortion roll calls using state-level means for white and African American opinion (see Table 4). As noted above, we are able to obtain reliable measures of state-level opinions in this issue area. In this framework, whites opinions were significantly related to Senators votes for 14 of 25 votes at the.05 level and 19 of 25 votes at the.10 level. In contrast, African Americans opinions were never related to votes at the.05 level and were related at the.10 level only twice. The marginal effects represent the change in the probability that a Senator casts a pro-life vote resulting from changing state white and African American opinion from the most pro-choice to the most pro-life values. When whites opinions were significantly related to Senators votes, they were estimated to change the probability of a prolife vote by an average of 70 percentage points, compared to only 9 percentage points for African Americans. Table 4 State-Level Models of Senate Abortion Votes Senate Marginal Effects RC # Whites African Americans Constant Whites Blacks N 105 th Congress RC13 6.29 + [3.70] 5.66 + [3.02] -16.61 ** [4.48] 0.49 0.54 58 RC71 5.99 [3.66] 3.16 [2.98] -12.18 ** [3.94] 0.44 0.30 58 RC129 5.81 [3.71] 4.28 [3.00] -13.60 ** [4.00] 0.44 0.40 58 RC167 8.54 * [3.85] 1.83 [3.07] -14.07 ** [4.13] 0.63 0.18 57 RC190 5.57 [3.65] 3.31 [2.93] -12.06 ** [3.90] 0.43 0.33 57 RC215 7.11 + [3.74] 2.05 [2.99] -12.84 ** [4.08] 0.51 0.22 58 RC105 6.66 + [3.69] 5.07 + [3.01] -16.20 ** [4.41] 0.51 0.48 58 RC176 12.83 ** [4.40] 0.36 [3.56] -17.71 ** [4.68] 0.80 0.04 53 RC277 5.99 [3.66] 3.16 [2.98] -12.18 ** [3.94] 0.44 0.30 58 RC282 5.16 [3.67] 3.24 [2.99] -11.50 ** [3.85] 0.40 0.34 57 106 th Congress RC148 9.55 * [3.91] 0.87 [3.11] -14.12 ** [4.12] 0.68 0.08 58 RC197 5.73 [3.60] 2.23 [2.91] -10.86 ** [3.80] 0.44 0.25 57 RC336 6.28 + [3.65] 3.48 [2.95] -13.49 ** [4.10] 0.50 0.35 57 RC337 6.28 + [3.65] 3.48 [2.95] -13.49 ** [4.10] 0.50 0.34 57 RC338 8.19 * [3.88] 3.85 [3.13] -16.67 ** [4.62] 0.60 0.39 57 RC340 8.33 * [3.87] 1.97 [3.21] -13.59 ** [4.05] 0.56 0.20 57 19

RC134 9.58 * [3.93] 0.62 [3.13] -13.83 ** [4.12] 0.69 0.05 57 107 th Congress RC160 11.36 * [4.47] 0.01 [3.50] -15.65 ** [4.78] 0.72 0.02 52 108 th Congress RC47 16.15 ** [4.69] -2.82 [3.62] -17.61 ** [4.77] 0.89-0.28 56 RC48 14.15 ** [4.64] -0.98 [3.51] -17.86 ** [4.91] 0.84-0.09 57 RC49 21.10 ** [5.31] -2.02 [4.05] -25.15 ** [6.11] 0.94-0.18 54 RC51 16.56 ** [4.60] -1.26 [3.74] -20.06 ** [4.94] 0.87-0.11 55 RC192 14.85 ** [4.61] -1.73 [3.54] -17.62 ** [4.76] 0.86-0.17 57 RC267 14.76 ** [4.87] -1.85 [3.71] -17.42 ** [4.86] 0.85-0.18 54 RC402 15.45 ** [4.48] -1.81 [3.61] -17.76 ** [4.43] 0.85-0.19 56 Standard errors in brackets; + denotes p <.10; * p <.05; ** p <.01 Figure 2: Impact of Preferences on Sample Abortion Vote 1 Prob. of Pro-Choice Vote 0.8 0.6 0.4 0.2 0 min 25th Percentile Mean 75th Percentile max whites African Americans This dramatic difference is illustrated in Figure 2, which portrays the probability that a Senator cast a pro-life vote on roll call 148 from the 106 th Congress, a vote with roughly average correspondence between Senators and both groups. The probability of a pro-life vote increases by about 67 percentage points as whites became most pro-life, from about 17 percent to about 84 percent. This probability changed far less for African Americans, moving only about 9 percentage points from 44 percent to 53 percent. 20

