Sorting All the Way Down

By Nick Davis (@ntdPhD) and  Jon Green (@_Jon_Green)

Political scientists frequently debate whether the mass public is ideological, exhibiting coherent worldviews with reach and consistency. Reach refers to links between attitudes across multiple issue dimensions -- an ideologically coherent liberal would take liberal positions on foreign policy, economic, and social equality issues. Consistency refers to these attitudes all moving in the same “direction;” a coherent ideological worldview implies broad agreement on which policies are more liberal than the status quo and which are more conservative. Fifty years ago, Philip Converse argued that the public’s attitudes didn’t meet those lofty standards. So, too, does a recent follow-up to that work. How, then, do people navigate political space without such belief systems? How can people with illogical combinations of policy preferences act in (relatively) predictable, rational ways?

One alternative is that policy preferences are a second-order concern for mass political behavior. Instead, citizens reason via their understanding of social and political group boundaries. While policy positions may, of course, give meaning to those boundaries, identification with groups seems to function as the primary fulcrum in politics. (Case-in-point: support for Donald Trump.)

Over the last decade, scholars have paid increasing attention to the sorting of Americans’ political and social identities. From Matthew Levendusky’s (2009) The Partisan Sort to Lilliana Mason’s (2017) Uncivil Agreement, political scientists have shown that partisans reliably connect their ideological, religious, and even racial identities together. The practical implications of this behavior are significant: sorting decreases out-group favorability, producing affective polarization, and increases behavioral rigidity – generating straight-ticket voting and undercutting preferences for compromise.

To what extent, however, does this sorting spread to other facets of social and political preferences? There is evidence that partisanship and ideology are strongly predictive of sexism, racism, and nativism (see previous WTHH work here and here). However, it is less clear whether these attitudinal constructs have effectively mapped onto the left-right divide together.  

In other words: Have attitudes toward many different social and political groups sorted onto a single left-right continuum? Is comfort with social hierarchy distinguishable from the broader American ideological spectrum?

The instruments

The Data for Progress What the Hell Happened survey included multiple question batteries that allow us to explore this question.

Social dominance orientation. How people think about social hierarchies and inequalities is an essential correlate of social and political preferences (see: here and here). The WTHH survey included four items commonly used to measure SDO (see this previous post). Respondents were asked the extent to which they agreed or disagreed with the following statements: 1) In setting priorities, we must consider all groups, 2) We should not push for group equality, 3) Group equality should be our ideal, and 4) Superior groups should dominate inferior groups.

Nativism. Nativism reflects how individuals think about immigrants and the social and economic consequences of immigration policy. As we outlined in a previous post, we operationalize nativism by culling down a number of questions. Here, nativism comprises seven items that query respondents about their attitudes regarding customs, language, naturalization and deportation policies, and the potential strain immigrants place on the economy.

Hostile Sexism. As Brian Schaffner writes here, hostile sexism is strongly related to political preferences -- a connection that Donald Trump has helped crystallize through his rhetoric and actions. We looked at responses to three questions that asked individuals to report the extent to which they agreed or disagreed with the following statements: 1) Most women interpret innocent remarks or acts as being sexists, 2) Women are too easily offended, and 3) Most women fail to appreciate fully all that men do for them.

Racism. Race and attitudes toward racial inequalities are a fulcrum with respect to political preferences (see: here, here, and this new book, here). Operationalizing Americans’ racial attitudes, however, is tricky. Here, we used items from the cognitive dimension of the Fear, Institutionalized Racism, and Empathy battery (FIRE). Respondents were asked whether they agreed or disagreed with the following statements: 1) White people in the US have certain advantages because of the color of their skin, 2) Racial problems in the US are rare, isolated situations, and 3) I am angry racism exists.

Partisanship. Partisan affiliation is asked using the branching instrument common to virtually all surveys. Respondents convey whether they consider themselves a “strong” or “not strong” partisan, a person who “leans” toward one party or the other, or a true Independent.

Ideology. Finally, liberal-conservative self-placement ranges from very liberal to very conservative, with pure moderates in the center.


To generate separate scales for nativism, social dominance, racism, and hostile sexism, we used graded response models (GRMs) -- a slightly more technical approach to generating latent constructs than traditional factor analysis.  We then re-scaled scores on the four indices to range from 0 to 1. Lower values on these scales convey attitudes that are stereotypically associated with the “left” (i.e. favoring equality and inclusion) and high values the “right” (i.e. favoring natural hierarchies, less inclusive of outsiders). Figure 1 plots respondent scores on these various constructs.

It is interesting to note, first, that the modal response is often a rejection of unflattering beliefs about groups. This pattern is probably not surprising given research on self-monitoring, or the degree to which persons express socially-desirable responses rather than their true feelings (see; here and here). While the distribution of nativism is bimodal, the pattern of responses across SDO, racism, and hostile sexism are more or less normal -- revealing minimal “herding” toward the poles, which would be indicative of sorting. Partisanship ranges from strong Democrats (1) to strong Republicans (7). While the sample contains more strong partisans than perhaps found in other surveys, the distribution of ideological self-placement is approximately normal -- most people tend toward moderation.


Because our primary question regards whether or not these constructs map onto a global left-right continuum, we first explored the correlation between these six items. Table 1 produces the Pearson’s correlation coefficients for these items. Values above 0.50 are generally considered to convey a strong positive relationship. In this case, none of the items display values less than 0.50, and many items are correlated above 0.70.

Table 1. Inter-item correlation matrix


We next ran an exploratory factor analysis (EFA) on the six items. All items loaded onto a single factor with loadings in excess of 0.70 -- a somewhat arbitrary, though generally accepted, threshold for acceptable commonality among items. Rotating the results produces no changes. Simply, these items very strongly related.

We then produced predicted scores for respondents, which are plotted in Figure 2. The distribution is bimodal, which conveys that there is significant stratification among these preferences. We interpret this as a global form of “sorting,” and, indeed, breaking these scores out by house vote preference illustrates the validity of the left-right continuum (Figure 3).


Finally, we pull out factor scores by different demographic groups. There is significant variation among white respondents. The peak among “right-leaning” orientations regarding social or traditional group hierarchies is much sharper than those among the “left.” In contrast, black respondents uniformly reject such positions; these respondents cluster to the left end on the continuum. Finally, we observe that, on balance, men appear slightly more “conservative” or “right-leaning” than women, which may be expected given the relationship between the patriarchy and traditional social hierarchies.



The study of political behavior involves understanding what animates ordinary citizens. While early attention focused on whether interconnected, instrumental belief systems populated by (or lacking) policy-linkages motivated behavior, recent research has instead elevated the role of social-group reasoning. The exercise presented here illustrates that there is tremendous sorting across a number of domains associated with group-level evaluations -- something one might call “mega identity politics.”

When people think about political choices, they appear to be drawing on a range of deeply-intertwined beliefs about political and social groups that tap into traditional social hierarchies. In some ways, then, the prevailing ideological cleavage in the United States regards an age-old definition of politics: who gets what, when, and how. Citizens on the left see inequalities and social hierarchies as problems; their peers on the right are less sensitive to these concerns.

Nick Davis (@ntdPhD) is an Assistant Research Scientist at the Public Policy Research Institute at Texas A&M University. You can find some of his other professional work at

Jon Green (@_Jon_Green) is a Ph.D. candidate in political science at Ohio State University and a co-founder of Data for Progress.