Context and White Crossover Voting

Table 1 reports descriptive statistics on the level of white crossover voting for each of the election units under investigation, organized by context and ordered by the racial makeup of election units within each context.  The highest level of white crossover occurred in the 1994 District B election (95.6 percent); the lowest level was in the 1986 District A contest (1.3 percent).  This range in white crossover voting across our election units provides sufficient variance in the dependent variable to identify conditions under which white voters cast ballots vote for black candidates.   

Mean white crossover among all 29 biracial elections was 25.0 with a standard deviation of 27.9.  Comparing the three racial contexts, we observe that the mean level of white crossover voting was lowest, 5.7 percent, in the white dominant context (i.e., in election units where less than 45 percent of the registered voters were black) and highest, 48.2 percent, in the black dominant context (election units where more than 55 percent of the registered voters were black) (48.2 percent).  In the racially competitive context (where blacks comprise 45 to 55 percent of the registered electorate) the mean level of white crossover was 16.1 percent.  These descriptive findings do not appear to support black threat theory as applied to white crossover voting; neither do they lend support to the modified black threat theory.

Table 1

The estimated effect of racial context at the election unit level is reported in Table 2.[5] As shown in Equation 1, contrary to what traditional black threat theory predicts, the relationship between black density and the level of white crossover voting is positive.  For every one percentage point increase in black density, there is an almost one-and-a-half percentage point increase in white crossover voting.  Findings reported in Equation 2 fail to support the prediction of a U-shaped relationship (a U-shaped relationship was predicted by the modified black threat theory).  The quadratic term is not statistically significant at the .05 level.  Thus, neither the traditional black threat theory nor its modification explain the observed pattern in white crossover voting across these 29 biracial elections.[6]

Table 2

Time and White Crossover Voting

The descriptive data reported in Table 1 suggest a temporal relationship between the year of an election and white crossover voting: most elections in the white dominant context took place earlier than those in the black dominant context.  For instance, three of the six contests in the white dominant context were held in the 1970s; six of the ten elections in the black dominant context occurred in the 1990s, and the earliest was in 1982.  This temporal relationship may provide an explanation for the positive relationship between black density and white crossover voting.  If racial attitudes among white voters became more liberal over time, resulting in increased white crossover voting, the relationship between black density and white crossover voting would be spurious.[7]

To test whether the increase in white crossover voting is attributable to time, a counter variable reflecting the year of the election was added to the regression analysis.  The time period under investigation encompasses eight election years, accordingly the value of the time variable ranges from 1 to 8.  As shown in Equation 3 of Table 2, the effect of time is not statistically significant.  Black density continues to be a reliable predictor of white crossover voting.  The empirical evidence fails to support the notion that greater crossover voting across these New Orleans elections was a function of time.

A Curvilinear Relationship Between Black Density and White Crossover Voting  

Findings reported in Table 3 indicate that the relationship between black density and white crossover voting is more accurately modeled as an upward sloping half-U than as a straight line.  (The quadratic term in Table 2, Equation 2, expressed the relationship as a full U-shape, but was not statistically reliable.)  The least squares fit for this upward sloping half-U curve is calculated in the following manner: Y=a(% Black Voters^b) where a and b are constants.  The dependent variable, white crossover voting, is a power function of the independent variable, black density.  Results indicate that this curvilinear model, reported in Equation 1 of Table 3, performs better (R2=.60) than the simple linear model (R2=.39) reported in Equation 2.

Table 3

This curved, upward sloping half-U relationship is largely due to voting in the election units where black density was lowest.  Mainly these were elections for the District A council seat which occurred in three of the four lowest levels of black density among all election units and yielded the three lowest rates of white crossover (see Table 1).  To determine whether the curvilinear relationship between black density and white crossover voting is due to these three District A elections, Equation 3 of Table 3 reports the results of a curvilinear model when these particular elections are excluded.  Based on a comparison of the power function reported in Equation 3 with that reported in Equation 1, we conclude that these three District A elections were largely responsible for the overall curvilinear relationship.  The R-squared for the power function when District A contests are excluded, reported in Equation 3 (.38), is 22 percentage points less than the R-squared for the power function for all 29 elections, reported in Equation 1 (.60). 