Generally speaking, Senators appear to represent their white constituents far more than their African American constituents when voting on abortion policy. Importantly, the results using the traditional method and our individual-level approach overlap considerably. For every vote registering a significant difference in correspondence at the individual-level, white opinion was significantly related to votes and African Americans opinions were not in the traditional framework. In fact, since the traditional framework found more significant substantive and statistical differences between whites and African Americans, the new individual-level correspondence approach may actually be more conservative, suggesting that disparities in representation we find elsewhere in this analysis may be even greater than estimated. Finally, returning to the individual-level approach, we modeled the selected LCCR votes for the 105 th -107 th Congresses. We do so to determine if the correspondence of white and African American opinion with Senator roll call behavior is greater for issues that tend to be salient to African Americans. In many instances, Senators votes were completely unrelated to preferences of both African Americans and whites (see Table 5). It is difficult to say whether this reflects a weaker nexus between these GSS survey items and the LCCR roll calls than was the case for abortion and general ideology items, or a lesser degree of correspondence in these issues domains generally. In either case, since the items are the same for both groups, the relative correspondence of each group remains meaningful. On these LCCR measures, Senators votes corresponded more to African Americans preferences than to whites preferences in 8 of 21 models (12 of 21 at the.10 level), while they corresponded more to whites preferences only twice. For these votes, on average, white opinion has virtually no connection to Senator behavior, but African Americans opinions increased the probability of Senators casting pro-civil 21

rights votes by about five percentage points. Thus, on the issues that are most salient to African Americans, they appear to be better represented in at least some instances. Table 5 Preference-Vote Correspondence on LCCR Key Votes LCCR Marginal Effects Test RC # White Opinion Black Opinion Race Constant Whites Blacks p-value N 105 th Congress, First Session RC3-0.02 [0.01] 0.11 ** [0.03] -0.14 ** [0.05] -0.03+ [0.02] -0.01 0.09 0.00 48,934 RC5 0.07 ** [0.01] 0.16 ** [0.03] -0.08 [0.05] -0.99 ** [0.02] 0.04 0.11 0.01 48,446 105 th Congress, Second Session RC5 0.01 [0.01] 0.10 ** [0.04] -0.04 [0.05] -0.84 ** [0.02] 0.00 0.07 0.01 48,934 RC8-0.02 * [0.01] 0.12 ** [0.03] -0.11 * [0.05] 0.08 ** [0.02] -0.02 0.09 0.00 47,152 106 th Congress, First Session RC2-0.01 [0.01] 0.04 [0.03] -0.12 * [0.05] -0.03 [0.02] 0.00 0.03 0.17 48,142 RC8-0.02 [0.01] 0.06 [0.03] -0.14 * [0.05] -0.045 [0.02] -0.01 0.05 0.02 46,792 RC9-0.01 [0.01] 0.04 [0.03] -0.09 + [0.05] 0.05 ** [0.02] -0.01 0.03 0.19 48,594 106 th Congress, Second Session RC3 0.01 [0.01] 0.09 ** [0.03] -0.13 ** [0.04] -0.32 ** [0.02] 0.01 0.07 0.01 47,970 107 th Congress, First Session RC3-0.02 [0.01] 0.04 [0.03] -0.09 + [0.05] -0.26 ** [0.02] -0.01 0.03 0.10 48,296 RC4 0.08 ** [0.01] 0.10 ** [0.04] -0.04 [0.05] -0.39 ** [0.02] 0.06 0.08 0.56 45,840 RC5-0.02 + [0.01] 0.04 [0.03] 0.01 [0.05] -0.13 ** [0.02] -0.01 0.03 0.11 48,296 RC5 0.07 ** [0.01] 0.12 ** [0.03] 0.04 [0.05] -0.27 ** [0.02] 0.06 0.09 0.19 48,078 RC5 0.12 ** [0.01] 0.11 ** [0.04] 0.18 ** [0.05] -0.39 ** [0.02] 0.09 0.09 0.77 45,840 RC5 0.00 [0.01] -0.06 ** [0.02] 0.20 ** [0.05] -0.16 ** [0.02] 0.00-0.05 0.00 47,676 RC6-0.01 [0.01] 0.05 [0.03] -0.11 * [0.05] -0.14 ** [0.02] -0.01 0.04 0.06 45,906 RC7-0.03 * [0.01] 0.07 + [0.04] -0.21 ** [0.05] -0.56 ** [0.02] -0.02 0.05 0.01 45,592 RC8-0.01 [0.01] 0.06 + [0.03] -0.14 ** [0.05] -0.21 ** [0.02] -0.01 0.05 0.04 47,715 RC9 0.01 [0.01] -0.08 ** [0.02] 0.19 ** [0.05] 0.14 ** [0.02] 0.00-0.06 0.00 47,274 RC9 0.10 ** [0.01] 0.10 * [0.04] 0.12 * [0.05] -0.04 * [0.02] 0.08 0.07 0.94 45,463 RC9-0.01 [0.01] 0.01 [0.03] 0.02 [0.05] 0.16 ** [0.02] 0.00 0.01 0.57 47,885 RC9 0.06 ** [0.01] 0.08 * [0.03] 0.04 [0.05] 0.06 ** [0.02] 0.05 0.06 0.74 47,673 Standard errors in brackets; + denotes p <.10; * p <.05; ** p <.01 22

