Is Defunding the Police a “Luxury Belief”? Analyzing White vs. Nonwhite Democrats’ Attitudes on Depolicing
After the killing of George Floyd in May 2020, a surprising number of Democrats embraced calls to “defund” the police. According to data from the 2020 Cooperative Election Survey, 35.4% of Democrats expressed support for reducing spending on law enforcement.
Even as violent crime surged across the country, many Democrats remained supportive of defunding, which was supposedly necessary to achieve racial justice and equity. But support for defunding and depolicing is actually higher among white (and Asian) Democrats than among black and Hispanic Democrats. Relatively stronger support among the former, more affluent groups has led some to suggest that these attitudes are “luxury beliefs” that the privileged can afford to adopt to signal their virtue because they do not have to suffer the consequences. The luxury beliefs thesis thus suggests that socioeconomic status (SES) drives support for depolicing. But it is also possible that a genuine moral-political ideology, not affluence, plays an important role. This report is an attempt to empirically test the luxury beliefs hypothesis. It ultimately finds support for both the SES and ideology-centered accounts.
Section 1 shows that support for defunding and depolicing policies is indeed greater among white and Asian Democrats than black and Hispanic Democrats. Ideological self-identification, however, is found to be a stronger predictor of support than household income and education.
Section 2 examines levels of support when socioeconomic and demographic variables are held constant. Doing so nearly eliminates the difference in levels of support between Asian vs. black and Hispanic Democrats. The gap in support between these groups and white Democrats, on the other hand, is moderated but not eliminated. The results further indicate that ideology is a much more important driver of the white vs. nonwhite support gap than all socioeconomic and demographic variables combined.
Section 3 finds that in areas with high levels of violent crime, support for defunding the police generally falls for all groups except white Democrats. This is true even when socioeconomic, demographic, and political background variables are held constant.
Section 4 examines whether local racial residential segregation—which may insulate white Democrats from crime in neighboring, heavily minority communities—explains the likelihood of supporting these policies. The data demonstrate little to no evidence for this theory.
Section 5 argues that white Democrats’ policing attitudes are subject to unique group-based moral pressures, including collective moral shame and guilt. I examine the effects of an index of racial liberalism—which is found to highly overlap with measures of white moral shame and guilt in past research—on support for defunding and depolicing as local violent-crime levels increase. Holding all other variables constant, as local violent-crime levels increase, the positive effects of racial liberalism on support for defunding and depolicing become stronger for white Democrats and weaker for nonwhite Democrats. The effects for white Democrats are generally significantly larger than those for nonwhite Democrats.
Amid the “racial reckoning” and months of protests that followed the May 2020 killing of George Floyd, progressive political activists, politicians, and media pundits increasingly called for “defunding”—or even “abolishing”—the police. What was meant by proponents of these slogans varied—with relative moderates conceiving it as a reallocation of some police spending toward social services—but all generally agreed that too much money was being spent on traditional policing. Traditional policing practices, the narrative suggested, were “racially biased” and disproportionately harmed people of color, and thus scaling them back, if not altogether abandoning them, was supposedly crucial for the advancement of racial justice and equity.
The idea resonated with a surprising number of self-identified Democrats. For instance, according to data from the 2020 Cooperative Election Survey, more than one in three Democrats (35.4% vs. 3.9% of Republicans) wanted their state legislatures to reduce spending on law enforcement, which is up from roughly 1 in 10 (10.2%) in 2018. An even larger share of Democrats (53.6% vs. 8.3% of Republicans) expressed support for cutting the number of police on the street by 10% and spending more on other public services.
What made these public opinion trends even more surprising was that they coincided with surging levels of violent crime in cities across the county. Those that regard the crime-reducing effects of policing as “common sense” or empirically undeniable are left scratching their heads. Against the backdrop of rising crime levels, how can those who purport to care about blacks and other disadvantaged minorities be in favor of reckless policies that undoubtedly endanger them?
Incredulity over the Left’s position on policing tends to beget cynical—though not necessarily incorrect—answers to the preceding question. One popular account, for example, is that of “limousine liberalism”—a pejorative term for the politics of upper-class (and typically white) people who sanctimoniously preach support for progressive policies (e.g., racially integrating and diversifying public schools) while eschewing the personal sacrifices they entail or require (e.g., enrolling one’s child in a racially diverse public school). “Limousine liberals” are status-motivated hypocrites—they support “virtuous” policies for their social-reputational benefits while saddling others with the costs of their enactment. Of course, this argument goes, if they expected to personally suffer the consequences of such policies, they’d be singing an entirely different tune.
A more sophisticated formulation of this argument is Rob Henderson’s luxury beliefs thesis. For Henderson, a doctoral candidate in psychology at St. Catharine’s College, Cambridge, luxury beliefs are “ideas and opinions that confer [social] status on the rich at very little cost, while taking a toll on the lower class.” In the past, he argues, “the wealthy displayed their social rank with physical status symbols.” But “as trendy clothing and other material goods become more accessible and affordable, less status is attached to them.” Luxury beliefs have thus “arisen as a new status symbol.” For Henderson, then, support for defunding the police is an example of a luxury belief that the socioeconomically advantaged adopt as a means of signaling membership in a higher or “enlightened” social class. Crucially, because of their greater material resources and relative insulation from crime, those who espouse these beliefs “are not harmed by them, and if they are, the damage they incur is not as severe” as for others. On the other hand, for the socioeconomically disadvantaged, who tend to reside in areas with higher crime rates, such a policy realistically threatens their personal safety. And if that safety is ultimately compromised, they are less able to cope with the fallout. It is no surprise that surveys show that support for defunding the police is greatest among those in the highest income categories and smallest among those in the lowest.
The luxury beliefs thesis is both intuitively appealing and consonant with earlier theoretical frameworks, such as Maslow’s “hierarchy of needs” and Inglehart’s conception of “postmaterialism.” It also accords with the recent work of Enke, Polborn, and Wu, which, echoing the findings of Inglehart, observed an association between material well-being and voting on the basis of moral values (vs. material self-interest). In the authors’ interpretation, rich voters can “afford” to prioritize their moral values (which they term “luxury goods”) over their material concerns, which explains why wealthy liberals overwhelmingly vote for Democrats despite the party’s “tax the rich” economic policies. Similarly, an earlier “Hidden Tribes” report on U.S. political polarization found that the voter groups that attach the greatest priority to moral issues (“progressive activists” and “devoted conservatives”) are also the richest. However, apart from survey data showing its correlation with income, no research has systematically tested the luxury beliefs (or goods) thesis as it relates to support for defunding the police. This report is the first to do so.
Theoretical Implications and Empirical Expectations
Henderson argues that the wealthy ultimately needed to adopt new (nonmaterial) status markers to better distinguish themselves from those in lower social classes. Accordingly, in the context of (de)policing attitudes, one implication of this thesis is that Democrats’ support for defunding and depolicing policies is generally superficial. That is, Democrats who support such policies do so primarily for social-reputational reasons and only to the extent that they are personally costless, or at least “affordable.” They talk the talk, so to speak, knowing that they’re unlikely to ever have to walk the walk.
By this account, socioeconomic status (SES) and material security are central enablers of pro-depolicing attitudes. First, those of higher SES are more likely to live in low-crime areas. Second, should they be victimized by crime, they are more likely to have the resources and support system to recover. Taken together, Democrats of higher SES are better able to avoid or afford the consequences of defunding and depolicing policies.
If the luxury beliefs account is true, there are several empirically testable implications that follow. First and most basic, Democrats’ support for defunding and depolicing policies should increase as a function of socioeconomic status. Second, we’d expect support to be greatest among the most socioeconomically advantaged Democrats—namely, whites and Asians. Additionally, differences in support between racial/ethnic groups should shrink when all indicators of SES are held constant. Finally, we’d also expect support to decline for all groups as local crime levels increase.
But what if the role of socioeconomic status is overstated, and these beliefs are in fact the product of moral-political ideology? On this view, people support defunding not because it is a status marker but because they genuinely believe it is “the right thing to do” in the face of “rampant” police racism. Beliefs grounded in moral conviction or “sacred values” may be less sensitive to considerations of material self-interest, so we might expect such individuals to maintain their support for depolicing even when it entails personal risk. Their ideological beliefs may lead them to place little weight on these policies’ unintended negative consequences. And if they do consider them, they are likely to downplay or rationalize them as morally necessary. By this account, material security is incidental rather than instrumental to the expression of pro-depolicing attitudes. Wealthier individuals may be more inclined to express them not so much because they are better able to “afford” the policy costs but because they are more likely to espouse a moral-political ideology that is less sensitive to and/or justifies them.
If ideology is more central to the expression of pro-depolicing attitudes than socioeconomic status, several of the empirical expectations specified earlier must be modified. First, the SES-adjusted effects of political ideology on support for defunding and depolicing policies will be stronger than the ideology-adjusted effects of SES. In other words, even when indicators of SES are held constant, political ideology should still strongly influence the likelihood of endorsing such policies. When ideology is held constant, though, the effects of SES should be comparatively smaller. Second, holding SES indicators constant is not expected to meaningfully account for group differences in support for defunding and depolicing policies. Instead, these differences are expected to be largely driven by differences in political ideology.
