Racial Context and Racial Voting in New York City Mayoral Elections Revisited

Transcription

1 Racial Context and Racial Voting in New York City Mayoral Elections Revisited Thomas M. Carsey Department of Political Science Florida State University Tallahassee, FL Paper presented at the Southern Political Science Association annual meeting, November 7-10, 2001, Atlanta, GA.

2 1 Racial Context and Racial Voting in New York City Mayoral Elections Revisited Race remains a dominant, if not the dominant, cleavage in United States politics. No other demographic or social factor divides voters more dramatically than does race, with 90% or more of African-Americans supporting Democratic candidates in most races and a Democratic candidate for president has not won a majority of the votes cast by whites since Numerous studies have shown, however, that the importance of race in U.S. politics extends beyond the influence racial identity has on individual-level decision making. Rather, individual preferences on various issues and their voting behavior have been shown to be affected by the racial make-up of the contextual environment within which people find themselves. Thus, for example, we can predict to some degree how an individual is likely to vote based on his/her race, but we can gain further insight into a person s voting behavior if we also know the racial make-up of the environment within which that person resides. This latter effect is generally referred to as a contextual effect. It is at the contextual level that the influence of race on voting behavior becomes less clear. On the one hand, we have a large and well respected literature that reports whites reacting negatively as proportion of blacks that reside in the same area in which whites reside increases. Whether based on irrational prejudice or some more rational notion of racial competition, this literature argues that whites perceive a racial threat when the size of the black population in the area increases which leads whites to vote against candidates sympathetic to black interests and/or express less support for policies dealing with such things as affirmative action and busing. In contrast, a second large and well-respected literature focusing on a wide variety of contextual factors other than race generally reports evidence of a positive contextual effect on the

3 2 political attitudes and behaviors of people within those contexts. Either through active interaction or more passive observation, this research finds that the partisan, social class, or ethnic make-up of a person s context shapes the attitudes and behaviors of individuals in a way that brings them more in line with the larger context. This paper brings together these two literatures together through an analysis of the 1989 and 1993 mayoral elections held in New York City, thereby updating and extending a previous paper which examined only the 1989 race (Carsey 1995). Using these two elections to build a bridge between these literatures may be particularly appropriate for several reasons. First, the politics of race tend to play out in dramatic fashion in our country s largest cities, and New York in particular has a history of race being politically salient. Second, the winner of the 1989 election, David Dinkins, was the first African-American elected mayor in New York. He ran for re-election in 1993 and lost. Thus, we have in both races a strong African-American candidate running which, combined with the historical nature of his first victory and term in office, serves to highlight the salience of race in these elections. Finally, Dinkins opponant in both years was white Republican Rudolph Giuliani. Though Giuliani was likely better known in 1993 than in 1989, the presence of the same two candidates in both races helps to minimize the impact that differences between the two elections might have on influencing the importance of race in affecting voting behavior. The Contextual Effects of Race Since V.O. Key (1949) scholars of the politics of race have overwhelmingly argued two things: 1) that whites respond to the racial make-up of their environments (that there is a

