In a recent Journal of Politics article, Alexander Kuo, Neil Malhotra, and Cecilia Hyunjung Mo make a very interesting and novel contribution to our understanding of partisan identification. In what’s particularly relevant to non-white minority groups, the authors argue that experiences of social exclusion on the basis of one’s racial/ethnic group membership can influence political identity. People can interpret individual experiences of exclusion as group exclusion. When one party is considered more exclusionary, these experiences can define which party best represents group interests, motivating greater attachment to/detachment from certain parties. Kuo et al. cite past research to establish the prevailing view of the Democratic Party as the party most beneficial to ethnic minority groups and the less exclusionary one. As a result, feelings of social exclusion should translate into greater identification with and support for the Democratic Party.
Kuo et al. offer observational and experimental evidence in support of this theory as it pertains to the political orientation of Asian-Americans. In their observational approach, modeling this group’s partisanship using 2008 National Asian American Survey (NAAS) data shows that all things else being equal, social exclusion–proxied by an indicator of whether a survey respondent reports having been a victim of racial discrimination–increases Democratic identification among Asians by three to four percentage points. The finding holds across several models. More convincing evidence comes in a laboratory experiment, where (Asian) participants are randomly exposed to feelings of social exclusion. This comes in the form of a racial microaggression against Asians: questioning U.S. citizenship status. Such a treatment that primes social exclusion is enough to make Asian-Americans much more favorable of the Democratic Party–and more negative towards the Republican Party–across several dimensions.
Clearly, some causal mechanism exists here for social exclusion. When proxied at a broader level, using a large representative sample of Asians to complement findings from a 114-subject experiment, traces of it affect partisanship as well. Kuo et al. focused only on Asian-Americans in their study. As the authors note in their suggestions for future research, the finding of social exclusion as an important factor begs the following question: does it influence partisanship of other minority groups? That motivated me to extend the modeling approach in Kuo et al.’s observational analysis to other groups–Latinos and blacks–using other survey data, while checking for the same relationship among Asians as well. With sizable minority group samples, the Collaborative Multi-racial Post-Election Survey in 2008 fits this extension of Kuo et al.’s approach and findings well (the 2016 data has not yet been made public).
With 1577 Latinos, 919 Asians, and 945 African-Americans in a nationally representative survey offered in six languages, the 2008 CMPS represents one of the best available sources to address the extent of social exclusion’s effect on partisanship at the subgroup level. The survey was conducted in the months following the 2008 election and spanned registered voters only. In trying to replicate and extend Kuo et al.’s findings and approach, I follow these author’s decisions for designing their models as close as possible (link paper). Just as they do, I’ll present four models in the regression tables for Asian, Latino, and black partisanship below.
- Regressing partisanship (excluding missing data) on each control
- Regressing partisanship (excluding missing data) on each control and ideology
- Regressing partisanship (with missing data coded as the midpoint) on each control
- Regressing partisanship (with missing data coded as the midpoint) on each control and ideology
To account for missing data on partisanship–responses of “Don’t know” or refusals when asked about party–I follow the two ways that Kuo et al. specify. I first recoded a 1-6 partisanship scale (strong Democrat, not strong Democrat, lean Democrat, lean Republican, not strong Republican, strong Republican) to create a 0-1 scale where 1 is most Democratic and 0 is most Republican. Missing data (made up of 507 respondents) is excluded in this first measure (party1). In the second measure of partisanship (party2), I recode observations with missing partisanship data to fall at the midpoint, a value of 0.5. In each table below, models 1 and 2 use party1, while Models 3 and 4 use party2. This approach produces more robust partisanship modeling, and in particular is helpful in accounting for missing data problems in different ways.
The crucial predictor of interest here is social exclusion. I use the CMPS question about whether an individual has “personally experienced discrimination, or been treated unfairly because of your race or ethnicity” as a proxy for social exclusion, a measure similar to what Kuo et al. use. For reference, 42 percent of Latinos, 50 percent of Asians, 63 percent of blacks, and 26 percent of whites report such experiences of racial discrimination. I then add in six main controls that closely match what the other authors used: an indicator for female gender, income (seven income groups coded on a 0-1 scale), education (six educational groups coded on a 0-1 scale), age group (16 age groups coded on 0-1 scale), an indicator for U.S. born, and religiosity (six church attendance rate reports coded on a 0-1 scale). A control for self-described ideology (seven ideological groups coded on a 0-1 scale, where 1 is most liberal) is included in Models 2 and 4
As Kuo et al. mention, coding dependent and independent variable values to fall between 0 and 1 allows for ease of interpretation: going from its variable’s lowest to highest value, each coefficient represents a 100*β increase in the dependent variable (Democratic partisanship). I also follow the authors’ approach for handling missing data (described in footnote 16 in the paper). This entails including two variables for each measure in each regression I run: 1) a variable with the actual values but with missing data recoded as 0 (e.g. actual education levels from 0 to 1, missing data as 0), and 2) a binary indicator for missing data (e.g. 1 if education level is missing 0 otherwise). In this way, I avoid listwise deletion of observations because of missing data for certain values.
