The Great Sorting · Affect
For sixty years Americans have rated social groups on a 0–100 scale — the police, labor, gays, immigrants, the rich. Which dividing line organizes those feelings used to be race. Now it is party.
Since 1964 the American National Election Studies has handed respondents an imaginary thermometer and asked them to register how warm or cold they feel toward dozens of groups. The instrument is crude and durable, and political scientists have spent a half-century mining the two readings that point straight at partisanship: how warmly you rate your own party and how coldly you rate the other one. That gap — affective polarization — has become one of the best-documented facts in the study of American politics.1
The thermometer battery also contains a quieter record. People rate groups that have nothing to do with party on their face: Catholics and Jews, the police and the military's critics, big business and labor unions, feminists and fundamentalists. Treat the whole battery as a single object and a structural question comes into view. Of all the ways Americans are divided — by race, by religion, by income, by party — which division best predicts how warm or cold a person feels toward these groups? And has the answer changed?
It has. For each year, every non-political target is run through a weighted analysis of variance against four cleavages at once, on the same set of respondents, so the four are directly comparable. The statistic is η²: the share of the spread in a group's ratings that a cleavage accounts for. Averaged across the groups, it draws a clean picture of which fault line organizes American social feeling.
Which cleavage organizes how Americans feel about social groups
Share of the explained variance in non-political feeling thermometers attributable to each cleavage, by year. The party band swells; race recedes.
Read in raw terms rather than shares, the reversal is just as plain. Across all thirteen groups, the variance party explains rose from 0.017 in 1964–84 to 0.088 in 2012–24, while race fell from 0.044 to 0.025. Party went from explaining about a third of what race did to roughly three-and-a-half times as much — a dominance ratio of 0.38 → 3.58. Hold the basket of groups fixed to the five asked across the whole span — blacks, whites, Jews, big business, labor — so the trend cannot be an artifact of newer, more partisan targets entering late, and the same reversal survives, attenuated: 0.019 → 0.061 for party against 0.050 → 0.026 for race, a ratio of 0.38 → 2.37. Either way the crossover lands at the same place: the year 2000.
The average hides the migration of specific groups into party's orbit. Tap any target below: the panels are tinted by the cleavage that best predicts feeling toward that group in the most recent reading, and the detail view shows the Democratic and Republican averages pulling apart while the four η² lines reshuffle underneath.
Thirteen groups, sorted into party's column
Each panel: the share of feeling explained by party (gold) versus race (blue) over time. Border color marks the dominant cleavage in the latest year. Tap to expand.
Average warmth, by party (0–100)
Variance explained (η²), by cleavage
The claim is about structure, not heat. Americans do not loathe Catholics or Asians the way they loathe the opposing party; on raw warmth, ratings of marginalized social groups remain far milder than the out-party rating, a point Iyengar and colleagues have stressed.2 What has changed is the axis. Knowing a person's party once told you little about how warm they were toward the police or toward labor; knowing their race told you more. Today party is the variable that sorts those feelings, and race, religion, and income have all receded behind it. Lilliana Mason's account of a partisan “mega-identity” — the stacking of race, religion, and ideology onto a single partisan team — predicts exactly this: as the other identities line up behind party, party inherits their power to organize affect.3
Race used to be the master cleavage of American social feeling. Around the millennium, party took the title and never gave it back.
The migration is legible group by group. The police were a cleavage of class and race through the 1970s and 80s; by the 1990s, partisanship predicted warmth toward them better than anything else, and the line has only steepened since. Feeling toward immigrants followed in the 2010s. Each group that crosses into party's column is one more former cross-pressure dissolved — one less way for an American's loyalties to point in two directions at once. That is what a sort is: not louder opinions, but fewer independent dimensions. The thermometer has been recording it all along.
Data. ANES Cumulative File, 1964–2024 (24 surveyed years), weighted by the cumulative full-sample weight. Thirteen non-political social-group thermometers; party, candidate, ideology, and institution thermometers are excluded as inappropriate for a “non-political” test.
Measure. For each group in each year, weighted η² (between-group sum of squares over total, frequency-weight convention) is computed for four cleavages — party ID (three categories, leaners folded in), race (three categories), religion (Protestant/Catholic/Jewish/Other-or-none), and income tercile — on the same respondents, those valid on the group and on all four cleavages. The four are then directly comparable. Year values average η² across available groups; the stacked figure normalizes the four to sum to one.
Honest limits. η² rises mechanically when a categorical has more or better-balanced cells; the party measure uses three groups as do race and income, so the comparison is fair, but the level should not be read as a causal effect size. The all-groups series mixes in late-entering, highly partisan targets (Muslims from 2004, immigrants from 1988); the constant-basket series removes that and is reported alongside as the conservative read. Income is the cleavage with the most item missingness; requiring all four cleavages valid shrinks early-year samples to a floor near 900 respondents.