The Bernie Effect: Modeling Black Political Support
As primary season pushes forward and Sanders’ path to the nomination begins to tighten, some are asking what happened to his coalition of young voters? So far, exit polls consistently suggest that Sanders is indeed winning the votes of younger Americans by large margins. However, Sanders’ coalition has fallen short of victory in numerous states, and his most significant losses have come on the back of African-American constituencies expressing overwhelming support for Biden - a result that was predicted by previous public opinion polling. This phenomenon has left many asking why the Black community hasn’t rallied behind Sanders (some answers are found here and here), while others continue to argue that, similar to national trends, Sanders’ support among African Americans is generational too. In other words, in spite of a growing narrative around Black support for Biden, there might also be a significant, younger segment of the Black community that has attached itself to Sanders' political movement. I believe that examining African American support for Sanders and Biden, specifically, the relationship between support and age, can offer some clues - for instance, South Carolina exit polls (as well as results from the National Black Voter Project) suggest older Black voters overwhelmingly supported Biden, but young Black voters still only broke to Sanders at roughly the same rate as Biden. So, what role does age play, among other factors, in explaining Black support for the two remaining candidates?
Turning to the 2019 National Black Voter Project (BVP), I attempt to model Black support for both Sanders and Biden through a step-wise approach, first examining the influence of socio-demographic factors, and then adding political measures and possible attitudinal variables that may influence candidate support. At first glance, a number of variables are correlated with support for both candidates (support is measured by a 0-100-scale feeling thermometer, where 100=warmest attitudes of support). Table 1 displays correlation coefficients for the relationship between various socio-demographic, political and attitudinal variables and support for either Sanders and Biden. (Generally, correlation coefficients <.2 are considered “very weak”, so here we are looking comparatively across variables to identify the most correlated relationships.)
When it comes to Sanders, Table 1 suggests that ideology (more liberal), political interest and the threat of Trump (defined as the threat that Trump’s presidency is destroying the country) are the three variables most correlated with warm feelings toward the candidate. Notably, age is only weakly correlated with support for Sanders, and the relationship is a positive one, suggesting little evidence of significant differences between younger and older Blacks in their levels of support.
On the other hand, when it comes to Biden, age, political interest, and the threat of Trump are all somewhat correlated with support. A quick look at correlations suggests at least part of the narrative around support for Sanders and Biden is correct: strong liberal identity is related to support for Sanders, and Black support for Biden is related to age. On the other hand, support for both candidates is also correlated with political interest and the threat of Trump. As it turns out, correlation does not always equal causation. Thus, the next step is to actually model support using multi-variable regression analysis to examine the relationship between each predictor of candidate support while also account for all of the other variables discussed.
Modeling Support
Results from multi-variable regression analysis both confirm insights gleaned from the correlation matrix, and offer additional information about Black support for the candidates. For each candidate, Sanders and Biden, separate models were specified predicting feeling thermometers measuring candidate support. To start, when modeling the effect of socio-demographic factors, gender, age, family income and church attendance all significantly predict support for Biden, and using standardized beta coefficients, we can see that age is the most powerful predictor followed closely by gender. Conversely, none of the socio-demographic measures are significantly related to support for Sanders. Again, these results suggest some of the current narrative around Black voters is correct when it comes to age and even religiosity; however, surprisingly, none of the assumed factors - such as age or income, predict support for Sanders. In other words, Black support for Sanders cannot be attributed to differences in age, income, or any of the other socio-demographic variables modeled. What happens if we also take into political factors?
Figure 1. Predicting Support for Sanders & Biden: Socio-demographic factors
Figure 2 present the results of our initial socio-demographic models with added political measures for ideology, political interest and mistrust in government. According to Figure 2, after adding additional controls for political variables, gender, age and church attendance all remain significantly associated with support for Biden. (While family income is no longer significant with the added controls, we see that our measure of college education moves into significance as holding a college degree is negatively associated with support for Biden.) Once again, none of our socio-demographic variables significantly predict support for Sanders. Of the political factors, expressing a strong liberal identity predicts support for both candidates, and, as we would expect, a liberal identity is the most powerful predictor of support for Sanders. Figure 2 also shows that political interest predicts support for Biden, and mistrust in government predicts support for Sanders. After considering socio-demographic and political explanations, our profile of Black Sanders’ supporters remains somewhat bare, as only ideology and mistrust in government significantly differentiates Black respondents’ attitudes toward Bernie Sanders. Our final models add attitudinal measures of candidate support.
Figure 2: Predicting Support for Sanders & Biden: Socio-demographic and political factors
In the final model specification, measures for attitudes toward the future of equality in America, a respondents racial group identity (or in the case of African Americans, their “linked-fate”), and the extent to which a respondent perceives Trump presidency as a threat to the country. Figure 3 present the results of our final specified models predicting support for Sanders and Biden while accounting for socio-demographic, political and attitudinal factors. As Figure 3 suggests, gender, age, education, church attendance and political interest are all confirmed as significant predictors of Biden support. Furthermore, negative perceptions of Trump as a threat to the country also matter, and in a powerful way. Interestingly, once controlling for all factors, ideology no longer significantly predicts support for Biden, suggesting that even the most liberal Black respondents were no more of less likely than other to express support (or distain) for Biden.
Figure 3. Predicting Support for Sanders & Biden: Socio-demographic, political and attitudinal factors
When it comes to supporting Sanders, a strong liberal ideology, as well as negative perception about Trump, are the only measures significantly associated with positive attitudes. Put another way, the only factors that differentiate Black respondents who support Sanders from those who do not are ideology and the threat of Trump. (Since attitudes of mistrust are no longer significant, it is likely mistrust is encompassed by attitudes toward Trump.) Here, it is important to mention that these models are NOT making a strict comparison, or forcing respondents to choose one candidate over the other. So, in all fairness, the factors we’ve discussed may work in different ways if African Americans are making a clear decision between the two candidates.
Nonetheless, past work claims comparative thermometer ratings are useful in determining the order of candidate preferences, so we can confidently conclude that the variables significantly associated with candidate support in our models are useful in understanding preferences for candidates in an electoral context. That leaves us with two main conclusions: 1) Black support for Biden is a reflection of gender, age, education, religion, political interest and attitudes toward Trump; and, 2) Black support for Sanders is primarily due to ideology and the threat of Trump.
So as the country continues to debate who Black primary voters supported and why it is most important for Sanders’s to consider how he can reach out to the Black community and create a campaign and message that appeals to more than just the most liberal. Similarly, Biden needs to find a way to build a bridge to the most liberal segments of the Black community, and address the likelihood that college-educated Blacks hold less than warm perceptions of him. It’s also clear that the threat of four more years of Trump is driving Black attitudes toward both candidates, and regardless of who ultimately wins the nomination, candidates that focus on the importance of removing Trump will continue to gain support. However, the fact that Trump is driving attitudes for both candidates also speaks to the importance of building a relationship with the Black community that extends beyond Trump - especially for a progressive candidate such as Sanders whose favorability relies solely upon ideology and the threat Trump presents. In other words, since Trump is driving Black attitudes toward both candidates in a significant way, the Democratic Party must have a plan to continue to appeal to the Black community if Trump is no longer in office and the immediate threat of his administration is removed.