This report ultimately finds support for both the SES- and ideology-centered accounts. In the main, both socioeconomic variables and ideology independently influence the likelihood of supporting defunding and depolicing policies, though the latter is relatively more important than the former. Further, whereas socioeconomic variables account for higher levels of support among Asians relative to black and Hispanic Democrats, ideology matters more for differences in support between the foregoing groups and white Democrats. Finally, and quite remarkably, local violent-crime levels significantly reduce support among nonwhite Democrats but have no effect, to significantly positive effects, on support among white Democrats. This discrepancy is not explained by conventional socioeconomic, demographic, and political background variables. Instead, the data indicate that white Democrats’ defunding and depolicing attitudes are powered by a unique group-based (or “privilege conscious”) moral ideology that boosts their support for such policies in high-crime areas.
Overview of Current Study
This report proceeds along five main empirical sections, each of which tests a different theoretical implication. The first tests whether, as the luxury beliefs thesis would predict, self-identified Democrats’ support for defunding and depolicing policies is indeed greatest among the most socioeconomically advantaged racial/ethnic groups, namely, whites and Asian Americans. It also tests and compares the expected positive effects of socioeconomic status and political ideology on the odds of supporting such policies. As expected, greater household income is found to significantly increase the odds of supporting defunding and depolicing policies—though these relationships are substantially confounded by educational attainment. White and Asian Democrats—the groups that score highest on all available indicators of SES—are significantly more likely than black and Hispanic Democrats, therefore, to support these policies. At the same time, ideological self-identification is found to be a stronger predictor of support than household income and education. And given that white and Asian Democrats are also more likely than their black and Hispanic copartisans to identify as liberal, the relevance of socioeconomic differences to explaining differences in support between these groups is questionable.
Section 2 attempts to resolve this uncertainty by testing whether and to what extent white and Asian Democrats’ relatively higher levels of support for defunding and depolicing is a function of their higher socioeconomic status. If material security facilitates the adoption of luxury beliefs, then Democrats of all racial/ethnic backgrounds—when matched on socioeconomic and demographic variables—should be equally likely to support defunding and depolicing policies. Indeed, when these variables are held constant, the difference in levels of support between Asian vs. black and Hispanic Democrats nearly vanishes. The gap in support between all of these groups and white Democrats, on the other hand, is moderated but not eliminated. The data instead indicate that differences in ideological self-identification alone are a much more important driver of the white vs. nonwhite support gap than all socioeconomic and demographic variables combined.
A possible reason for the weaker influence of SES on white vs. nonwhite differences in support is taken up in section 3, which tests whether the odds of Democrats’ support for defunding and depolicing policies are conditional on local violent-crime levels. Under the luxury beliefs hypothesis, the odds of support would be expected to fall across all racial/ethnic groups as local violent-crime levels increase. Although the greater material security of white and Asian Democrats may work to sustain their levels of support in high-crime residential contexts, there is less reason to expect differential patterns (e.g., odds of support fall for black and Hispanic Democrats but hold steady for white and Asian Democrats) when all measures of socioeconomic status are held constant. Perhaps surprisingly, this section finds that in areas with high levels of violent crime, support for defunding the police generally falls for all groups except white Democrats. This is true even when socioeconomic, demographic, and political background variables are held constant.
Section 4 briefly tests a potential explanation for the relative resilience of white Democrats’ defunding and depolicing attitudes in areas with relatively high levels of violent crime. Specifically, it examines whether local racial residential segregation—which may insulate them from crime in neighboring, heavily minority communities—explains the likelihood of supporting these policies. But the data demonstrate little to no evidence for this theory. By and large, white Democrats in relatively high-crime areas have similar odds of supporting defunding and depolicing policies, irrespective of the degree of local residential segregation.
A fifth and final empirical section explores whether the relative insensitivity of white Democrats’ defunding and depolicing attitudes to local violent-crime levels is at least partly attributable to the influence of a unique underlying moral ideology. Drawing on my dissertation work, I make the case that white Democrats’ policing attitudes are subject to unique group-based moral pressures, including collective moral shame and guilt; these pressures function to constrain the expression of self-interested policing policy preferences while promoting those that are perceived to protect people of color from institutional racism. Thus, as threats to self-interest (or personal safety) are likely to be greatest in high-crime areas, so, too, is the importance of moral pressures for supporting for defunding and depolicing policies. As a partial test of this account, I examine whether the effects of an index of racial liberalism—which is found to extensively overlap with measures of white moral shame and guilt in past research—on white Democrats’ odds of supporting defunding and depolicing become stronger as local violent-crime levels increase. I also perform the same analysis for nonwhite Democrats—for whom racial liberalism typically has less moral significance—and thus the effects should be significantly weaker. Both of these hypotheses are generally supported by the data. Holding all other variables constant, as local violent-crime levels increase, the positive effects of racial liberalism on support for defunding and depolicing become stronger for white Democrats and weaker for nonwhite Democrats. Further, outside of areas with relatively low levels of violent crime, the effects for white Democrats are generally significantly larger than those for nonwhite Democrats.
1. Examining Racial/Ethnic Differences in Democrats’ Support for Defunding and Depolicing Policies
If support for defunding and depolicing policies are luxury beliefs whose adoption is facilitated by material security, we would expect levels of support for these policies to significantly vary across different racial/ethnic groups. Because white and Asian Democrats are socioeconomically better off than their black and Hispanic counterparts—and those of higher socioeconomic status are less likely to be victimized by crime and are thus less likely to bear the consequences of these policies—we would expect the former to be significantly more supportive than the latter. Also, as Henderson argues, when the well-to-do are victimized by crime, they are better positioned to recover from it. They can thus “afford” such policies in ways that those of lower socioeconomic standing cannot.
Table 1 below, which reports data from the 2020 Cooperative Election Study (CES), shows that white and Asian Democrats do indeed fare significantly better than their black and Hispanic counterparts across all available indicators of socioeconomic status. Figure 1A illustrates the effects of two of these SES indicators, namely family income and educational attainment, on the probability of a Democrat endorsing reduced state spending on police, cutting the street police presence by 10% and spending more on public services, and both policies collectively. The left panel graphs the baseline odds of endorsement at varying levels of family income. As with the data cited by Henderson, we observe generally linear relationships such that the odds of endorsing each outcome increase at higher family-income categories. At the same time, the right panel indicates that these effects are heavily confounded by education: when educational attainment is held constant, the positive effects of income are substantially moderated. Greater income still matters, but the average effects of higher income are a small fraction of their baseline size. In contrast (though not shown), adjusting for family income has little to no effect on the attitudinal influence of education. Holding family income constant, moving from a high school education to a BA degree predicts a nearly 26-point increase (18.2% → 44.2%) in the odds of endorsing reduced police spending, a nearly 28-point increase (38.6% → 66.2%) in the odds of favoring a 10% reduction in the street police presence, and a 27-point increase (13.7% → 40.9%) in the odds of supporting both policies. This suggests that support for defunding and depolicing may actually be driven more by nonmaterial factors associated with higher education (e.g., moral values, political interest, openness)—and perhaps higher education itself—than it is by material security.
Summary Statistics for Various SES Indicators and Ideological Self-Identification, by Race/Ethnicity
The Baseline and Education-Adjusted Effects of Family Income on the Likelihood of a Democrat Supporting Defunding and Depolicing Policies
While greater education and (to a lesser extent) family income matter for the expression of pro-defunding and pro-depolicing attitudes, the data also suggest that ideological self-identification may be relatively more important. As shown in the right panel of Figure 1B, when both educational attainment and family income are held constant, moving from the “moderate” to “liberal” category on a 7-point measure of ideological self-placement (i.e., 1=very conservative, 7=very liberal) predicts 22.1-, 22.9-, and 22-point increases in the odds of favoring reduced spending on police (17.7% → 39.8%), decreasing the street police presence (40.5% → 63.4%), and both policies (13.3% → 35.2%), respectively. In contrast, when ideology and family income are held constant, a shift from a high school education to a BA degree corresponds to 22.8-, 18.0-, and 7.9-point increases in the odds of these outcomes, respectively. Further, when ideology and education are held constant, moving from the bottom (< $20,000) to the top ($200,000+) family-income category raises the odds of these outcomes by just 10.2, 5.3, and 11.7 points, respectively. Thus, while all of these variables independently positively influence the probability of giving pro-defunding and pro-depolicing responses, political ideology (followed by education) appears to matter the most.
The Baseline and SES-Adjusted Effects of Ideological Self-Placement on the Likelihood of a Democrat Supporting Defunding and Depolicing Policies
White and Asian Democrats both score highest on measures of socioeconomic status and are also most likely to identify as “liberal,” so it follows that they will be the strongest supporters of defunding and depolicing policies. Both the CES as well as data from the American National Election Study (ANES) show that this is indeed the case, at least as of 2020. Figure 2 below, which graphs the CES data (2014–20), displays the percent of Democrats in each racial/ethnic group who prefer that their state legislatures reduce spending on law enforcement. Between 2014 and 2018, the share of Democrats who gave this response was low and differed only modestly across racial/ethnic groups. For instance, in 2018, 12.1% of white Democrats supported decreasing police spending, as compared with 9.4% of black Democrats, 6.9% of Hispanic Democrats, and 9.1% of Asian Democrats. By 2020, though, and likely in response to the Summer of Floyd, these figures had jumped to 40% of white Democrats and 36% of Asian Democrats, compared with 24.6% of black Democrats and 25% of Hispanic Democrats.
