Do White In-Group Processes Matter, Too? White Racial Identity and Support for Black Political Candidates Gregory A. Petrow University of Nebraska Omaha John Transue University of Illinois Springfield Timothy Vercellotti Western New England University Keywords: vote choice, white racial identity, prejudice, racial resentment, Barack Obama, race and elections Scholars find that negative evaluations of blacks lead whites to vote against black political candidates. However, can an in-group psychological process have the same effect? We consider white racial identity to be a strong candidate for such a process. We argue that the mere presence of a black candidate cues the identity, reducing support for these candidates among whites. We test this hypothesis on vote choice in seven instances. Five of them involve simple vote choice models: the 2008 and 2012 Presidential elections, and three elections in 2010: The Massachusetts Gubernatorial election, black candidates for the U.S. House, and black candidates for the U.S. Senate. The other two are tests of the notion that white racial identity reduced President Obama’s approval, thus reducing support for all Democratic Congressional candidates in the 2010 Midterm and 2012 Congressional elections. We find support for these notions in all seven cases, across these seven elections, using four different survey research datasets, and four different measures of white identity. Comparisons with other presidential elections show that white identity did not significantly affect mono-racial elections. Furthermore, we find the white identity and racial resentment results to be very 1
104
Embed
· Web viewDo White In-Group Processes Matter, Too? White Racial Identity and Support for Black Political Candidates Gregory A. Petrow University of Nebraska Omaha John Transue University
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Do White In-Group Processes Matter, Too? White Racial Identity and Support for Black Political Candidates
Gregory A. PetrowUniversity of Nebraska Omaha
John TransueUniversity of Illinois Springfield
Timothy VercellottiWestern New England University
Keywords: vote choice, white racial identity, prejudice, racial resentment, Barack Obama, race and elections
Scholars find that negative evaluations of blacks lead whites to vote against black political candidates. However, can an in-group psychological process have the same effect? We consider white racial identity to be a strong candidate for such a process. We argue that the mere presence of a black candidate cues the identity, reducing support for these candidates among whites. We test this hypothesis on vote choice in seven instances. Five of them involve simple vote choice models: the 2008 and 2012 Presidential elections, and three elections in 2010: The Massachusetts Gubernatorial election, black candidates for the U.S. House, and black candidates for the U.S. Senate. The other two are tests of the notion that white racial identity reduced President Obama’s approval, thus reducing support for all Democratic Congressional candidates in the 2010 Midterm and 2012 Congressional elections. We find support for these notions in all seven cases, across these seven elections, using four different survey research datasets, and four different measures of white identity. Comparisons with other presidential elections show that white identity did not significantly affect mono-racial elections. Furthermore, we find the white identity and racial resentment results to be very similar in terms of their robustness and apparent effect sizes. This indicates in-group evaluations, and those that focus on out-groups, operate independently of one another.
1
2
Whether or not race-related feelings and beliefs cost black candidates the
votes of whites in U.S. elections is a very important question for political behavior
scholars. Most blacks are the descendants of slaves, and the legacy of slavery
continues, as blacks suffer in comparison to whites in most of the ways the two
racial groups can be compared (see Kinder and Dale-Riddle 2012 for a summary).
A political avenue by which blacks may redress this inequality is by electing other
blacks to office; that is, if the white majority will elect them. Of significant
concern is that most of the progress blacks have made since the 1960s in holding
public office has been in majority-minority contexts. While blacks are 12% of the
U.S. population, they only hold 2% of all elected offices (e.g. Kinder and Dale-
Riddle 2012).
The academic study of white racial prejudice and voting is expanding
quickly. This literature can be divided into two eras: the pre-Obama era, and the
post-Obama era. Before the election of President Obama, there were relatively
few studies of this question, and they tended to yield mixed results. Some
scholars found that whites were not reluctant to vote for black candidates
(Bullock 2000, Citrin, Green and Sears 1990, Highton 2004, Voss and Lublin
2001), while others found they were reluctant (Bullock and Dunn 1999, Gay 1997,
Reeves 1997, Terkildsen 1993). The election of President Obama has generated
more interest in this question. Most scholars find that he did lose support among
whites due to racial factors, or racial prejudice (Tesler and Sears 2010; Highton
2011; Jackman and Vavreck 2010; Lewis-Beck, Tien and Nadeau 2010, Pasek et
*p<.05, **p<.01See Appendix C for full coefficient reportAll dependent variables coded as 2 party vote1984 to 2000: Analysis of American National Election Study data. The white identity measure is the dichotomous “Feel close to Whites” measure.2008: Cooperative Congressional Election Study Data. See Appendix A for white identity wording.2012: American National Election Study data. See Appendix A for white identity wording.1984 standard errors clustered on counties; 1988 and 1992 on census tracts; 1996 and 2000 counties; 2008 and 2012 Congressional DistrictsStarting with 1992, all models employ sample weightsControl variables in all models: Party identification, ideological identification, economic evaluations, gender, education, income, age, church attendance, and region (Deep South) Control variables in all models but 2008: Egalitarianism and limited government.Control variables in 2008 only: Support withdrawing from Iraq and a carbon tax.Control variable in 2012 only: Oppose Health Care Reform Act
16
We model the relationship between white identity and voting for the
Democrat for President for the elections of 1984 to 2000, and then 2008 and
2012. 6 The 2004 and 2008 ANES lacked identity measures. For 2008 we analyze
data from the Cooperative Congressional Election Study. One can see in the top
row of the table only the latter two coefficients are negative and statistically
significant – whites with stronger levels of white identity voted against Barack
Obama in 2008 and 2012. We include racial resentment as a control variable
when available (Tesler and Sears 2010, Kinder and Sanders 1996). It led whites
to vote against Michael Dukakis in 1988, Bill Clinton in 1992, and then Barack
Obama in 2008 and 2012. We also report the bivariate coefficients between white
identity and vote choice for comparison purposes.
We plot the white identity coefficients (from the full models) and their
corresponding 95% confidence intervals in Figure 1. One can see that before
2008, the coefficients centered around zero, and had very wide confidence
intervals. However, in 2008 and 2012, the coefficients center below zero, and the
confidence intervals stop before reaching the zero point, which reflects statistical
6 When available in the data, all of the models include this set of control variables:
Party identification, ideological identification, age, income, education, gender,
church attendance, and region (the Deep South). When available, we also
controlled for economic evaluations and political values. Given the political
dynamics of the 2008-12 elections, and data availability, we also elected to
control for relevant policy-related evaluations, such as evaluations of the
Affordable Care Act.
17
confidence that these results hold for the population. In other words, the hyper-
racialized thesis does not hold, as white identity only correlates with vote choice
when Barack Obama is the candidate. The null hypothesis does not hold as well.
The coefficients are in the units of logits, and thus are not intuitively
interpretable. To provide a sense of the substantive impact we translate the
Obama coefficients into predicted probabilities of voting for him in these two
elections given stronger levels of white identity, while holding the values of the
other variables in the analyses at their mean levels. We report the resulting
predicted probability plots in Appendix D, as Figures D1 for 2008 and D2 for
2012. To summarize those results, we find that white racial identity reduced the
predicted probability of the Obama vote in 2008 by about 30% (from 80% to 50%),
and in 2012 by about 10% (from 45% to 35%).
