By Nick Davis (@ntdPhD)

1. Introduction

Political candidates spend an enormous amount of time, energy, and money claiming that their vision for the country most closely aligns with the preferences of most voters. This often involves some sort of lukewarm appeal to “American values” balanced against the standard body of policies for which the party they represent is known. In turn, the combination of these ideas is often carefully calibrated to speak to various intra-party constituencies in an attempt to motivate a broad spectrum of supporters. At their core, however, most campaigns strategically develop a platform rooted in where they perceive voters are located within the broader political spectrum.

With the 2020 election looming, the pundit class will inevitably argue that the clearest path to victory requires threading an almost impossible needle that gives enormous deference to the political “center.” According to this wisdom, if primaries are ultimately about satisfying the demands of the party’s base, then the general election requires tacking back to the center to earn the trust of “moderates.” This folk logic, more or less encapsulated in the median voter theorem and this tweet, is such a core feature of election commentary that it is impossible to avoid the sort of punditry that presumes that elections hinge on the whims of these voters – 2016 or academic research be damned.

This tension over where candidates ought to run involves squaring how elites and ordinary citizens think about politics. Questions about the fit between elite and mass preferences are tricky to resolve, in part because voter sophistication is questionable (e.g. here, here), but also because legislators don’t do a very good job benchmarking the preferences of their constituents (here). Further, such debates intersect with questions over whether voters possess a set of interconnected policy positions -- that they think ideologically. While the proportion of voters who possess fully-coherent belief systems is modest, it is probably fair to acknowledge that the mass public isn’t completely devoid of sorted preferences. That is, for most voters, their policy preferences correspond with their chosen political party’s. Thus, if anything, the “disconnect” between mass and elite ideology probably doesn’t involve voters desiring more moderation inasmuch as it involves a preference for specific combinations of non-centrist preferences.

If we operate under the assumption that most elected officials and candidates make some attempt to faithfully represent the average preferences of their constituents, then I think two questions are intriguing in the run up to the 2020 elections. First, are voters well-sorted across the pressing issues that structure partisan conflict? If candidates claim to be running on the sort of values that their copartisans possess, then what is the nature of those preferences? Second, to what extent does the political “middle” differ from rank-and-file partisans? If deference is to be paid to the middle, then what, precisely, does the middle look like?

To calibrate how we think about the preferences of the electorate in 2020, I turned to the What the Hell Happened Survey, which provides useful information about the public’s priorities. For this exercise, I explore how attitudes across three major domains that were predictive of voting in 2018, economic preferences, gender, and beliefs about the persistence and nature of racism in America, and a fourth category of attitudes about criminal justice policies intersect. While a growing body of work in political science conveys that Republicans have sorted asymmetrically compared to Democrats (see here and here), I will show that Democratic voters are well-sorted with respect to these domains. In other words, most voters do not divorce matters of racial and economic justice. However, in the case of self-described “moderate” and “liberal” Democrats, we find compelling evidence that, while both groups of individuals lean to the left, operational ideology among the former is less progressive than the latter.

2. Instruments

The survey instruments used in this project ask individuals whether they agree or disagree with a given statement. Possible responses range from “strongly agree” (1) to “strongly disagree” (5). “Don’t know” (6) responses were recoded as missing. I then calculated a series of graded response models (GRMs) using the instruments associated with the various domains. These effectively allow us to predict the extent to which respondents possess “liberal” or “conservative” attitudes on the respective indices. Each scale possesses a mean of 0 and a standard deviation of 1. Lower (higher) values on each scale convey more liberal (conservative) positions. Mean values (0) shouldn't necessarily be interpreted as true moderation, per se, but represent the cut-point from which we'll work later when juxtaposing scores on one scale against another (for more information on this, see the small Appendix at the end of the memo).