Percent of Democrats Supporting Decreased State Spending on Law Enforcement, by Race/Ethnicity
Figure 3 shows comparable ANES data from 1984 to 2020. In 2016, 9.4% of 2016 white Democrats vs. 11.1% of black Democrats (and 7.1% of Hispanic and 0.6% of Asian Democrats) thought that federal spending on “dealing with crime” should be decreased. Four years later, in 2020, support for this position spiked, receiving support from 22.4% of white Democrats and 19.8% of Asian Democrats, compared with just 14% of black Democrats and 13.4% of Hispanic Democrats. Coincidently, support for increasing federal spending on dealing with crime fell 22.7 and 24.8 points, respectively, among white (from 57.2% to 34.5%) and Asian Democrats (69.9% to 45.1%) while declining by just 8.4 points among black (68.4% to 60%) and 12.2 points among Hispanic Democrats (68% to 55.8%).
Percent of Democrats Supporting Decreased Federal Spending on Dealing with Crime, by Year and Race/Ethnicity
Police spending is not the only policy area in which white and Asian Democrats now “out-woke” their black and Hispanic counterparts. As shown in Figure 4, as of 2020, white (22.9%) and Asian (32.2%) Democrats are also significantly less likely than black (39.1%) and Hispanic (41.4%) Democrats to support a 10% increase in the number of police on the street.
Democrats’ Support for Increasing Street Police Presence, by Race/Ethnicity
Indeed, Figure 5, which summarizes the data from the 2020 CES, reveals that nearly 60% of white Democrats and 50.7% of Asian Democrats expressed support for decreasing the number of police on the street by 10% (dark grey bars). Among both black and Hispanic Democrats, however, these numbers fall to roughly 42%. Further, whereas 35.9% of white Democrats and 30.4% of Asian Democrats endorse both defunding and depolicing policies (dark blue bars), just 17.8% of black and 19.8% of Hispanic Democrats do.
All told, as the luxury beliefs thesis would predict, the racial/ethnic groups that rank highest on measures of SES (i.e., white and Asian Democrats) are also the groups most supportive of defunding and depolicing policies. Whether SES alone accounts for these differences is the question I turn to next.
Democrats’ Support for Defunding, Depolicing, and Both Policies
2. Do Socioeconomic Differences Actually Explain Racial/Ethnic Differences in Support for Defunding and Depolicing Policies?
While suggestive, the fact that Asian and white Democrats tend to fare better on measures of socioeconomic status does not necessarily mean that socioeconomic factors explain differences in support for defunding and depolicing. Indeed, these groups were also the most likely to identify as liberal, which is more independently predictive of support than even educational attainment. A more rigorous test of this SES-centered hypothesis thus requires examining levels of support when all indicators of SES are held constant. In other words, if white and Asian Democrats’ higher levels of support for defunding and depolicing policies are fundamentally driven by their relative socioeconomic advantage, we would expect between-group gaps in support to largely, if not entirely, dissipate when all groups are matched on socioeconomic and demographic indicators. This prediction is tested below with a series of logistic regression models that regress each of the three outcome variables on the categorical race variable and differing combinations of socioeconomic, demographic, and political covariates.
Column (a) of Figure 6 reports the baseline odds of endorsement of each policy for each of the four main racial/ethnic Democratic groups. Column (b) reports each group’s odds of endorsement when the following are held to their median/modal values (shown in parentheses): age (52), family income ($60,000–$69,999), education (“some college”), stock ownership (none), census division (South Atlantic), type of residential area (suburb), marital status (married), number of children in the household (0), home ownership (owner), sex (female), and immigration status (parents, grandparents, and respondent all born in the U.S.). In the end, holding these socioeconomic and demographic indicators constant moderates but hardly eliminates white vs. black/Hispanic differences in the odds of supporting defunding and depolicing. The size of these covariate-adjusted differences, which are all statistically significant at the p < 0.001 level, averages roughly 10.5 points between white and black Democrats (down from a baseline average of 16.7 points) and 11.9 points between white and Hispanic Democrats (down from a baseline average of 15.9 points). Meanwhile, differences in support between Asian vs. black and Hispanic Democrats are, in fact, largely eliminated and no longer statistically significant at even p < 0.05. They now average just 0.5 points for black Democrats (down from a baseline average of 10.5 points) and 1.9 points for Hispanic Democrats (down from a baseline average of 9.7 points). Interestingly, the gap between white and Asian Democrats—which averaged 6.2 points at baseline—grows to an average of 10 points when socioeconomic and demographic indicators are held constant. This suggests that the (smaller) baseline difference between white and Asian Democrats is at least partly due to the latter’s relatively higher socioeconomic status (see Table 1).
Explaining Racial/Ethnic Group Differences in Democrats’ Support for Defunding and Depolicing Policies
Model (c) adjusts only for ideological self-placement (“liberal”) and strength of Democratic Party identification (“strong Democrat”). These variables account for slightly more of the average baseline difference between white vs. black (avg. baseline vs. avg. adjusted difference=16.7 vs. 9.8 points) and white vs. Hispanic (avg. baseline vs. avg. adjusted difference=15.9 vs. 10.7 points) differences than model (b), which adjusts for all SES and demographic indicators. The results indicate that differences in support between white and nonwhite Democrats are a function of political orientation and do not merely reflect socioeconomic and demographic differences.
Furthermore, whereas holding all socioeconomic and demographic indicators constant enlarges the average baseline white vs. Asian difference (avg. baseline vs. avg. adjusted difference=6.2 vs. 10.0 points), holding ideology and strength of party identification constant all but eliminates it (avg. baseline vs. avg. adjusted difference=6.2 vs. 1.2 points).
Socioeconomic and demographic differences explain most of the differences in support for defunding and depolicing between Asian Democrats and black/Hispanic Democrats, but they account for only a small share of those between the latter groups and white Democrats. Further, the relative SES advantage of Asian Democrats appears to explain why their baseline levels of support generally approach those of white Democrats. Indeed, when matched on SES indicators, the white-Asian gap widens.
What remains to be explained, then, is why SES accounts for more of the difference between Asian and other nonwhite racial/ethnic groups than it does for white vs. nonwhite Democrats. One possibility, which I further explore below, is that white Democrats of all socioeconomic backgrounds are less threatened by violent crime than nonwhite Democrats.
3. Assessing the Role of Local Violent-Crime Levels
Another way to test the luxury beliefs thesis is to examine whether support for defunding and depolicing policies varies as a function of local violent-crime rates. Indeed, conservatives often suggest that those on the woke left are siloed in predominantly white and upscale communities that do not suffer the consequences of the policies they advocate for. This could explain why white Democrats are more supportive of depolicing than nonwhite Democrats, who may be at greater risk of criminal victimization, which defunding and depolicing policies threaten to exacerbate. If relative immunity from the consequences (i.e., increases in violent crime) facilitates support for these policies, we would expect to find that support among Democrats, including white Democrats, falls (rises) in places with higher (lower) levels of violent crime. Alternatively, as Henderson suggests, if those with higher SES are more able to “afford” the consequences of these policies in high-crime locales, levels of support among white and Asian Democrats may remain stable as crime increases. But if so, we’d expect that, when matched on all socioeconomic and demographic indicators, all groups will be less likely to support depolicing as crime increases, because the matched groups would be more or less equally capable of absorbing the policies’ costs.
Unfortunately, one of the most readily available and widely used sources of data to measure local crime rates—the FBI’s Uniform Crime Reporting (UCR) county-level data set—is notoriously unreliable. First, not all police agencies report data, and thus estimates for many (typically small) counties must be imputed, which is likely to induce measurement error. Second, when a police agency spans several counties, crimes are distributed according to population. This is a problem insofar as crime is not evenly distributed by population. Finally, the most recent UCR data are from 2016, which means that I’d be using counties’ estimated 2016 violent-crime levels to measure 2020 policing attitudes. While levels of violent crime in a county in one year are likely to be highly correlated with those in subsequent years (e.g., a county that has relatively high levels of violent crime in 2016 is unlikely to have relatively low levels in 2020), this measurement strategy is far from ideal.
Zip code–level violent-crime estimates from CrimeGrade.org offer a potentially superior alternative. These estimates are outputted from a proprietary machine learning tool that leverages a combination of existing crime data and more than 180 indicators of an area’s characteristics to generate “crime scores” for zip codes across the United States. The company claims to regularly update these estimates, and it boasts the ability to “tell police departments where their common reporting errors cause issues.” Beyond this, though, the company provides virtually no information about its underlying data and estimation methodology. As such, its reliability is uncertain.