Our next case involving white identity affecting vote choice involves another
African American executive – the governor of Massachusetts, Deval Patrick.
18
Governor Patrick was first elected in 2006 and then re-elected in 2010. Here we
have two dependent variables: a pre-election measure of support for him (in late
October)7, and a post-election self-reported vote (see Table 2).8
Table 2: White Racial Identity among Whites, Democratic Vote Choice and Candidate Preference for Massachusetts Governor in 2010 among Whites, Logistic Regression
Variable Patrick Candidate Preference 2010
Patrick Vote 2010
White Identity^ -.38** (.14)
-.27* (.13)
Intercept^ -.26 (.77)
.77 (1.51)
N^ 390 204Pseudo R2 .42 .47Bivariate White Identity -.17* -.15
7 If the election for governor were held today, and the candidates were: Deval
Patrick, the Democrat; Charles Baker, the Republican; Tim Cahill, the
Independent; or Jill Stein, the Rainbow Party/Green Party Candidate, for whom
would you vote?
[Asked to those who answer Don’t Know or Undecided] At this moment do you
lean more towards: Deval Patrick, the Democrat; Charles Baker, the Republican;
Tim Cahill, the Independent; or Jill Stein, the Rainbow Party/Green Party
Candidate? Patrick supporters in question 1 and Patrick leaners in question 2
coded as 1; all others, coded zero.
8 For whom did you vote [in the 2010 Gubernatorial election]? Was it: Deval
Patrick, the Democrat; Charles Baker, the Republican; Tim Cahill, the
Independent; or Jill Stein, the Rainbow Party/Green Party Candidate? Patrick
coded 1; all others, zero.
19
coefficient (.08) (.09)**p<.01, *p<.05Data source: 2010 Massachusetts State Survey.See Appendix C for full coefficient reportResults incorporate sample weights for gender, age, and race.Models cluster standard errors on counties.^Control variables in these two models: Massachusetts right direction, party identification, education, income, female and age.
The survey interviewers at Western New England University who conducted
the Massachusetts State Survey re-interviewed more than 200 of the respondents
from the pre-election survey to ask them whom they actually voted for (the white
identity measure is from wave 1). This allows us to analyze both the pre-election
candidate preference (from wave one), as well as the self-reported vote choice
(from wave two). In both cases, stronger white identity predicts opposition to
Governor Patrick. In the last row of the table we report the bivariate coefficient
for comparison purposes.
To assess the substantive magnitudes of these relationships, we estimate
how the predicted probabilities of Patrick’s pre-election support and Election Day
vote choice share vary according to stronger levels of white racial identity. When
estimating this relationship, we hold the other variables in the analysis constant
at their means. We find that moving from the weakest level of white identity to
the strongest reduces Patrick’s support by about 15% on both measures. Because
half of all respondents chose the strongest white identity category, this
represents a meaningful reduce in his electoral support. We display these results
visually in Appendix D, Figure D3.
20
We now turn to an analysis of voting for Congress in the 2010 midterm
elections. We consider elections for the U.S. Congress in which black candidates
oppose white ones. In Table 3 we report results from the 2010 Congressional
elections. We begin with white voters in states with black U.S. Senate candidates
(South Carolina, Georgia and Florida). All black candidates were also Democrats,
and the dependent variable is respondents’ self-report of supporting the
Democratic candidate for U.S. Senate in an October pre-election survey
conducted by the ANES. In columns one and two, we include a term for the
interaction between the three states with U.S. Senate contests featuring black
candidates and white racial identity. The results are collinear, and so we mean-
center the interaction components to reduce collinearity (e.g. Jaccard and Turrisi
2003).9 In the first column the model includes only the three variables of interest
–white identity, black U.S. Senate candidates, and the interaction. However, the
second column reports the model we are the most interested in, because that
model includes a bevy of control variables. We expect the interaction coefficients
to be negative. We find that they are. Compared to elections without black
candidates, and compared to elections with black candidates but where
9 Some scholars find that centering items at their means before multiplying them
together in an interaction is not an advantageous collinearity solution. However,
Paccagnella (2006) finds that in the specific case of cross-level interactions,
centering is a good solution. We of course estimate such an interaction here. Our
collinearity statistics indicate that the centering solved the collinearity problem.
21
respondents endorse the lowest white identity category, whites with stronger
white identities will be more likely to oppose black candidates.
Table 3: White Racial Identity among Whites and 2010 U.S. Congress Candidate Preference among Whites, Logistic Regression Variable Democratic
U.S. Senate Candidate Preference
Democratic U.S. Senate Candidate Preference
^
Democratic U.S. House
Candidate Preference
Democratic U.S. House Candidate
Preference^
U.S. Congress InteractionsWhite Identity X Black U.S. Democratic Congressional candidate
-1.17* (.49)
-1.37**(.47)
-2.60*(1.22)
-2.34** (.86)
Interaction Effect ComponentsBlack Democratic U.S. Congressional candidate
-.32 (.32)
2.08* (1.04)
1.63* (.64)
2.18** (.63)
White Identity -.29 (.20)
.00 (.22)
-.34 (.18)
-.04 (.25)
Political PredispositionsDemocratic Party ID
.46** (.10)
.64** (.10)
Liberal Ideology ID .23 (.17)
.13 (.15)
Racial Resentment -.01 (.05)
.01 (.04)
Intercept -.02 (.16)
.59 (2.37)
-.19*(.08)
-.53 (1.46)
N 596 581 790 771Pseudo R2 .02 .50 .01 .50
**p<.01 two tailed test, *p<.05 two tailed testData source: 2010 ANES. Results incorporate ANES sample weight.See Appendix C for full coefficient reportU.S. Senate models cluster standard errors on States, U.S. House models cluster standard errors on Congressional Districts.All interaction terms centered at 0 to reduce collinearity.
22
^Control variables: Presidential approval, economic stimulus evaluations, Democratic legislative agenda evaluations, education, income, gender, region (South), church attendance and age.U.S. Senate elections, all three black candidates were DemocratsU.S. House elections, 16 major party candidates were black, 14 of them
Democrats
We perform similar analyses for U.S. House elections in 2010. We find that
the two interactions between white identity and black U.S. House candidates are
also negative and statistically significant. We plot the results from the table to
see how the predicted probability of supporting black Democratic candidates
varies with levels of white identity (for the model including the control variables).
We set the other variables in the analyses to their means. We begin by plotting
the results from the U.S. Senate vote choice model. We use the second U.S.
Senate vote choice interaction term from Table 3 to plot the predicted probability
of voting for black Democratic U.S. Senate candidates. We then use the white
identity coefficient from the same table to plot how the predicted probability of
supporting non-black Democratic candidates for the U.S. Senate varies with levels
of white identity. This results in Figure 2.