2.1. Racial attitudes

I operationalize racial attitudes using four instruments: two measures traditionally included in “racial resentment” scales (Generations, Favors) and two items from the “FIRE” battery (Institution, System). Although the two sets of instruments were not originally intended to be used as part of a common scale, they nevertheless load well onto a common factor within an exploratory factor analysis framework. (It is true the inter-item correlations are lower for black, relative to white respondents; however, the four instruments still load acceptably well onto common factors, implying that dimensionality is low.) Figure 1A illustrates the distribution of scores on this index broken out by partisanship. Predictably, there are prominent differences among partisans with respect to recognizing that racial inequalities exist -- Democrats are more likely than Republicans to acknowledge the existence of such discrimination; pure Independents are split.

  • [GENERATIONS] Generations of slavery and discrimination have created conditions that make it difficult for African Americans to work their way out of the lower class.

  • [FAVORS] Irish, Italian, Jewish, and many other minorities overcame prejudice and worked their way up. Blacks should do the same without any special favors.

  • [INSTITUTION] White people in the U.S. have certain advantages because of the color of their skin.

  • [SYSTEM] Racial problems in the U.S. are rare, isolated situations.

2.2 Economic Attitudes

The index of economic attitudes is comprised of eight items that ask individuals about various issues of public policy that loosely touch the economy. While the items range across a broad spectrum of issues, their interitem correlations are robust in magnitude. According to Figure 1B below, again, we observe prominent partisan differences.

  • [PUBLICINT] Would you support or oppose the creation of a publicly-owned Internet company to fill coverage gaps in rural, urban, or remote areas that currently lack robust Internet access?

  • [GREENJOB] Would you support or oppose giving every unemployed American who wants one a job building energy-efficient infrastructure?

  • [POLFEE] Would you support or oppose a policy to charge pollution fees on companies that emit high levels of greenhouse gasses?

  • [PUBLICGEN] Would you support or oppose having the government produce generic versions of life-saving drugs, even if it required revoking patents held by pharmaceutical companies?

  • [BOND] Would you support or oppose giving every American a $5,000 savings account at birth that they can access when they turn 18, paid for by raising taxes on estates worth $10,000,000?

  • [FREECOLL] Would you support or oppose a law that would raise taxes on income in excess of $200,000 per year by 5 percent in order to cover college tuition for all students, while capping the rise in tuition at the rate of inflation for all colleges accepting such funding?

  • [WEALTH] Would you support or oppose a tax on wealth greater than $100 million?

  • [M4A] Would you support or oppose expanding Medicare such that it becomes the main health insurance provider for all Americans?

2.3. Gender Attitudes

Three items are used to conventionally construct an index of attitudes toward traditional gender roles. These items reflect a general “hostility” toward women in the sense that they imply women fail to appreciate men and misinterpret their (men’s) motives. Figure 1C illustrates that these attitudes are approximately normally distributed for pure Independents and diverge prominently among Democrats and Republicans who are more and less likely to reject these sentiments.

  • [REMARKS] Most women interpret innocent remarks or acts as being sexist.

  • [OFFEND] Women are too easily offended.

  • [APPRECIATE] Most women fail to appreciate fully all that men do for them.

2.4. Criminal Justice Attitudes

Finally, our index of criminal justice attitudes incorporates a number of different issues. While partisan differences exist on this scale, they are admittedly more modest than the distribution of attitudes across the other scales, presumably because marijuana attitudes cross-cut traditional partisan cleavages. Still, with the addition of bail, solitary confinement, and ICE instruments, Figure 1D illustrates that Democrats and Republicans are moderately sorted with respect to these attitudes.

  • [ICE] Would you support or oppose defunding Immigration and Customs Enforcement (ICE)?

  • [BAIL_item] Would you support or oppose ending cash bail for individuals awaiting trial, and instead only detaining individuals pre-trial if they were a violence or flight risk?

  • [MARREP] In states where marijuana is legal, would you support or oppose requiring that any tax revenues collected from its sale be re-invested in communities disproportionately affected by the War on Drugs?

  • [MARAM] Some people are currently incarcerated for marijuana-related offenses in states where marijuana has since been legalized. Do you support or oppose releasing these people from prison?

  • [MARLEG] Would you support or oppose fully legalizing marijuana at the national level?

