Election Night Preview: 2024 General
it’s a tossup whether this tossup turns out to be a tossup.
The U.S. general election is on Tuesday. Here’s what’s happening:
We will elect presidential electors, in what some are already calling “the fifth most important election of your lifetime.” (The electoral college will meet on December 14 to elect the president.)
We will elect 34 out of 100 U.S. Senators.1 The Democrats currently control the Senate by a narrow margin: 51-49. If the Senate ends up tied 50-50, the winner of the presidential election provides the tiebreaker.
We will elect all 435 Representatives to the U.S. House. The House is currently controlled by Republicans by a very narrow margin, 220-212. Speaker Mike Johnson would love to expand his majority, because having so few votes to spare makes everything a mess—but, at this point, he’d be lucky to keep it at all.
In the states, the Republicans are hoping and praying to break the Democratic trifecta in Minnesota. Democrats are taking aim at the Republican trifecta in New Hampshire.
There are a number of abortion referenda, which we covered yesterday in Election Night Preview: Abortion Edition.2
There’s lots more going on, including lots of interesting ballot referenda and a number of competitive state supreme court elections. For a comprehensive listing, and to find elections that you particularly care about, see Bolts Magazine’s 2024 General Elections Cheat Sheet.
Of special interest to many De Civ readers at that link: lots of ballot measures about ranked choice voting, redistricting, mail voting, voter ID, and other nuts-and-bolts democracy stuff!
Blah blah blah you know there’s an election and you know the stakes: if the wrong lizard gets in everyone you love will die, but if the right lizard gets in it’s ice cream for dinner for the next four years. People don’t read the De Civ election preview to find out what’s going on; people read this column every two years because they want to know who’s going to win!
Who’s Going To Win?
Josh Shapiro!
Cocaine Mitch, because he’s resigning after!
Whatever company makes Xanax!
Half of gamblers!
Man, I don’t know who’s gonna win!
Last election, the throughline of my preview was really simple: Vibes Are Red —> Vibes Are Wrong —> Trust the Data —> Embrace Uncertainty. This year, everybody who’s anybody has already embraced uncertainty. I don’t need to deliver my usual rant against The Vibes, because The Vibes are dead and all the demons are here. A few people on both sides are projecting confidence, but the pervasive feeling that one candidate was going to win—Harris in September, Trump in October—has dissipated.
Data provides no succor. Polls show a tied race. They’ve been close to tied since Harris entered the race. Then, after the Walz/Vance debate,3 they became truly tied. The past month, we’ve been living through the closest sustained tie in my memory, and (from what I hear) the memory of every other polling analyst I respect. Then, somehow, in this final week of the election, they’ve become even more tied.
I truly thought they were out of room to be any more tied. Then, during the very last editing pass on this article, Silver Bulletin published its final forecast for the 2024 presidential race. I have never, ever—not even in this already insanely tight race—seen numbers this close:
The last time the polling averages were very close on Election Day was 2004, but, even then, they weren’t anything close to this tied. Just before the election nightmare of Florida 2000, polls were not tied; George W. Bush led the final national average by 3 points. Final electoral college projections haven’t shown a count this close since 1972—and not because they actually were close in 1972, but because CNN’s Harry Enten hasn’t gone back further than that to check.
However, I think most readers misunderstand what a tied race means for Election Night. I’ve heard people expecting another long night where we don’t know the outcome until the early morning—or perhaps even later in the week. That’s a mistake! A long night is very possible, but not quite likely. This is because the polls are probably wrong.
Hitting the Bullseye (is Hard)
Polls are always wrong. When you try to find out what several million voters are going to do based on asking just a few hundred of them, the laws of mathematics tell you that—miraculously!—you are very likely to get an answer that is very close to correct… but you are almost certainly going to miss by a little bit, because a few hundred people are not going to exactly match a much larger group. This is called the margin of sampling error.
For most polls, the margin of error is about 3 points. If you’re measuring the distance between two candidates, you have to double that, since the margin of error affects each candidate. In other words, if a poll says “Tie,” but has a margin of sampling error of +/-3, the true result could be anything from Trump+6 to Harris+6. Worse still: even after you account for the margin-of-error, one in twenty polls will produce an “outlier,” which gives a result that is so wrong it falls outside the margin of error. The Washington Post’s infamous Biden+17 Wisconsin poll in 2020 is a good example of an outlier.4
You can reduce the size of the possible error by asking more people, but not by very much. You cannot reduce the likelihood of an outlier.5 This is a mathematical law governing perfectly conducted polls.