Given the limitations and uncertainties of both the county-level and zip code–level violent-crime indexes, I first test whether each is meaningfully predictive of self-reported past-year crime victimization. The results of these prediction models are visualized in Appendix B.2. Briefly, both raw indexes are significantly predictive of reporting past-year victimization. However, and particularly in the zip code–level models, log-transformed scales show the best fit to the data. For this and other reasons, these are the scales I use in the prediction models that follow.
Figure 7A shows the standardized average crime index scores for the zip codes and counties of Democrat respondents. Consistent with the “limousine liberal” stereotype, white (and, to a lesser extent, Asian) Democrats do indeed tend to reside in zip codes and counties that have relatively lower levels of violent crime. And, as suspected, Figure 7B shows that white and nonwhite Democrats in the lowest family-income categories nonetheless reside in areas with significantly different levels of violent crime. Once again, this could explain why socioeconomic variables are less relevant to gaps in support between white and nonwhite Democrats.
Violent-Crime Levels in Zip Codes and Counties of Democrat Respondents, by Race/Ethnicity
Log Zip-Code and County Violent Crime, by Family-Income Category
The question at hand is whether white Democrats’ greater support for defunding and depolicing is a function of these “sheltered” or low-crime residential contexts. I test this possibility with a series of logistic regression models. The first examines the attitudinal effects of zip-code violent crime for white vs. nonwhite Democrats as a whole by regressing each of the defund, depolice, and “both” variables on a white x zip-code violent-crime score interaction term. A second compares the effects for distinct racial/ethnic groups by substituting a categorical “race” variable for the white/nonwhite dummy. As a check on the reliability of the zip code–level estimates, I fit two additional models in which the FBI UCR’s county-level violent-crime estimates are substituted in the interaction term. Finally, I fit covariate-adjusted (i.e., “kitchen sink”) models that hold all SES, demographic, and political background indicators to their median/modal values. Further, to account for the possibility that the effects of crime are confounded by local population size, and to account for the possibility that group differences in support stem from differential rates of crime victimization (i.e., black Democrats may be more likely to be victimized by crime in high-crime zip codes than others), I also control for zip-code/county population size, as well as a zip-code/county-level violent crime x past-year victimization interaction term.
Figure 8A graphs the estimated baseline (dark blue bars) and covariate-adjusted (light blue bars) average marginal effect (AME) of a standard deviation increase in log zip-code violent crime on white and nonwhite Democrats’ probability of support for the policies. The estimated effects for white Democrats are entirely at odds with expectations. In the baseline models, the effects on each outcome are significantly positive, which indicates that white Democrats are actually more likely to support defunding and depolicing as local violent-crime levels increase. Indeed, on average, a 1-standard-deviation increase in log zip-code violent crime predicts a 2.6-point (p= 0.001) increase in the probability of supporting decreased state spending on law enforcement, a 1.4-point (p=0.030) increase in the probability of favoring decreases in the street police presence, and a 2.4-point (p=0.001) increase in the odds of endorsing both policies. These effects are reduced to statistical insignificance when all other covariates are held constant, but they remain positive, which is revealing, especially when examining the patterns for nonwhite Democrats.
For nonwhite Democrats, the effects of violent crime on the odds of support are significantly negative and significantly different from those of white Democrats across all outcomes. Specifically, a 1-standard-deviation increase in log zip code violent crime predicts a 2.1-point (p=0.005) decrease in the probability of supporting decreased spending on law enforcement, a 2.6-point (p=0.009) decrease in the probability of supporting a 10% decrease in the street police presence, and a 2.1-point (p=0.003) decrease in the probability of supporting both policies. Crucially, these effects remain significantly negative—and significantly different from those of white Democrats—when all control variables are held constant.
Average Effects of Zip-Code Violent Crime on the Probability of a White vs. Nonwhite Democrat Endorsing Defunding and Depolicing Policies
Figure 8B plots the baseline (solid lines) and covariate-adjusted predicted probabilities (dashed lines) of support for each group and outcome at varying levels of zip-code violent crime. We see that in zip codes with the lowest (i.e., –3SD) relative levels of violent crime, a white Democrat is expected to have a 33.0% chance of supporting decreases in police spending, a 55.8% chance of supporting a decrease in the street police presence, and a 29.4% chance of supporting both policies. In zip codes with the very highest (i.e., +3SD) relative rates of violent crime, these odds jump to 48.2%, 64.5%, and 43.7%, respectively. Although these differences are reduced to insignificance when adjusting for controls, it is noteworthy that white Democrats in high-crime zip codes do not become even nominally less likely to support either or both policies than those in low-crime zip codes. Among nonwhite Democrats, by contrast, those in high-crime zip codes are always significantly less likely to endorse these policies than those in low-crime zip codes. In the lowest-crime zip codes, unadjusted predicted odds of support for nonwhite Democrats are virtually identical to those of whites. A nonwhite Democrat in such locales is estimated to have a 35.5% chance of supporting decreased police spending, a 54.4% chance of supporting a decrease in the street police presence, and a 29.5% chance of supporting both policies. However, when moving to zip codes with the highest relative levels of violent crime, these figures fall to 22.8%, 38.7%, and 16.9%, respectively.
Local violent-crime rates influence the defunding and depolicing attitudes of nonwhite Democrats more than they do for white Democrats. In the lowest-crime zip codes, support for defunding and depolicing is virtually the same for white and nonwhite Democrats, but in higher-crime zip codes, white Democrats are more likely to support and nonwhite Democrats more likely to oppose these policies. Finally, when all control variables are held constant (i.e., the kitchen-sink models), the positive effects on white Democrats’ odds of support are reduced and no longer statistically distinguishable from zero for any outcome. The negative effects on the odds of support among nonwhite Democrats, however, are similar, if not larger, in magnitude and are still statistically significant at the p < 0.01 level for all three outcomes. They are also still significantly different from those for white Democrats. In other words, even when all socioeconomic, demographic, and political background variables are fixed to the same values across groups, nonwhite Democrats in high-violent-crime zip codes are still significantly less likely to support defunding, depolicing, or both policies than their white copartisan neighbors.
Predicted Probabilities of Supporting Defunding and Depolicing Policies at Various Levels of Zip-Code Violent Crime
Figures 8C and 8D graph the average effects and predicted probabilities, respectively, from the models in which a measure of county-level violent crime is substituted for the zip-code index. These estimates are very similar, if not identical, to those in the zip-code models.
Average Effects of Log County Violent Crime on the Probability of a White vs. Nonwhite Democrat Endorsing Defunding and Depolicing Policies
Predicted Probabilities of Supporting Defunding and Depolicing Policies at Various Levels of the County Violent-Crime Index
Next, I examine the heterogeneity of effects across the three main nonwhite racial/ethnic Democratic groups (blacks, Hispanics, and Asians). Figure 9A graphs the average baseline (dark blue bars) and covariate-adjusted (light blue bars) effects of zip-code violent crime on the odds of favoring defunding, depolicing, or both policies for each of the four main racial/ethnic groups. For black Democrats, the pattern we observe resembles that of nonwhite Democrats as a whole. Both the baseline/unadjusted and covariate-adjusted effects of violent crime are negative and statistically significant at the p < 0.05 threshold or lower across all outcomes. Figure 9B, which plots the unadjusted and covariate-adjusted predicted odds, shows that in zip codes with the lowest relative rates of violent crime, black Democrats have a roughly 32.8% (adjusted=40.5%), 54.3% (69.4%), and 27.0% (37.5%) chance of supporting defunding, depolicing, or both policies, respectively. In zip codes with the highest relative rates of violent crime, these odds drop to 19.9% (26.8%), 35.3% (48.8%), and 13.1% (19.6%), respectively. For Asian Democrats, the baseline effects are negative but indistinguishable from zero for all outcomes. The covariate-adjusted effects, on the other hand, are much larger and are significant at the p < 0.05 threshold or lower. In fact, the adjusted effect on Asians’ support for decreasing the street police presence is the largest effect observed for any group in any model. Holding all other variables constant, an Asian Democrat has an estimated 76.4% chance of supporting this policy in zip codes with the lowest relative levels of violent crime, as compared with a 36.9% chance in zip codes with the highest—a nearly 40-point difference. Finally, and quite curiously, the baseline effects of violent crime for Hispanic Democrats are all positive but near zero, while the adjusted effects are (with one exception) negative but near zero. Given that their baseline levels of support for these policies are very similar to those of black Democrats (for whom the effects of violent crime are all significantly negative), these null and directionally inconsistent results are somewhat surprising.
Average Effects of Zip-Code Violent Crime on Each Racial/Ethnic Group’s Probability of Endorsing Defunding and Depolicing Policies
Predicted Probabilities of Supporting Defunding and Depolicing Policies at Various Levels of Zip-Code Violent Crime
The results from the county-level violent-crime models, which are shown in Figures 9C and 9D, are again similar to those obtained in the zip-code models. One clear exception, though, is the effect of county-level violent crime on Hispanic Democrats’ odds of supporting a decrease in street police presence. Whereas the effects of zip-code violent crime on this outcome were positive but close to zero, those of county-level violent crime are both significantly negative and relatively large (Baseline AME= –5.3, p=0.012; adjusted AME= –5.9, p=0.002). More precisely, a Hispanic Democrat in a county that falls 2 standard deviations below the mean of the violent-crime index is expected to have a 55.4% chance (adjusted odds=68.7%) of supporting a decrease in the street police presence, as compared with a 34.4% (45.2%) chance in a county that scores 2 standard deviations above the mean.