  • [SOLITARY] Would you support or oppose ending the practice of solitary confinement in the United States?


3. Sorting on Criminal Justice and Economic Policy

We begin by looking at the extent to which voters are sorted on criminal justice and economic policy. Recalling that the quadrants are drawn using the mean value for the two indices, the lower-left and upper-right quadrants more or less reflect the natural “homes” for Democratic and Republican voters. For all scatterplots, we added a tiny amount of random noise known as jittering to help clarify respondents with identical configurations of attitudes. The estimates were weighted by the survey’s sampling weights; point estimate size varies as a function of those weights, such that larger bubbles correspond to greater prevalence in the population.

Looking at voters in the 2018 midterm election, we see tremendous clustering among people who voted for a Democratic House candidate (r=0.57; conventionally, values above 0.30 are considered evidence of a modest relationship, while values above 0.50 convey a strong relationship). People who voted for Republican House candidates are also well sorted, but the relationship is a bit noisier (r=0.43). The difference between these relationships is, in turn significant. Although there is still important variation among Democratic voters, Figure 2 provides evidence that criminal justice and economic preferences converge among those drawn to Democratic candidates.


Data for Progress has tried to emphasize over recent months that younger voters are, generally, more open to progressive policy solutions (see here, here). To the extent that voters’ preferences may differ as they age, we also broke this scatterplot out by age cohort. Millennials voting for Democratic candidates in 2018 were the most well-sorted age demographic among the group (75 percent are below the means on both scales). Gen Xers, Boomers, and members of the Silent Generation are also surprisingly well-sorted, hovering around 70 percent in the lower-left quadrant. In contrast, for persons voting for Republican candidates, the strength of these relationships are weaker and do not vary much across cohorts (save for Millenials, where the relationship between the two scales is particularly weak).


As a final way of exploring the intersection of these preferences, we also looked at the combination of these attitudes by vote choice across three elections: Presidential vote choice in the 2012 and 2016 elections and 2018 House vote choice. While a prior post digs into these categories in greater depth here, I broke the eight permutations of choice into two panels by virtue of whether the respondent cast a Democratic or Republican House ballot in 2018.

The first panel illustrates that consistent Democratic voters far outnumber any of the other combinations of voters who voted Democratic in 2018. Given the relative scarcity of the other permutations of voters, it would be unwise to extrapolate too much from this analysis. Still, it seems that, among the other combinations of choice from 2012 to 2018, about half of 2018 voters who reported a Democratic House vote are positioned below the midpoint of the criminal criminal justice index. Further, the majority of these voters also cluster toward the more liberal end of the economic policy scale.


Among Republicans, a somewhat more confusing picture emerges. Again, consistent Republican voters overwhelm the other combinations of voters. However, for persons who reported previously casting a vote for Obama either in 2012 or 2016 and cast a Republican vote in 2018, the joint combination of preferences across criminal justice and policy issues actually skews “liberal.” In other words, their preferences on these issues notwithstanding, these persons were still driven to vote Republican in 2018 for some other reason(s).

4. Race, Gender and the Economy

We move next to exploring the intersection of attitudes about race and gender and policy preferences. Breaking the scatterplots out by House vote choice in 2018, the Figure 5 illustrates two things. First, in the case of economic policy, there is significant heterogeneity within both groups of voters when we incorporate attitudes about racial inequalities (Democratic voters, r=0.41; Republican voters, r = 0.28) and hostile sexism (Democratic voters, r=0.29; Republican voters, r = 0.08). Beginning with the top-left graph, voters who reported casting a ballot for Republican candidates in 2018 were almost uniformly likely to score above the mean on the racial attitudes index. While there is moderate spread across the distribution of economic policy preferences, few Republican voters acknowledge that racial inequalities are institutionalized. Democratic voters are more well-sorted; 58 percent of these respondents fall below the mean on both scales. In the top-right graph, Republican attitudes about gender exhibit modest dispersion, which indicates that there is no real evidence of sorting here (which, normatively, is probably good in that a nontrivial proportion of Republican voters reject hostile sexism). In contrast, approximately 66 percent of Democrats exhibit more “liberal” preferences across both indices.