23
For the black U.S. Senate candidates, the solid line with the negative slope
indicates that accompanying each stronger level of white identity is a statistically
significant decrease in support for black Democratic candidates compared to the
weakest category of white identity (feeling “not at all” close to whites). In
elections with black U.S. Senate candidates, white identity appears to reduce the
predicted probability of supporting such candidates by about 60%. In contrast,
when Democratic candidates are not black, the line is flat and dotted, indicating
there are no statistically significant changes from the weakest category of white
identity to the strongest.10
10 The dotted flat line indicates that when Democratic U.S. Senate candidates were
not black, on average they received about 30% of the vote nationally. We note
that at the national level, Democrats regularly receive less than half of the vote
from whites, with supermajorities from non-whites generating the winning
margins when they sometimes occur.
24
As with the other analyses, we plotted the effect of white identity on the
predicted probability of supporting a Democratic candidate for the U.S. House,
for voters with both black and non-black major party candidates. This results in
Figure 3. We generate the results for this figure by holding all other variables at
their means. As with the U.S. Senate results, we represent the negative slope of
the predicted probability of support for black U.S. House candidates with a solid
line, indicating that the lower levels of white support, corresponding with
stronger levels of white identity, are statistically different from the weakest white
identity level. The predicted probability of supporting non-black U.S. House
Democratic candidates varies only slightly with stronger levels of white identity,
and we find no statistically significant differences, as reflected by the flat and
dotted line. Therefore we can reject the null hypothesis of no effect of white id,
and also hypothesis one, which claimed that higher white identification
diminishes support for all Democratic candidates. The evidence supports
hypothesis two – black candidates activate white racial identity among whites.
25
As with the results from 2008 and 2012 for the Obama vote, and 2010 for
the Patrick vote, we consider how the distribution of the white identity variable
could affect that variable’s contribution to the final election outcome. When
responding to this survey item, 55% of whites choose the two strongest white
identity categories. Similarly to the other operationalizations of white identity,
this measure reflects large numbers of whites’ willingness to claim a strong racial
identity, which then causes the identity to have a substantive effect on the final
vote outcome.
Presidential Approval as a Mediating Variable
Our previous results estimated the direct effect of white identity on vote
choice. However, a potent indirect effect exists as well – people’s approval levels
of the President affects whether or not they vote for the President’s party’s
candidates for Congress (e.g. Tufte 1975). Since we found that white identity led
26
whites to vote against Barack Obama, we test whether it also led them to approve
of him less. Given Kinder and Dale-Riddle’s (2012) and Tufte's (1975) claims, we
test whether white identity affected support for all Democratic Congressional
candidates across the country. Following the structure of our earlier analysis, we
begin by testing whether white identity affected Presidential approval of previous
Presidents. We run regressions in which Presidential approval is the dependent
variable, and we report our results in Table 4. If the racialization hypothesis
holds, we expect white identity will increase approval for Republican presidents,
and decrease it for Democratic presidents. Alternatively, if a black president
activates white identity, then we would expect to see that it only affects
Presidential approval when Barack Obama is President. In the last row of the
table we report the bivariate white identity coefficients for comparison purposes.
27
Table 4. White Racial Identity among Whites and Presidential Approval in 1984 thru 2000; 2008 thru 2012: OLS
*p<.05, **p<.01See Appendix C for full coefficient reportAnalysis of American National Election Study data for all election years but 2008; and in 2008, the Cooperative Congressional Election Study. In the ANES before 2010, the white identity measure is the dichotomous “Feel close to Whites” measure. In the CCES and the ANES in 2010 and 2012, it is the ones we report in Appendix A.1984 standard errors clustered on counties; 1988 and 1992 on census tracts; 1996 and 2000 counties; 2008, 2010 and 2012 Congressional DistrictsStarting with 1992, all models employ sample weightsControl variables in all models: Party identification, ideological identification, economic evaluations, education, income, age, marital status, church attendance, region (South)Control variables in all models but 2008 and 2010: Egalitarianism and limited governmentControl variable in 2010 and 2012: Oppose Affordable Care ActControl variable in 2008 only: Support Iraq withdrawControl variables in 2010 only: Oppose Democratic legislative agenda and oppose the economic stimulus bill
28
In this table we report predictors of Presidential approval for the election
years of 1984 through 2000, as well as 2008 to 2012. We do not report 2004
because the ANES dropped measures of identity in that survey. We find in the
top row of the table only one statistically significant result before 2010: In 2000,
whites with higher levels of white identity approved more of Bill Clinton. Given
our first hypothesis, we would expect the coefficient to be negative. However,
when Barack Obama is President in 2010 and 2012, the coefficient is negative and
statistically significant with a two-tailed test at the p<.01 level in 2010 and p<.05
level in 2012. We plot the coefficients and 95% confidence intervals from the fully
specified models in Figure 4.
We now consider the empirical implications for Congressional vote choice
when white identity reduces the Presidential approval of a black President. We
29
know from a vast trove of previous findings that Presidential approval affects
electoral support for the Congressional candidates from the President’s party
(e.g. Campbell 1993). One way to test this kind of mediating hypothesis is with a
Structural Equation Model (SEM) in which white identity affects approval in the
first stage of the model, approval affects vote choice in stage two, and an indirect
effect on electoral support for all Congressional Democratic candidates is
manifested through lower approval. This leads us to Table 5, in which we report
the results of four models. In the first two, we analyze 2010 ANES data, using the
October pre-election horserace question as the dependent variable. The first
column is for the U.S. House results, and the second for the U.S. Senate. In the
latter two columns we analyze the 2012 ANES data, using self-reported vote
choice as the dependent variables.
30
Table 5: Indirect Paths of White Racial Identity among Whites and White Candidate Preference for the U.S. Congress in 2010 and U.S. Congress Vote Choice in 2012, Structural Equation Models
Endogenous Variables Exogenous Variables
U.S. House 2010^
U.S. Senate 2010^
U.S. House 2012^
U.S. Senate 2012^
President Obama ApprovalWhite Identity -.12**
(.04)-.12** (04)
-.04** (.01)
-.05** (.02)
Racial Resentment -.06** (.02)
-.06** (.02)
-.09** (.03)
-.08** (.03)
R2 .68 .68 .68 .68Democratic Congressional Voting
President Obama Approval
.04** (.01)
.33** (.11)
1.00** (.18)
.51** (.08)
R2 .48 .61 .52 .69N 883 883 3,193 3,193
Indirect Paths to Democratic Congressional Voting thru Approval
White Identity
Racial Resentment
-.01*(.00)-.00* (.00)
-.04* (.02)-.02* (.01)
-.04** (.01)-.09** (.03)
-.02** (.01)-.04* (.02)
**p<.01, *p<.05^ 2010 results for October 2010 candidate preference; 2012 results for post-election vote choiceData sources: 2010 and 2012 ANESError terms of President Obama Approval and Democratic Congressional Voting correlated for all models but U.S. House 2010Results incorporate ANES sample weight.2010 models and 2012 U.S. House model cluster standard errors at the state level; 2012 U.S. Senate model clusters standard errors at the Congressional District level U.S. House Candidate Preference Model 2010: chi2 = 94.50 (p<.01) df=56, CFI = .78, TLI = .69, RMSEA = .028U.S. Senate Candidate Preference Model 2010: chi2 = 118.03 (p<.01) df=55, CFI = .64, TLI = .48, RMSEA = .036U.S. House Vote Choice Model 2012: chi2 = 376.81 (p<.01) df=102, CFI = .60, TLI = .47, RMSEA = .032U.S. Senate Vote Choice Model 2012: chi2 = 425.88 (p<.01) df=105, CFI = .48, TLI = .34, RMSEA = .031See Appendix C for tables with complete coefficient reports for control variables
31
In all four models we use SEMs to consider the indirect path of white
identity operating on Congressional vote choice, as mediated by Presidential
approval. In other words, we report results for two endogenous variables –
Presidential approval, and also Congressional vote choice. Both variables are
ordinal, so we use Mplus to adjust for the variables’ not being interval. We also
employ the ANES’ survey weights. We report the indirect effect coefficients at
the bottom of the table.
