Thread by @Nate_Cohn

Some early theories on went went wrong with the pre-election polling this year
nytimes.com/2020/11/10/ups…

Before we go into what went wrong, let's just call a spade a spade here: this was a bad polling error. It's comparable to 2016 in size, but pollsters don't have the excuses they did last time.
This year's polls would have been *way* worse than 2016 with a 2016 methodology

There are really two halves of polling: the quality of the sample you get, and the adjustments you take to improve the representativeness of your sample.
Since 2016, pollsters got better at the adjustments, but the underlying sample got worse

To my mind, the easiest way to see it: the crosstabs on white voters nationwide, particularly whites without a degree
In 2016, the polls *did* show Trump doing way better among those voters.
In 2020, they did not. And they were dead wrong.
Same story with seniors.

So what happened? How did the samples get that much worse over the last four years, especially among white voters without a degree and seniors?
At this stage, it's really just speculation. We'll know more later--it's too early for an autopsy.
But here's some early speculation

Let's start with the most interesting bit of theorizing I encountered, from always interesting @davidshor.
He thinks it's the pandemic: Dems took it seriously, stayed home and started responding to polls more. GOP did not.

This theory's almost too smooth to be true, but it's elegant and fits a lot of things together.
Remember those studies that said Biden does better in a COVID hotspot?
Well, Biden didn't do better in COVID hotspots.
So maybe... that just means the polls were biased by COVID

Another interesting thing it helps explain.
Our 10/19 battleground polls were... pretty accurate!
So were the state polls conducted from Jan-Mar 20.
Now, maybe these polls were just as bad--and Trump actually led big back then!
Or maybe... the bias started since then

Shor also put forward some data that's at least plausibly consistent with it. The increase in Dem response clearly happens before the pandemic, in the primary. But maybe it held because of the primary

There are some other theories worth consideration, too.
It's possible that four more years of Trump did four more years worth of damage to the credibility of media/polls, creating a 'hidden Trump' vote that didn't really exist in 2016

On the flip side, another possibility is the resistance: the huge surge of political activism on the left. We know political engagement is correlated with survey response. Maybe the folks now donating 100 millions to Senate races are now taking polls way more than four years ago

And a final possibility--and ultimately a falsifiable one--is the turnout.
There were a lot of polls showing no LV/RV gap or even a D turnout edge. Not sure that will hold up with final data.
It certainly didn't hold up in FL, where we have great data already.

Anyway, I do think it will take a while to get some better answers here. If you're looking for a polling autopsy, then I'd say the 'body' of polling is still at the scene of the crime and won't make it to the morgue for a bit.

But the fact that we don't have an autopsy wouldn't stop the police from acknowledging the existence of a dead body, and we don't need to shy away from the obvious, either: the poll results were quite bad, and the final numbers won't fundamentally change that