4 3 contextual effect of black population sizes or densities on the attitudes and behaviors of whites), and, 2) that this effect is negative. This affect has been demonstrated in the South in general by Key (1949), in Southern presidential voting a generation later (Wright 1977), and nationally in presidential voting (Huckfeldt and Kohfeld 1989). The effect has been reported at the county level (Wright 1977; Giles and Buckner 1993) and the state-level (Huckfeldt and Kohfeld 1989). Finally, the salience of race in urban electoral politics has been documented by numerous scholars (e.g. Katznelson 1981; Huckfeldt and Kohfeld 1989; Browning and Marshall 1986; Kleppner 1985; Browning, Marshall, and Tabb 1984). In virtually all of these settings, the reaction of whites to increasingly larger black populations is one of hostility or at least opposition to the policy preferences and candidates typically supported by African-Americans. In contrast, a substantial amount of evidence suggests that the impact of the social context within which citizens find themselves generally positively influences their political attitudes and behaviors (e.g. Berelson, Lazarsfeld, and McPhee 1954; Tingsten 1963, Putnam 1966; Huckfeldt and Sprague 1987, 1995; Huckfeldt 1983, 1984; Johnson, Shively, and Stein 2000, 2001; Jackson and Carsey 2002). There is some debate as to the mechanism that produces these affects. Huckfeldt and Sprague (1995) for example devote substantial attention to interpersonal interaction. Johnson et al (2001) focus more on what they call social connectedness. Others suggest that citizens simply receive cues from those around them independent of direct social interaction (Burt 1987). Finally, some suggest that a self-selection process may be at work in which citizens sort themselves into different contexts (Achen and Shively 1995; Carsey 1995). If the factors driving this sorting process are correlated with political variables like candidate preference or policy views, Achen and Schively (1995) show that this sorting can create the

5 4 appearance of a contextual effect where none exists. It is debatable whether one should define this sort of selection process as not being evidence of a contextual effect itself (Carsey 1995; Huckfeldt 1983). If residential choice is driven in part by the context (racial or otherwise) of an area, and that choice subsequently manifests itself politically, it may be reasonable to interpret that process as a context-driven political process. Regardless of the mechanism at work, this literature seems to suggest that context measured in any terms other than race seems to produce a positive impact on individuals. Is Race the Exception to the Contextual Rule? Carsey (1995) produced the first evidence that the contextual affects of black population sizes on whites was not necessarily negative. In that study, Carsey found that the probability that a white voter reported voting for David Dinkins, a black candidate, for mayor of New York in 1989 was positively influenced by an increase in the percentage of the population in that white voter s precinct that was black. Carsey also found that the same effect held at the borough (county) level in the city, and that when precinct-level and borough-level measures of black population percentages were both included that both retained their positive and significant effects on the reported voting behavior of whites. Carsey also reports similar precinct-level effects on white voters in the 1987 mayoral race in Chicago that involved the re-election of Harold Washington, an African-American. Others have also recently found evidence that the racial context of a place might produce similarly non-hostile responses among whites. Voss (1996) for example finds no evidence that higher proportions of blacks in a parish (county) in Louisiana produced greater white electoral support for former Ku Klux Klan leader David Duke. What

6 5 negative affect there is Voss argues was isolated in the most rural parishes and that actually in urban centers, larger black populations are associated with less white support for Duke (see the response by Giles and Buckner (1996) and rejoinder by Voss (1996) for further discussion). Johnson et al (2000) also find that whites have a more positive affective evaluation of blacks as a group as the percentage of blacks in their city council district increases. 1 In their study, Johnson et al. (2000) conduct telephone interviews in Houston, asking respondents to rate several groups, including Whites and Blacks, on a 1 to 10 feeling thermometer. They limit their analysis to 51 white respondents interviewed by white interviewers for whom complete data is available. The dependent variable for this group was their rating of Whites minus their rating of Blacks. Johnson et al found that whites living in city council districts in Houston (of which there are nine) that had larger black populations were more likely to have more positive evaluations of blacks relative to whites than were those whites living in districts with smaller black populations. Importantly, they find that this affect is conditioned by how well connected the respondent is with his/her neighborhood (measured by the response time to a question about the racial make-up of their neighborhood). This suggests a real contextual effect that is not simply a function of a residential self-selection process. There are several reasons why the findings of these studies differ from the larger body of negative findings reported by others regarding the contextual effects of race. First, the analysis reported by Carsey and Johnson et al. measure racial contest at a level of aggregation smaller than the county while the bulk of the negative findings are measured at the county or state level 1 They further find that this affect is non-linear and is mediated in part by the degree of connectedness to the community, but the general pattern of their findings is one of a positive effect.