In the first row we report the coefficients for the effects of white identity on
presidential approval in 2010 and 2012, with coefficients estimated separately for
the Senate election models and the House election models. White identity
correlated negatively with President Obama’s approval in all four models, with
the coefficients statistically significant at the p<.01 level with a two tailed test.
We include racial resentment as a control variable, and it, too, correlates
negatively with the President’s approval.
We next consider how President Obama’s approval impacts voting in
Congressional elections. All four coefficients are substantively large and
statistically significant at the p<.01 level, indicating that in the Congressional
elections in both 2010 and 2012, people’s approval levels of the President were a
large force in determining which party to support in the election. At the bottom
of the table we find that in all four cases the indirect effect coefficients for white
identity on Democratic Congressional vote choice are all negative and statistically
significant at the p<.01 or p<.05 level with a two tailed test. We also present the
32
same indirect effects for racial resentment and find that they are generally of the
same magnitude, and also statistically significant in all four cases.
At the very bottom of the table we report model fit statistics. The root mean
square error of approximation (RMSEA) statistics all indicate that the models fit
the data well (an RMSEA of below .05 is considered good model fit; MacCallum,
Browne and Sugawara (1996)). We expect that the unexplained variance for
Presidential approval and Congressional vote choice will be correlated. We know
that Presidential approval and Congressional voting share a strong relationship,
and we acknowledge that even after modeling the causes of both, as we do here,
other factors probably remain unaccounted for by the model. Those factors are
manifest in the error terms for the two variables, and they should be related to
one another. Therefore, we correlate the errors of the two endogenous variables.
When we fail to do so, the model fit suffers in three of the four models. Please see
Appendix E where we describe the different model variations we estimated, and
where we also report the model fit statistics for the second-best fitting models
that we estimated.
The next results we report are for the magnitudes of the relationships
between white identity and support for Congressional Democrats. To present the
magnitudes of the results in a more accessible format, we plot how the predicted
averages of supporting Democratic U.S. House and Senate candidates in the
ANES 2010 pre-election survey, and the predicted vote choice in the ANES 2012
post-election survey, varied according to levels of white identity. Because the
pre-election Democratic candidate support variables are coded as 0 (supporting a
33
Republican or third party candidate) or 1 (supporting a Democrat), the average is
akin to a predicted probability of supporting the Democratic candidate. We
report this result as a range between 0 and 100 to ease interpretation. The
variance in support for these candidates, given levels of white identity, reflects
the indirect path of white identity on the election support, as mediated by
Presidential approval.11 These values are averaged over all of the other variables
in the models.
We report the resulting predicted probability plots in Appendix D, Figures
D4 (for 2010) and D5 (for 2012). In Appendix D we also include some text to
guide the interpretation of the figures. To summarize those results, we find that
in 2010 white identity reduced the predicted probability of supporting Democratic
U.S. House and Senate candidates by about 20% (from about 60% to 40%), and in
2012 by about 15% (from about 55% to 40%).
DISCUSSION AND CONCLUSION
There is little research on white identity and vote choice, and there is none
that tests for how variations in racial cues lead to its activation. There are studies
of how white racial identity relates to other constructs. Policy and/or social
attitudes—not vote choice—are the dependent variables in Wong (2010), Kinder
11 We estimated the models in Table 3 using Mplus. To estimate the predicted
averages, we took the coefficients for white identity and Presidential approval
from the Mplus output and fed them into Stata’s Structural Equation Model
program and then used the Predict command to generate the predicted score
averages and variances which we report here.
34
and Kam (2009), Citrin and Sears (2014), Wong and Cho (2005) and the
experiments reviewed in Hutchings and Jardina (2009). Vote choice is a
dependent variable in Tesler and Sears (2010) and Kinder and Dale-Riddle (2012),
but those authors do not test for the independent influence of in-group
identification except to the extent that their unidimensional scales (racial
resentment and ethnocentrism, respectively) pick it up. Jardina (2014, p 138 and
251) comes closest to our analyses. She tests for the influence of white identity
and controls for racial resentment. She also concludes that white in-group
identity affected vote choice in the 2012 presidential election. However, without
the replication of 2008 and the comparison to the mono-racial elections in the
1984 – 2000 elections, her vote choice analysis does not include variation in what
we claim activates white identity: a black candidate running against a white
candidate.
Thus, this study is the first to test for relationships between white racial
identity and vote choice across a range of elections. The central contribution of
this study is that white in-group evaluations lead whites to oppose black
candidates. White identity involves no explicit out-group evaluations at all. All of
the results we observe in this study, in which whites penalize black political
candidates, are due solely to whites’ affinity for their own group – their desire to
benefit their own group, without reference to any out-groups. This view
corresponds with the work of the psychologist Marilynn Brewer (1999) – in-group
“love” in and of itself can lead to out-group “hate.” To clarify both our meaning
and hers, “hate” involves the actions that the in-group takes against the out-
35
group. Previous scholars have found that negative feelings and attitudes toward
out-groups lead to acts against them, but Brewer (1999) argues that positive in-
group evaluations in and of themselves can lead to the same outcomes. Our
findings here support that view. This work builds on the work of scholars like
Sears and Tesler (2010) and Piston (2010) who find that negative evaluations of
other groups, vis-à-vis our positive evaluations of our own group, can lead whites
to vote against black candidates. Our work extends their analyses to an
additional causal process: in-group identity. The social identity literature shows
that in-group processes are distinct from out-group processes (Brewer 1999).
That is, there are two dimensions—attitudes toward the in-groups and attitudes
toward out-groups—rather than one dimension, where attachment to an in-group
can only mean greater distance from out-groups.
Our analyses show this empirically. Racial resentment and white in-group
identity have independent effects; they are not two ends of a single spectrum. In
1992 racial resentment is statistically significant in predicting voting against Bill
Clinton for President, but white identity is not. However in 2008 and 2012 white
identity is statistically significant and substantively important in the same
elections when racial resentment is also statistically significant. The magnitudes
of the results for the two factors are also comparable, and the white identity
results are at least as robust as the ones for racial resentment. 12 Racial
resentment tends to be a variable that deflates the impacts of other racial factors,
12 Both variables are coded between 0 and 1, and so the coefficients are
comparable.
