What You Say You Want

Ask somebody what they are looking for in a partner and you will get an answer. The answer will be fluent, considered, and in some tension with who they actually end up interested in.

The clearest demonstration is a speed-dating study. Participants, students at an American university, filled in measures of their ideal partner preferences before the event: how important physical attractiveness was, how important earning prospects were, and so on. Then they met between nine and thirteen other participants in short rounds and rated their romantic interest in each one.

Two findings came out of it.

The first is the one that made the paper famous. The stated ideals showed the familiar pattern that evolutionary accounts predict: men rated physical attractiveness as more important, women rated earning prospects as more important. But in the live data, there was no sex difference. Attractiveness and earning prospects predicted interest for both groups, at similar strength, once there was an actual person sitting opposite.

The second is the one that matters more. A given participant's stated ideals did not predict which of the people they met they were interested in. Everybody had a type. Nobody's type forecast their evening.

Why would people be so poor at this?

Because self-report about causes of your own judgements is generally unreliable, and has been known to be since the 1970s. People do not have privileged access to the machinery producing their preferences. They observe their own conclusions and then generate a plausible account of why, and the account can be entirely wrong while feeling like a memory. This is not a fact about romance; it shows up in consumer choice, hiring decisions and moral judgement.

Now the caveats, because the strong version of this finding is not supportable and the strong version is what circulates.

The original sample was students at one university, in a speed-dating format, which is a specific and unusual context. And a later direct replication with a larger sample did find that attractiveness and earning prospects predicted interest, again with no sex difference. Related work found that stated preferences predicted interest in photographs but not in live interaction, which suggests the format is doing real work.

So the claim that stated preferences predict nothing is too strong. The claim that survives is more modest: a person's list of requirements is a weak guide to who they will actually like when they meet them.

Modest, and consequential, because of what gets built on top of it.

Every dating platform collects stated preferences, because stated preferences are what a form can collect. Age ranges, height, education, religion, whether you want children, how important various things are to you. The filters and the matching questionnaires all run on self-report about preferences.

Which means the industry is constructed on exactly the variable that predicts least well, not through carelessness but because it is the only variable available before two people are in a room together. That is the problem the next lesson runs into, and it does not have a software solution.