7 6 (though Carsey finds significant positive borough-level affects as well). This suggests that social context may operate on individuals differently at different levels of aggregation. Measuring context at the neighborhood level may be capturing a more proximate measure of the mix of people with whom a respondent more readily interacts. The so-called racial threat that so many others have reported may stem from anxiety, prejudice, or fears of competition directed at relatively faceless groups of people who are close enough to be threatening but not close enough to interact with. Voss argues, for example, that David Duke received much more support among whites residing in largely white suburbs of cities with large black populations rather than among whites that lived within those same cities. Measuring social context at a more local level may be better at capturing the positive impact of social interaction that drives most of the positive findings in the literature regarding class, partisan, and ethnic contexts found in the literature. The second possible cause of this set of divergent findings regarding racial context is the focus on large central cities. Carsey and Johnson et al focus exclusively on major cities while Voss s conclusions rest primarily on comparing results from the largest cities in Louisiana to the suburbs and rural areas. It is possible that the contextual effects of race play out differently within large central cities than it does across counties or states. This paper does not attempt to resolve all of the issues raised thus far regarding the contextual effects of race. Instead, this paper is designed to explore whether unexpectedly positive results uncovered by Carsey (1995) for the 1989 mayoral election in New York remained four years later. The findings here will answer whether the 1989 results were a fluke and/or whether four years of experience with their first every African-American mayor altered how whites responded to the racial make-up of their neighborhoods.

8 7 Data and Methods The data for this study come from two exit polls conducted on election day in New York City in 1989 and The first was conducted by CBS/New York Times and the second was conducted by Voters Research and Surveys. 2 In each survey, samples were collected using a twostep process. First, a probability sample of voting precincts was selected with a probability of being selected proportionate to the number of voters in each precinct. Second, within each precinct, individual voters were selected systematically throughout the day at a rate that gave all voters in a precinct the same chance of being interviewed. Each election featured the same two candidates: David Dinkins and Rudolph Giuliani. Dinkins, an African-American Democrat, had previously served as president of the borough of Manhattan before being elected mayor in He had worked his way through the Democratic party machine in New York, and had defeated incumbent Ed Koch in the Democratic primary. Giuliani, a white Republican, was a less well known former federal prosecutor before running for mayor in A Catholic of Italian ethnicity, Giuliani represented the more liberal wing of the Republican party. Kaufmann (1998) notes that race had long been politically salient in New York City politics leading up to the 1989 election. Mayor Koch had increasingly antagonized liberal whites and minorities with his growing conservatism on racial issues. In addition, several prominent instances of race-related violence coupled with a weakening economy had raised racial tensions. This environment was reinforced by Dinkins campaign focus on racial healing while Giuliani stressed law and order. 2 These polls were made available through the Inter-University Consortium for Political and Social Research, study numbers 9493 and 6249, respectively.

9 8 By 1993, Kaufmann (1998) notes, a few things had changed, not the least of which was that Dinkins was now the incumbent. However, many of the same circumstances prevailed. The economy was still sluggish and had actually worsened. Additional acts of racial violence had taken place. Kaufmann (1998) argues that racial tensions were even more salient in 1993 than they had been 4 years earlier. However, Kaufmann s overall conclusion is that race was quite salient in 1989 and maybe even more so in 1993, and that the politics of race in New York City for both elections was structured primarily in terms of inter-group conflict. It would seem from this account that the traditional racial threat prediction of a negative contextual effect of black population size on white voting behavior would be a reasonable expectation. The dependent variable in each model is the vote choice of the respondent. It is coded as a 1" for those who reported voting for Dinkins and a 0" otherwise. Given the dichotomous nature of the dependent variable, a logit or probit model would be appropriate, and logit is employed in this study. 3 The key explanatory variable in this analysis is the percentage of a white respondent s neighborhood that is black. Following Carsey (1995) and Johnson et al. (2001), neighborhood is operationalized as the voting precinct. This measure is computed as the percentage of the respondents in the exit poll in each precinct who self-identified as Black to a question about racial identity that included options for White, Black, Hispanic, Asian, and Other. 4 Computing contextual measures by aggregating respondents from a survey is a common method of generating such variables (Sprague 1976; Weatherford 1980; Carsey 1995). The average number of respondents per precinct across the two polls is 45. One precinct in each 3 All analysis was performed using the statistical program Stata The 1993 poll used the label Hispanic/Latino instead of just Hispanic.