36
but white in-group identity’s associations with voting in elections with black
candidates remains large and robust. This is evidence they are conceptually
distinct. Racial resentment is an attitude, which focuses on the out-group. White
racial identity correlates with racial resentment (as we report in Appendix B), but
it also leads to a whole constellation of beliefs about the in-group.
As with the effects of racial resentment on vote choice, black political
candidates cue (or activate) white racial identity (Kinder and Dale-Riddle 2012,
Jardina 2014). Tesler and Sears (2010), in their Table 3.1, statistically isolate the
effect of racial resentment from other elements of Presidential vote choice models
across several elections. They persuasively demonstrate that racial resentment
becomes more influential in contests involving Barack Obama. We find that while
white racial identity does not correlate with vote choice in most elections, it does
when black candidates are on the ballot. Obama's elections did activate white
racial identity. We find the most satisfying theoretical basis for this explanation
comes from Kinder and Dale-Riddle (2012, 25) when they write,
"In the short run, which aspects of identity and attitude become
important--which are activated--depend on political circumstances….
Even more effective is to embody membership, as Barack Obama
embodied race in 2008. … Whatever Obama said about society and
government and about problems and policies, at the end of the day,
every time American voters caught a glimpse of him, he was black."
Our analysis ends in 2012, but readers will likely wonder about its
applicability to 2016. We found that white identity affected support for
37
Congressional candidates even though President Obama was not a candidate in
those elections. Sides and Ferrell (2016) found that white racial identity
correlated with support for Donald Trump in the Republican primaries. This leads
us to wonder if white in-group identity’s influence will persist after President
Obama leaves office. Some analysts suggest that one reason for Donald Trump’s
victory is the activation of white in-group identity. However, the 2016 election is
confounded because while the election included both racial rhetoric and charges
of racism, it was also a referendum on a sitting black President. We do not know
whether white in-group identity can be activated by rhetoric only.
We tested four hypotheses for possible effects of white identity – white
identity would lead white voters to oppose all Democrats, no Democrats at all,
only when the identity was cued by a black candidate, or indirectly by lowering
President Obama’s approval. We estimated the effects of white identity on
Presidential vote choice and Presidential approval in Presidential elections going
back to Ronald Reagan’s re-election in 1984, and only for President Obama do we
find negative effects. This leads us to reject hypotheses zero (the null) and one
(white identity would lead whites to oppose all Democrats). We also find effects
for white identity on voting in Deval Patrick’s 2010 gubernatorial re-election
campaign in Massachusetts, and for U.S. Senate and U.S. House when candidates
are black. Hypothesis two thus stands – that black candidates cue the identity
and make it salient, causing white voting discrimination. Furthermore, we find
support for hypothesis three (the indirect effect on vote), because in 2010 and
2012 white identity apparently reduces President Obama’s approval, which then
38
indirectly reduces support for Democratic candidates in the Congressional
elections.
We note several limitations to these findings. First, the 2010 analyses rely
solely on pre-election candidate support measured in October. As a result, it is
possible that for the actual vote choice, the results may have been weaker.
However, we believe our interpretation is valid for several reasons. First, we
replicate the 2010 indirect effect of white identity on voting for Democratic
Congressional candidates using the 2012 data, which does involve vote choice.
Second, the pre-election candidate support in 2010 was measured less than a
month before the election, and so it is likely that the vast majority of voters had
already made up their minds as to whom they would vote for.
Second, in results we do not report here due to space limitations, the
correlations that we find between white identity and Congressional candidate
support in 2010 are not statistically significant in 2008 or 2012. We speculate
that in Presidential election years, the campaign and turnout stimulus of the
Presidential election swamps the effects of white identity on Congressional vote
choice (e.g. Campbell 1993). However, in midterm elections, in which
information about candidates is much less plentiful, a black candidate’s racial
group membership serves as a stronger cue. Or perhaps the effects of white
identity under a black President affect Congressional elections when the public
cannot vote on the President directly, but are focused on the President when he is
on the ballot.
39
Third, we find white identity correlates negatively with both pre-election
support for Deval Patrick, and also the vote choice, as reported by respondents
after the election. The vote choice result is especially notable given the small
sample size. Unfortunately, a full slate of control variables is not available in
those models; most importantly, the model lacks ideological identification and
racial resentment. While white racial identity is our only racial variable in the
models, the results are consistent with the results from the more complex models,
and so we have some confidence in the results.
A fourth limitation of the study is that the black Democrats running for the
U.S. Senate in 2010 were non-competitive candidates. Starting with Florida, that
Senate race featured a strong white Independent candidate, former Republican
Governor Charlie Crist, who opposed both Republican candidate Marco Rubio (of
Cuban-American descent) and Democratic candidate Kendrick Meek (who is
African-American). The Election Day vote shares were: Rubio with 49%, Crist
with 30%, and Meek with 20% (Federal Election Commission 2010). In the races
for U.S. Senate in Georgia and South Carolina, the margin of victory for the white
Republican candidate over the black Democratic candidate was 19 percentage
points in Georgia and 34 percentage points in South Carolina (Federal Election
Commission 2010). Because all three Senate races were non-competitive, then
the Senate results we find can only be generalized to non-competitive U.S. Senate
races with black Democratic candidates. However, we find apparent effects for
white identity in the Presidential elections of 2008 and 2012, and the
Massachusetts Gubernatorial election of 2010, and these elections were
40
competitive. Perhaps the apparent effect of white identity in U.S. Senate
elections with black Democratic candidates would be even stronger in competitive
elections.
WORKS CITED
American Association for Public Opinion Research. 2011. Standard Definitions:
Final Dispositions of Case Codes and Outcome Rates for Surveys. 7th
edition. AAPOR.
Ansolabehere, Stephen. 2008. Cooperative Congressional Election Study 2008.
[computer file] (Study Global ID hdl:1902.1/14003).
Arriola, Kimberly R. Jacob and Elizabeth R. Cole. 2001. “Framing the
Affirmative-Action Debate: Attitudes toward the Out-Group Members and
White Identity.” Journal of Applied Social Psychology 31(12): 2462-2483.
Barone, Michael and Chuck McCutcheon. 2011. The Almanac of American Politics
2012. Chicago: University of Chicago Press.
Bobo, Lawrence, and Gilliam, Franklin D. Jr. 1990. “Race, Sociopolitical
Participation, and Black Empowerment.” American Political Science Review
84(2):377-93.
Bollen, Kenneth A. 1989. Structural Equations with Latent Variables. New York:
Wiley.
Branscombe, Nyla R.. Michael T. Schmitt and Kristin Schiffhauer. 2007. “Racial
Attitudes in Response to Thoughts of White Privilege.” European Journal of
Social Psychology 37: 207-15.
41
Brewer, Marilynn B. 1999. “The Psychology of Prejudice: Ingroup Love or
Outgroup Hate?” Journal of Social Issues, 55(3): 429-44.
Brewer, Marilynn B. and Wendi Gardner. 1996. “Who is This ‘We’? Levels of
Collective Identity and Self Representations.” Journal of Personality and
Social Psychology 71(1):83-93.
