How to Detect a Crappy Poll

I frequently comment on podcasts and elsewhere about some of the methodological problems with current opinion polling, but rather than get into the arcana of sampling difficulties, statistical weighting, question design, etc., let me suggest a simpler method of instantly detecting a crappy poll: see if it’s from NBC News.

The latest NBC News poll just out finds that 45 percent of voters approve of Joe Biden’s performance in office, while 47% of registered voters say they prefer Democrats to control Congress, while 46% want a Republican-controlled Congress. The numbers tilt a bit closer to other polls showing much wider Republican margins when the domain moves from registered voters to likely voters, though the general conditions this year—and some of the “mood” numbers in this poll—make it hard to qualify “likely” voters.

But the real clue that this is a crappy poll is the issue that the poll reports is the first choice for most important for midterm voters: “Threats to democracy” (at 20 percent, just ahead of inflation at only 18 percent). There are virtually no actual voters who volunteer this answer: it is clearly a pre-calculated choice designed to tickle the media-programmed Pavlovian slobbering of liberal voters who answer the poll, and provide the media with the anti-GOP headlines they want.

The poll also reports that Americans are—wait for it!—deeply polarized! Large majorities of each party distrusts the other, and think the other party is a threat to America. Gee—what profundity. Findings like this—with percentage numbers attached!—indicate we are in the presence of no ordinary pollster.

If you are a glutton for punishment you can read the entire 24-page poll results here, and what you notice is that is has “only” 29 questions (plus a “statistical purposes only” panel), but many of the questions have six or seven parts, which means a respondent is answering maybe 75 questions in all. I especially like Question 16, which asks respondents to rank their trust in several public offices, including their state legislatures. I’ll bet less than 10 percent of respondents can even name their state senator or lower house member.

A poll this long is worse than useless.

Notice: All comments are subject to moderation. Our comments are intended to be a forum for civil discourse bearing on the subject under discussion. Commenters who stray beyond the bounds of civility or employ what we deem gratuitous vulgarity in a comment — including, but not limited to, “s***,” “f***,” “a*******,” or one of their many variants — will be banned without further notice in the sole discretion of the site moderator.