Average Effects of Log County Violent Crime on Each Racial/Ethnic Group’s Probability of Endorsing Defunding and Depolicing Policies
Predicted Probabilities of Supporting Defunding and Depolicing Policies at Various Levels of County Violent Crime
In sum, the data show no evidence that white Democrats are more likely to support defunding and depolicing because they tend to reside in low-crime areas. To the contrary, white Democrats in zip codes and counties with the highest relative levels of violent crime are actually more likely to support these policies. Even if these positive relationships are largely the work of other variables, the best that can be said is that white Democrats’ defunding and depolicing attitudes are effectively indifferent to local violent-crime levels. However, the same cannot be said of nonwhite Democrats as a whole, and black Democrats in particular. In the lowest-crime areas, odds of support among nonwhite Democrats are very similar, but in higher-crime areas, levels of support among nonwhite Democrats drop significantly. Even when matched on all socioeconomic, political, and demographic covariates, nonwhite Democrats in higher-crime zip codes and counties are still significantly less likely than white Democrats to favor defunding and depolicing. This is particularly surprising given that when their relative socioeconomic advantage is statistically neutralized, the patterns for Asian Democrats closely approximate those of nonwhite Democrats as a whole. And yet, for whatever reason, support among white Democrats holds relatively steady.
We are thus left with the question of “why”—why are white Democrats in higher-crime areas more likely to support these policies than nonwhite Democrats of the same socioeconomic backgrounds and political leanings? In the section that follows, I briefly consider the potential explanatory role of racial residential segregation.
4. Assessing the Role of Racial Residential Segregation
Thus far, we have seen that even in higher-crime areas, white Democrats remain uniquely supportive of defunding and depolicing. It is possible, however, that, even in these locales, white Democrats are still insulated from violent crime due to high levels of racial segregation within the county or zip code. If so, we would expect white Democrats’ support for these policies to fall in high-crime areas with low degrees of racial residential segregation. Unfortunately, as far as I can tell, the zip code–level residential segregation data needed to test this hypothesis are not available. Therefore, I rely on county-level estimates from the American Community Survey (2015–2019), which I interact with the county-level violent crime variable. I then compute the predicted odds of endorsing each or both policies in counties with comparatively very high (+2SD) and low (-2SD) rates of violent crime at varying levels along the white-nonwhite residential segregation index.
The results are shown in Figure 10. First, we see that the odds of white Democrat support for decreased state police spending are indeed somewhat greater in high-crime counties with relatively high degrees of residential segregation and lower in high-crime counties with relatively low degrees of residential segregation. For instance, a white Democrat in a county that is +2SD on the county violent-crime index and +2SD on the county segregation index is estimated to have a 49.4% chance of supporting defunding. In contrast, in counties with the same level of crime (+2SD) but that fall two standard deviations below the mean of the segregation index, the odds are 43.4%. This 6-point difference, however, is not statistically significant (p=0.423). Moreover, the same pattern of results is even less evident in the remaining two outcomes. Finally, as shown in the bottom three panels of Figure 10, holding all socioeconomic, demographic, and political control variables constant eliminates virtually all differences in support between counties across the segregation index.
Predicted Probabilities of a White Democrat Endorsing Defunding and Depolicing Policies in Relatively Low vs. High-Violent-Crime Counties with Differing Degrees of Racial Residential Segregation
These findings suggest that residential segregation in and of itself is not a very meaningful buoy of white Democrats’ support for defunding and depolicing in higher-crime areas. At the same time, the plausibility of this relationship—coupled with the limitations of the segregation data—cautions against making any strong conclusions. The most that can be said is that the relationship is not clearly manifest in the present data, which is not to say that it doesn’t exist or wouldn’t emerge in other data. Only additional research can resolve this question.
5. Testing an Alternative Theory
If residential segregation does not explain white Democrats’ support for defunding and depolicing policies in relatively high-violent-crime areas, then what does? What is it that enables or incentivizes them to endorse policies that are likely to compromise the personal safety of themselves and others?
Questions like these were the focus of much of my dissertation work. And, following from the theory I outline and advance therein, I propose that the answer to this question is the same as for why many white Democrats support race-based affirmative action policies, which can also impose personal costs. Specifically, I contend that the defunding and depolicing attitudes of white Democrats are subject to unique group-based moral considerations and pressures that make them less sensitive to, or more tolerant of, personal risk. These group-based moral pressures are founded on guilt over racial inequality and the illegitimate advantages they perceive themselves and other white Americans as enjoying and shame over white America’s (im)moral legacy and continued racist tendencies. These collective emotions are able to inspire the adoption and expression of pro-outgroup (and, at times, anti-ingroup) attitudes and policy preferences. While a deeper theoretical discussion of these two collective emotions is beyond the scope of this report, it is necessary to briefly examine the theoretical underpinning of these constructs.
Recent theory characterizes collective guilt as a negative self-focused emotion that, generally speaking, follows from appraisals of personal moral inadequacy and, more specifically, from appraising the self as directly or indirectly complicit in past and/or present wrongdoing(s) against an outgroup. Some research finds that ingroup members who feel guilt report feeling angry or frustrated, which, in turn, engenders a desire for moral self-reformation. One means of coping with or resolving feelings of moral inadequacy is to support policies (or candidates) that promise to repair or compensate a victimized outgroup(s). For instance, whites who feel guilty because they believe they have benefited from their ingroup’s mistreatment of blacks can feel better about themselves by voting for black political candidates and supporting pro-black affirmative action policies.
Collective shame is related to guilt, but it is a more ingroup-focused emotion that relates to a global negative description (e.g., “racist”) of an ingroup’s moral character or to the perception that an ingroup is a serial violator of important moral values. “Ashamed” ingroup members are likely to report anger toward, or frustration with, their ingroup and a desire to morally distinguish themselves from it, and to challenge and reform it. They are also likely to feel a strong moral obligation to uplift or protect the victims of their ingroup’s cruelty as well as outgroups that resemble them (e.g., people of color). Thus, whites who score high on collective shame are much more likely to support equity-oriented pro-black policies, such as affirmative action, reparations, and race-based government assistance.
Guilt and collective shame are highly overlapping, but there are subtle differences between the two. “Guilty” whites are motivated by a need for personal psychological relief (via moral self-improvement), while “ashamed” whites seek to atone for, and morally distance themselves from, white America’s racist essence. The guilty are bothered by their own personal moral shortcomings; the ashamed are disgusted with those of white Americans as a whole.
Because guilt is self-focused, it will motivate pro-outgroup attitudes and behaviors only to the extent needed to resolve psychological distress. Moral shame, by contrast, is further reaching and more enduring because the moral “mission” isn’t complete until white America’s racist character—and the harms it has inflicted and continues to inflict on people of color—is fully corrected or neutralized. Until then, the “ashamed” must cope with their membership in an ignoble ingroup—one that, on account of their skin color, they can’t hide or escape from. By supporting policies that benefit nonwhite outgroups, though, they can at least signal to themselves and to others that they are morally different from “other” white people. Thus, my research finds white moral shame to be not only strongly predictive of support for pro-black policies but also of support for higher immigration levels, open borders, favoritism toward non-European vs. European immigrants, and antiwhite/pro-nonwhite sentiment.
Recent survey data suggest that feelings of race-based moral shame and guilt (to a lesser and more uncertain extent) abound among white Democrats. For instance, referring to Table 2 below, a nationally representative March 2021 YouGov survey reveals that nearly 80% (vs. 20.9% of white Republicans) indicate feeling “ashamed for the racist tendencies of white people,” while a similar proportion report that whites’ treatment of black people sometimes makes them think that white Americans are “racist and mean.” These figures approximate those observed in an August 2020 Prolific survey of white Americans that I conducted as part of my dissertation work, which, unlike the YouGov survey, included measures of guilt. For instance, in the Prolific survey, 69.7% of white Democrats (vs. 17.7% of Republicans) agreed that they feel “guilty about the social inequalities between white and black people,” while another 74.3% (vs. 28% of white Republicans) said they feel “guilty for the behavior of white Americans toward black people.” Given that the underlying sample was not representative of the wider white American population, though, these estimates must be treated with caution. That said, data from a nationally representative September 2020 Voter Study Group (VSG) survey show that just under half (47.7%) of white Democrats (vs. 5.8% of Republicans) agree that “white people should feel guilty about racial inequality.” Thus, a conservative estimate is that small to large majorities of white Democrats are at least somewhat “ashamed” of white America, while substantial minorities to large majorities express feelings of “guilt.”
