Evaporative Cooling of Group Beliefs
A reader's summary of a metaphor from the rationalist blogs: why a group can become more extreme without anyone being persuaded, simply because the least convinced members leave first. What the image claims, where its pieces come from, and how far the evidence goes.
The term at a glance
Evaporative cooling of group beliefs is the idea that a community united by a belief can radicalize through selective departure alone. In a cooling cup of tea, the fastest molecules escape as steam, so the average energy of the liquid left behind drops. The analogy turns this around: in a group, the members with the weakest commitment leave, and the average conviction of those who remain goes up. It is a close relative of the purity spiral and of groupthink, but the mechanism is different. Nobody has to be pressured to conform; the people who would have disagreed are simply no longer there.
Origin
The phrase comes from Eliezer Yudkowsky, who wrote a short post titled “Evaporative Cooling of Group Beliefs” in December 2007 on the blog Overcoming Bias, later carried over to LessWrong. The post is brief and informal, an essay by analogy rather than a research paper. Its central example is a story drawn from social psychology rather than from physics, and the physics is there only to name the pattern.
History and context
The case Yudkowsky leans on is Leon Festinger, Henry Riecken and Stanley Schachter's 1956 book When Prophecy Fails. The authors had infiltrated a small group in the American Midwest that expected a flood to destroy the world on a set date and a flying saucer to rescue the faithful. When the date passed with nothing happening, the book reports that the members who were least committed, those who had kept their jobs and families outside the group, drifted away, while the core intensified its belief and began to seek publicity. Festinger used the episode to develop his theory of cognitive dissonance. The selective-departure reading is an additional lens on the same story, not Festinger's own thesis.
A second ancestor is the economist Albert Hirschman's Exit, Voice, and Loyalty (1970). Hirschman asked what members of a firm, party or state can do when it declines: leave, speak up, or stay loyal. He noted that exit can quietly damage an organization by removing exactly the critics who might have forced it to improve. The evaporative image is that argument run on beliefs instead of customers.
A third, empirical neighbour is group polarization. Cass Sunstein and others have documented that like-minded groups, when they discuss, tend to end up more extreme than their members began. Selective exit adds a different route to the same place, one that works even if nobody's views shift inside the room. The online version is easy to recognise: a forum splits after a dispute, a moderate subreddit empties, a chat server is left to its most intense regulars. It joins the echo chamber as one of the stock descriptions of how online groups narrow.
Main ideas
The moderates leave first
When a group's shared belief meets bad news, the members who are least invested walk out quietly. Those who stay are, on average, the most committed, so the group's average belief rises even though nobody changed their mind.
Exit is cheaper than voice
Arguing with a group costs time and standing; leaving costs a muted notification. Hirschman's 1970 framing explains why the people best placed to correct a group are often the first to stop trying.
The remainder reads as proof
Each departure leaves a denser room of believers, which feels like confirmation. A smaller, more certain group can sincerely conclude that the leavers were never really committed.
It feeds on itself
As the average rises, the next tier of moderates finds the room more uncomfortable and goes too. The process stops only when the group is small and uniform, or when something outside intervenes.
Platforms make exit nearly free
Muting, unfollowing and leaving a server take a click, and a replacement community is a search away. The cost of leaving, the brake on the process, is close to zero online.
Critique
- It is a metaphor, not a measured law. The original post offers no data on how often groups behave this way. Real departures are noisy: people leave for boredom, conflict or a job change, not only for doubt.
- The evidence from the founding case is thin. When Prophecy Fails studied one small group, through participant observation, and its accuracy has been disputed. Critics have argued that the observers shaped events and that the portrait of intensified belief is overstated.
- Exit can also cool a group down. Sometimes the people who leave are the zealots, and a group becomes calmer and more pragmatic. Nothing in the model says which tail departs first; that depends on what the group punishes.
- New members refill the room. Groups recruit. If newcomers are more moderate than the core, they dilute the concentration; if they are more extreme, they speed it up. A closed-system picture misses this.
- It is hard to test from outside. Showing the effect would take records of who left and what they believed beforehand. Observers usually see only the extreme remainder and infer the departures, which is the shape of survivorship bias in reverse.
Impact
The image has become a common shorthand in online discussion, used to explain how a community's tone sours after a controversy, why a once-broad movement ends up speaking only to itself, and why leaving a toxic space can feel both right and costly. Its practical lesson for anyone who runs or belongs to a group is plain enough: criticism that arrives as a departure cannot be answered, so a group that wants to stay calibrated has to make staying and disagreeing cheaper than leaving. For a reader, the useful question about any intense community is not only what its members believe but who used to be in the room.