Democrats and Republicans today seem like members of rival Veronese families. Not only do partisans distrust and disparage the out-party, they also dislike the idea of anyone in their family dating an out-party supporter. A Democratic Romeo and a Republican Juliet in 2020 might just be more star-crossed than their fictional counterparts.

Some of this animosity is grounded in genuine disagreements over politics and policy. But much is not. Among the most perverse sources of hostility is stereotyping: Americans misperceive their political opponents, believing parties are composed of stereotypical cutouts.

For example, Democrats believe that 50% of Republicans support white nationalism—just 9% do—and Republicans believe that only 70% of Democrats say they love America—96% do.

And just as Americans hold stereotypically biased beliefs about partisans’ attitudes and behaviors, Americans believe partisans look like their caricatures. Republicans, for instance, believe that 39% of Democrats are black—the actual number in 2015 was 24%—and that 38% are gay, lesbian, or bisexual (vs. 7% in reality). Likewise, Democrats believe that 44% of Republicans are age 65 or older (vs. 21% in reality) and that 44% also earn $250,000 or more per year (just 2% do). Crucially, correcting misperceptions like these causes partisans to like the other side more.

How do incredibly inaccurate—and seemingly consequential—stereotypes like these develop? The leading theory is that stereotyping is a consequence of humans’ relentless need to simplify the world. People reflexively use mental shortcuts, or heuristics, to simplify the myriad judgments they make every day.

One of the most important shortcuts is the representativeness heuristic (hereafter referred to as “the RH”). Like all heuristics, the RH makes judgments easier by substituting a harder question with an easier one. For instance, people judge party composition—said to drive party identity itself—by thinking in terms of “representative” traits, i.e., those which most strongly distinguish Democrats from Republicans. In doing so, people often overlook important information relevant to the actual, harder question and arrive at a suboptimal answer.

To estimate the share of Republicans who earn at least $250,000 annually, people need to consider not just how representative of Republicans high-income people are—that is, how much more likely it is for one of these high earners to be a Republican than a Democrat or independent—but also that less than 2% of Americans earn $250,000 annually. Recognizing this, people should conclude that there are relatively few über-rich Republicans, even if the rich tilt toward the GOP. When using the RH, people ignore information about base rates and rely simply on their perceptions of how distinctive the traits are of party identities.

If the RH causes people to stereotype groups like this, we ought to observe its distinctive effects on their reasoning when they categorize others. In a new paper, we present a series of experiments documenting just this—when people engage in stereotyping, the things they say and do show all of the hallmarks of using the RH.

The first of these peculiarities is the conjunction fallacy—concluding that some subset of Event A (e.g., drawing a royal flush in poker) is more likely than Event A itself (drawing any straight flush). Kahneman and Tversky famously demonstrated this consequence of the RH with “the Linda Problem,” a vignette about a recent philosophy major who was active in anti-nuclear demonstrations and concerned with discrimination and social justice. When people are asked whether Linda is more likely to be “a bank teller” or “a bank teller who is active in the feminist movement,” large majorities routinely select the latter option simply because Linda is described in a manner highly representative of feminists. Yet bank tellers who are active in the feminist movement make up a small subset of the broader category of bank tellers.

We appropriate the Linda Problem with a character named James, whose race, religion, sexuality, and ideological orientation randomly vary such that he can be more or less representative of either party. Sure enough, when James has more attributes that are representative of the Democratic Party—being secular or liberal but especially being black or gay—respondents are more likely to say that James is “a salesman who is a Democrat” instead of just “a salesman.” When his attributes become more representative of the GOP, he is more likely to be “a salesman who is a Republican.” And when he holds a mix of attributes (or bland, “control condition” attributes), people are least likely to commit the conjunction fallacy.

Predicted probablility

Note: Survey respondents saw a profile with a mix of randomly chosen attributes that you see above. These attributes color snap-judgments people make about others, to the point of causing impossible conclusions.

Another experiment addresses how the RH causes stereotyping by making representativeness more salient for randomly selected participants. The outcome of interest is participants’ responses to questions about the percentage of party supporters belonging to party-stereotypical groups, like the percentage of Republicans who earn at least $250,000 per year. Essentially, this experiment tests whether stereotyping worsens when we make the RH more available to participants.

In the treatment condition, we expose participants to “representative information,” e.g., the percentage of Americans earning at least $250,000 per year who are Republicans (50%) as opposed to Democrats or Independents, and pay them a bonus simply to reproduce this number at the end of the survey. In a second condition, we expose participants to groups’ base rates, e.g., the percentage of Americans who earn at least $250,000 per year (1%)—information that should improve perceptions but that people are said to overlook in favor of the RH. Finally, we compare responses in these conditions to a control condition in which we simply administered the outcome measures.

If representativeness fuels stereotyping, we ought to observe the most biased perceptions in the group exposed to “representative information”—and, indeed, this is exactly what we observe below.

Note: Respondents in the Base rates condition were paid a bonus to remember information about the percentage of the US adult population belonging to each of the two party-stereotypical groups. Respondents in the RH condition were paid a bonus to remember the percentages of the two groups belonging to the party. The dependent variable is the respondent’s perception of the percentage of the party belonging to each of the two groups.

Our final experiment on representativeness confirms its heuristic use in political stereotyping. It also offers a glimmer of hope for reducing stereotyping’s ill effects. Asking the same types of questions—e.g., “What percentage of Democrats are black?” or “What percentage of Republicans are age 65 or older?”—we show that stereotyping increases when people face heavier cognitive loads. When we put participants under time pressure, their responses to these questions were significantly larger, indicative of stereotyping. By contrast, when we asked participants to wait 15 seconds before answering, their responses were less biased. This suggests that stereotyping is deeply ingrained in our automatic mental architecture, but it also suggests stereotyping is far from immutable—simply granting oneself time to think may be a solution. But to that end, it is important to note that people are only likely to reflect on their own perceptual shortcomings when they are aware of them.

In sum, we see all the artifacts of the representativeness heuristic one would expect to see if it fueled stereotyping—the conjunction fallacy, base rate neglect, and greater use of representativeness among the most cognitively pressured. This suggests that while there is much to disagree about in today’s politics, in many cases, partisans’ gut-level sense of alienation from the other side could use a gut check.