Divided by Identity on the Left? Partisan Spillover and Identity Politics Alignment

2.1 Study 1. What Is the Trend in IP Alignment in the Mass Public?

To look at IP alignment over time and test whether different explanations explain the trend, I look at a variety of affective and policy measures. I first look at an affective measure in the American National Election Studies (ANES) Cumulative File. The ANES consistently asks a series of feeling thermometer questions related to a variety of groups – Blacks, Hispanics, Asians, feminists, the poor, fundamentalists, gays, Jews, Muslims, people on welfare, and illegal aliens. This captures affective attitudes to groups on race, gender, religion, and class dimensions. The feeling thermometer scores are scaled 0 to 100, with 50 explicitly defined as neutral. Anything over 50 is a positive evaluation, while anything under 50 is a negative evaluation. In terms of IP alignment, the following would count as IP alignment on the left: positive evaluations for groups typically associated as Democratic constituents (Blacks, Hispanics, Asians, feminists, Jews, Muslims, the poor, people on welfare, and illegal aliens) and negative evaluations for constituent groups perceived as Republican (fundamentalists). The opposite would count for IP alignment on the right.

The top section of Table 1 shows average feeling thermometer scores by Democratic and Republican affiliation for the periods 1998 to 2006, and 2008 to 2016. From 2008 to 2016, the average feeling thermometer for Blacks, Hispanics, Asians, Jews, and poor people are well over 50 for both Democrats and Republicans. Looking at just these thermometers, there would be no grounds for calling Democrats less unified. Republicans are favorably rating groups that are not typically associated with the Republican party, so if anything, it is Republicans who are more divided. There are some thermometers in which there is a partisan split, with the average for one party above 50 and the average for the other party below 50 in the expected directions. These include the thermometers for feminists, fundamentalists, gays, people on welfare, and Muslims. For feminists, gays, and people on welfare, the Democratic distance from the neutral score of 50 is larger than the Republican distance from 50 – that is, Democrats are warmer to these constituent groups than Republicans are colder. Only for Muslims and fundamentalists might it be said that Democrat “feeling” is less intense than Republican feeling, with Republican distance from 50 greater than Democratic distance from 50. The only thermometer that is below 50 for both Democrats and Republicans is the thermometer for illegal aliens. So only with respect to fundamentalists, Muslims, and illegal aliens could Democrats be described as potentially lagging behind Republicans.

Switching focus to change over time, the change in Democratic thermometer scores over the two time periods is greater in the liberal direction than the change in Republican thermometer scores in the conservative direction for all of the thermometers except for Muslims and the two thermometers that deal with non-IP groups (i.e. poor and environmentalists). The last two lines of the first part of Table 1 show the average change by party for all thermometers except the gay and environmentalist thermometers. I exclude the thermometer for gays since it may be an outlier in that it has experienced the most dramatic positive change in the liberal direction for both Democrats and Republicans over the past 30 years. I also exclude the environmentalist thermometer from the average since it is not related to IP. The summary shows that the change for Democrats in the liberal direction is greater than for Republicans in the conservative direction, which is consistent with partisan spillover leading to a trend of greater IP unity.

As a harder test, I convert the thermometer ratings to z-scores to look at dispersion around the average in each survey year. If the right is really more united than the left, one would expect more extreme views in the tail of the distribution associated with right-leaning opinion than the tail of the distribution associated with left-leaning opinion. Looking at z-scores by year takes out of consideration thermostatic effects – i.e., the tendency of public opinion to track against the party that holds the presidency. Since the z-score is comparing the dispersion around the average for each survey year, the fact that the average changes thermostatically each year is irrelevant. Looking at z-scores also allows me to exclude party from the analysis to avoid the problem of sorting. One problem with looking at thermometer averages by party in cross-sectional data is that it does not take into account the possibility that people with positive evaluations of IP are sorting into the Democratic party while people with negative evaluations of IP are dropping out. Underlying mass opinion may not have changed, only the sorting of individuals into party. By looking at the average of those in the bottom and top of the thermometer distribution without including party in the analysis, the analysis excludes the possibility of partisan sorting and allows me to see whether there is actual change in mass public opinion and whether it is in the right or left tail that this change is occuring.