The Prevalence of Expressions of White Moral Shame and Guilt by Party ID
I believe the theories above and related findings have important implications for understanding white Democrats’ attitudes toward defunding and depolicing policies. They suggest that considerations of self-interest will be relatively weaker determinants of policy preferences that implicate blacks and other “persons of color.” Indeed, especially after events like the killing of George Floyd, concerns of self-interest are likely to be less cognitively salient than—if not altogether overshadowed by—perceived group-based moral obligations and the need to affirm or distinguish themselves as “good white people.” In other words, white Democrats may be so focused on race-based moral concerns that they do not consider self-interest or personal safety. And if they do, there’s no guarantee that considerations of self-interest will prevail. Threats to self-interest may be downplayed, denied, or even rationalized. For instance, as part of my dissertation research, I ran an experimental study on a large sample of white Americans that, in part, sought to increase support for race-based affirmative action by inducing feelings of white shame and guilt. Even though the question measuring support for affirmative action explicitly mentioned the risk of discrimination against whites, those in the “shame/guilt” condition still expressed significantly higher levels of support, and these differences were largely mediated by my measures of moral shame and guilt. As one 38-year-old “strong Democrat” female participant explained:
I believe that African-American voters and citizens are still suffering the discriminations of past generations, not to mention ongoing discrimination today; I believe that governmental efforts including hiring preference for African-American applicants is the beginning of a process to improve the long-standing inequality. I do feel fear about my ability to compete in a professional world where I am at an explicit disadvantage, but I realize that I have to let go of that fear and recognize that many Black applicants have felt and suffered that disadvantage for decades.
Moral pressures can cause white Democrats to overlook, justify, or downplay the potential personal costs of both race-based affirmative action policies and defunding and depolicing policies. Objectively speaking, such policies may marginally increase their risk of criminal victimization, but assuming this consequence is even appreciated (and it generally may not be), it is the price to be paid for living up to their group-based moral obligations and, thereby, addressing feelings of group-based shame and/or guilt. Nonwhite Democrats, by contrast, do not carry the same moral burden or approach policing policies with the same group-based moral ideology. They may support defunding and depolicing policies when the price is negligible (such as when residing in a low-crime area), but they lack the moral drive necessary to sustain such support when the personal risks are higher. After all, they are on no moral mission and have nothing to prove to themselves or to others. So why risk it?
But if group-based moral considerations and emotions help to sustain white Democrats’ support for defunding and depolicing policies in high-crime areas, what evidence of this would we expect to observe in the data? Presumably, these moral constructs should have more impact on support in higher-crime areas—that is, in the areas where the costs of crime are higher and have to be outweighed by something stronger. In low-crime areas, moral conviction may still matter somewhat, but it is probably not a necessary condition for support.
Unfortunately, a proper test of this hypothesis requires measures of white shame and guilt, neither of which is included in any waves of the CES. The 2020 CES, however, does include four measures of racial attitudes, three of which were found to be highly correlated with white moral shame and guilt in the Prolific survey I ran as part of my dissertation work. These measures and—where available—their bivariate correlations with my dissertation’s indexes of shame and guilt are shown in Table 3. The three CES racial attitude items correlate at roughly r=0.74 with moral shame and at r=0.63 with guilt. Thus, though not a perfect proxy for shame and guilt, those who score high on the three items will also tend to score high on the shame and guilt indexes. The fourth item did not feature in my dissertation survey, though a similar measure of the perceived severity of existing antiblack discrimination correlated at r=0.681 with moral shame and at 0.581 with guilt. Averaging and combining the four items produces a sufficiently reliable index (α=0.896), which I hereafter refer to as “racial liberalism” in the analyses that follow.
With only a proxy measure of moral shame and guilt, we are forced to slightly modify the earlier prediction. While the effects of racial liberalism on white Democrats’ support for defunding and depolicing policies are still expected to be stronger at higher local levels of violent crime, we would also expect them to be stronger than those for nonwhite Democrats. White Democrats’ responses to the racial liberalism indicators are likely to substantially reflect group-based feelings of responsibility for racial inequality and black disadvantage—but for nonwhite Democrats, the same responses do not have the same moral implications. For instance, when a nonwhite Democrat “strongly agrees” that white people “have certain advantages because of the color of their skin” or that slavery and discrimination hinder black social mobility, it may simply be an impassive assessment of empirical fact—one with fewer and less obvious moral implications. As such, a nonwhite Democrat is likely to be less morally inhibited from supporting or opposing policies purely on the basis of material self-interest. On the other hand, for a white Democrat, this acknowledgment is likely to coincide with feelings of group-based shame, guilt, and responsibility for the origins and persistence of white advantage—feelings that limit or weaken the pull of self-interest and encourage the expression of support for policies that are perceived to benefit or protect blacks and other racial/ethnic minorities. While maintaining or increasing levels of police spending and the street police presence may objectively enhance their personal safety, it would also (in their view) lead to increased police harassment and violence against people of color—and what “good” liberal white person wants that on his/her conscience?
Bivariate Pearson R Correlation Coefficients Between Indicators of Racial Liberalism and Indexes of White Moral Shame and Guilt
To test the foregoing predictions, I begin by interacting the averaged racial liberalism scale with race and the zip-code violent-crime index. While holding all other control variables constant, I next calculate the average effects of a standard deviation increase in racial liberalism on the odds of endorsing each policy outcome at varying levels of zip-code violent crime. Estimates for white Democrats and nonwhite Democrats overall are shown in Figure 11A. With one exception, they generally align with expectations. For white Democrats, the effects of racial liberalism on the odds of supporting decreased police spending grow by an average of 2.4 points for each standard deviation increase in zip-code violent crime. In zip codes with the lowest relative levels of violent crime, each standard-deviation increase in racial liberalism is expected to result in a 20-point increase in the probability of endorsing this policy. In zip codes with the highest levels of violent crime, though, the same swing in racial liberalism is expected to produce a 34.5-point increase in these odds, which amounts to a statistically significant (p=0.007) 14.5-point growth in effect size.
For nonwhite Democrats, the pattern runs in the opposite direction. On average, each standard-deviation increase in violent crime corresponds to a nearly 3.5-point decrease in the effects of racial liberalism on the likelihood of a nonwhite Democrat supporting reduced police spending. In zip codes with the lowest relative levels of violent crime, a standard-deviation increase in racial liberalism is expected to increase the odds of endorsement by nearly 33.4 points, as compared with just 12.5 points in zip codes with the highest—a nearly 21-point diminution that is significant at the p < 0.05 level (p=0.010). Consistent with expectations, the effects of racial liberalism on defunding support in zip codes with the highest relative levels of violent crime are 22 points larger (p < 0.001)—or nearly 2.8 times the size—for white vs. nonwhite Democrats. This difference is further illuminated in Figure 11B, which plots the predicted odds of support for those at the top of the racial liberalism distribution (i.e., roughly 0.92 of a standard deviation above the Democrat mean). Specifically, in zip codes with the lowest relative levels of violent crime, a maximally “woke” nonwhite Democrat (predicted odds=52%) is 10 points more likely to support defunding than his/her white Democrat counterparts (42%). In zip codes with the highest relative violent-crime levels, however, maximally “woke” white Democrats (57.4%) are nearly 28 points more likely to be pro-defunding than their nonwhite counterparts (29.6%).
Average Effects of Racial Liberalism on White and Nonwhite Democrats’ Odds of Supporting Defunding and Depolicing Policies at Varying Levels of Zip-Code Violent Crime
Predicted Odds of Supporting Defunding and Depolicing Policies for Maximally “Racially Liberal” White and Nonwhite Democratic Respondents at Varying Levels of Zip-Code Violent Crime
For the “depolicing” model, the patterns of the estimates are still in the anticipated direction, but high- vs. low-crime differences in the effects of racial liberalism are much smaller and approach significance only for nonwhite Democrats. More precisely, in zip codes with the lowest relative levels of violent crime, a standard-deviation increase in racial liberalism corresponds to 26- and 34.9-point increases, respectively, in the odds of white and nonwhite Democrat support for decreasing the street police presence. In zip codes with the highest relative levels of violent crime, these odds go up by just 1.5 points (p=0.662) for white Democrats while dropping by 14.4 points (p=0.063) for nonwhite Democrats. While the average effects of racial liberalism in such zip codes are still larger for the former (27.6 points) than the latter group (20.5 points), this difference is not distinguishable from zero (p=0.160).
The results in the “support both” model generally fall between those of the previous two models. On average, a standard-deviation increase in zip-code violent crime enlarges the effects of racial liberalism on the likelihood of supporting both defunding and depolicing policies by 1.9 points for white Democrats while shrinking them by just over 3 points for nonwhite Democrats. When moving from zip codes with the lowest to highest relative levels of violent crime, the average effects of a standard deviation increase in racial liberalism grow by 11.5 points (22.9 → 34.4, p=0.045) for white Democrats and decline by 18.3 points for nonwhite Democrats (38.2 → 19.9, p=0.003). Thus, in the zip codes with the highest relative violent-crime levels, the effects of racial liberalism on the odds of endorsing both policies are roughly 14.5 points larger (p=0.002) for white than nonwhite Democrats. In practical terms, Figure 10B shows that in the lowest-crime zip codes, a maximally racially liberal white and nonwhite Democrat has an estimated 40.4% and 49.7% chance of endorsing both policies, respectively, as compared to a 53.2% and 29.6% chance in the highest-crime zip codes. Put differently, whereas the odds of support for maximally racially liberal white Democrats trails that of their nonwhite counterparts by 9.3 points at the bottom of the violent-crime index, they exceed them by 23.6 points at the top of it.