We also note that when the three District A elections are excluded, a linear model (R2=.36, Equation 4) is virtually as good as the curvilinear model (R2=.38, Equation 3).  Because the low levels of white crossover voting in District A are the source for the curvilinear relationship between black density and white crossover voting, a closer examination of these three contests is in order.

White Crossover Voting in a White Dominant Election Unit  

If the level of white crossover voting reflects the level of white perception of black threat, as traditional black threat theory suggests, results thus far indicate that among white voters those in District A had the strongest perception of black threat.  This brings up a perplexing question.  Given the relatively low black density level in District A (this district was the only election unit dominated by white voters across the entire time-period under investigation), how might whites in this district have had a strong perception of black threat?  According to traditional black threat theory, black threat should have been stronger, hence white crossover lower, in electoral units with a comparatively larger black cohort.

One possible explanation is that the black threat for District A white voters reflected a generalized rather than a district-specific threat.  It may be the case that the citywide transition from white to black political control enhanced the racial sensitivity of white voters in District A, thereby keeping crossover voting among whites in this district low in elections to all offices, citywide as well as for the District A seat.  This was not the case, however.  In the 1986 District A council election, for example, an estimated 1.6 percent of white voters crossed over.  In the simultaneous mayoral primary, white voters in District A crossed over at a rate of 35.7 percent.  Though white voters in this district were less supportive of the black mayoral candidate than were white voters generally (overall, white crossover voting in the 1986 mayoral primary was 51.3 percent), the 34.1 percentage point difference in the level of crossover voting between the mayoral and council contest among District A whites voters suggests that the low level of crossover in elections for the District A council seat likely was related to the candidate choices offered to voters rather than to any generalized perception of black threat.

During the last two decades only four black candidates competed for the District A seat, all in three open-seat elections.  While none of these candidates received more than an estimated 0.6 percent of the white vote, they also did poorly among black voters – receiving as little as 1.3 percent, and no greater than an estimated 16.6 percent of the black vote.  Though “strategic” voting by blacks in these elections cannot be dismissed (i.e., these candidates may have been the preferred candidates of black voters but did not receive many of their votes because these candidates were viewed as having little chance of being elected), the low level of support for these candidates may be due at least in part to candidate characteristics.

White Crossover Voting in Racially Competitive and Black Dominant Contexts

When we turn to elections in the racially competitive context, we note that the evidence is inconsistent with the modified black threat theory.  White voters in this context exhibited a higher (not lower) level of support for black candidates by comparison to white voters in the white dominant context.  (Only the difference between the racially competitive, and the black dominant contexts is as predicted by the modified theory – a higher level of white crossover voting in the black dominant context.)  Greater than predicted white crossover voting in the racially competitive context, relative to the white dominant context, suggests that white voters in the competitive setting accommodated to the fact that black candidates were now more likely to win. 

Elections in the black dominant context tended to exhibit the highest levels of white crossover voting.  This indicates that at least some whites voted in a strategic way for a black candidate when there was little realistic chance of electing a white candidate.  If white voters decided to strategically cast ballots for particular black candidates, how did these voters distinguish among black candidates to ascertain the one most favorable?  We address this question by considering the effect of newspaper endorsement and incumbency.

Newspaper Endorsement, Incumbency, and White Crossover Voting

Previous research indicates that the strength of black candidates can influence the candidate choices of white voters (Thernstrom and Thernstrom, 1997: 295-297).  The media are potentially a major cue-giver in biracial elections, and no cue is more explicit than a newspaper endorsement.  Black candidates who have an endorsement from a major newspaper are likely to enjoy higher levels of white support (Llorens, Parsons and Perry, 1997: 110; Bullock, 1985: 193-194; Watson, 1984: 167-169).   Evidence suggests that incumbency also is important for black candidates.  Black incumbents are likely to have more opportunities to engage in credit-taking activities and are more likely to have resources through which to attract white support by comparison to black nonincumbents (Bullock, 1985; Watson, 1984).