The bottom section of Table 1 looks at the average thermometer score of those with a z-score higher than 0.5 and less than −0.5. This is essentially the bottom 30% and top 30% of the distribution. The same story as above holds. On the left the average thermometer scores are generally farther from the neutral score of 50 than the right for most thermometers (the exceptions being Muslims, illegal aliens, and fundamentalists). Looking at change over time also yields the same story as above. Comparing 1988–2006 to 2008–2016, there is greater change on the left than the right for most thermometers, the exceptions being Blacks, Muslims, and environmentalists. The last two lines of Table 1 shows the average for change in all thermometers in the left and right tails (again excluding the gay thermometer as an outlier and the environmentalist thermometer as non-IP-related). This shows a greater degree of change for the tail of the distribution associated with the left, rather than the right.

It still remains to be seen whether the left is aligned when looking at multiple thermometers simultaneously. I select the thermometers for Blacks, Hispanics, feminists, and the poor to look at simultaneously.^[1] An individual is IP aligned in a given year if on at least three of the four thermometers they have a z-score at least half a standard deviation in the liberal or conservative direction.^[2] For instance, positive z-scores greater than 0.5 on the Black, Hispanic, feminist, and poor thermometers would count as leaning liberal, and having a liberal opinion on at least three of these four thermometer scores would count as IP alignment on the left. This coding scheme gives a fairly easy interpretation to the resulting measure of those coded as 1—percent of the population that is IP aligned either on the left and right.

Pooling the data (Figure 1) allows a comparison of the trend in alignment before and after Obama became president. From 1988 to 2004, alignment on the right exceeds alignment on the left looking either at all races or just Whites. This would seem to confirm the standard narrative of a left divided by IP issues. However, from 2008 to 2016, left alignment overall is about the same as right alignment. Left alignment for Whites lags behind the right, but the gap has diminished over time.

Figure 1:

Aligned percentage: Black, feminist, Hispanic, poor thermometers (pooled, ANES).

These results complicate the narrative of a left divided by IP issues. On the left, higher levels of alignment of Blacks (Figure 1) is counteracting the lower alignment of Whites, so that overall differences between the left and right when all races are considered together are minimal by the Obama and Trump years. There is still some justification for calling the left more divided since the aggregate score conceals the difference between Whites and Blacks on the left. This is a greater problem for the left because Blacks are a much more substantial proportion of the left than the right. By comparison, Figure 1 shows that the religious (those defined as going to church at least once a month or more) on the right are slightly less aligned than everyone on the right, but this difference is not nearly as large as the difference between Blacks and Whites on the left. Still, the striking fact is that alignment of Whites on the left is increasing over time, so that the gap with Whites on the right has decreased.

These trends over time do not match up well with explanations of IP alignment that predict continued asymmetry. The proportions of fixed and fluid personalities in the population have not shifted significantly (see Appendix 2, Supplementary Figure 4). The left has become more racially and religiously diverse as time has passed. The left has also not caught up to the right in terms of message infrastructure according to Skocpol (2016) and Hertel-Fernandez (2016). The trend here also does not correspond to thermostatic effects. From 2008 to 2016, the entirety of the Obama administration, one would expect public opinion to have swung thermostatically against the liberal direction. The trend here is more consistent with the view that increasing polarization and partisan spillover has prompted higher levels of left alignment, especially among Whites on the left.

2.2 Study 2: Panel Analysis

Although the trend over time in the cross-sectional data is suggestive that partisan identity is causing left IP alignment to catch up, I further test the idea that partisanship has motivated increasing IP unity on the left by looking at the Democracy Fund Views of the Electorate Research (VOTER) panel. The VOTER survey, conducted online by YouGov, started in 2012 with a nationally representative sample of adults. VOTER has re-polled these initial respondents, with 8000 respondents recontacted in 2016 and 5000 in 2017. With the VOTER panel data, the analysis below can look at the same individuals over time, see whether they are becoming more aligned, and tease out the factors that explain why someone who is initially unaligned in IP flips to aligned in the later time period.

I start with feeling thermometer measures in VOTER, then perform a second analysis looking at policy attitude measures. For the feeling thermometers, I look at z-scores as a harder test than just looking at the raw thermometer scores, since the average on most feeling thermometers are above 50. Judging whether one is left- or right-aligned by whether they are above or below 50 on the thermometer would massively favor left alignment. Looking at the z-score, I am instead using distance from the average as a measure of being on the left or right.

Figure 2a looks at the change in z-score on a variety of group feeling thermometers from 2012 to 2017, broken down by party identification in 2012. Self-identified Democrats in 2012 improved their z-scores five years later on feeling thermometers for Blacks, Hispanics/Latinos, Asians, gays/lesbians, Muslims, and Jews, while also feeling colder toward Christians. In comparison, both self-identified Republicans and those with no party identification in 2012 saw decreases five years later in their z-scores on feeling thermometers for groups thought of as Democratic constituent groups.

