Lessons In Humility

On 2020 Failures And 2024 Polling

Should anyone care what I have to say about American politics?

It’s probably the question I wrestle with the most now, reminded constantly of 2020 and the plainly terrible way I conducted myself that entire cycle. I’m less pained by the fact that I was wrong - live in the predictive spaces long enough and you’ll be wrong often enough - but the absolutely horrible way I treated any idea I wasn’t infallible. Genuine intellectual debate was dismissed as hackery, genuine Democratic hackery was passed off as brilliance, and I acted in a manner completely unbecoming of anyone with a public platform. And none of that gets erased by a good 2022.

2022 was, objectively, a good year for me, but it was also a year where there were only 4 genuinely difficult to predict Senate races and only a couple of genuinely tossup Governorships. Hell, even including Blake Masters in that list might be generous, so 3. Despite a House topline that was nearly bang-on, the weird and wacky nature of the year meant that individual seat hit rates were merely fine, not excellent. If you listened to me you certainly made money, as readers of my work for TheLines did, but the honest truth is you made money on my 2020 work too, because Biden did win and more importantly he won Georgia, which was the singular Die On This Hill stand I’ve made in America that paid off. That doesn’t mean 2020 wasn’t an absolute fucking failure on my part.

The honest truth is that 2024 has been an incredibly weird cycle, and the paralysis of the speed of the recent newscycle has allowed me to justify not writing about it here. It’s worked, in that I’m not pot committed to easily screenshot-able terrible takes, but it’s also not really true anymore that the state of the race is uncertain as to justify delay. I’ve written that Kamala is a slight favorite to win, and nothing’s changed that for me. Donald Trump very easily can win this election, and Democrats who are getting over their skis about their chances are engaging in a mild version of what I did in 2020. It’s fun to play out a good run of form to the logical conclusion, and yes, if the Trump campaign continues to fuck around and make stupid mistakes and focus on dumb shit, they’ll lose. It’s also pretty clear that no matter what Trump is posting on Truth Social there are Republicans who have realized that there’s a path to hitting Kamala being too left wing, using the various stances she took in 2019 to put her firmly to Joe Biden’s left in that primary, so even if the candidate can’t get on message the billions of dollars in campaign and PAC money will be on it.

What is less clear is whether there’s any reason to give a shit about the polls. That said, for good or for ill, there’s probably few people in the world who have thought more about how bad the polling was in 2020 than me, so as Democrats abound in optimism, let’s have an honest conversation about the polls, 2020, and whether getting excited about a D+8 national Marquette poll or those New York Times/Siena poll is just doing the same thing and expecting a different response. Because it’s me, let’s make it six thoughts, why don’t we.

There’s No Real Reason To Think The Polls Will Be Wrong The Same Way Again …

As a general rule, elections after big polling misses historically have not seen the same polling bias again. In 2015 the UK polls overestimated the left, which caused many to ignore how bad of a campaign Theresa May was running. Then the polling error flipped, Jeremy Corbyn beat his polls, and Labour nearly governed out of 2017. Famously (to massive nerds), a certain faction of Labour supporters used that miss as industrial strength copium in 2019. Turned out the polls were right then.

Australia 2019 saw the polls overestimate the left, a fact that buoyed a certain brand of poll denial by Scott Morrison loyalists. By the end, the polls were nearly bang on the money. Head to America, and 2012’s polls being low on Democrats was (incorrectly) used as proof that Hillary Clinton was in fine shape. At a basic level, pollsters have to fix their problems, because if they can’t, they don’t get their contracts extended. Most pollsters either work for campaigns/pressure groups or media outlets, and in both cases there is a direct financial incentive for the polls to be as accurate as possible. Being right can make you rich, and being wrong can take it all away.

Generally speaking, polling misses are as likely to invert as they are to continue. In certain, specific sub-national jurisdictions there are consistent biases (you can set your watch to Alberta Conservatives beating their polls), but nationally, generally, the pollsters will correct.

… But I Also Said This In 2020

Let’s not beat around the bush - getting a bigger error in the same direction the next cycle over is also pretty unprecedented. It might be that American polls are broken in a way that, say, Australian or Canadian polls aren’t. Overconfidence that there’s no way the polls are wrong the same way again is a hope and not a fact, especially in the face of polls like Marquette.

There’s still reason for progressives to be worried the polls are too good to be true. It’s not super likely, given pollsters’ incentives, but it’s definitely possible America is an international outlier. It’s also true that one of the biggest pieces of “evidence” in favour of polling being right is crap, which leads to my next point.

Midterm Polling Accuracy Doesn’t Correlate To Presidential Years

One of the mistakes that I made in 2020 was thinking that 2018 had somehow shown the polls were fixed. In 2018, there were 8 high profile statewide races in either swing states or what could at least generously be called that - Ohio, Wisconsin, Pennsylvania, Arizona, Georgia, Nevada, Texas, and Michigan. In 2018, Democrats beat their polls in 5 of those 8 races, and in 7 of those 8 races they either beat their polls or underran them by less than 2 points, on average. In all 8 of those states, Democrats underran their polls in 2020.

In Wisconsin, a state where Democrats only underran their polling by 1% in the Governor’s race in 2018, they underran by 7.8% in 2020. In Nevada, where they overperformed their polling in 2008, 2010, 2012, 2016, and 2018, they underran by 2.9%. Michigan saw them underrun their polling by 5.1% despite Stabenow only underrunning by 1.9% in 2018, and Governor’s polling being slightly more accurate.