…and, hey guess what… no poll is perfectly conducted. Some are actually rather badly conducted, especially now that nobody answers calls from unknown numbers on their cellular phones. Polls (usually) get you in the vicinity of the correct answer, but only the vicinity. Averaging all your polls together helps. If one perfect poll misses a little bit to the left, another will miss to the right, and the midpoint between them will (probably) be close to the truth. If one pollster consistently hits a little further left than the rest, you can adjust for that, too. It’s like if you had a paper dartboard on the wall, then took it down: a home inspector could probably figure out roughly where the bullseye was just by looking at the holes the darts left in your wall—but, without a lot of very clear holes, he probably could not figure it out exactly.
As a result, even polling averages miss. In presidential elections since 1999, polls collectively have been biased by, on average, 2.4 points:
How It Might Go Down
In 2016 and 2020, large polling bias underestimated Republicans. But in 2012 and 2000, polls underestimated Democrats. You noticed in 2016 and 2020 because the polling errors turned both elections from cakewalks into nailbiters. You probably didn’t notice in 2012, though, since all it did was make Mitt Romney lose harder.
On the other hand, polls aren’t always wrong. They were within just about a point in the 2004 and 2008 presidentials, in both of the most recent midterm elections (2018 and 2022), and plenty of others. They could be right again!
Before you read on, you might want to turn on some appropriate background music.
Polls Are Right: Another Nailbiter
If the polls are correct (let’s call this a statistical bias of less than 1.5 points in either direction), then, yes, it’s going to be a long night. Trump leads by less than 1.5 points in Georgia, Nevada, North Carolina, and Pennsylvania. Harris leads by less than 1.5 points in Michigan, Wisconsin, and the national popular vote. That’s every battleground except Arizona. If polls are pretty much spot-on, anyone could therefore win any of them.
Polling bias is going to be different in each state, too. Even if the polls are pretty much spot-on overall, some states will be biased Trump+0.2 and others will have a Harris+3 bias (so don’t count out Arizona!). There’s a lot of random regional and demographic variation in polling error that you just can’t account for in advance. If the polls are right, we are just going to have to sit and watch each of the battlegrounds count. Some states might tilt one way or another and get a race call on Election Night. Others may stay too close to call until literally every vote has been counted—which means waiting for mail vote, absentee ballots, and even provisional ballots to get counted. This should take less time than it did in the 2020 fiasco… but we’re still talking days, not hours.
If this is what happens, it’s genuinely unknowable who would win the presidency, except by doing the hard work of actually counting every vote. Here’s a plausible version of the map on Wednesday morning:
In this scenario, Harris would still be favored to win the national popular vote, but only very slightly. Trump could easily win the national popular vote in a nailbiter race. In fact, the nailbiter is where we can get the weirdest outcome: if it’s this close, Trump could plausibly win the national popular vote while losing the electoral college. This is obviously the funniest possible outcome, as well as the most karmically satisfying, although, given Trump’s contempt for the rule of law, it is also probably the most dangerous.
Likewise, control of the House of Representatives will likely be unknown for days if not weeks, just like in 2022.
However, the Senate will likely go to Republicans in this scenario: they pick up seats in West Virginia, Montana, and mayyyyybe Ohio, plus mayyyybe one more of the battlegrounds6 and prrrrobably hold the seats they’re defending7 to land on 52 or maybe 53 Republican seats. Things could go south for them, but it’s pretty unlikely.
Abortion opponents probably win one or two ballot measures.
A nailbiter Election Night is about 40% likely.
Let’s put that in perspective: when an NFL team plays a third down, they have about a 40% chance of converting the third down into a first down. It happens often, but it does not happen most of the time. In other words, you wouldn’t be surprised if it happened, but it probably won’t.
Polls Missed Red: Trump Wins
As we’ve seen, the average polling miss is between 2 and 3 points. It wouldn’t be remotely surprising to see a normal miss in about this range. Call this a normal polling error.
In 2016 and 2020, polls thought Trump was losing, but they modestly missed Trump’s true support, and we ended up with very close races. However, in 2024, the polls already show Trump in a very close race. If they’re underestimating him again, then Trump just wins, and we probably know pretty early. Here’s the most likely map:
It was really easy to make this map: just give all the battlegrounds to Trump. This sounds simplistic but, if a polling error misses Trump again, this is the most likely outcome by far, although it is possible that Harris lucks into holding one or even two of the battlegrounds by happenstance. Won’t save her. Trump wins.