Figure 2:

(a) Mean change in z-score 2012 to 2017, feeling thermometers (VOTER data).

(b) Aligned percentage: Black, Hispanic, Muslim, gay thermometers (VOTER data).

To analyze IP alignment on multiple thermometers simultaneously, I look at the subset of thermometers for Blacks, Latinos, gays/lesbians, and Muslims. Unfortunately, the thermometer for feminists was not asked in 2012, but the four thermometers used here do capture divisive IP issues related to race and religion in the contemporary era.^[3] As before, someone is on the left if their z-score on an individual thermometer is greater than 0.5, and they are on the right if their z-score is less than −0.5. A person is coded as aligned on the left if their z-score is greater than 0.5 on at least three of these thermometers in a given year. A person is coded as aligned on the right if their z-score is less than −0.5 on at least three of the thermometers in a given year. Based on this measure, IP alignment for self-identified Democrats and Republicans in 2012 is about the same in 2012 and 2017 (see Figure 2b). In terms of change over time from 2012 to 2017, the only group for which there is significant improvement in alignment is White Democrats.

I employ a lagged regression to see which factors in 2012 are correlated with a change in alignment in 2017. The binary dependent variable is whether or not one is aligned on the left in 2017 in one regression (Table 2, column 1), and whether or not one is aligned on the right in 2017 in a separate regression (Table 2, column 2). Similar to Egan (2019), I include as an independent variable the value of alignment on the left and right in 2012. The inclusion of this lagged variable means that the regression can explain the change in alignment from 2012 to 2017 – that is, the newly aligned and unaligned in 2017.

I then include other independent variables from 2012 to test for the possibilities covered in the literature review section of this paper. I include variables for party and ideology in 2012 to test my main hypothesis that party spillover leads to improvements in alignment. I include racial dummies (African American, Latino, Asian, and other), which is relevant to understanding how much racial homogeneity within parties matters. Authoritarian/fixed personality is identified by the standard four questions on child rearing as defined by Hetherington and Weiler (2016; see Appendix for wording). Unfortunately, these questions were not asked in 2012, only in 2016. However, since personality is supposed to be a stable trait, the value in 2016 should be close to its value in 2012 for the same individuals in a panel.

Potential exposure to conservative/liberal infrastructure is measured by a variable indicating how closely the respondent follows and is aware of politics. This variable should capture some exposure to conservative or liberal messaging. The measure is overly broad, as other factors account for interest in politics, but this fact makes the overall regression a more conservative test of whether party and ideology matter. As further measures of the potential influence of messaging infrastructure, I also include a measure of how interest in politics changed from 2012 to 2017, as well as a measure of Internet use in 2012, since the Internet may expose the respondent to messages from political entrepreneurs.

Finally, I include control variables for gender, age, family income, and Church attendance. In addition, I include a variable for change of income between 2012 and 2017, since more dire economic circumstances may enhance competition and threat from out-group members.

The results of the logit regression are reported in Table 2. Column 1 looks at IP alignment on the left, while column 2 looks at IP alignment on the right. Being a Democrat in 2012 significantly leads to a greater probability of being aligned on the left. Fluid personality is also statistically significant, though it does not have as large an effect as party. Race and measures of exposure to messaging are not significant. For someone who was not left aligned in 2012, the average marginal effect of being a Democrat in 2012 contributes 11 percentage points to the probability of one becoming left aligned in 2017. For instance, the predicted probability of a White fluid liberal male who was not IP aligned in 2012 becoming left aligned in 2017 goes from roughly 16 percent to 30 percent if one was a self-identified Democrat in 2012.

For IP alignment on the right, Republican identification and conservative ideology are significant under one-tailed tests, and fixed personality is significant under a two-tailed test for explaining whether one is right aligned in 2017. Exposure to messaging infrastructure and race are not significant (Table 2, column 2). For someone who was not right aligned in 2012, the average marginal effect of Republican party identification is 4 percentage points to the probability of being right aligned in 2017. In addition, the average marginal effect of conservative ideology is 6 percentage points to the probability of being right aligned in 2017. In summary, party matters for increased IP alignment on both the right and left, but to a larger degree for the left. This is consistent with the idea that the right has been previously better aligned so that the marginal effect of party on alignment is not as impactful as it is for Democrats.