To summarize, Senator behavior generally corresponds much more closely to the preferences of whites than African Americans. This was true for aggregate voting measures over seven congresses, and in the specific domain of abortion policy over four congresses. However, this is not always the case as votes sometimes correspond to both groups about equally. Furthermore, in many instances, Senators votes on issues identified as particularly salient for African Americans corresponded more closely with African Americans preferences than with whites. Explaining the Disparity Turnout Assuming we wish to address these racial disparities in representation, we must know what causes them. We take a first step in this direction by examining whether whites are better represented because they are more likely to vote in elections. 6 Factually, we know that turnout differs substantially by race (e.g. Verba, Schlozman, and Brady 1995). According to the GSS, 71.4 percent of white respondents reported voting in the most recent presidential election, compared to 62.8 percent of African Americans, a difference significant at the.01 level. We also know that voters are better represented than nonvoters in the decisions elected officials make at higher levels of aggregation (Martin 2003; Griffin and Newman 2004). There are at least three reasons for this. First, voters are more likely to engage in other forms of political participation that more precisely communicate their preferences (Verba and Nie 1972). If officials try to represent their constituents, they will probably represent views that are communicated most often (e.g., Verba, Schlozman, and Brady 1995). Second, voters tend to select representatives who are like-minded, creating a natural correspondence between representatives and voters that may or 6 Another possible explanation, the racial composition of the Senate, cannot be explored without the election of more African American Senators. 23

may not extend to non-voters. Third, Senators may fear the ability of voters to sanction them, so they may be most responsive to voters preferences (Bartels 1998). Since African Americans are less likely to vote and non-voters are underrepresented, this may account for part of African Americans underrepresentation. To investigate this possibility, we first ask whether Senators votes correspond more with voters than non-voters using the individual-level approach. We do this by modeling W- NOMINATE scores as a function of all respondents ideologies, a turnout variable (coded 1 for voted, 0 otherwise), and an interaction between turnout and ideology. If voters are better represented, this interaction term will be positive and significant. As Table 6 shows, Senators votes correspond significantly more with voters ideologies than non-voters in all seven congresses at a.05 level. Thus, differences in turnout may account for part of the racial gap in representation. Table 6 Disproportionate Representation of Voters Congress 101 st 102 nd 103 rd 104 th 105 th 106 th 107 th Ideology 0.001 0.002 0.005 0.021 ** 0.023 ** 0.025 ** 0.026 ** [0.005] [0.004] [0.004] [0.005] [0.005] [0.005] [0.004] Ideology*Turnout 0.020 ** 0.018 ** 0.019 ** 0.013 * 0.013 * 0.013 * 0.011 * [0.005] [0.005] [0.005] [0.006] [0.005] [0.006] [0.005] Turnout -0.089 ** -0.087 ** -0.108 ** -0.104 ** -0.102 ** -0.096 ** -0.104 ** [0.022] [0.022] [0.021] [0.026] [0.022] [0.024] [0.021] Constant -0.205 ** -0.286 ** -0.334 ** -0.146 ** -0.135 ** -0.187 ** -0.225 ** [0.019] [0.019] [0.018] [0.022] [0.019] [0.021] [0.018] N 34,050 40,370 42,780 51,369 54,908 60,562 64,413 Robust standard errors in brackets + denotes p <.10; * p <.05; ** p <.01 24

However, this evidence that Senators correspond more to voters in general does not demonstrate that they correspond more to African American voters than to African American non-voters, or that Senators are equally reflective of African American and white voters. Both kinds of disparity seem likely, given prior work showing that African Americans are "captured" by the Democratic Party, and that both parties are therefore less responsive to them (Frymer 1999). To evaluate whether whites are "rewarded" more for voting than are African Americans, we estimate another individual-level model of Senators W-NOMINATE scores, including whites and African Americans ideologies, turnout, white voters and African American voters ideologies, and all appropriate lower order effects. If whites and African Americans are rewarded for voting, compared to non-voting whites and African Americans, the voter parameter estimates for both races should be positive and statistically significant. Table 7 shows that this is always the case for whites, but never true for African Americans. In fact, the parameter estimate for African American voters is never even close to statistical significance except in the 101 st Congress, when the t score was 1.58. 7 It appears that voting provides whites an additional boost in representation, but does not provide the same benefit for African Americans. These results also imply that the ideologies of white voters are more likely to correspond with their Senators roll call decisions than are the ideologies of African American voters. While the parameter estimates for white voters and African American voters are often similar in magnitude, the confidence intervals around the African American 7 As a robustness check, we also estimated the model reported in Table 6 by race. For every Congress, we found a greater correspondence between the decisions of Senators and the ideologies of white voters (compared to white non-voters), but not a greater correspondence between Senators and African American voters (compared to African American non-voters). 25