10 9 poll had only 25 respondents, but the remaining all had at least 35 respondents available to compute this measure. The measure ranges from 0% to 89% in 1989 and from 0% to 95% in Not surprisingly most whites reside in areas made up of mostly white people, with roughly 75% of whites in these two polls voting in precincts that are only 10% or less black. A number of control variables are also included in the models. These include dummy variables for party identification, ideological self-placement, and religious affiliation. Also included is a categorical variable for education level that ranges from a low of 1 for those who did not complete high school to a high of 5 for those with some education beyond a college degree. A categorical income variable is also included, which ranges from 1 to 6. 5 Finally, one additional contextual measure was included to tap the social status of the precinct: the mean education level of precinct respondents. Given the nature of the debate on whites responses to the racial make-up of their surroundings, this paper focuses only on white respondents. 6 Findings Before turning to the results of the logit analysis, presenting some simple descriptive statistics regarding the relationship between race and vote choice in these two elections is warranted. Table 1 reports the percentage of respondents who reported voting for Dinkins for 5 The associated income values are somewhat higher for each category in 1993 as compared to The distributions across the six categories, however, are roughly the same across the two polls with the exception of the second highest category. Twenty-two percent of respondents in 1989 placed themselves in category 5, which included those with incomes from $50,000 to $100,000. In 1993, only eight percent of respondents put themselves in category 5, which included those with incomes from $75,000 to $99, This set-up replicates exactly the structure of the analysis presented in Carsey (1995) for the 1989 election.

11 10 mayor in each year. In both 1989 and 1993, barely more than a quarter of the whites surveyed indicated that they voted for Dinkins, despite the fact that nearly half of whites who were interviewed were self-identified Democrats. In contrast, 90 to 95 percent of African-Americans reported voting for Dinkins. Dinkins also did well among Hispanic voters, but the differences between the behavior of whites and blacks makes it clear that race was a salient political divider in both election years. Table 1 is important for two reasons. First, even though the focus of this paper is on the contextual effects of race, readers should not loose sight of the sharp distinction between the voting behavior of whites and blacks that exists at the individual level. Race plays a major role in NYC electoral politics outside of any contextual effect. Second, this paper will focus on the positive impact that increasing black racial densities have on white voting behavior, but readers should not loose sight of the basic fact that more than 70% of whites surveyed did not vote for Dinkins. Turning next to assessing the impact of racial context on white voting behavior, table 2 presents the results of estimating the same logit model for both 1989 and The first thing to note is that the findings are quite similar across the two elections, with the caveat that most of the coefficients operating on the individual-level predictors are somewhat larger in absolute value and the overall model fit is somewhat stronger in The similarity likely reflects the similarity in electoral circumstances. The sharper coefficient estimates and model fit in The astute reader will (eventually) discover that the results presented here and in Carsey (1995) for the 1989 CBS/NYT exit poll are extremely similar, but not identical despite the identical structure of the model. The differences emerge because the analysis presented in Carsey (1995) used the data originally deposited with ICSPR in May of The current analysis presented here, however, uses the data as updated in February of 1992, an update that resulted in 54 fewer usable cases for this analysis.