Modeling Asian, Latino, and Black Party ID
I first wanted to confirm what Kuo et al. initially found: do significant effects of exclusion on Asian partisanship hold? If such effects are present across several high quality surveys (2008 NAAS data and now 2008 CMPS data), then it lends all the more credence to the theory of social exclusion influencing partisanship. As shown in the below Table 1 that subsets observations to just Asian American respondents, the proxy for social exclusion does indeed significantly affect partisanship.
Across the four models–regardless of how I handle missing data in the dependent variable or whether I include an ideology control–past experience of social exclusion increases Asian identification with the Democratic Party by five to seven percentage points (significant at the p<.05 level)–to a greater degree than what Kuo et al. found. This positive and significant relationship between racial victimization and Democratic partisanship supports the overarching theory of social exclusion’s impact on political identity for Asians, and confirms what Kuo et al. found in their similar analysis using other survey data (their controls differed a bit from mine).
Before moving on to identical tests but for other minority groups, some other demographic determinants of Asian party identification are worth touching on. Unsurprisingly, ideology significantly shapes partisanship. Higher levels of education also correspond with greater identification with the Democratic Party. The relationship between religiosity and lesser Democratic partisanship is not as robust as Kuo et al. find–including ideology in a specification cuts the effect in half and makes this variable insignificant.
Outside of the significant result for exclusion, the most notable takeaway from these models is the strong effect of income group level: Asian-Americans are between 12 and 17 percentage points less Democratic going from the lowest to highest income level. This stands in stark contrast to what Kuo et al. found using the NAAS survey data, which shows hardly any correlation between income and partisanship among Asians. The authors interpret this result as a challenge to income-based explanations of political orientation, such as the one Gelman et al. (2009) establish. This relationship will be an important one to test when more recent high quality (and large observation) survey data on Asians is released.
It now seems clear that social exclusion affects partisan inclinations among Asians–but is the same force as potent among other minority groups in America, who of course can also suffer experiences of discrimination in social life? In checking for how extensive this mechanism is, I start by estimating the the effect of social exclusion on partisanship for Latinos in Table 2 below. This follows the same steps and model specifications as before: models 1 and 2 remove individuals who don’t report partisanship while models 3 and 4 code them at the midpoint of the partisanship outcome, and models 1 and 3 exclude ideology as a control while models 2 and 4 include them.
Just as with Asian Americans, social exclusion appears to affect the partisanship of Latinos in the U.S. as well. Importantly, it follows the theorized direction–growing closer to the Democratic Party, considered as the less exclusionary party relative to the GOP. Effect sizes for reported experiences of racial victimization increasing Latino Democratic attachment range from four to six percentage points, all of which attain significance at the p<.01 level. This constitutes the most notable finding in all of this analysis. The same social exclusion effect on partisanship that Kuo et al. initially found might indeed be more widespread. If report of racial discrimination is a true exogenous force (more on this later), then it could play a key role in partisanship development of minority groups who encounter marginalization.
Some other results from Table 2 are worth mentioning. Female Latinos are roughly six points more Democratic than male ones are, and older aged Latinos are also more Democratic. As with Asians, income group level once again proves a significant negative correlate of Democratic Party attachment with all other factors held constant–going from the lowest to highest income decreases Democratic partisanship by 12 to 18 points, depending on the model. Educational and religiosity level are also significantly negatively correlated with degree of Democratic identification.
Finally, I turn to examining the same relationships from above–and the power of racial victimization on partisanship in particular–for African Americans in Table 3. I drop the indicator for U.S.-born individuals in these specifications as this variable is less relevant for blacks (when I include it, results don’t change). The below regression takes the same approach as before in predicting Democratic partisanship.
Unlike for Asians and Latinos, the proxy for social exclusion has very little effect on the partisanship of African Americans and does not attain significance. Given the much more uniform and rigid structure to African American partisanship, stemming from much deeper roots in American politics that could be passed on from each generation unlike with Latinos and Asians who are more likely to be relative newcomers to the U.S., it’s not too surprising that exclusion has little additional effect on this group’s partisanship. A very stable variable does not get easily influenced by a (presumably) exogenous factor.
Additionally, unlike among Asians and Latinos, income does not prove a consistently strong negative correlate of Democratic partisanship among blacks. Specifically, the income variable loses its significance and half of its effect size going from Models 1/2 to 3/4. The other aspect of socioeconomic status in education, however, does prove a strong and significant negative predictor of Democratic attachment: blacks in the highest educational bracket are between 12 to 16 points more attached to the Republican Party than are blacks in the lowest educational group. Age also shapes black partisanship, as going from the youngest to oldest age group increases Democratic identification by 12 to 16 percentage points (this pattern evokes a similar age-based one found in 2016 voting behavior).