parameters are so great that they cannot be distinguished from zero. White voters are better represented than African American voters. 8 Table 7 Race, Representation, and Turnout Congress 101 st 102 nd 103 rd 104 th 105 th 106 th 107 th White 0.005 0.002 0.007 0.024 ** 0.026 ** 0.028 ** 0.028 ** Ideology [0.005] [0.005] [0.005] [0.006] [0.005] [0.005] [0.005] Black -0.013-0.001-0.002 0.005 0.007 0.009 0.013 Ideology [0.010] [0.010] [0.010] [0.012] [0.010] [0.011] [0.009] Black 0.051 0.013 0.071 0.028 0.031-0.011-0.023 [0.046] [0.045] [0.045] [0.054] [0.048] [0.050] [0.044] White Ideology 0.019 ** 0.020 ** 0.020 ** 0.015 * 0.013 * 0.015 * 0.012 * * Turnout [0.006] [0.006] [0.005] [0.007] [0.006] [0.006] [0.006] Black Ideology 0.019 0.006 0.012 0.006 0.014 0.010 0.007 * Turnout [0.012] [0.012] [0.012] [0.014] [0.012] [0.013] [0.011] Turnout 0.017 0.063 0.042 0.131 * 0.087 0.127 * 0.134 * * Race [0.055] [0.053] [0.054] [0.064] [0.057] [0.059] [0.052] Turnout -0.090 ** -0.098 ** -0.116 ** -0.128 ** -0.118 ** -0.119 ** -0.128 ** [0.025] [0.024] [0.023] [0.029] [0.025] [0.027] [0.024] Constant -0.215 ** -0.288 ** -0.349 ** -0.152 ** -0.142 ** -0.185 ** -0.220 ** [0.022] [0.021] [0.020] [0.025] [0.022] [0.023] [0.021] N 34,050 40,370 42,780 51,369 54,908 60,562 64,413 Marginal Effects White Ideology 0.028 0.014 0.042 0.146 0.155 0.164 0.169 Black Ideology -0.080-0.006-0.011 0.030 0.047 0.055 0.076 Prob. Equal 0.100 0.730 0.390 0.140 0.110 0.130 0.150 Diff. 0.108 0.020 0.053 0.116 0.108 0.109 0.093 Diff. # 0.123 0.096 0.103 0.174 0.120 0.135 0.126 % change -0.122-0.792-0.485-0.333-0.100-0.193-0.262 Robust standard errors in brackets; + denotes p <.10; * p <.05; ** p <.01 # from Table 2, all differences significant at p <.01 8 Similarly, Radcliff and Saiz (1993) found that African American voting strength within states did not correspond with state-level policies favorable to African Americans. 26

To determine whether the gap in correspondence we observed in Table 2 and Figure 1 is attributable to differences in turnout among racial groups, we compare the marginal effects for white and African American ideology from Table 7, where turnout is controlled, and Table 2, where it is not. Table 7 reports the difference in the marginal effect of white and African American ideology from Table 2, as well as the marginal effects (and their difference) simulated from the estimates in Table 7, together with the percentage change in their difference. Based on these results, we can conclude that accounting for turnout helps to close the racial gap in representation we have observed. While there is wide variation in the extent to which the representation gap narrows across Congresses (from a low of 10% to a high of 79%), the weighted average of this change is 31%. That is, turnout accounts for about a third of the gap. Apparently, the reason that the gap closes somewhat is that when turnout is controlled, the marginal effects for white ideology decline much more sharply than do the marginal effects for African American ideology. Thus, a substantial portion of the gap is attributable to the representation gains of white voters. Moreover, accounting for turnout makes us quite a bit less confident that the estimate of white ideology s effect is distinguishable from the estimate of African American ideology s effect, as the coefficient comparison p values rise from less than.01 before turnout is controlled to greater than.10 afterward. In sum, the ideologies of white voters correspond better with the roll call decisions of their Senators than do the ideologies of both white non-voters and African American voters. On the other hand, African American voters ideologies do not correspond better with their Senators behavior than do the ideologies of African American non-voters. These differences help to explain about one-third of the gap in representation between whites and African Americans. 27