12 11 may reflect the greater familiarity voters had with the candidates and/or the reported higher intensity of group conflict in 1993 compared to 1989 (Kaufmann 1998). Turning to the primary variable of interest, table 2 reports that as the percentage of respondents in a precinct that are black increases, the probability that a white voter will support Dinkins for mayor also increases. The effect is statistically significant in both 1989 and The coefficient estimate itself is somewhat larger in 1989 ( =.024) than in 1993 ( =.018), but simple t-tests of each model suggest that the difference between these two estimates is not statistically significant. 8 In both instances, this positive effect is uncovered after controlling for a series of strong individual-level predictors of voting behavior as well as the average education of the precinct. Thus, Carsey s (1995) finding of a positive effect of increasing black population densities on white support for Dinkins appears not to be just an artifact of that particular election as four years later the same affect holds. Interpreting logit coefficients directly is not particularly intuitive. So, to facilitate their interpretation, I computed the predicted probability of a white respondent voting for Dinkins across the entire range of the percentage of a precinct s population that is black. These predicted probabilities are plotted in figure 1, and were computed by holding all other factors in the model at their means and allowing the percentage of the precinct that is black to vary from 0 to 100. Figure 1 is meant to serve as a heuristic device for understanding the impact of the racial make- 8 Specifically, after running each logit model, I performed a t-test of whether the coefficient on the percent black variable was equal to what it was estimated to be in the other equation. Thus, for the 1989 sample, I tested whether the parameter on percent black was significantly different from.018 and in the 1993 sample I tested whether the same parameter was significantly different from.024. In neither case could I reject the null hypothesis of no difference. Note that this is a conservative test because it compared an estimated parameter to a fixed value which ignores the estimated standard error around that value.

13 12 up of a precinct on white voting behavior remember that most white voters voted in precincts comprised mostly of whites. Nevertheless, figure 1 clearly demonstrates the substantive importance of black population densities on white voting behavior. Moving across the range of the percentage of a precinct s population that is black from 0% to 90% (the observed range in the data), the predicted probability of a white person voting for Dinkins for mayor increased from.20 to about.69 in The same increase produced a shift in likely support for Dinkins in 1993 from.14 to.46. Recalling that table 1 reported that 27-28% of whites reported voting for Dinkins in these two elections, the predicted ranges here are substantively dramatic. Limiting our focus to those precincts in which most whites live, moving from those with 0% black populations to those with 20% black populations increases the predicted probability of whites voting for Dinkins by about 9 percentage points in 1989 and 5 percentage points in The curves presented in figure 1 invite two additional interpretations. First, the curve for 1993 starts lower than does the one for 1989 and remains below it across the entire range of precincts. This stems from the larger negative constant estimated for 1993 and the larger in absolute value negative coefficients operating on a number of the individual-level independent variables that are included in the model. Substantively, this reflects the lower probability overall of whites voting for Dinkins, particularly white Catholics. Second, the curve for 1993 is obviously flatter than the curve for This is wholly a function of the smaller coefficient estimated for the effect of the percentage black in a precinct on white voting behavior. This suggests concluding that whites responded less to the racial make-up of their neighborhoods in 1993 than they did in 1989, despite the hightened level of racial conflict reported in 1993 (Kaufmann 1998). The difference between the two years is greatest at

14 13 higher levels of percent black. This interpretation might be consistent with the notion that voters were more familiar with the two candidates in 1993 such that their policy positions and Dinkins performance as mayor were more important in determining who voters would support than was the racial cue sent by a voter s context. However, such an interpretation is risky on at least two fronts. First, recall that while the estimated coefficients are different from each other, that difference is not statistically significant. Second, very few whites reside in precincts that are substantially black, meaning that the predicted difference between the voting behavior of whites illustrated in figure 1 at higher levels of percent black in the precinct are based on relatively few observations. In fact, if the models are re-run including only those whites who voted in precincts that are 20% black or less, the size of the coefficient estimate on the percentage black variable increases in both years and actually becomes larger in 1993 than in 1989 ( =.044 and =.08 in 1989 and 1993, respectively). As noted above, one possible reason for the differences in findings between the results reported here and the majority of published studies that uncover a negative effect of black population densities on white voter behavior is the level of aggregation at which the contextual effect is measured. The findings here measure context at the precinct level whereas most of those studies measure context at the county or state level. As reported in Carsey (1995), the 1989 exit poll also recorded the borough in which a respondent voted. Computing a borough-level measure of percent black and including that in the logit model produces a positive coefficient estimate operating on this variable and the positive effect of precinct level black densities continues to remain significant. Thus, the findings in New York for 1989 do not appear to be