We would have our first inkling early, when Florida, North Carolina, and Georgia were all called surprisingly quickly for Trump. The win would be all-but-confirmed by an early race call in Wisconsin (or possibly Pennsylvania), probably around 10:00 or 11:00 P.M. Central Time (11:00/Midnight Eastern). Trump would still be a few electoral votes short at that point, but the final result would be a foregone conclusion, and we’d just be waiting for West Coast counts to come in to confirm the diagnosis.
Republicans probably win the House in this scenario, although error size matters here. Democrats have a good gerrymander advantage right now and can very possibly hold the House even if Republicans win the national House popular vote by up to 2 points.8 Look no further than the 2022 midterms for an example: Republicans won the popular vote by nearly 3 points, yet they barely won control after a weeks-long battle.
A normal polling error in this direction helps the Republican Senate map, too, though maybe not as much as you expect. 52 seats becomes a firm floor, and 55 seats becomes a reasonable thing for Republicans to hope for with a lot of luck. However, Democratic candidates are far enough ahead in many of their Senate races that even a normal polling error doesn’t make most of them easy for Republicans.
If the Republican majority is 52 seats or less. The Senate filibuster is in absolutely no danger. In my judgment, the Republicans do not pose a serious threat to the filibuster unless the Republican Senate majority reaches 55 seats or more under a Republican president.9
With a normal polling error favoring Team Red, abortion opponents likely win 2 or 3 ballot measures, and could even luck into a surprise 4th win.
A normal polling error favoring Team Red is about 25% likely.
Let’s put that in perspective: when a Major League Baseball player steps up to the plate, he has about a 25% chance of getting a hit. (League-wide batting average was .248 last year.) It’s pretty unlikely… yet it still happens, on average, 16 times in a 9-inning game! When your guy hits a single down the right-field line, you cheer, but you don’t faint from the sheer shock of it.
Neither should you be shocked by a normal polling error giving Team Red a strong night.
Of course, Team Red has had strong nights (relative to polling) in three of the past five elections.10 I don’t think anyone’s shocked when this happens anymore.
Know what might be shocking?
Polls Missed Blue: Harris Wins
You should not assume that polls will err in Trump’s favor for a third time running. Polling error is pretty random to begin with, and, as we’ve seen, it goes both ways. The specific reason pollsters missed Trump so hard in 2016 seems to have been that they weren’t weighting their samples by education level, because education level wasn’t electorally important until the 2016 realignment. They missed Trump again in 2020, and actually missed worse, but there’s good arguments that the reason they missed had a lot to do with the craziness of the pandemic. Pollsters have fixed that, and, for the most part, they ended up having a really strong 2022 midterm. Even though the Vibes said “red wave,” the polls were always very skeptical of that, and landed very close to the mark on the day. If the polls are fixed, they might still miss, but they could miss in either direction.
That’s not all.
Pollsters, terrified of missing Trump three times in a row (which might discredit their profession forever in the eyes of the public), may have overcorrected. They might miss a strong night for Team Blue this time just because they’re watching too closely now for a strong night for Team Red. Pollsters, by and large, truly don’t care who wins the election; they care far, far more about their own reputations, because their reputation for accuracy is how they get business, which is what keeps their doors open. After missing Trump twice, that gives pollsters perverse incentives to avoid missing Trump at all costs—even if that means increasing the odds of missing Harris.
It therefore would not be remotely surprising to see a normal polling error: a 2-3 point miss—in Harris’s favor. Here’s the map:
It’s very possible that Trump holds on to some combination of Arizona and Georgia in this scenario and perhaps even one of the other battlegrounds, but it doesn’t save him. Harris wins.
The first sign of this would be weak results for Trump in Florida, where he ought to win by about 6 points. The state might still be called for him, but the weakness would likely be obvious by 8:00 Central (9 Eastern), with an overall result of Trump+3 or Trump+4. This would be followed by a call in North Carolina for Harris some time around, probably, 9:00 or 10:00 Central, depending how big the error is. (These timing estimates are very guessy.) Winning North Carolina does not quite seal the deal for Harris, but Trump’s road back from losing North Carolina is nearly impossible. A race call in Wisconsin or Michigan plus Georgia would officially change the night from a race to a pre-mortem. We might not see every plausible Trump path officially taken off the table until past midnight Central, and the AP call could take even longer, but the wind would be out of Team Trump’s sails quite early in the evening.