To check the robustness of this finding beyond affective measures, the VOTER panel also allows for examination of IP alignment on attitude measures related to race, immigration, religion, and class redistribution. In 2012 and 2016, respondents were surveyed on the standard four questions on racial resentment related to African Americans; three questions related to immigration; two questions related to taxing the wealthy and governmental regulation of business; and a question on abortion. This gives four distinct attitude positions related to African Americans, immigration, class, and religion.

The four standard racial resentment questions are coded as 0 for the strongest liberal position, 0.5 for a neutral position, and 1 for the strongest conservative position, and are combined to form one score ranging from 0 to 1 to measure respondent’s attitudes on African American inequality (see appendix for wording of the survey questions in this section).

On immigration VOTER asks whether illegal immigrants are a drain on American society; whether there should be a path to citizenship for the undocumented; and whether immigration to the United States should be more difficult. The most liberal position was coded as 0, a neutral position as 0.5, and the most conservative position as 1. Moderately liberal or moderately conservative positions were coded as 0.25 and 0.75 when appropriate. These measures were combined to form one measure on attitudes toward immigration.

For attitudes on class redistribution, VOTER asks whether respondents favor raising taxes for those making over $200,000 and how much government regulation is desirable. Liberal responses were coded as 0, neutral responses as 0.5, and conservative responses as 1. These scores were combined to form a measure of class redistribution.

Finally, VOTER asks whether respondents think abortion should be illegal in all circumstances, some circumstances, or legal in all circumstances. The liberal position (legal in all circumstances) was coded as 0, neutral responses (some circumstances) as 0.5, and the conservative response (illegal in all circumstances) as 1.

From 2012 to 2016, the change in average scores on these four measures are shown in Figure 3a. Attitudes for all party and non-party identifiers liberalized in relation to African American racial resentment and immigrants, but self-identified Democrats in 2012 became relatively more liberal in their attitudes than non-Democrats. Abortion also became more polarized, with Democrats becoming more liberal and Republicans and nonparty identifiers moving in the opposite direction. Strikingly, it is only with the one non-IP issue – taxation and government regulation – where both Democrats and Republicans became less polarized. Self-identified Democrats became slightly more conservative, while all others became slightly more liberal, contrary to the notion that class redistribution issues would be more likely to unite the left.

Figure 3:

(a) Attitude change in liberal direction, 2012 to 2016: African Americans, immigration, abortion tax/regulation (VOTER data).

(b) Aligned percentage: Black inequality, immigration, tax/regulation, abortion (VOTER data).

To look at the measures simultaneously, I create a dummy variable coding someone as left aligned in a given year if their scores on racial resentment, immigration, class redistribution, and abortion are below 0.5 for at least three of the measures. I create another dummy variable coding someone as right aligned in a given year if their answers are above 0.5 on at least three of the measures. As seen in Figure 3b, those who self-identified as Republicans were better aligned in 2012 compared to Democrats, consistent with party asymmetry theories. By 2016 however this asymmetry in alignment disappears. Both White and African American Democrats improved their alignment score in 2016 compared to 2012, allowing Democrats to catch up to Republicans in alignment.

As before I run a logistic regression to see which factors predict left and right IP alignment in 2016. For this analysis, I use the same independent variables in the regression for the thermometer scores above. The results are reported in Table 2, columns 3 and 4. Both Democratic and liberal identification in 2012 are significantly associated with a higher probability of being left aligned in 2016. African American and fluid personality are also significantly associated with left alignment in the expected directions, while none of the variables potentially associated with message infrastructure are significant. If one was not left aligned in 2012, the average effects margin for being a Democrat in 2012 is 17 percentage points to the predicted probability of being newly left aligned in 2016. The average effects margin for being a liberal in 2012 is 16 percentage points to the probability of being newly left aligned in 2016.

On the right, party identification and conservative ideology are significant in the expected direction. Consistent with the view that better right infrastructure may contribute to IP alignment, political interest is positively associated with right alignment. Being Black was negatively associated, as expected, while fixed personality was not associated with right alignment. The average effects margin of being a Republican in 2012 is 9 percentage points to the probability of being newly right aligned in 2016; the average effects margin for being conservative in 2012 is 5 percentage points.

In summary, the second panel analysis shows that prior party identification leads to better IP alignment on both sides of the political spectrum, but with a greater effect for Democrats. Fluid personality also contributed to better alignment on the left, but Democratic affiliation exerts an independent effect, and this effect is stronger than fluid personality. Both panel analyses, with different specifications of IP alignment, show evidence for the partisan spillover hypothesis. Controlling for other factors, Democratic partisanship is associated with an increase IP alignment for the same individuals at a later time period.