Knowable Errors, Georgia, And 2022

The Runoff Result Was Always Obvious

There are two kinds of predictive errors - ones where the process was correct, and for reasons beyond your control, you got it wrong, and ones where there was obvious flaws in the methodology, a clear reason for the error, and one that was - generally speaking - clear before the event itself.

The classic example of the latter is how I didn't think there was any chance Donald Trump could win an election in 2016 - that he was too extreme, and too insane to win. I had no justification for the claim, and while there was a polling miss, it wasn't a miss of a size that should have justified my confidence. And yet, in my cocoon of liberalism at the University Of Ottawa, I was sure he would lose. It was an error, obviously, and one that should have been obvious with foresight.

Contrary to that, I actually don't think the quant work that many did this year was particularly wrong, process wise. The polls, supposedly better after fixing their 2016 problems, were worse - and in Texas, where the polls had overstated Republicans in both 2016 and 2018, the miss went the other way, so it's not like a correction for past polling misses would be some easy answer. The errors of so many - obviously, myself included - are obvious with the benefit of hindsight, but I maintain that there wasn't much that could have been done about it beforehand. With the information we had, Florida should have gone blue. It's just that our inputs were shit.

…

The concept of "knowable error" is one I talk about often - was the error something you could have reasonably foresaw, or was it just bad luck? If it was a knowable error, then your process needs to be fixed, but if it was just bad luck, the question needs to become whether or not whether it was a one-off source of bad luck or whether the amount of variance in your project renders quantitative analysis meaningless.

Readers of mine will know my view - US polling is so broken that I do think attempts to do quantitative work is going to fail. Yes, the polls in the Georgia runoff were correct in aggregate - but that was a bit of a mirage. Most of the polls were either D+5 or more, with some others showing Republican leads. Far from the polling averages converging near where the bulk of the polling lay, the averages were a position where there was almost no singular polls - but a whole host of bad polling got cancelled out to the right spot. That isn't a sign of health, it's a sign of luck.

The tone and tenor of my writing in the past - the arrogance that dripped off my past work - was a fucking disaster, and I should have seen it at the time. That was absolutely a knowable error on my part, and something that I'm trying - desperately - to avoid doing again. I will have some strong opinions, but there's a difference between that and willful arrogance. That said, this is not just an exercise in self hatred, so let's get to it.

The Georgia runoff results were obvious for two fucking months, if you didn't see that, that's on you, and we shouldn't have to pretend it was some hard call.

…

I literally never thought the GOP would win the runoffs, as much as I was terrified of going out on a limb with pro-Democratic sentiment again after everything that had happened on November 3rd. That said, what happened on November 3rd was obvious right away - Trump managed to draw out a cadre of super low propensity non-college whites which turned the electorate everyone expected into one that nobody saw coming - and the consequence of it was obvious too. Those people weren't going to turn out again in a damn runoff - and, shocker, they didn't. It was obvious if you looked at the 2008 Senate runoff, where the share of the vote in counties in Metro Atlanta rose, and the turnouts in the rural south was trash in 2008.

And then Nate Cohn told us that Black turnout - as a share of the electorate - was down at 27% in the general election, the lowest by their math since 2006. My immediate reaction to that was that the election was clearly D favoured now, because Black turnout was going to be up - and probably up a bunch - in a runoff. Oh, wait, what's that? That's exactly why Democrats won? Yep, sounds about right, but, see, I was being delusional when I said this.

Pointing out that runoffs and general elections were completely different - that their dynamics and their symmetries were totally different - that was just "word salad," and I was missing the important facts - that Perdue beat Ossoff in the general, and almost won outright. I was missing that crucial fact, because, you know, those votes get to be banked or some shit. Oh, wait, they don't, my bad. It was wrong that I saw differential turnout patterns - the same ones that happen every single fucking year that doesn't include a Presidential election - of non-degree whites staying home and college whites voting - and going, yeah, the GOP are in trouble.

And then we got the early vote data, which consistently showed a Blacker electorate than 2020, corrected for every possible factor - but again, it was still a Tilt R or Tossup race, all the smart people said. The night that friend of Scrimshaw Unscripted Lakshya Jain finally - after days of my bullying - moved the race to Lean D he got a torrent of Doomers at his gates.

And then as the GOP didn't do what it needed to in the early vote, we still had people saying it was a Tossup. Hell, I respect the Elections Daily crew for at least having the balls to go to Lean R, even though they were wrong. The runoffs happened in the exact way that everyone who actually understood the reasons behind the GOP's previous overperformance in Georgia runoffs said they would - the electorate was more educated, rural whites didn't show up, and the party of educated whites did better. That trade used to be death for Democrats. Now, it's a win. And this was obvious to everyone.

It was, also, obvious that suburban reversion wasn't going to suddenly turn the increasingly blue suburbs back around, as anyone who had even a passing knowledge of International elections would understand - but yet, here we were, still stuck in this nonsensical position of having to pretend that suburban reversion was possible. Or, at least, others were - I never pretended.

What happened in Georgia was a knowable error, but it seems like everyone has just moved on from it. Far from wanting to learn the lessons of it, it seems like everyone is just content to go into 2022 without having learned any of the lessons that Georgia shows us. Just don't be surprised when the nonsense masquerading as analysis comes up short again in 2022 - and yes, if you bank on rural whites turning out like a Presidential year in a midterm, your failure will be a Knowable Error.