With a normal polling error at their back, the Democrats would also very very likely win control of the House of Representatives. The GOP is not as well gerrymandered at the moment. They shoulda taken my recommendation to abolish gerrymanders when they had the chance! Now they face a real risk of a few years under minoritarian rule.
In this scenario, Senate control is a real contest, but it favors the Democrats. The GOP probably still wins the race in Montana to land on 51 seats. The question, however, becomes whether they can still defend their seats in Florida, Texas, and Nebraska if there’s a normal polling error favoring the Democrats. They’d still be favored in all three, but not by very much—and they would have to hold all three, which is tough. If the Democrats could pick off any one of them and/or hold Montana, they’d retain the Senate. (If the GOP is unlucky, Dems could pick off two for an even larger majority.)
Democrats are likely to end the legislative filibuster if they gain control of the presidency and retain the Senate. The only members of their caucus who publicly oppose ending the filibuster are Sens. Manchin and Sinema… and they are both retiring this cycle.
With a normal polling error favoring Team Blue, abortion opponents likely win zero ballot measures.
A normal polling error favoring Team Blue is about 25% likely.
Once again, this is about the same chance an MLB player has of getting a hit during an at-bat. It’s pretty unlikely… yet it still happens routinely and is in no way surprising.
Historic Miss: Bloodbath
Polls could also miss by 4 points or more. This is rare, but hardly unprecedented: it happened in the last presidential election! Pollsters are grappling with rock-bottom response rates, complicated internal modeling (to deal with the response rates), and mortal fear of slipping on the pollster ratings chart. Most have been weighting their polls to the results of the 2020 election, which means they will miss any major changes caused by newly-registered voters.
It’s possible (though very unlikely) that all this will add up to a very big miss.
If a major polling error happens and it favors Trump, New Hampshire, Minnesota, New Mexico, and Virginia are in play. The House is theirs. The Senate is 55-57 Republican seats. I really don’t know what it would mean for abortion amendments, but it would be good for the Fetal Rights Gang.
If a major polling error happens and favors Harris, Florida, Maine’s Second District, Texas, Ohio, and Iowa are all in play. The House is theirs. The Senate ends up at 46-47 Republican seats, as Democrats expand their current majority. Obviously all abortion amendments fail.
Oddly, because the House is already so close (and so gerrymandered), even a huge tide like this probably does not deliver a “wave” election (defined as an election where the House shifts by 25 seats or more). However, this is the only scenario where we might see a wave.
This is unlikely. Despite 2020, polls don’t miss by this much very often. I put the chances at 10%. That’s a 5% chance of a Republican bloodbath and a 5% chance of a Democratic bloodbath.
Put this in perspective: when an MLB player steps up to the plate for an at-bat, he has about a 5% chance of hitting a home run.11 This is a rare event… but not a stunning one.
Put this all together, and it turns out that we have about a 40% chance of another nailbiter… but about a 60% chance of a fairly clear outcome fairly quickly. 60/40 is just within tossup territory, so let me put it this way: it’s a tossup whether this tossup election turns out to be a tossup.
A Brief Note on Senatepocalypse
Those of us who fear the consequences of losing the Senate filibuster will want to take a moment to assess the odds of, specifically, the filibuster being eliminated.
On the Republican side, the filibuster is in serious peril only if they perform exceptionally well. It’s likely endangered in the Bloodbath scenario (which happens 5% of the time), as Republicans would have enough seats to ignore the pro-filibuster moderates in their caucus. The filibuster might be in danger if the Republicans get a Normal Polling Error (which has a 25% chance of happening), but only at the very high end of Republican performance—perhaps in one-tenth of Normal Polling Error situations (or a 2.5% chance overall). Add that up and we find that Republicans have perhaps a 7% or 8% of ending the filibuster after this election.
On the Democratic side, because the Democratic Senate caucus has already committed to ending the filibuster, the filibuster is in serious peril in every scenario where they win the Senate.
In the bloodbath scenario (which happens 5% of the time), Democrats win the Senate. No need for further analysis.
Democrats will probably win the Senate in the Normal Polling Error scenario (which happens 25% of the time), but there are some situations where Republicans could scrape by with exactly 51 seats—perhaps in two-fifths of Normal Polling Error situations. The Democrats win the Senate the other three-fifths of the time in a Normal Polling Error scenario. 25% times 60% = 15% of the time overall, Democrats win the Senate thanks to a normal polling error in their favor.
