A few days ago, the watchword of the pollster-watchers was "herding". Polls adjust based on underlying assumptions (models) of the electorate, and these assumptions can sharply shift the reported results. Many have hypothesized that the pollsters, feeling burned by underestimating Trump's support in 2016 and 2020, are now overcorrecting to show a tight race so they don't look foolish in the event that we have our third straight tight race.
To this possibility, a lot of responses online took the form of "report the data you cowards!" If the data was showing Harris with a larger than "expected" lead (say, because she's cleaning up amongst women still furious about the fall of Roe), then don't hide from your own conclusions -- report them!
And then, right as that call of "don't be a coward" was cresting, the extremely highly regarded Selzer poll in Iowa put Harris over Trump by three in a state virtually nobody had in play. And that, in turn, generated a wave of "now I'm not saying I expect Harris to win Iowa, but ...."
Turns out, we're all cowards too.
No pollster is perfect, but Selzer's reputation for accuracy is well-earned. As you can see, in the last seven statewide Iowa races, Selzer's biggest miss was 5 points (2018 Governor), and more often she nails it to within a point or two.
So I'll add my voice to the chorus, but basically to echo Scott Lemieux: I'm also not going to venture a prediction that Harris wins Iowa, but even Iowa being closer than expected (say, Trump +2 rather than his 8-9 point margin from the last two elections -- this would equate to Selzer's largest recent "miss") augurs very, very well for Democrats across the country (and in contested House races in Iowa, for that matter).
3 comments:
Headline very funny, good work.
There is a distinction b/w 'report the true data' and 'treat the true data as strong predictive evidence', so idk if this is really revealing so much ~hypocrisy/superstition/etc as a strict reading of the post suggests? And in particular that biggest miss was a report of approximately this?
Post a Comment