The exit poll asked which issue mattered most to your vote and offered seven options. I have the instrument in front of me. The seven were drafted in June, tested in July, and fielded in November, and the thing that dominated the last three weeks of the campaign is not among them. It would have gone in under "other", which took four percent.
The published finding was that the economy was the top issue, at thirty-one percent.
I have used exit polls for a decade and I am not going to argue they are worthless — they are the only instrument that asks voters anything on the day, the fieldwork is hard, and the people who run them are candid about the limits in a way the people quoting them are not. What I want to say is narrower.
A closed-response question does not measure priorities. It measures the fit between a respondent's priority and a list somebody wrote months earlier, and then it ranks the words on the list.
Watch what that does. Two respondents choose "the economy". One means she cannot make rent. One means he is angry about a tariff. They are recorded identically, and the aggregate — thirty-one percent — is then reported as an explanation, and within a week it is a strategy. The word absorbed two incompatible things because the word was on the list and the things were not.
Now the direction of the error, which is what makes this worse than noise. Broad, durable words like the economy are the ones a drafter can be confident about in June, so they are always on the list. Specific, late-breaking concerns cannot be, by construction. So the instrument systematically over-reports the stable and under-reports the recent, in a survey whose entire purpose is to explain a thing that just happened.
An open-ended version exists. It is more expensive to code, the results are messier, and the coding introduces its own judgments — I have done that coding and it is not a clean escape. But the messiness is the finding. When you ask people in their own words what decided it, you get a distribution with a long tail, and the tail is where an election actually turns, and the closed question is a machine for flattening exactly that.
I want to be careful about my own position, since I have quoted the thirty-one percent myself in writing and did not attach any of this to it. The reason I did not is instructive: the number was convenient, it agreed with what I already believed, and the caveat would have cost me a paragraph. That is the whole mechanism by which a bad number stays in circulation, and it does not require anybody to be fooled.
