About those claims that the election forecasts hurt the Democrats in November

One reason that I’m skeptical of these claims of depressed voter turnout is that I’m old, and I remember the 1980 election. People blamed the Democrats’ poor performance in the west coast on the fact that the election had been called for Reagan before the polls had closed in those states. So the message seems to be: If the Democrats think they’re gonna lose, they won’t turn out, and if the Democrats think they’re gonna win, they won’t turn out. I guess it’s possible that Democrats are more sensitive than Republicans to the perceived closeness of the election, but what I really think is going on is that most of the people doing these analyses are Democrats, and they’re implicitly acting under the assumption that Republicans’ behavior is fixed and Democrats’ behavior is changeable. I connect this to the so-called fundamental attribution error in psychology. I could swear I wrote a blog post ten or fifteen years ago about this general point (the idea that historians often seem to give their own country more agency when writing about past events; one example is the various British historians who blamed World War 1 on . . . Britain), but now I can’t find it . . .

14 thoughts on “About those claims that the election forecasts hurt the Democrats in November

    • Ben:

      You might be right, but point 3 of that post (and the subsequent article) was about survey response, not voter turnout. Less than 10% of people respond to surveys, whereas about 60% of eligible voters turn out to vote. So it makes sense that whether you respond to a survey is subject to all sorts of influences, while the voter turnout decision would be more stable.

      • Well, it doesn’t take a huge change to cross the kind of thresholds which seem to matter in elections in any case. Is that fair to say? It’s small percentage-point swings right?

        And 100 choose 60 > 100 choose 10, so in terms of ways to get 60%, I think there are less ways to get 10% than 60% so I don’t see why we’d assume the 10% is more wiggly than the 60% just from lookin’ at the numbers.

        Or in r, the second curve looks wider:

        > curve(dbeta(x, 11, 91))
        > curve(dbeta(x, 61, 41), add = TRUE)

        So in a population of 100, it seems like there’s more variation in 60% of people turning out than 10%.

        • Ben:

          What’s relevant is relative variation not absolute variation. Take those curves and divide the first one by 0.1 and divide the second one by 0.6 and you’ll find the second curve is more variable.

        • If we pose the problem in terms of people not-turning out to vote, then the relative rescaling doesn’t help and I’m not sure how it’s a different problem.

        • Ben,

          I dunno, I just think that whether someone votes is much more predictable than whether someone responds to a poll. I agree that most people don’t respond to polls at all, so in that sense it’s very predictable—but I think the pool of respondents varies a lot from day to day, much more than the way there is uncertainty in voter turnout. I’ll have to think further about how to formulate this idea.

  1. Andrew said, “One reason that I’m skeptical of these claims of depressed voter turnout is that I’m old, and I remember the 1980 election.”

    I’d say you’re middle-aged. I’m old, and I remember the 1948 election, when Dewey-Warren ran on the Republican ticket, vs Truman-Barkley on the Democratic. We had a baby-sitter who gave us Dewey-Warren buttons. (I thought that Dewey Warren was one person, with first name Dewey, and last name Warren. But, speaking of vice-presidential candidates — Harold Stassen had been a contender for vice-presidential candidate on the Republican ticket. I remember my father referring to him as St. Assen.)

    • Rm:

      In a primary election with many similar candidates and no party cues, I think polling results can feed back into voter behavior in many ways. In a general election with two candidates from opposing parties, not so much. See this article from a few years ago, “Why Are Primaries Hard To Predict?”

  2. Someone should tell the machine learning people this…! From the intro of a paper I happened to be reading today (https://arxiv.org/abs/2102.08570): “For example, election forecasts impact campaign spending and affect voter turnout, hence influencing the final election outcome” (then cites the Westwood, Lelkes, and Messing paper https://www.journals.uchicago.edu/doi/abs/10.1086/708682).

    The intuition you give for why it matters more in primaries makes sense.

Leave a Reply to Andrew Cancel reply

Your email address will not be published. Required fields are marked *