15 strictly a function of the level at which racial context is measured. 9 Unfortunately, the 1993 exit poll does not include a borough code. 14 Conclusions The results presented in this study replicate and extend those reported by Carsey (1995), showing a significant positive effect of black population increases at the precinct level on white voting behavior in the Dinkins/Giuliani elections for mayor of New York City in 1989 and As the percentage of the population in a precinct that was black increased, the probability that whites voting in those precincts would vote for a black candidate, David Dinkins, increased. The effect holds despite numerous individual-level controls and remains largely unchanged from one election to the next. Many questions remain, but the implication first drawn in Carsey (1995) that,... the contextual effects of race may not be so different from the contextual effects of factors like partisanship, ethnicity, or social class as we might have believed (p. 228) is reinforced by the findings presented here. These results are surprising given the heightened sense of racial conflict in New York surrounding the 1989 and in particular the 1993 election and the historic nature of Dinkins candidacy in 1989 and incumbency in 1993 if a racial threat existed to provoke a negative reaction among whites, these two elections seemed well primed to produce it. The data employed for this study cannot satisfactorily arbitrate between the traditional interpretation of the results shown here as contextual effects versus the competing notion that it 9 Recall that Voss (1996) also reports a diminished support among whites for David Duke in urban counties in Louisiana as the percentage of residents that are black increases.

16 15 is strictly a function of residential self-selection (Achen and Shively 1995). However, adding Johnson et al. s (2000) findings that the influence of the racial make-up of a neighborhood on white attitudes toward blacks is conditioned by the degree to which a person is connected to his/her community reinforces the assertion that, The same social interaction process believed to produce the positive contextual effects of other factors found in the literature is likely the process producing [the positive findings reported here] (Carsey 1995, p. 228). There remains a need for further research to identify when and under what circumstances we might expect to find positive or negative contextual effects related to race, and improved data is needed to sort out traditional contextual effects from residential self-selection effects. The findings presented here, however, add to the small but growing body of work which argues that long-established prediction of a hostile political response by whites as the concentration of African-Americans increases, the racial threat hypothesis, can no longer be uniformly assumed in U.S. politics. References Achen, Christopher H. and W. Phillips Schively Cross-level Inference. Chicago: University of Chicago Press. Berelson, Bernard R., Paul F. Lazarsfeld and William N. McPhee Voting. Chicago: University of Chicago Press. Browning, Rufus P., Dale Rogers Marshall, and David H. Tabb Protest is Not Enough. Berkeley: University of California Press. Browning, Rufus P. and Dale Rogers Marshall eds. "Black and Hispanic Power in City Politics: A Forum." PS 19: Burt, R. S Social Contagion and Innovation: Cohesion versus Structural Equivalence. American Journal of Sociology, 92: Carsey, Thomas M The Contextual Effects of Race on White Voter Behavior: The 1989