In comparing models across Tables 1-3, an interesting result arises: the effect size of ideology on partisanship among blacks is two to three times smaller than that for Asian and Latino partisanship. This might owe to the fact that roughly one-third of African-Americans tend to self-identify as ideologically conservative, while their partisan inclinations are much more uniformly leftward.
Accounting for Region
Region type, an increasingly important variable for political orientation and one that can serve as a rough proxy for cosmopolitanism (which Kuo et al. ignore but say could be important), might also be related to perceptions of social exclusion and party identification. The above models for Asian and Latino partisanship were thus re-run to include indicators for urbanicity (living inside a central city relative to living outside one). This indicator did not diminish the exclusion variable’s effect on partisanship, but did prove significantly positive for Latinos: across the four models, Latinos living in central cities were 6 to 8 percentage points more Democratic.
The Endogeneity Issue
A notable problem with this analysis is the ambiguous causal direction–do instances of social exclusion (i.e. discrimination) lead to greater Democratic identification, or are Democrats more likely to interpret certain experiences as acts of discrimination? Kuo et al. also confront this issue in their first observational study, which motivates their experiment where they can apply a clear exogenous activation of social exclusion–and clarify exclusion as a driver of Asian Democratic attachment. Without that approach, it’s not so easy to clarify the causal direction.
Despite these concerns, Kuo et al.–in their study of Asian political identity–still consider recalled experiences of discrimination as “temporally preced[ing] the dependent variable,” partisanship, and thus constituting an exogenous variable. Similarly, they argue this variable captures recollection of specific discriminatory incidents but not attitudes on feelings about discrimination. Nevertheless, the endogeneity issue here remains somewhat unsolvable. The factor of social exclusion should not be discounted due to this important flaw however, as Kuo et al.’s experimental results clearly show something with racial discrimination as a driver of partisan identity is at play here. But as it pertains to using observational data, I thought of one way to get around this issue, an explanation for which I describe below.
The idea of reverse causation stems from Democratic partisanship potentially translating into greater sensitivity to discrimination, while Republican partisanship possibly doing the opposite. Partisanship attachment levels likely relate to knowledge level of party position differences, such as those on discrimination. Knowledge of party positions could influence how individuals recall and perceive past social experiences–those with greater knowledge see their partisanship shape their recall more. Given the power of partisanship for coloring all kinds of perceptions, this would not be unordinary. Individuals with greater awareness would presumably be more likely to know each party’s position on discrimination. If people determine whether they have experienced discrimination according to their partisanship level, they would also likely know the varying partisan sensitivity levels to discrimination. This amounts to knowledge of party differences, which a political awareness proxy–such as whether someone follows political news–might capture.
The logic here is a bit convoluted, but this could lead to an informative test that speaks to the (potential) causal arrow between exclusion and partisanship. Specifically, I test the moderating effect of awareness by seeing whether social exclusion retains significance in a model with the political news indicator included. I do this for this for the Asian and Latino models of partisanship (where social exclusion proved significant). If significance goes away and/or the effect of social exclusion decreases, then based on the logic above, the relationship might be more about partisanship coloring recall of possible discriminatory experiences. However, for both the Asian and Latino models of partisanship, there is no moderating effect of the awareness proxy on social exclusion. If awareness determines the extent to which partisanship could be used in processing past social experiences, this result–a significant exclusion effect at high and low levels of political awareness–lends at least some support for social exclusion influencing partisanship rather than the other way around.
The significant effects of the social exclusion proxy on Asians and Latinos are particularly important. As noted before, these minority subgroups are more likely to be immigrant newcomers to the U.S. Indeed, the 2008 CMPS shows that even among a registered voter universe, 43 percent of Latinos and 71 percent of Asians were born outside the U.S. This could mean less integration into U.S political life and less potential for potent family socialization influence. In turn, that leave partisanship–the most powerful shaper of political behavior–less stabilized for many Asians and Latinos than it is for other Americans. The contours of these minorities’ partisanship is thus still developing. If racial discrimination and associated experiences of social exclusion can significantly shape political identity, these phenomena–and related conceptions of political parties as social inclusive or exclusive–could prove especially important. By extension, defining the partisanship of rapidly growing minority populations could also mean defining the future of American politics.
Two limitations should also be kept in mind (outside of those already mentioned). The CMPS data used here encompasses registered voters only–perhaps these dynamics (for social exclusion or other demographic determinants) play out differently when increasing the population scope to unregistered and thus less politically active minority individuals. Secondly, the data used in the above analyses comes from 2008. While partisanship and its determinants often don’t change much over time, it’s not out of the question that the most recent election season may have heightened the importance of certain factors. Evaluating these same dynamics but with 2016 CMPS data is thus all the more important.