Figures 12A-B present the results for black, Hispanic, and Asian Democrats. Estimates for these groups are underpowered and thus very noisy. Nonetheless, all adhere to the same substantive pattern. Across all outcomes, the estimated effects of racial liberalism for each of these three groups become weaker—though not always (statistically) significantly so—as zip-code violent crime increases. Furthermore, and without exception, in zip codes with the highest relative violent-crime levels, the effects of racial liberalism for white Democrats are always at least nominally if not significantly larger than those for any of the three nonwhite racial/ethnic groups.
Average Effects of Racial Liberalism on Each Racial/Ethnic Group’s Odds of Supporting Defunding and Depolicing Policies at Varying Levels of Zip-Code Violent Crime
Predicted Odds of Supporting Defunding and Depolicing Policies for Maximally “Racially Liberal” Democrats at Varying Levels of Zip-Code Violent Crime
In sum, the results are broadly consistent with the notion that group-based moral pressures help to sustain white Democrats’ support for defunding and depolicing policies in high-crime areas. First, for white Democrats, the effects of racial liberalism—an index that greatly overlaps with measures of white moral shame and (to a lesser extent) guilt—become stronger (and, in most cases, significantly so) as local violent-crime levels increase. Second, despite being matched on all socioeconomic, demographic, and political background variables, the effects of racial liberalism in areas with the highest relative levels of violent crime were varyingly but consistently (and generally significantly) larger for white than for nonwhite Democrats. Indeed, and contra to the pattern for whites, the effects of racial liberalism for nonwhite Democrats become weaker at higher relative levels of violent crime.
Taken as a whole, these results suggest that racial attitudes are a more important driver of pro-defunding and pro-depolicing attitudes for high-crime-area-residing white Democrats than nonwhite Democrats. They also suggest that the racial attitudes of white and nonwhite Democrats are qualitatively different. Building on my dissertation work, I contend that these differences are, at least in part, the result of group-based moral considerations and emotions that are largely specific to whites. More specifically, I argue that collective shame, guilt, and a sense of ingroup responsibility for the past and continued effects of racism function to constrain or neutralize the otherwise dominant influence of material self-interest on white Democrats’ race-related policy preferences. Accordingly, to the extent that these moral constructs permeate the racial attitudes of white Democrats, they will be less likely to consider or will be less deterred by the potential personal costs of defunding and reducing the police presence in high-crime areas. Nonwhite Democrats are not subject to the same moral cross-pressure—they are less morally distracted or constrained in opposing policies that threaten their personal well-being. Of course, there may be alternative explanations for the patterns observed here, one of which I briefly discuss below.
6. Summary/General Discussion
Are pro-defunding and pro-depolicing attitudes “luxury beliefs”? Do Democrats who endorse such positions do so only to signal higher social status, and only insofar as they are relatively shielded from the consequences? This report took a first stab at empirically addressing these questions. Ultimately, the data it parses can give only mixed, qualified, and generally uncertain answers. On one hand, indicators of socioeconomic status, such as family income and educational attainment, were indeed predictive of Democrats’ support for defunding and depolicing policies. And holding these and other demographic variables constant virtually eliminated gaps in support between black and Hispanic vs. Asian Democrats. As the luxury beliefs thesis would predict, support among these groups also tends to fall—in some cases substantially—in zip codes and counties with relatively high levels of violent crime.
On the other hand, despite the emphasis the luxury beliefs thesis gives them, the data show that socioeconomic differences are far from the whole story. First, when adjusting for education, family income becomes a relatively weak predictor of support for defunding and depolicing policies, which calls into question the importance of material security in facilitating the expression of these attitudes. Second, the independent effects of education and family income on support were relatively smaller than those of ideological identification. Income and education, along with other socioeconomic and demographic variables, also collectively explained only a portion of the gaps in support between white vs. black and Hispanic Democrats. Additionally, holding all socioeconomic and demographic indicators constant enlarges the comparatively modest baseline white vs. Asian gaps, which further suggests that socioeconomic status is more important for the expression of pro-defunding/depolicing attitudes among Asian Democrats (and nonwhite Democrats in general) than among white Democrats.
The data reveal no evidence that white Democrat support for depolicing is due to their tendency to reside in low-crime areas. Indeed, the odds of white Democrats’ support for defunding and depolicing were found to be greatest in areas with the highest relative levels of violent crime. When all socioeconomic, demographic, and political variables are held constant, white Democrats in these residential contexts are still significantly more likely to support these policies than their nonwhite counterparts.
The reasons for white Democrats’ relative indifference or insensitivity to local violent-crime levels are not obvious. One intuitively plausible explanation—and the one most consistent with the luxury beliefs thesis—is that white Democrats in high-crime areas tend to live in racially segregated residential communities that buffer them from the crime in adjacent areas. Surprisingly, however, this was not clearly borne out in the data. White Democrats in counties with both relatively high levels of violent crime and high degrees of residential segregation were somewhat less likely to support decreased police spending than their counterparts in high-violent-crime, less segregated counties, but these differences were modest and did not approach statistical significance. Their direction was also reversed when it came to support for decreasing the street police presence, with those in relatively high-crime and weakly segregated counties having marginally but insignificantly greater odds of support. Finally, when all control variables were held constant, there were virtually no differences in white Democrats’ support for either or both of the policies in counties with relatively low vs. high degrees of residential segregation. Given the limitations of the data, it is possible that there are meaningful differences. With a larger sample or a more reliable, individual-level indicator of residential segregation, such effects might be present; these tests should thus be an object of future research.
It is also possible that white Democrats in high-crime areas are still socioeconomically better off than their nonwhite counterparts in ways that are not captured by conventional measures of family income, education, and stock and home ownership. While plausible, this account doesn’t explain why the adjusted attitudinal effects of violent crime were almost always significantly negative for Asian Democrats, who tend to outperform their white counterparts on various indicators of socioeconomic status (including those controlled for).
A third and final possibility I considered, which follows from my dissertation work, is that white Democrats experience unique group-based moral considerations and pressures that make them less sensitive to or more tolerant of personal risk. In other words, a “privilege conscious” moral ideology constrains the expression of self-interested policing policy preferences (i.e., sustaining or increasing levels of police spending and the size of the street police presence), especially for white Democrats residing in areas with relatively high levels of violent crime. Much as feelings of collective shame, guilt, and moral responsibility for racial inequity can cause whites to overlook, rationalize, or downplay the potential personal costs of race-based affirmative action policies, they can also cause them to overlook, rationalize, or downplay those of defunding and depolicing policies (e.g., greater risk of criminal victimization). In both cases, the underlying moral imperative is essentially the same—rectifying or preventing racial discrimination and furthering racial equity. Thus, in the context of post-Floyd policing, this account holds that for white Democrats, considerations of self-interest are less (if at all) salient relative to the moral urgency of protecting blacks and other communities of color from a perceived racist institution that so regularly victimizes them. It is a trade-off of (some) physical security for the moral and psychological security of being a “good” white person who stands up for the racially oppressed.
Nonwhite Democrats, by contrast, typically hail from historically victimized or disadvantaged racial/ethnic groups and thus do not confront the same group-based moral pressures. They have no historical sins or privilege to atone for, nor are they as frequently accused of “racism.” Other moral and social pressures, such as the need to signal intergroup or ingroup racial solidarity, may still shape policy preferences, but they are unlikely to be strong enough to override self-interest when policies impose costs on personal safety. As such, the defunding and depolicing policy attitudes of nonwhite Democrats are less morally inhibited or distracted from the pursuit of self-interest. Where such policies do not meaningfully or realistically threaten self-interest, such as in low-crime residential contexts, support for them is likely to be greater, as social pressures and concerns of racial solidarity will assume greater attitudinal influence. But where such policies realistically threaten self-interest, such as in high-crime residential contexts, support for them is likely to fall, as these other concerns and considerations will be outweighed by those of personal safety.
Unfortunately, a complete test of this theoretical account was not possible with the current data. However, this account suggests that group-based moral pressures (collective shame and guilt) are more necessary for defunding and depolicing policy support in high-crime areas, and measures of some of these moral constructs significantly overlap with available measures of racial liberalism. Thus, we predicted that the effects of racial liberalism on white Democrats’ support would strengthen as local violent-crime levels increase. For nonwhite Democrats, on the other hand, these moral constructs do not carry the same personal moral significance or constrain preferences—racial liberalism does not implicate themselves or their ingroups in moral wrongdoings and, therefore, does not impose a moral duty to correct them. Accordingly, for nonwhite Democrats, the effects of racial liberalism on support for defunding and depolicing policies were expected to weaken in residential areas with relatively high levels of violent crime. Finally, and following from these two divergent predictions, the effects of racial liberalism on support in high-crime areas were expected to be significantly stronger for white than nonwhite Democrats.