In addition to economy policy, we also explore how race and gender attitudes intersect with criminal justice policy preferences. While there is significant heterogeneity within both groups of voters when we incorporate attitudes about racial inequalities (Democratic voters, r=0.50; Republican voters, r = 0.20) and hostile sexism (Democratic voters, r=0.46; Republican voters, r = 0.25), again, Democrats exhibit modest sorting across these indices.  59 percent of Democratic voters held “liberal” positions on both race and criminal justices issues; 60 percent held “liberal” attitudes toward gender and criminal justice matters.

5. Sorting Among “Moderate” and “Liberal” Democrats

Our analysis to this point has primarily focused on differences among respondents who reported casting votes for Republican or Democratic candidates. We want to end this analysis, however, by drilling down into the policy preferences of different types of Democrats. Namely, how do self-described moderate and liberal Democrats differ with respect to how their economic policy preferences relate to beliefs about race and gender?

This is an important question, particularly with respect to thinking about how the 2020 Democratic presidential primary will shake out. Consider this statement quoted in a recent 538 piece on the importance of moderates to Democratic primary hopefuls in 2020.

“I want a Democrat who cares about and will act on climate change, immigration and criminal justice reform because they are real justice issues, not because they are convenient new weapons in the culture wars,” said Michael Wear, a self-described moderate Democrat who directed outreach to religious groups for Obama’s 2012 campaign. “And I want a Democrat who will at the very least respect religious freedom and our national disagreement around abortion.

On the one hand, acting on climate change, immigration, and criminal justice conveys that that this prospective voter ought to be “operationally” liberal. Yet, respect for religious freedom and pro-life advocates cross-cut the above sentiments. This circles us back to the tensions that Ahler and Broockman lay out in their work I referenced in the introduction to this piece. Mr. Wear is not a conventional moderate in the sense that he prefers centrist positions, but because, presumably, he is balancing cross-cutting preferences. Thus, the trouble with equating moderation with neutrality.

Figure 6 breaks our scatterplots apart by whether a respondent was a self-identified moderate or liberal Democrat. If we consider the lower-left quadrant to be the natural “home” of left-leaning persons, Figure 6A illustrates that there are some modest differences between self-identified liberal and moderate Democrats. Whereas 77 percent of liberal Democrats score below the mean on sexism and economic preferences, only 55 percent of moderate Democrats are distributed in that quadrant. This pattern persists in Figure 6B. 70 percent of liberal Democrats exhibit a combination of “liberal” operational ideology on perceptions of institutional discrimination and economic preferences, while only 45 percent of moderate Democrats convey such preferences. Finally, in Figure 6C, liberal Democrats exhibit extremely well-sorted criminal justice and economic preferences (85 percent), while roughly 60 percent of moderate Democrats exhibit consistently “liberal” preferences.

It is important to note that the individual scales we’re looking at here are effectively unidimensional. That is, we’re not attempting to ram together cultural, foreign policy, criminal justice, and tax policy together. That’s where scale construction gets substantively messy and where making claims about “moderation” can fall apart because of the problem of cross-cutting preferences averaging out in the aggregate. Instead, by looking at these scales individually, we get a chance to observe where some of the cross-cutting issues lie for Democrats. Although moderate and liberal Democrats hold somewhat similar views when looking at race, gender, and the economy, there are some sincere differences to how these respondents approach the intersection of criminal justice and economic policy beliefs.

Unlocking why some Democrats eschew the “liberal” label even as they are “operationally” liberal on average is beyond the scope of the present project, although my colleagues have written a bit about this elsewhere. But it is hard to imagine that some of these differences won’t have an important role to play in the upcoming primary.


