The Great Sorting · Tolerance
For half a century the General Social Survey has asked whether unpopular speakers should be allowed to speak. Which side would deny them a platform used to be a settled question. It is now a question of who is at the podium.
The Stouffer tradition in survey research built political tolerance into a simple instrument: would you allow a controversial person — a Communist, an atheist, a militarist — to give a speech in your community? For decades the lesson was tidy. Education raised tolerance, and the political left, by reputation, supplied it. Whatever a liberal thought of the speaker, the liberal defended the speech.
That portrait has come apart, and not uniformly. Run the GSS’s six targets through the partisan divide across three eras and the change is legible as a single picture: the question is no longer which party is more tolerant but tolerant of whom.
The partisan tolerance gap, by target and era
Difference in the weighted share who would let each speaker speak. Blue = Democrats more permissive; red = Republicans more permissive.
Two readings deserve emphasis. First, the early uniformity is real: across racists, militarists, Communists, atheists, gays, and (where asked) a Muslim cleric, Republicans in 1976–1994 were the marginally more permissive party on every target. The "left is more tolerant" story was always about ideology, not party — self-identified liberals did rate higher — and even that has now inverted for one target.
The modern question is not whether the left or right is more tolerant. It is whose speech each will defend — and whose it will not.
Second, the racist is the hinge. He is the one speaker whose platform the left has actively withdrawn: the share of Democrats willing to allow the speech fell while Republicans held, and the ideological gap flipped from +6 points (liberals more permissive, 1976–94) to −5 points (conservatives more permissive, 2012–24). On the other targets — the Communist, the atheist, the gay man, the Muslim cleric — Democrats edged ahead. Tolerance did not rise or fall so much as it became conditional on the identity of who is speaking.
The era averages compress a steady reorganization. Each panel tracks the share of Democrats and Republicans who would allow that speaker, year by year.
Would you let this person speak in your community?
Weighted share saying yes, by party. Democrats and Republicans.
Stouffer read tolerance as a civic muscle that education strengthened and exposure trained. Sullivan, Piereson, and Marcus complicated that, arguing tolerance is only tested against a group you actually dislike — "put up or shut up."2 The GSS now shows each coalition putting up for different speakers and shutting down others. That is not the disappearance of tolerance. It is its sorting: like affect, like trust, like the vote, the willingness to extend a civil liberty has lined up behind party and target rather than behind a general principle.
Data. GSS 1976–2024, weighted by weight_use. The measure is the "allow to speak in your community" item (spk*) for six targets; tolerant = would allow. Party folds leaners into Democrats (partyid 0–2) and Republicans (4–6); ideology uses polviews (liberal 1–3, conservative 5–7).
Wording matters. The targets are not symmetric. The racist item names "a person who believes Blacks are genetically inferior"; the Muslim item names "a Muslim clergyman who preaches hatred of the United States" — loaded phrasing that depresses tolerance for everyone and is asked only from 2008. Reading the racist and Muslim rows as clean mirror images would overstate the case; the racist trend and the ideological sign-flip are the load-bearing facts.
Honest limits. GSS 2021+ waves are mixed-mode (web redesign); the racist reorganization predates 2021 and is visible across multiple pre-pandemic waves, so mode change does not drive it. Cells with fewer than 40 weighted respondents are suppressed.