Bullock, Charles S. III. 2000. “Partisan Changes in the Southern Congressional
Delegation and their Consequences.” in David W. Brady, John F. Cogan, and
Morris Fiorina (eds.), Continuity and Change in House Elections, pp. 39-64,
Stanford, CA: Stanford University Press.
Bullock, Charles S. III, and Richard E. Dunn. 1999. “The Demise of Racial
Redistricting and the Future of Black Representation.” Emory Law Journal
48: 1209-53.
Campbell, James. 1993. The Presidential Pulse of Congressional Elections.
University of Kentucky.
Carmines, Edward G. and James A. Stimson. 1989. Issue Evolution: Race and the
Transformation of American Politics. Princeton University Press.
Carmines, Edward G. and James A. Stimson. 1986. “On the Structure and
Sequence of Issue Evolution.” American Political Science Review 80(3): 901-
920.
Carmines, Edward G. and James A. Stimson. 1981. “Issue Evolution, Population
Replacement, and Normal Partisan Change.” American Political Science
Review. 71(1): 107-118.
42
Carter, Robert. 1990. “The Relationship between Racism and Racial Identity
Development among White Americans: An Exploratory Investigation.”
Journal of Counseling and Development 69: 46-50.
Carter, Robert, Janet E. Helms and Heather L. Juby. 2004. “The Relationship
between Racism and Racial Identity for White Americans: a Profile Analysis.”
Journal of Multicultural Counseling and Development 32 (January): 2-17.
Chow, Rosalind M., Brian S. Lowery and Eric D. Knowles. 2008. “The Two Faces
of Dominance: The Differential Effect of In-Group Superiority and Out-Group
Inferiority on Dominant-Group Identity and Group Esteem.” Journal of
Experimental Social Psychology 44: 1073-81.
Citrin, Jack and David O. Sears. 2014. American Identity and the Politics of
Multiculturalism. Cambridge University Press.
Citrin, Jack, Donald Phillip Green and David O. Sears. 1990. “White Reactions to
Black Candidates: When Does Race Matter?” The Public Opinion Quarterly
1(Spring): 74-96.
Coates, Ta-Nehisi. 2014. “Bill Clinton was Racialized, Too.” The Atlantic.
feelings, less admiration for blacks, greater linked fate with whites, the belief that
Obama favors blacks over whites, the belief that Obama is a Muslim, racial
resentment, and more positive white stereotypes over blacks. One coefficient for
feelings of sympathy for blacks is not statistically significant.
American National Election Study Construct Validity Variables
“feelings toward blacks, whites and Hispanics”:I'd like to get your feelings toward some of our political leaders and other people who are in the news these days. I'll read the name of a person and I'd like you to rate that person using something we call the feeling thermometer. Ratings between 50 degrees and 100 degrees mean that you feel favorable and warm toward the person. Ratings between 0 degrees and 50 degrees mean that you don't feel favorable toward the person and that you don't care too much for that person. You would rate the person at the 50 degree mark if you don't feel particularly warm or cold toward the person. Still using the thermometer, how would you rate the following groups:
58
BlacksWhiteHispanics
59
“Feelings of sympathy for blacks”:How often have you felt sympathy for Blacks? Always, about half the time, some of the time, or never?
“Feelings of admiration for blacks”: How often have you felt admiration for Blacks? Always, about half the time, some of the time, or never?
“Linked fate with whites”:Do you think that what happens generally to white people in this country will have something to do with what happens in your life? Yes or no?
“the belief that Obama favors blacks over whites”:Do the policies of the Obama administration favor whites over blacks, favor blacks over whites, or do they treat both groups the same?
“the belief that Obama is a Muslim”: Now we would like to ask you some questions about the religion of the presidential candidates. Would you say that [Obama] is Protestant, Catholic, Jewish, Muslim, Mormon, some other religion, or is he not religious? Codes so 1=Muslim, 0=rest.
60
Table B2: Cooperative Congressional Election Study 2008 Bivariate Regression Coefficients between White Identity and Other Constructs
McCain Anger
Perceived Anti-Black Discrimination
White Pride
Black Anger
McCain Feeling Thermometer
Obama Feeling Thermometer over McCain
Too Little White Influence
Too Little Black Influence
Racial Resentment
More angry at blacks than whites
Racial Threat
Blacks Cause Racial Tension
Logit Ordered Probit
Ordered Probit
Ordered Probit
OLS OLS Ordered Probit
Ordered Probit
OLS OLS Logit Ordered Probit
-.21* (.09)
-.10* (.04)
.22** (.06)
.12* (.06)
2.67* (1.12)
-5.74* (2.60)
.22** (.06)
-.14** (.04)
.22** (.04)
.12** (.04)
.22* (.11)
.18** (.04)
744 749 670 693 713 665 748 745 752 666 728 751**p<.01, two tailed test; *p<.05, two tailed test
We report results from 10 bivariate analyses which demonstrate that the white identity variable
correlates with 10 constructs that one expects to be related to a valid measure of white identity. First,
whites with stronger white identities reported feeling less anger toward John McCain. They were less
likely to perceive that blacks face discrimination. Third, whites with stronger white identities feel more
white pride, and also, greater anger toward blacks. These whites also feel warmer feelings for John
McCain as measured with a feeling thermometer, and stronger white racial identities correlate with
warmer feelings for McCain in comparison to Barack Obama. Stronger white identities correlate
positively with whites believing that whites have too little influence in society, and negatively with the
belief that blacks have too little influence. The variable predicts higher levels of racial resentment, as
61
well as a greater likelihood of answering that one feels angry at blacks, relative to whites. The CCES
measured racial threat by asking if people thought black civil rights leaders were “moving too fast,” and
stronger white racial identities predict that one agrees. Finally, stronger white identities increase the
likelihood that whites will answer that blacks are mainly responsible for creating racial tension.
62
APPENDIX C
Table C1. White Racial Identity among Whites and Democratic Presidential Vote Choice in 1984 thru 2000 and 2008 thru 2012: Logistic Regression
*p<.05, **p<.01All dependent variables coded as 2 party vote1984 to 2000: Analysis of American National Election Study data. The white identity measure is the dichotomous “Feel close to Whites” measure.2008: Cooperative Congressional Election Study Data. See Appendix A for white identity wording.2012: American National Election Study data. See Appendix A for white identity wording.1984 standard errors clustered on counties; 1988 and 1992 on census tracts; 1996 and 2000 counties; 2008 and 2012 Congressional DistrictsStarting with 1992, all models employ sample weightsDeep South: Respondents in former Confederacy states of South Carolina, Mississippi, Florida, Alabama, Georgia, Louisiana, Texas, Virginia, Arkansas, Tennessee, and North Carolina.
64
Table C2: White Racial Identity among Whites, Democratic Vote Choice and Candidate Preference for Massachusetts Governor in 2010 among Whites, Logistic Regression
Variable Patrick Candidate Preference 2010
Patrick Vote 2010
White Identity -.38** (.14)
-.27* (.13)
Massachusetts Right Direction16
1.46** (.36)
1.61** (.51)
Republican Party ID -1.27** (.10)
-1.39** (.19)
Education -.00 (.08)
-.25 (.14)
Income .23 (.27)
.11 (.56)
Female -.68** (.27)
-.78 (.57)
Age .04 (.15)
.20 (.25)
Intercept -.26 (.77)
.77 (1.51)
N 390 204Pseudo R2 .42 .47
**p<.01, *p<.05Data source: 2010 Massachusetts State Survey.Results incorporate sample weights for gender, age, and race.Models cluster standard errors on counties.
