The Cultural Realignment of State White Electorates in the 21st Century

I use the term “white” to refer to people who are white and not of Latinx ethnicity. The focus on white state electorates is due to the divergent trends between whites and nonwhites, the greater homogeneity across nonwhite electorates, and some empirical limitations. All three reasons are discussed more fully below.

Republican party platforms “have asserted the fundamental right to life of the unborn; have criticized efforts to extend civil rights protections to homosexuals [and] have called for an expansion of public religious expression, particularly in the schools…. Democratic platforms, in contrast, have asserted the fundamental right of women to choose an abortion, have strongly supported the Equal Rights Amendment, and have called for expanded protection of the rights of homosexuals. Moreover, although there have been clear differences between the parties’ stands on cultural matters since 1980, these differences have grown ever larger over time” (Layman 2001, pp. 122–123).

For example, Miller and Schofield (2003, 2008) claim there is long-term periodicity driven by underlying instability of party competition in multidimensional space while Claassen (2015) emphasizes how demography and demographic change shape the political landscape and composition of the activist pool. of the antecedent causes, there remains agreement regarding the importance of activists to the process of party realignment.

“The processes of political change proceeds under the handicap of considerable friction. Preexisting party attachments may have a durability that contributes to the lag…. A new generation or so may be required for one outlook to replace another” (Key 1959, p. 204).

To be sure, against these two advantages is the disadvantage of not observing individual-level change. For example, while the results below clearly show that the partisan alignments of state white electorates have changed over time to come into alignment with preexisting cultural attitudes, the individual-level processes, if any, by which this has come about remain unobserved.

On the question of the micro-level causal interdependencies between cultural issue preferences and party identification, the empirical evidence is mixed. Carsey and Layman (2006) analyzes abortion policy preferences and finds that even among people for whom the abortion issue is salient, the effect of party identification on abortion policy preferences is stronger than the effect of abortion policy preferences on party identification. Goren and Chapp (2017) finds almost an exact opposite pattern. Rather than people brining their policy preferences into alignment with their party attachments, Goren and Chapp (2017, p. 124) reports that “most people ground their partisan identities in judgments about the frontline issues in the culture war [abortion and gay rights].”

As in the case of party identification, if there is a contemporaneous correlation between PRESVOTE and PID or CULT, then assessing the direction of influence becomes much more complicated.

While 16 years proves to be ample time for the analyses, it is important to note that there is nothing especially significant about 2000, either theoretically or empirically, compared to 1996, 2004, or some other starting point. The key is that the period of time encompassed (a) be long enough to reveal what, if any, realignment has taken place and (b) cover at least part of period during which cultural issues have risen to prominence in American politics.

For example, as the 2016 presidential election approached Pew (2016, p. 6) found that “[w]hile the GOP has made gains overall and among key groups…blacks and Hispanics are as likely to identify as Democrats or lean Democratic today as they were four or eight years ago.”

There is also much more homogeneity in preferences across states for African Americans compared to non-Latinx whites. In the 14 states with more than 250 African American voters, the Democratic margin of victory ranges from 70 to 93% points with a standard deviation of about 6 points. In contrast, there is much more heterogeneity among non-Latinx whites with a range of − 11 to 63 points and a standard deviation four times larger than that for African Americans, 24 points. Thus for African Americans, and possibly Latinxs, variation across states—the focus of this paper—may be less important than variation over time, especially very long periods of time, from reconstruction to the present day.

I coded party identification with the NAES data into five categories: strong Democrats (− 100), not strong Democrats, including “leaners,” (− 50), pure independents (0), not strong Republicans (+ 50), and strong Republicans (+ 100).

These polls were funded by a consortium of news organizations and conducted in all 50 states in 2000, 2004, and 2008. In 2012 and 2016 state exit polls were only conducted in selected states. Strong and not strong partisans are not differentiated in these surveys so I code party identification from the exit polls as a trichotomy with Democrats (− 100), independents (0), and Republicans (+ 100).

