What people say they want and what they choose

People can describe what they are looking for in considerable detail. Studies of what they then choose find the description predicts remarkably little.

The speed-dating findings

The cleanest evidence comes from studies where participants stated their preferences in advance and then met a series of people under controlled conditions.

The consistent result is that stated preferences predicted actual choices weakly, and that the traits people said mattered most were not the ones associated with who they selected.

Follow-up work found that after meeting someone, people revised their stated ideals in the direction of that person — the preference adjusted to the choice rather than the reverse.

Which suggests that the checklist is partly a post-hoc account rather than a decision rule.

Why this happens

Introspective access is limited. People do not have direct insight into the processes producing their preferences, a general finding in psychology that applies here with force.

The criteria are not comparable. Attraction is a response to a whole person in a specific moment, and it does not decompose into the attribute list people use to describe it.

Stated preferences are partly social. What people report wanting is influenced by what seems reasonable to want, which is not the same as what moves them.

Context does a great deal of work. The same person encountered in different circumstances produces different responses, which no attribute list captures.

The parts that do predict something

Not nothing, and worth separating out.

Similarity in values and background is associated with relationship stability across a good deal of research, more reliably than similarity in personality.

People do tend to pair with those similar to them on education, background, political outlook and religiosity, and those similarities predict outcomes better than shared hobbies do.

Perceived responsiveness — the sense that the other person understands and cares — is among the more robust predictors of attraction developing, and it is about interaction rather than attributes.

Repeated exposure. Familiarity tends to increase liking, which is well documented and is why so many relationships begin among people who were already around each other.

What follows practically

Loosen the specification. A detailed list of required attributes is filtering on criteria that do not predict what you will actually respond to, while excluding people who would.

The items worth keeping are the ones with real consequences: children, location, religion, general values, whether they are kind.

Weight interaction over profile. How a conversation goes carries more information than the attributes it revealed, and it is available only by having it.

Meet more people in fewer categories rather than fewer people in more. Since prediction is poor, sampling is more informative than filtering.

Be suspicious of the type. People who describe having a type frequently describe a series of relationships that failed the same way.

Where a pattern is present and the outcomes have been consistently poor, the pattern is worth interrogating rather than serving.

The attractiveness question

Handled directly, since it is the elephant in most of this research.

Physical attractiveness predicts initial interest strongly and reliably, more so than people report it does.

Its predictive power declines with acquaintance. Studies of people who know each other well find much lower consensus on attractiveness ratings than studies of strangers, which suggests that knowing someone changes how they look.

Which is consistent with ordinary experience and worth stating: the strong effect exists at first meeting and substantially weakens afterwards.

The uncomfortable implication

If preferences predict poorly and compatibility is difficult to forecast, the search model of dating is partly mistaken.

The alternative model — that a workable relationship is constructed between two reasonably compatible people who both invest in it — has more support in the outcome research and is much less appealing as a story.

It implies that the difference between relationships is less about who was selected and more about what happened afterwards.

Which is not an argument for settling. It is an argument that the extended search for a person who matches a specification is optimising something that does not determine the outcome.