There were 81 black candidacies among the 29 biracial elections under investigation.  Seven of these were incumbents, and 11 had an endorsement from New Orleans’ major newspaper, the Times-Picayune.  Mean white crossover for the seven black incumbents was 32.6; for the 74 black nonincumbents, 7.0.  A t-test for equality of means indicates a statistically significant difference (t=6.29, p<.001).  This suggests that black incumbency enhanced the level of white crossover voting.  Mean white crossover voting for the 11 black candidates with an endorsement was 30.5; for the remaining 70 black candidates, 5.8.  A t-test reveals that these means are significantly different (t=4.40, p<.001).  This suggests that newspaper endorsement of black candidates increased the level of white crossover voting.

Table 4 reports results of two regression equations that estimate the effects of black density and black candidate strength, measured by incumbency status and endorsement by the Times-Picayune.[8]  Because six of the seven black incumbents were endorsed by the Time-Picayune, black incumbency and endorsement were entered into separate equations.  In addition, white incumbency status, a measure of white candidate strength, was entered into these equations.  A variable measuring election type, whether primary (coded 0, N=22) or runoff (coded 1, N=7) was introduced as a control because the level of white crossover voting is reported to be lower in runoff, than in primary elections (Bullock, 1985). 

The effect of incumbency is statistically significant, as is the effect of endorsement.  The findings indicate that newspaper endorsement may be somewhat more important in increasing crossover voting among whites than is incumbency.  The equation with black incumbency (Equation 1) yields an adjusted R-squared of .49; the equation with newspaper endorsement (Equation 2) yields an adjusted R-squared of .52.  Importantly, black density is a reliable, positive predictor of white crossover in both equations.

Table 4

 

CONCLUSION AND DISCUSSION  

Our goal was to test both traditional black threat theory and its modern modification as applied to white crossover voting.  To do so we examined white crossover voting across 29 biracial elections for municipal office in New Orleans from 1977 through 1998.  Black density does matter.  Our empirical analysis demonstrated that black density influenced the level of crossover among white votes.  Contrary to prediction based on traditional black threat theory, the relationship between black density and white crossover voting was not inverse: as the black density of the election unit increased, white voters voted in greater proportions for black candidates.

Our analysis uncovered a nonlinear relationship between black density and white crossover voting.  Contrary to prediction based on the modified black threat theory, the shape of this relationship was an upward half-U, instead of a full U-shape.  The major difficulty with the modified black threat theory is that it fails to account for the effect of an election unit’s racial makeup in non-black contexts.  The modified black threat theory correctly predicts high white crossover in the black dominant context, but does not accurately predict for the other contexts.  The white dominant context can seriously limit the likelihood of electing black candidates, and the racially competitive context may reduce the chances of electing white candidates.  As a consequence, changes in the racial makeup of an election unit can have considerable effect on white voter support for black candidates.

Our findings showed that the level of white crossover voting was lowest in the white dominant context (instead of in the racially competitive context as predicted by the modified black threat theory).  One explanation for the low level of white crossover in the white dominant context is that white strategic voting for a black candidate is unnecessary in this setting, given the small chance that a black candidate will be elected.  Furthermore, our investigation into white crossover voting in District A suggests that low white support for black candidates in the white dominant context may be due, at least in part, to the lack of strong black candidates -- only one black candidate in the white dominant context received newspaper endorsement, and there were no black incumbents.  This particular context may dissuade many otherwise viable blacks from seeking elective office.

Contrary to a confrontational political style suggested by black threat theory, the upward curved relationship between black density and white crossover voting points to an accommodative and strategic adjustment of white voters to a changing racial context.  In the racially competitive context, the likelihood of a black candidate being elected increased.   Stronger black candidates competed for office: five black candidates competing in this context received newspaper endorsement and three black candidates were incumbents.  Correspondingly, the likelihood of a white candidate being elected in this context decreased.  In this setting, white voters may seek out those black candidates perceived as most friendly to white group interests.  Sidney Barthelemy, New Orleans’ second black mayor, for instance, was considered such a candidate in the 1986 mayoral election (this election took place when the percentage of registered voters citywide was 51.3 percent, i.e., a racially competitive context). (for a further discussion of Barthelemy’s appeal to white voters see Hirsch, 1997: 279-280; Perry, 1997: 185-197; Persons, 1993: 197).