6. Conclusion

In a previous post at DFP, we explored the idea that traditional social hierarchies were providing a powerful conduit for sorting people among the traditional “left” and “right” in American politics. Unlike that analysis, one of the goals of this post was to explore how real policy preferences intersect with social and racial attitudes with respect to voting behavior. We also sought to illustrate the extent to which there are partisan differences across this sorting, as well as clarify what differences exist between moderate and liberal Democrats.

Moving forward, I would make two speculations. First, while prior research argues that sorting among the right is more prominent than sorting among the left, the evidence presented here implies that across different social and economic dimensions, voters who preferred Democratic candidates in 2018 are reliably better sorted than those who preferred Republicans across this particular set of attitudes and policies. This is important. It suggests that there is meaningful convergence among Democratic beliefs, if not priorities. A popular work in political science describes Democrats as a loose collection of social groups that are not bound to central, hierarchical, and ideological principles like Republicans. This is sometimes used to explain why Democrats have a difficult time settling on a unified governing agenda. However, to the extent that this sorting is emblematic of convergence regarding issue positions and awareness of social problems, it would seem that greater consensus among Democrats exists than is usually attributed.

Second, the patterns presented here have implications for thinking about how candidates might portray themselves during the 2020 presidential primary. Based on the available evidence here, Democratic candidates who “run left” find themselves with a wellspring of support among the operationally liberal. On the one hand, defaulting to bridge-building or tacking back to the center would seem to discount the progressive wellspring of the party and be counterproductive with respect to its goals. However, there is evidence that, at least with respect to criminal justice issues, self-identified “moderate” Democrats are, indeed, less operationally liberal. That probably matters in that candidates who have a shaky progressive record on criminal justice issues -- for example, Joe Biden and, to some extent, Kamala Harris -- may appeal to these less liberal Democrats.

On balance, however, I think the outlook with respect to Democratic voters and operational ideology is optimistic. In 2018, almost 85 percent of persons who reported casting a Democratic vote had criminal justice and economic preferences that both converged toward the liberal end of those respective scales. In other words, at least with respect to matters of economic and racial justice, the modal Democratic voter does not divorce those issues. While campaign messaging is tricky and often context-dependent, it would seem to me that arguments for moderating Democratic candidates’ tone regarding race and gender in favor of strong economic messaging fly in the face of what the available data says about the present state of the Democratic Party’s voting base. In fact, there is broad convergence with respect to matters of social and economic justice.

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

Appendix: A Technical Note about Interpreting the GRM Scales

As I briefly outlined under the Section 2 heading, graded response models are a flexible way of constructing scales. Unlike simple additive scales used elsewhere, GRMs estimate a latent construct, along with a probability that a respondent will receive a specific “score” for each scale item. This latent variable modeling approach is particularly flexible with respect to missing data because it neither requires imputation nor arbitrary choices regarding coding missing responses. Instead, when calculating the probability of the latent trait, the model simply ignores the missing data from the calculation. Given the unidimensionality of our individual sets of instruments, graded response models produce latent scales that do not correspond neatly to additive ones. To illustrate this point, the table below takes our four indices and conveys the mean score associated with a categorical typology derived using an “additive” approach (literally just adding scores on all of the instruments together and then averaging across the items). The categories liberal, moderate, and conservative are rooted in whether a person’s additive score was below, above, or right at the “neutral” midpoint, which could prove problematic in that the “moderate” category confuses cross-pressured persons with offsetting liberal and conservative preferences with people who literally answered at the midpoint on all items. Indeed, this exercise is instructive because it illustrates that the score 0 on our latent indices (i.e. the mean of the GRMs) always falls between the liberal and moderate categories of the additive scale. In turn, this means that drawing cut-points at the quadrant coordinates (0,0) in this memo is probably a conservative approximation of the liberal-ness of operational ideology, thus highlighting the trouble with additive models in that “moderate” rarely corresponds to “neutral.”

Figure A1. Mean values on graded response models at liberal, moderate, and conservative categories derived from additive models. Estimates weighted by population weights provided by DFP.

Figure A1. Mean values on graded response models at liberal, moderate, and conservative categories derived from additive models. Estimates weighted by population weights provided by DFP.