16 Do you feel things in Massachusetts are generally going in the right direction, or
do you feel things have pretty seriously gotten off on the wrong track? Right track,
wrong track, or do you not know?
65
Table C3: White Racial Identity among Whites and 2010 U.S. Congress Candidate Preference among Whites, Logistic Regression
Variable Democratic U.S. Senate Candidate
Preference
Democratic U.S. House Candidate
PreferenceU.S. Congress InteractionsWhite Identity X Black Democratic U.S. Congressional candidate
-1.37** (.47)
-2.34** (.86)
Interaction Effect ComponentsBlack Democratic U.S. Congress candidate
2.08* (1.04)
2.18**(.63)
White Identity .00 (.22)
-.04 (.25)
Political and Economic EvaluationsObama Approval .24^
(.14).13
(.11)Negative Economic Stimulus Evaluation
-.31* (.15)
-.40** (.12)
Negative Economic Evaluation
-.02 (.15)
-.15 (.13)
Disapprove Democratic Legislative Agenda
-.29** (.11)
-.15 (.08)
Political PredispositionsDemocratic Party ID .46**
(.10).64** (.10)
Liberal Ideology ID .23 (.17)
.13 (.15)
Racial Resentment -.01 (.05)
.01 (.04)
DemographicsEducation .13
(.22).13
(.18)Income -.05
(.04).01
(.03)Female -.05
(.31)-.00 (.37)
Deep South -2.30* (.92)
-1.35** (.50)
Church Attendance .03(.08)
-.07(.06)
Age .01 .00
66
(.01) (.01)Intercept .59
(2.37)-.53
(1.46)N 581 771Pseudo R2 .50 .50
**p<.01 two tailed test, *p<.05 two tailed test, ^p<.10, two tailed testData source: 2010 ANES. Results incorporate ANES sample weight.U.S. Senate models cluster standard errors on States, U.S. House models cluster standard errors on Congressional Districts.All interaction terms centered at 0 to reduce collinearity.Deep South: Respondents in former Confederacy states of South Carolina, Mississippi, Florida, Alabama, Georgia, Louisiana, Texas, Virginia, Arkansas, Tennessee, and North Carolina.U.S. Senate elections, all three black candidates were DemocratsU.S. House elections, 16 major party candidates were black, 14 of them Democrats
67
Table C4. White Racial Identity among Whites and Presidential Approval in 1984 thru 2000; 2008 thru 2012: OLS
Political Predispositions and EvaluationsRepublican Party Identification
.29** (.02)
.32** (.02)
.34** (.02)
-.28** (.03)
-.08** (.01)
.26** (.03)
-.28** (.03)
-.26** (.02)
Conservative Ideological Identification
.13** (.03)
.06** (.02)
.07* (.03)
-.19** (.04)
-.04** (.01)
-.04 (.07)
-.11 (.07)
-.05* (.02)
Egalitarianism -.28** (.03)
-.07 (.04)
-.07 (.06)
.16* (.07)
.03 (.02)
Not available
Not available
.03 (.03)
Limited Government
-.01 (.02)
.10** (.03)
-.03 (.05)
-.18* (.07)
-.04* (.02)
Not available
Not available
-.45** (.10)
Retrospective Economy
.48** (.05)
.28** (.04)
.45** (.04)
.43** (.05)
.03** (.01)
.40** (.11)
.37** (.06)
.28** (.04)
Prospective Economy
.23** (.05)
.06* (.03)
.17** (.06)
.21** (.07)
.02 (.02)
.12 (.09)
.11** (.03)
Current Economy
.26** (.03)
Withdraw Iraq -.71** (.16)
Ck Oppose Affordable Ca
-.44* (.06)
-.38** (.04)
68
Care ActOppose Democratic Legislative Agenda
-.10** (.03)
Oppose Economic Stimulus Bill
-.25** (.04)
Demographics and ControlsEducation -.01
(.01)-.09** (.02)
-.09** (.02)
-.05 (.03)
.01 (.01)
-.04 (.05)
.09* (.03)
-.03(.02)
Income .01 (.01)
.01* (.01)
.00 (.01)
.01 (.01)
.01** (.00)
.02 (.04)
-.01 (.01)
-.01**(.00)
Age -.00 (.00)
-.01** (.00)
-.00 (.00)
.00 (.00)
.00* (.00)
-.00 (.00)
-.00 (.00)
-.00(.01)
Church Attendance
.04 (.02)
.07** (.02)
.06** (.02)
-.03 (.04)
-.02** (.01)
.11** (.03)
-.01 (.03)
-.04**(.01)
Deep South .15 (.08)
.19** (.07)
.11 (.08)
.10 (.12)
-.03 (.02)
.13 (.14)
.08 (.10)
-.08(.05)
Face to face interview mode
.07(.06)
Intercept 1.14** (.29)
2.05** (.28)
.65 (.38)
2.32** (.48)
.56** (.14)
.24 (.40)
10.16** (.39)
3.85** (.31)
R2 .50 .37 .41 .57 .36 .55 .68 .68N 1490 1601 1513 691 1047 586 883 3176*p<.05, **p<.01Analysis of American National Election Study data from all election years but 2008; and in 2008, the Cooperative Congressional Election Study. In the ANES before 2010, the white identity measure is the dichotomous “Feel close to Whites” measure. For the CCES and the ANES in 2010 and 2012, see Appendix A.1984 standard errors clustered on counties; 1988 and 1992 on census tracts; 1996 and 2000 counties; 2008, 2010 and 2012 Congressional DistrictsStarting with 1992, all models employ sample weightsDeep South: Respondents in former Confederacy states of South Carolina, Mississippi, Florida, Alabama, Georgia, Louisiana, Texas, Virginia, Arkansas, Tennessee, and North Carolina.
69
Table C5: Indirect Paths of White Racial Identity among Whites and White Candidate Preference for the U.S. Congress in 2010 and U.S. Congress Vote Choice in 2012, Structural Equation Models
EvaluationsPositive Retrospective National Economic EvaluationsPositive Prospective National Economic Evaluations
Not available
Not available
Not available
Not available
-.12* (.06)
-.06 (.06)
Income .02 (.02)
-.02 (.02)
.00 (.01)
.01 (.01)
Age .00 (.01)
.01 (.01)
.03 (.02)
.07** (.02)
Female -.01 (.03)
-.05 (.22)
-.07 (.10)
.11 (.09)
Education .02 (.02)
.05 (.12)
.10 (.06)
.05 (.06)
Deep South -.12** (.03)
-.51* (.22)
-.26 (.19)
.01 (.20)
Church Attendance
-.01(.01)
.02(.04)
-.11** (.04)
Face to Face Interview Mode
-.41** (.10)
-.09 (.13)
Threshold -4.96** (1.01)
-1.63 (1.31)
.96 (.82)
.82 (.74)
R2 .48 .61 .52 .69N 883 883 3,193 3,193
Indirect Effects on Democratic Congressional Voting thru Approval
White Identity
Racial Resentment
-.01*(.00)-.00* (.00)
-.04* (.02)-.02* (.01)
-.04** (.01)-.09** (.03)
-.02** (.01)-.04* (.02)
**p<.01, *p<.05^ 2010 results for October 2010 candidate preference; 2012 results for post-election vote choiceData sources: 2010 and 2012 ANESError terms of President Obama Approval and Democratic Congressional Voting correlated for all models but U.S. House 2010Results incorporate ANES sample weight.