In the end, the above predictions were generally borne out in the data. For all three policy outcomes, the effects of racial liberalism on white Democrats’ likelihood of endorsement become stronger—and, for two of these outcomes, significantly so—with increases in local violent crime. The effects of racial liberalism on nonwhite Democrats’ odds of endorsement consistently shrink as local violent-crime levels increase—a pattern that holds for each of the three racial/ethnic groups under study. Though matched on all control variables, the effects of racial liberalism on endorsement in high-crime areas were always stronger (and generally significantly so) for white than for nonwhite Democrats. Though by no means dispositive, these findings suggest that white Democrats continue to support defunding and depolicing policies even in high-crime areas in part because of a “privilege conscious” or “white savior” moral ideology, which constrains self-interested policy preferences and promotes those that are perceived to protect or benefit blacks and other marginalized people of color.
It may be true that woke attitudes are luxury beliefs that some express to signal their membership in a higher social class, but this motive may not be uniformly applicable, or may operate differently for people of different racial/ethnic backgrounds. In context of pro-defunding/depolicing attitudes, a desire for moral self-affirmation—the assurance that one is a “good person”—seems to be a stronger motivator for white than nonwhite Democrats. And, crucially, the motive for white Democrats may not necessarily be to signal virtue to others but rather to prove their moral worth or probity to themselves, to prove that they are not like other “racist,” indifferent whites. Thus, whereas nonwhite Democrats may be likely to express support (which they are likely to jettison when inconvenient) for defunding and depolicing primarily as a means of fitting in with or conforming to group norms and (white) upper-class moral sensibilities, white Democrats may be more likely to do so on basis of genuine moral-ideological conviction.
Of course, even if white Democrats are sincere in their support for defunding and depolicing policies, Henderson is still correct that these policies disproportionately harm the socioeconomically disadvantaged, many of whom are “persons of color.” But if this is all that is meant by luxury beliefs—that they harm those they are meant to help—then it would not be a novel observation, nor would it be particularly relevant to contemporary political phenomena.
An alternative and more general interpretation of these results, which follows from the recent work of political scientist Eric Kaufmann, is that political ideology is a more important and all-encompassing social identity for white than nonwhite Democrats. The policy attitudes of liberal white Democrats exhibit greater ideological consistency or coherence than those of liberal nonwhite Democrats. And because ideology is more central to the social identity of the former, this consistency persists even when policies impose personal costs. Thus, by this account, racial liberalism has stronger positive effects on the defunding and depolicing attitudes of white Democrats in high-crime areas because they are more ideologically consistent and because political ideology is a more important component of their identity. The question, however, is why political ideology is more important and totalizing for liberal white than nonwhite Democrats. The results presented here are robust to controls for education and political knowledge, which suggests that differences in political sophistication—a critical antecedent of ideological consistency—do not appear to be decisive.
Kaufmann suggests that political ideology’s greater identity centrality for white Americans stems, at least partly, from their relatively weaker racial/ethnic attachments. The void of a strong “ascribed” identity—one inherited through birth—liberates whites to pursue “achieved,” or self-constructed, identities, including ideological progressivism. Nonwhites, by contrast, derive meaning and purpose from their racial/ethnic identities, and thus feel less need for all-encompassing political substitutes. Assessing the validity of this account—including whether it holds true for all nonwhite subgroups—is beyond the scope of this report. But it does not necessarily conflict with the theory I have outlined. First, white Democrats’ weaker racial/ethnic attachments may result from the shame they feel toward “whiteness.” Indeed, as discussed earlier, the desire to morally distance and distinguish oneself from a perceived ignoble ingroup (by, for example, adopting “woke” attitudes) is one of the hallmarks of group-based moral shame. Second, and relatedly, the absence of a strong positive “white” identity does not preclude the existence of a strong negative or “ingroup-critical” white identity. This antiwhite identity may be subsumed under a broader progressive ideology, but it is likely an important—if not the defining—feature of it. After all, race and racism are central to contemporary progressivism’s cosmology of human affairs, relations, and outcomes. Finally, as I expound on in my dissertation, ideological orientations are likely to condition whites’ susceptibility to negative, white-focused moral appraisals and emotions. For instance, white liberals are far more likely than conservatives to attribute existing racial disparities mostly, if not entirely, to past and/or enduring white racism. As such, they are (naturally) also more likely to experience feelings of shame and/or guilt when racial categories and inequities are made salient (such as after the killing of George Floyd).
Limitations and Future Research
Beyond those that apply to all correlational studies (e.g., relationships that cannot be shown to be causal), this report has at least two critical limitations. The first, which most severely affects analyses of the race x violent-crime interactions, is a lack of statistical power. While 23,000–28,000 observations are more than sufficient for detecting small between-group differences in policing policy support, this sample size is less than sufficient for detecting small between-group differences across different zip codes and counties. Zip codes in the 2020 CES, for instance, contained a median of four Democratic respondents (range=1–28), while counties had a median of 60 (range=1–749). Consequently, estimates of the attitudinal effects of violent crime for each racial/ethnic group are very noisy. The hope is that the aggregation of observations in zip codes and counties with similar levels of violent crime allows some meaningful signal, however faint, to emerge. And though I believe this signal is present, the level of noise doesn’t permit any confident inferences regarding the heterogeneous attitudinal effects of violent crime. Combining data from earlier CES waves doesn’t really help matters, either, as gains in statistical power are offset by decreases in effect size (because Democrats’ support for defunding and depolicing was substantially lower in 2016 and 2018). Future researchers thus have two conceivable (and rather inconvenient) options for improving the reliability of these estimates: independently acquire an even larger sample (i.e., where N is in the hundreds of thousands, if not millions) or patiently await the release of additional CES waves (and hope that existing attitudinal trends persist and that the same or similar measures are included).
A second limitation concerns this study’s measures of violent crime. These data likely suffer from considerable measurement error, and their time-limited nature only compounds this issue. Specifically, the county-level estimates were derived from aggregations of the 2014 and 2016 UCR violent-crime data, while CrimeGrade’s zip code–level estimates are regularly updated and not time-differentiated, which can be a problem to the extent that levels of violent crime in any given area can vary from year to year. Consequently, rather than modeling the effects of 2020 violent-crime levels on 2020 attitudes, the analyses modeled the effects of earlier (2014, 2016) and later (2022) violent-crime levels on 2020 attitudes. However, crime levels within counties and zip codes are correlated across time, so the degree of bias is unlikely to be large enough to render estimates useless. In fact, measurement bias can depress estimates, in which case this study is likely to have understated the attitudinal effects of violent crime. Still, researchers should aspire for measures that are maximally bias-free. Until more reliable annual zip-code and county-level estimates become available, researchers interested in further testing this relationship may have to make do with aggregated multiyear and/or cross-sectional measures of local violent crime.
Finally, additional research is needed to better understand the nature of white Democrats’ policing policy attitudes. One possible experimental approach is to test whether, as earlier suggested, manipulating group-based feelings of guilt and shame affects whites’ support for defunding and depolicing policies. Another can examine whether, or to what extent, white Democrats’ support for these policies rests on overestimates of their support among their nonwhite counterparts. White Democrats—perhaps because they are mostly exposed to the sentiments of social media activists—may believe there are high levels of black and Hispanic support for defunding and depolicing policies, and these (mis)perceptions may, in turn, inform their own support for them. Future research can test this possibility by examining whether informing white Democrats of the actual rates of black and Hispanic Democrats’ support reduces or moderates their endorsement of “woke” policing policies.
As far as I’m aware, this report constitutes the first systematic attempt at empirically testing the implications of Henderson’s luxury beliefs theory (and similar theories) as they relate to policing policy attitudes. My hope is that it won’t be the last. If the luxury beliefs theory is to become a viable empirical framework, rather than a mere culture war “own,” it behooves its proponents to formulate and test falsifiable predictions. If nothing else, I hope this report helps to move the ball closer to this end and that it encourages other researchers to move it still closer.
Appendix A. Replication Materials
All replication materials—including do-files and both raw and working data sets—can be accessed via https://osf.io/cq8rt/.
Appendix B. Violent-Crime Measures
Appendix C. Weighted and Unweighted Model Results for Transformed and Original Violent-Crime Indexes
Appendix D. Supplemental Graphs
About the Author
Zach Goldberg is a Paulson Policy Analyst who recently completed his PhD in political science from Georgia State University. His dissertation focused on the “Great Awokening”—closely examining the role that the media and collective moral emotions played in recent shifts in racial liberalism among white Americans. At MI, his work will touch on a range of different issues including identity politics, criminal justice, and understanding the sources of American political polarization. Some of Goldberg’s previous writing on identity politics in America can be found at Tablet and on his Substack.
In the summer of 2020, Goldberg joined MI President Reihan Salam, Columbia University Professor Musa al-Gharbi, and Birkbeck College Professor and MI Adjunct Fellow Eric Kaufmann for a conversation on the “Great Awokening”—the strong leftward shift among white liberals on issues of racial inequality and discrimination, immigration, and diversity that has been taking place since 2014.