17 16 New York City Mayoral Election. Journal of Politics, 57: Giles, Michael W. and Melanie A. Buckner David Duke and Black Threat: An Old Hypothesis Revisited. Journal of Politics, 55: Huckfeldt, Robert "The Social Contexts of Ethnic Politics: Ethnic Loyalties, Political Loyalties, and Social Support." American Politics Quarterly, 99: Huckfeldt, R. Robert Political Loyalties and Social Class Ties: The Mechanisms of Contextual Influence. American Journal of Political Science, 28: Huckfeldt, Robert and Carol Weitzel Kohfeld Race and the Decline of Class in American Politics. Urbana, IL: University of Illinois Press. Huckfeldt, Robert and John Sprague Networks in Context: The Social Flow of Political Information. American Political Science Review, 81: Huckfeldt, Robert and John Sprague Citizens, Contexts, and Social Communication: Information and Influence in an Election Campaign. New York: Cambridge University Press. Jackson, Robert A. and Thomas M. Carsey Group Effects on Party Identification and Party Coalitions Across the United States. American Politics Research, 30: Johnson, Martin, W. Phillips Schively, and Robert M. Stein Accessibility and Contextual Explanations of White Racial Attitudes. Paper presented at the Annual Meeting of the American Political Science Association, August 31-September 3, Washington, D.C. Johnson, Martin, W. Phillips Schively, and Robert M. Stein Contextual Explanations of Presidential Vote Choice. Paper presented at the Annual Meeting of the Midwest Political Science Association, April 19-22, Chicago, Illinois. Katznelson, Ira City Trenches. Chicago: University of Chicago Press. Kaufmann, Karen M Racial Conflict and Political Choice: A Study of Mayoral Voting Behavior in Los Angeles and New York. Urban Affairs Review, 33: Key, V. O Southern Politics in State and Nation. New York: Vintage Press. Kleppner, Paul Chicago Divided: The Making of a Black Mayor. DeKalb, IL: Northern Illinois University Press. Sprague, John "Estimating a Boudon Type Contextual Model: Some Practical and Theoretical Problems of Measurement." Political Methodology, 3:

18 17 Putnam, Robert Political Attitudes and the Local Community. American Political Science Review, 60: Tingsten, Herbert Political Behavior: Studies in Election Statistics. Translated by Vilgot Hammarling. Totowa, NJ: Bedminster Press. Voss, D. Stephen Beyond Racial Threat: Failure of an Old Hypothesis in the New South. Journal of Politics, 58: (with Comment by Michael W. Giles and Melanie A Buckner, pp , and Rejoinder by D. Stephen Voss, pp ). Weatherford, M. Stephen The Politics of School Busing: Contextual Effects and Community Polarization. Journal of Politics, 42: Wright, Gerald C., Jr "Contextual Models of Electoral Behavior: The Southern Wallace Vote." American Political Science Review 71:

19 18 Table 1: Percentage of Respondents Who Reported Voting for Dinkins by Major Racial Group, 1989 and 1993 Group Whites 28% 27% Blacks 90% 95% Hispanics 70% 66% Note: Data from the 1989 CBS/NYT exit poll and the 1993 VRS exit poll

20 19 Table 2: Logit models predicting the probability of whites voting for Dinkins for New York City mayor, 1989 and 1993 Independent Variables Democrat.610 (3.15)**.883 (3.51)** Republican (-2.64)** (-3.65)** Conservative (-1.71)* (-3.15)** Liberal (6.36)** (6.98)** Education.191 (2.38)**.118 (1.17) Income (-1.53) (-1.21) Jewish (-2.62)** (-4.17)** Catholic (-2.37)** (-3.05)** Percentage Black in the Precinct.024 (4.17)**.018 (2.47)** Average Education in the Precinct.649 (3.46)**.967 (4.09)** Constant (-5.56)** (-5.09)** Model properties Model Chi-Square % in modal category 71% 71% % correctly predicted 76% 83% Proportional reduction in error N-size Note: Cell entries are logit coefficients with z-scores in parentheses. Data for 1989 comes from the CBS/NYT exit poll. Date for 1993 comes from the VRS exit poll. * p<.1, ** p<.05 (2-tailed);

21 Figure 1: Predicted effect of precinct-level black population density on the probability of whites living in those precincts voting for Dinkins, holding other factors at their means 20