Like party identification from the NAES I code five categories in the CCES data: strong Democrats (− 100), not strong Democrats, including “leaners,” (− 50), pure independents (0), not strong Republicans (+ 50), and strong Republicans (+ 100).

This involved two steps. To put the CCES white state partisanship scores on the exit poll scale, I first regressed white partisanship as measured in the exit polls on white partisanship as measured in the CCES. Then, I computed the predicted exit poll partisanship based on the parameter estimates for all state-years for which CCES data was available. This rescaled version of the CCES white state partisanship scores remains perfectly correlated with the original version of the CCES measure, but its “units” are in terms of exit poll partisanship. Then I repeated the process for the NAES data. Finally, I averaged the available white state partisanship scores by year. Thus for each presidential year from 2000 to 2016 I produced a state-level measure of white partisanship based on the available survey data, all of which was scaled in the same units.

I employed the mi routine in Stata (StataCorp. 2017), using a multivariate normal model and Bayesian iterative Markov Chain Monte Carlo (MCMC) procedures. See the Appendix for further details.

See the Appendix for a discussion of two other plausibly important issues (immigration and guns) that are not included in the analysis.

Because the yearly sample sizes are smaller with the PEW data, following Highton (2011), which is also a state-level analysis (as opposed to Tesler (2012, 2016), which analyzes the individual-level data) I pool all of the PEW surveys over the period of time the four racial attitudes questions were asked from 1987 through 2000. This could put the racial attitudes measure at a disadvantage vis-à-vis the other two attitudinal measures, however as I will show the lack of much apparent effect of racial attitudes on white partisanship is not matched by a lack of apparent effect of racial attitudes on white presidential voting.

A factor analysis of all of the items (cultural, economic, racial) together confirms that there are three distinct dimensions. For one of the three extracted factors, the average loading for the cultural items is .85 compared to .06 for the economic items and .28 for the racial items. On the second factor, the average loading for economic items is .71 compared to .02 for the cultural items and .01 for the racial items. On the third factor, the average loading for racial items is .70 compared to .03 for economic items and .09 for cultural items.

Unlike the measures of state partisanship and cultural issue attitudes, constructing measures of state white economic and racial attitudes in the presidential election years between 2000 and 2016 was not possible due to lack of suitable items. While this places some limits on the analyses that may be undertaken, it does not interfere with testing the key hypotheses about long-term cultural realignment.

To investigate whether there is regional variation that might be driving the results, I reestimated the two models including interactions between the independent variables and a dummy variable coded 1 for the 11 former Confederate states and 0 for nonsouthern states. None of the interactions reach conventional levels of statistical significance and tests of the joint significance of the set of interactions (in both models) indicate that null hypothesis of no regional variation in effects cannot be rejected with confidence (p = .39 for the parameter estimates of partisanship in 2016 and p = .61 for the parameter estimates of cultural issue attitudes in 2016).

Ideally, in addition to estimating the effects of the independent variables in 2000 on the subsequent presidential elections as in Table 3, I would have also estimated a set of models where presidential vote in year (t) is regressed on previous presidential vote (t − 4) and the independent variables measured at time (t − 4). As described earlier, estimating these models was not possible. However, the Appendix provides additional estimates based on the same logic. The results reinforce those reported in Table 3.

To be sure, this is not to say the evidence showing that ordinary citizens sometimes adopt or change their preferences to those of their preferred parties is wrong (e.g., Lenz 2009, 2012; Margolis 2018a, b). However, the notion that either partisanship or issue positions are always more central may be in need of revision (Highton and Kam 2011).

The four intervals are 2000 to 2004, 2004 to 2008, 2008 to 2012, and 2012 to 2016.

A model of linear change (by state) with time fits the data better than one that also includes a higher order polynomial (time²).

In supplementary analyses I found that whether the immigration item is included or excluded has no effect on any of the substantive results.

An alternative would be to rely on the CCES for the 2012 and 2016 state white presidential vote estimates. As shown in the Appendix, the results are almost identical.