Finally, the Democrats still have a chance at winning the Senate in a nailbiter night. It would require some very good luck, but the Democrats found good luck in spades in 2022 midterms. Moreover, they won’t end the filibuster unless they win both the Senate and the presidency. In a nailbiter, that isn’t likely. Democrats might come out of a nailbiter (which happens 40% of the time) with the presidency and the Senate one-eighth of the time, if they’re lucky. That works out to 5% of all elections.
Add this up and the Democrats have around a 30% chance of ending the filibuster. As we’ve seen, things that happen 30% of the time are not uncommon.
Combined, then, the filibuster has maybe a 35% chance of biting it (at the hands of one party or the other) in this election. That’s better than one-in-three. For filibuster-likers, this is a real risk.
Signs and Portents
This is as far as hard data built into reliable analysis of polling data, fundamentals, and demographics can carry us. The race looks like the closest presidential election in decades, maybe even in history.12 However, there is, as always, a lot more uncertainty than most people realize, and way more ways it could play out than the “Rust Belt / Sun Belt” map burned into most people’s heads.13 If it tilts either way—which it very well might—it could be over fast.
This is not very satisfying. Every cycle, after I paint the full picture of plausible outcomes, I then take a dangerous look into the sheep’s entrails to try to scry where exactly things are headed. In 2022, with a little wisdom but a lot of luck, I correctly discerned the red ripple that was coming, and I predicted the exact number of House seats the GOP ultimately won. I almost certainly won’t be that close this year, but I might (might!) be able to give you some directional hints.
In 2022, I looked at three sources of information:
Jon Ralston’s analysis of the Nevada early vote. Ralston was and remains the only early-vote analysis with proven predictive value.
Apparent outlier polls. As I explained then, pollsters, who care more about their reputations than anything, tend to stop publishing polls that look like outliers in the final weeks of an election. They stick close to whatever the current average is. This is called “herding.” It seems very clear that there is a lot of herding going on in this election. The few pollsters brave enough to publish polls that look like outliers sometimes turn out to be picking up on real movement in the race—movement the herders also saw, but were too scared to report. On the other hand, most polls that look like outliers turn out to just be outliers!
The Vibes. Specifically, I argued, the Vibes are wrong more often than they are right. Worse, the Vibes tend to corrupt polling averages. Therefore, you should tend to bet against the Vibes.
In 2022, these indicators all pointed the same direction: Red Ripple. In 2024, however, they are all over the place.
Dead Vibes: Null
The Vibes are null. They were Trumpy vibes right up until last week, when they suddenly dropped dead. The best quantitative measurement we have of the Vibes is betting markets, and they are now all tied up after leaning Trump last week: PredictIt is 53-53, Kalshi was exactly 50-50 last night (now 54-46 Trump, no clue why), and Polymarket is 60-40 Trump, but largely on the strength of a single insane gambler. My personal sense of the Vibes is that they now point slightly toward Harris, but not strongly enough to bet against them. Also, this final Vibe Shift came late enough in the campaign that it probably can’t corrupt the polls.14 We can extract no signal from The Vibes.
Ralston’s Early Vote: Null
A very strong opening from Republicans, who started voting early in massive numbers in Nevada this year, has scrambled the warp plots of everybody in Nevada politics, including the oracular Ralston. If Republicans are mostly cannibalizing their Election Day vote, they won’t hold their lead as bluer Democratic mail and Election Day voting comes in. However, if their E-Day vote holds up on the strength of new voters, they will hold their lead.
Since this has never happened before, not even Ralston knows what to expect. In the end, Ralston puts his faith in the famous Reid Machine (named for the late Sen. Harry Reid’s remarkably effective get-out-the-vote operation in Nevada) to get Harris just over the line. However, for the first time ever,15 Ralston has been forced to add decimal points to his prediction, because he thinks Harris is going to win by just 0.3 points. Whatever the margin of error is on a Ralston prediction, it’s a lot higher than a third of a point.
Quite frankly, since Ralston is (as he admits) a strong Democratic partisan, I would discount this a little bit. My (weak) suspicion is that Ralston is banking a little too much on unrealized hope, not data, and he’s the first to admit that’s possible.
Still, even if you take this Ralston Report at face value, what he’s reporting is a dead heat, not a Harris lead. We can extract no signal from the early vote.