72
2010 models and 2012 U.S. House model cluster standard errors at the state level; 2012 U.S. Senate model clusters standard errors at the Congressional District level U.S. House Candidate Preference Model 2010: chi2 = 94.50 (p<.01) df=56, CFI = .78, TLI = .69, RMSEA = .028U.S. Senate Candidate Preference Model 2010: chi2 = 118.03 (p<.01) df=55, CFI = .64, TLI = .48, RMSEA = .036U.S. House Vote Choice Model 2012: chi2 = 376.81 (p<.01) df=102, CFI = .60, TLI = .47, RMSEA = .032U.S. Senate Vote Choice Model 2012: chi2 = 425.88 (p<.01) df=105, CFI = .48, TLI = .34, RMSEA = .031Deep South: Respondents in former Confederacy states of South Carolina, Mississippi, Florida, Alabama, Georgia, Louisiana, Texas, Virginia, Arkansas, Tennessee, and North Carolina.
73
APPENDIX D
74
Predicted Probability plots for Congressional Elections
75
Consider Figure D4. When whites report feeling “Not at all” close to
whites, for both the House and Senate, the probability of declaring support
for the Democratic candidate appears to be about 60%. The predicted
probability of support then appears to drop (although not monotonically),
reaching about 40% for both the House and Senate at the “Very” close to
whites category. In other words, whites who indicated that they felt “very
close” to whites were statistically less likely (at the p<.05 level) to report
support for Democrats, compared to whites who responded they felt “not at
all” close to whites. The same holds true for those feeling “moderately” and
“extremely” close to whites. While it appears that those who answered that
they felt “a little” close to whites also reported lower levels of Democratic
candidate support, those differences are not statistically significant – as we
indicate with the dotted line. The analysis indicates that white identity
reduces the predicted probability of support for Democratic U.S. House and
Senate candidates in the 2010 midterm elections by about 20%.
76
In Figure D5 we plot analogous results from the analysis of 2012. In
this case, self-reported vote choice from the 2012 ANES survey is the
dependent variable. The weakest category of white identity is for those
whites who answered “Not at all” to the question of how important being
white is to their identities. The top line has a negative slope, indicating that
whites with progressively stronger white identities were less likely to self-
report voting for Democratic U.S. Senate candidates – representing the
indirect effect of white identity on vote choice, operating through
Presidential approval. Whites who responded to the white identity question
with “a little” did not have a statistically significantly lower likelihood of
voting for Democratic candidates compared to the weakest white identity
77
category (as reflected by the dotted line). However, those who chose the
three strongest white identity categories (moderately, very and extremely)
were statistically less likely to vote for Democrats (as reflected by the solid
line).
The line below that is for the predicted probability of voting for
Democratic U.S. House candidates in 2012. The slope of the decreases in
the predicted probabilities given stronger levels of white identity is nearly
parallel compared to the decreases for the U.S. Senate. The white identity
categories with the statistically significant reductions in the predicted
probability of candidate support are identical to those of the U.S. Senate
results. It appears white identity reduced the predicted probability of
voting for Democrats for the U.S. House and Senate in 2012 by about 15%
(from 55% to 40%).
78
APPENDIX E
Additional SEM model configurations
For the 2010 SEMs, we estimated 16 different model configurations for both
the U.S. House and Senate vote choice models (32 different models total).
For the 2012 SEMs, we estimated 12 different model configurations for
each (24 total). In Tables 5 and C5 we report the results for the four models
with the strongest overall model fits.
The 2010 SEMs
In the “full model” we allowed Affordable Care Act evaluations, economic
evaluations and evaluations of the 2010 Democratic legislative agenda to
directly affect vote choice. In the next model we dropped health care from
the vote choice model (forcing the full effect to operate through Presidential
approval). Then we dropped the economic evaluation variables. Then we
dropped the Democratic legislative agenda evaluations. This results in four
different models. We then ran these four models clustering the standard
errors at the state level, and then again at the Congressional District level
(resulting in eight models). We then ran these eight models two more
times, correlating the endogenous errors, and then not correlating them (for
a total of 16 models).
For the U.S. House model, the next-strongest model fit correlated the errors
and clustered at the state level, dropping the Affordable Care Act variable
and the economic evaluation variables from the vote choice model. The
resulting model fit statistics are: chi2 = 93.71 (p<.01) df=56, CFI = .78, TLI
79
= .68, RMSEA = .028. The white identity indirect effect coefficient is not
statistically significant at the p<.05 level in this model. The chi-square
statistic indicates this model is a better fit, while the TLI indicates that the
model we report in the table is a better fit. The other model statistics are
identical.
For the U.S. Senate model, the next-strongest model fit clustered the
standard errors at the state level, did not correlate the endogenous errors,
and dropped health care and economic evaluations from the vote choice
model. The model fit statistics are: chi2 = 118.59 (p<.01) df=55, CFI = .64,
TLI = .47, RMSEA = .036. The white identity indirect effect coefficient is
statistically significant at the p<.05 level in this model. The chi-square from
this model indicates a worse fit compared to the one we report, as does the
TLI, while the others are identical.
The 2012 SEMs
In the “full model” we allowed evaluations of the Affordable Care act and
economic evaluations to directly affect vote choice. In the next model we
dropped the health care variable from the vote choice model (forcing the
full effect to operate through approval). In the next model we dropped
economic evaluations from the vote choice model. This results in three
models. We then ran them clustering at the state level, and then the
Congressional District level (resulting in six models). We then ran those six
models correlating the endogenous error, and not doing so (resulting in
twelve models).
80
For the U.S. House model, next strongest-fitting model is identical to the
model we report in Table 5, but the Affordable Care Act variable is allowed
to affect vote choice directly. The model fit statistics are: chi2 = 377.311
(p<.01) df=102, CFI = .59, TLI = .46, RMSEA = .029. The white identity
indirect effect coefficient is not statistically significant at the p<.05 level in
this model. All of the model fit statistics we report here indicate a worse fit,
compared to the model we report in the table.
For the U.S. Senate model, the next-strongest fitting model is identical to
the one in Table 5, excepting that the economic evaluations are allowed to
affect vote choice directly. The model fit statistics are: chi2 = 428.67
(p<.01) df=105, CFI = .47, TLI = .31, RMSEA = .032. The white identity
indirect effect coefficient is statistically significant at the p<.05 level in this
model. All of the model fit statistics we report here indicate a worse fit,