It is true that in 2022, the polls underestimated Democrats in high profile Senate races. In the 8 even vaguely competitive Senate races, Democrats beat their polling averages in every single one (albeit by 0.1% in Ohio). The errors ranged from small (Ohio, a 1.1% miss in North Carolina that didn’t change the outcome or the pre-election narrative) to large (a 5.4% miss in Pennsylvania that showed entirely fictional Ozmentum, a 7% miss in New Hampshire that showed a competitive race that never was), but they were clear. In Michigan and Texas, the two other states that had Governor’s races, Democrats underran Beto’s polls by less than a point while Gretchen Whitmer outperformed by nearly 6%. It’s a nice set of facts to remind people that Democrats can beat their polls. It’s not a good argument that the polls have suddenly fixed their issues, because getting a midterm right is not the same thing as getting a general election right.

Beware Sweeping Claims About The Polls

There are, widely agreed, 7 competitive states in this Presidential election. I think North Carolina is a level off of the core 6 swing states, and I am not under any circumstances letting myself indulge Texas or Florida hope this cycle, so there’s 7 states. I’m going to include Ohio in this analysis not because I think Kamala can win it. I have Ohio Safe GOP Presidentially. She will not win it. But Sherrod Brown exists, so for the purposes of this exercise, there’s a Group Of 8 states this year, splittable into two groups of 4.

In the Rust Belt 4, the average polling miss in 2020 was Democrats underrunning their polls by 6%. The four biggest misses of the Group Of 8 were all Rust Belt states, and there were two errors above 7%. In the Sun Belt 4, there was an average error of 2.3%. Now, not every pollster missed by that much (Marquette’s Wisconsin error was 4%, to pick a pollster everybody is arguing about this week), but that doesn’t change the basic fact that you should judge Sun Belt polls less harshly than Rust Belt ones. Plainly, there does seem to be some evidence for the idea that a D+2 poll in Georgia or Arizona is worth more than a D+2 in Ohio or Michigan.

We Probably Need To Keep A More Open Mind Than We Realize

With days to go till the 2020 election I wrote something to the effect of “Donald Trump would need a substantially bigger poll miss than the one he got in 2016 to even have a chance to win.” Turns out, of course, he got an even bigger than 2016 miss and nearly did win. Susan Collins won in the race that still haunts my dreams, Florida not just didn’t flip to Biden as the polls consistently showed but gave Trump an extra 2% on the margin, and Colorado swung 9 points left on the same day California swung just under a point right.

None of this is dispositive - using “wacky things can happen” to argue in favour of any specific wacky outcome is nonsensical. It’s a flavour of nonsense that I have used before, and it’s enough that you can make pseudo-intellectual arguments sound fine. But as a rule, we should probably have more space for seemingly wacky ideas, given the last two cycles.

Should We All Freak Out About Marquette/NYT?

One of the things that I think is true about 2024 is that there is a universe in which a robust Democratic win - something like 2020 redux or 2020 + North Carolina, maybe with Sherrod Brown thrown in - isn’t that hard to explain afterwards. I’m not saying I’m predicting it, and I actively think this will be decided by thousands of votes, but it wouldn’t be that hard to see a decent sized Democratic win.

Joe Biden won by 4.5% against Donald Trump, and since that time Donald Trump led an insurrection and his Court appointees provided the votes to overturn Roe. It’s not hard to see how Kamala does roughly Biden 2020 margins with working class whites and racial minorities, the suburban trends continue, and it’s a ~D+6 environment and Dems win the House.

Now, by the same token it’s not hard to see how Democrats lose, either - the economy was better for most people in 2017-2019 than it’s been in 2021-2024, and people don’t blame Trump for the COVID economy. It’s not a crazy theory - Pierre Poilievre just won an incredibly socially liberal inner city Toronto seat because economic conditions (read:housing costs) meant that voters voted for the Conservatives despite hating the social values of the average Conservative. One of the things that the post-2020 election results have shown is that as much as the suburban shift is real, it’s not a guarantee that Democrats can never lose.

Plainly I don’t think Democrats are winning all three of Pennsylvania, Wisconsin, and Michigan by 4. I don’t think Harris is really tracking for an easy, clear by 11PM (even if the networks can’t quite call it yet because of mail ballot delays) victory that the Times polling and Marquette are showing. I don’t think it’s impossible, and when I see people confidently saying that the polls will be wrong the same way again I get an intense deja vu of Labour supporters saying that as Corbyn’s campaign burned in 2019. But I’m also not nearly as confident that the polls won’t be wrong again as I was in 2019.

The other part of this is that I’ve always been a believer in Kamala. The case for Harris optimism - that there’s a narrow anti-Trump majority that’s a bit bigger than in 2020 because of Dobbs and Jan 6th - is compelling. But Kamala Harris’ appeal in the 2020 primary was also compelling and she didn’t even make it to Iowa. I have always been higher on Harris than the market, both because I find her ideological position (left of Obama, right of Sanders) close to mine and because I’ve always found her personality to be compelling and not aggravating. What can I say, I love a politician who loves Motown as much as she does. I am aware of the fact that I have overestimated her appeal before. I am terrified I’m potentially doing it again.

The honest truth is that I’m absolutely terrified. I think Kamala’s the favourite, but I can absolutely hear an argument that she's not. If it turns out that her polls right now represented some sort of sugar high that ran out sometime after the DNC and before Labour Day I wouldn’t be shocked. I don’t think it is, but I also thought last time would be easy. As we saw, it wasn't.