This post went up at 8:30am on Friday; I mostly wrote it Thursday morning, but I did tweak it some Friday morning. It might be a little schizophrenic as a result.
As I write this, we sit in a state of uncertainty. Will Biden pull through and win? He only has to get Nevada, Pennsylvania, or Georgia to do so... if the media calling Arizona for him was right. Trump needs to take all four to win. But Trump has beaten expectations before.
And he's done so again. In my first draft of this, I complained that "[e]ven if Biden wins, the results we're seeing don't accord with the predictions of most pollsters and forecasters." But on reflection, I don't think that's true. Maybe my FiveThirtyEight buy-in is too high, but I'm pretty sure I saw Nate Silver say (either on Twitter or their election liveblog, though of course I can't find it now, so maybe it was Nate Cohn or Dave Wasserman or someone) that if the likely final results (306 electoral votes for Biden) had come in on election night, people would not have been decrying this as another big polling miss. Indeed, if you look at FiveThirtyEight's ballswarm of maps, you can see the likely final result right there:
It hovers at what I would guess is about halfway between the median outcome and the biggest possible Trump win. So no blowout, but nothing highly improbable either.
That said, it seems like even Silver would agree things weren't great:
In the end, polling averages will probably "call" the winners of all but 1-3 states correctly, along with the winner of the popular vote, which should wind up at Biden +4/+5. That's not great. It's better to look at *margins* and some of the margins were off.
— Nate Silver (@NateSilver538) November 5, 2020
In a follow-up, he adds, "the polls will have done mediocrely, but not terribly" and that "[t]he Democrats' performance in Congress *would* (correctly, IMO) have been seen as disappointing, and some of the hot takes would be about the contrast between that and Biden's performance."
My home state of Florida is a good example of all of this. FiveThirtyEight gave Biden a 69% chance of winning, with an expected final vote share of 50.9%. Biden's final takings were 47.8%. (Trump got 51.2%.) That doesn't seem too off. But if you look at the last few polls of Florida, you see this:
Biden in fact hit exactly where the last few polls said he would! ABC News seemed to get this one right, while Siena College/NYT and Quinnipiac had Trump way down. Poking around I can see that these two gave third-party candidates around 2-3%, and I assume the rest were undecideds. In actuality, all third-party candidates together took 0.9% of the final vote. Isn't this very similar to what happened in 2016? How did we not see it coming?
I don't know enough about forecasting to really know what went wrong. But it seems to me as though too much trust was placed in the idea of an even split in unallocated voters, and older polls with a higher rate for Biden might have been right when they were taken, but wrong when the election came around, and thus given too much weight? (I have one conservative friend who swung at the last minute from Biden to Trump because of the Barrett nomination. He had given money to the Biden campaign a month ago!)
Anyway, we can handwring about this, but I was curious to see Florida across the board. As mentioned before, I live in Florida's "most flippable" House district. In 2018, the results were 53/47, advantage Republican; this year the Republicans held the seat at 55/45, so we actually trended more Republican. (And, admittedly, FiveThirtyEight only gave the district a 20% chance of flipping; Democratic candidate Alan Cohn never polled too well.) Two House seats in Florida did flip-- but from Democrat to Republican. These were seats both labelled by FiveThirtyEight as districts where the Democrats were "favored" with an 80% chance. Of course, you should expect that things given an 80% chance will not happen 20% of the time, so this isn't impossible, but the margins seemed off across the board.
I thought the overperformance was bad, so I plotted it; on average, Republicans outdid the FiveThirtyEight prediction by 1.8%.* They outperformed in 17 of 25 competitive districts,† but underperformed in eight of them. Not as bad as my kneejerk reaction made it feel, but the districts they flipped, they overperformed by over 5%. If you look at the polling in those two districts, though... there was almost none! The last poll in FL-27 was in October, and the last poll in FL-26 was way back in July... and had the Republican up the amount he actually won by. I am sure making predictions about districts with little or no polling is tricky, but it seems to me then that an 80% certainty was misplaced.
Anything above the dotted line is a better than expected outcome for
Republicans. You'll note that Democrats really only did better than expected
in Republican strongholds where even a 5% overperformance would make no
difference. Republicans overperformed everywhere else, including where it
mattered most.
What does all this mean? I'll be honest, and say I don't really know. There's a lot to be written about polling in general, I'm sure, as well as polling in Florida. To me, though, it seems to speak to an underestimating of a particular kind of Trump supporter/Republican that keeps Florida locked "red" despite an apparent "purpleness," and thus an underestimation of a particular kind of American who keeps helping deliver (or nearly deliver, in 2020) this country to Trumpism despite everything.
One thing is certain: the mass repudiation of Trumpism many of us expected just did not take place.
* My final vote tallies come from the Tampa Bay Times as of Thursday morning. Provisional ballots are yet to be assessed, I think, but for the most part Florida counts everything on election day.
† In FL-2 and FL-25, Republicans ran unopposed.
No comments:
Post a Comment