Black density, certainly, is not the sole factor that stimulates white strategic crossover.  In order to strategically protect group interests through supporting a black candidate, white voters need to be able to distinguish among black candidates.  They therefore need information.  White strategic crossover is facilitated by newspaper endorsement as well as by incumbency status, with endorsement likely being the stronger voting cue.  Casting a ballot for a white candidate who lacks a realistic chance of winning can be considered a wasted vote, and newspaper endorsement can identify to white voters those black candidates most “friendly” to white group interests.

END NOTES

1.  Compatible with this particular expectation, Vanderleeuw (1990) found a curvilinear relationship between the racial makeup of New Orleans’ city council districts and racially divided voting.  Voting was most polarized in racially competitive contexts, but less so when blacks were the dominant electoral force.  Vanderleeuw’s study, however, focused on racially divided voting, of which white crossover voting is one aspect.

2.  Twenty-two of these are primary elections, seven are runoffs.  Louisiana employs a bipartisan primary system.  All party candidates for the same office appear on the same ballot.  A runoff between the top to vote-getters is held if no candidate receives a majority of the primary vote.  

3.  Voters register according to race, which gives us a better idea of who votes then do population figures.  We use voter registration figures as our measure of an election unit’s racial makeup.  Citywide, there were 428 precincts in 1977; of these 80 were at least 90 percent black and 143 were at least 90 percent white.  There were 445 precincts in 1998, of which 130 were at least 90 percent black and 76 were at least 90 percent white.  Precinct population size is limited.  The city limits to about 1,000 the number of registered voters within a precinct; beyond this threshold, a precinct is divided.

4.  Another important feature of King’s EI procedure is that it allows researchers to model aggregation bias (that violates one assumption of King’s basic model) through the use of covariate(s) in extended models.  To discover possible aggregation bias, EI provides useful graphic diagnostics (King, 1997: 158-196; 282-291).  All estimates of precinct-level white crossover in the 29 elections under investigation were checked for possible aggregation bias based on King’s recommended diagnostics.  Extended models were employed for 18 of our elections.

The EI software is provided by King through his web page at http://Gking.harvard.edu.  The EI procedure first computes turnout rates for each racial group both at the election unit and at the precinct level.  Three precinct-level variables are used to estimate white turnout – the number of registered votes, the proportion of black registered voters, and the proportion of registered voters who signed in to vote in each precinct.  After white turnout is computed, the percentage of votes received by black candidates is used to calculate the level of racial crossover voting.

5.  Equations presented in Table 2 are based on an ordinary least squares model, which we use for interpretive purposes (see Burden and Kimball, 1998).  Equations based on a weighted least squares model (weighted by the inverse of the standard error) yielded very similar results, and are available from the authors.

6.  Some previous research suggests that the positive relationship between black density and white crossover voting may be explained by social interaction theory, which posits that the degree of spatial concentration of blacks determines the amount of racial interaction that whites have with black individuals (Carsey, 1995; Huckfeldt and Sprague, 1995; Kinder and Mendelberg, 1995; Huckfeldt, 1986).  As racial interact increases, “the probability that those [white] individuals will adopt political attitudes and behaviors similar to those [blacks] with whom they interact increases” (Carsey, 1995: 223).  The election unit level data employed in the present study, however, are not appropriate for testing social interaction theory.  The black density of a relatively large geographic area, such as a city or council district, may not relate to the level of racial interaction, which can better be measured at the neighborhood level (Huckfeldt, 1986).

7.  We do not mean that white voters necessarily changed their opinions on racial issues.  In aggregate, changing attitudes can result from the outmigration of whites with more conservative racial attitudes and/or the inmigration of whites with more liberal attitudes on race.

8.  As in Table 2, equations presented in Table 4 are based on an ordinary least squares model.  Results based on a weighted least squares model (weighted by the inverse of the standard error) were very similar, and are available from the authors.

REFERENCES

To the top

Back