John Sides has a Monkey Cage post called Democrats are gay, Republicans are rich: Our stereotypes of political parties are amazingly wrong, which is really just a precis of an as-yet unpublished article by Douglas J. Ahler and Gaurav Sood called The Parties in Our Heads: Misperceptions About Party Composition and Their Consequences. (I should say that Matt Yglesias wrote about it over at Vox a month ago, but I don’t generally read his stuff these days.) Mr. Sides starts his note thusly: Here’s a quiz question for you: What percentage of Democrats identify as gay, lesbian or bisexual? Here’s another: What percentage of Republicans make more than $250,000 a year? I’d love to know what numbers Gentle Readers would guess for those two questions, ideally before reading the rest of this note.
My guesses were that 2% of Democrats so identified, and perhaps 10% of Republicans were over that mark. I had very little confidence in my numbers, but that was what I came up with off the top of my head. The actual numbers according to the article tell us that 6.3% of Democrats are LGB (the article sometimes includes T and sometimes not, which is bad and wrong but probably doesn’t shift the numbers much) and that only 2% of Republicans have quarter-million-dollar paychecks. I wouldn’t be altogether surprised if people in general under-stereotype their own party and over-stereotype the other, although the research appears to show that in general people over-stereotype both. Still and all, for my over-stereotyping of the Other Party, considering that (according to the table on p. 46) 90.9% of respondents overestimated the correct answer, and (according to the terrific plots on pp. 47-48, if I’m reading them right) 75% of Republicans estimated the percentage of “rich” Republicans as 20% or more, I think it wasn’t a bad guess.
My main complaint, though, is about the use of averages in both the paper and the Monkey Cage post. Some Gentle Readers will remember how much I loathe the use of averages when you don’t mean to average things, and in particular, I hate attempts to average people. Mr. Sides says “On average, Americans thought that 32 percent of Democrats are gay, lesbian or bisexual.” In the paper’s abstract, Mr. Ahler and Mr. Sood write that “people think that 32% of Democrats are LGBT”. There are graphs and tables that similarly use the average—sometimes with a confidence interval. I find the highlighting of this average number as utterly and completely perplexing. What difference would it make in the study if that number were 30? Or 35? Or 40? What’s important is that they were overestimating, and overestimating by a lot—highlighting the average just gives a totally false sense of confidence that people really put that 32 number down, which of course most of them do not. It also makes that number seem important, which, again, it is not. What is important is that people consistently overestimate the extent to which the Parties are demographically different from each other, and different from the country at large. The 32% average of estimates of LGB Democrats is meaningless, as is the 38% average of estimates of $.25M/yr Republicans. I believe these numbers convey less than thirty-or-more would. That’s why the graphs are so good—here, I’ll reproduce them (this is smashed together from the two graphs).
Now, I know you can’t use those graphs in an abstract or the lede to a blog post, but that doesn’t mean you should use averages.
Tolerabimus quod tolerare debemus,