(Bonus: Nevada allows mail-ins to arrive through Friday, so we can expect a long drawn-out counting process and major doubts of the election’s legitimacy if Nevada is decisive.)
(Apparent) Outlier Polls: Leaning Republican?
How about polls that look like outliers?
As of Friday, no single clear signal was leaping out at me, so I starting putting together sort of a silly rough way to see if there was a signal.
I went through the Silver Bulletin polls for the competitive states and picked out all the polls that weren’t herding.16 These are the polling firms that are courageous. Not all of them are truly reporting outliers, though. Some polling firms simply have strong house effects. (That is, their “aim” slants consistently a little bit to the left or to the right, often due to internal sampling decisions.) Once you adjust for these house effects, some of these (unquestionably courageous) polls turn out to agree with the herd after all.
At that point, what you’re left with is a collection of true apparent outliers—which could be picking up on a genuine signal of movement that the herders are missing. Are we seeing anything like that in 2024?
There might be some signal here:
If you just count this up, which is a stupid way to analyze this, there are 14 outlier polls pointing in the GOP direction and 6 pointing in the DEM direction. That could be a signal of real GOP strength… or it could just be a random artifact of the specific ways I chose to slice up this data. For example, two very good NYT/Siena polls favoring Democrats just missed the cut on this chart after I made the house effect adjustment.
I do think there might be something to the idea that Republicans are being underestimated in the Sun Belt (GA, NV, NC, AZ), where all the outliers favor them. That looks like signal, and it lines up nicely with some NYT/Siena polls from last month that also showed Republican strength in the Sun Belt (but weakness in the Rust Belt).
Last week, with this data in hand, I was thinking that the race was tilting Republican. I was waiting only for J. Ann Selzer’s final Iowa poll to confirm my suspicions.
The Selzer Poll: Leaning Democratic
The Selzer poll is perhaps America’s most famous outlier, because Selzer has a solid track record of publishing apparent outliers that turn out to be exactly right. When pollsters thought Obama was in a close race for the 2008 Iowa caucuses, Selzer found Obama’s imminent decisive win. When pollsters thought Romney was still close, Selzer found Obama winning Iowa by 5, everyone (including me), shouted “outlier,” and Obama won by 6. When pollsters thought Clinton was competing for Iowa, Selzer showed Trump winning by 7. Most cried “outlier,” but Trump won by 9. When pollsters thought Biden was way ahead nationally, Selzer reported that Trump was still winning Iowa by 7, and he won by 8.2.
This is a miraculous track record. You have to go all the way back to 2004 for a Selzer “outlier” that turned out to be actually wrong. (She reported Kerry winning Iowa by 5. Bush won by less than a point.)
So if you’re doing outlier analysis, you want Selzer’s take before you publish. If my theory of Republican overperformance was correct, I expected Selzer to show Trump winning Iowa by 9 to 12 points. If it was incorrect, I expected Selzer to show a Trump lead in the 5 to 8 range. If there were actually a normal polling error favoring Harris nationally, I expected Selzer to show Trump winning Iowa by 1 to 4 points.
The final Selzer poll showed Harris winning Iowa by 3.
If that is actually true—if Harris wins Iowa by 3 or more points—the election is almost certainly over. Iowa isn’t decisive, but Iowa voters look and think a whole lot like Wisconsin and (to a lesser extent) Michigan and Pennsylvania voters.17 Those states are decisive—and they are almost certainly much bluer than Iowa. Moreover, for Harris to lead in Iowa implies such a massive national polling miss that you have to assume the battlegrounds are all long gone, that Harris has won Florida, and that she may be about to add Ohio and Texas to her pile.
This actually aligns with a pattern I was noticing but discounting in my focus on swing-state outliers: Harris has been performing pretty well in polls of non-Rust Belt swing states. In Kansas, where Trump “should” win by 15 based on 2020, a recent poll found him up only 5 points with likely voters. In Ohio, where Trump won by 8 in 2020, a string of good pollsters found him up only 6 (although he’s also had some stronger Ohio polls since). In Arkansas,18 Trump “should” lead by 28, but actually led the last poll by 15. Democrats have also performed well in largely unnoticed polls about competitive House districts. It could be that pollsters are herding in the high-visibility swing states, but are more relaxed about releasing “outlier-looking” results in states nobody’s paying attention to—which means these strong non-swing state polls could be picking up on real Harris strength that the herded swing-state polls are missing.
I wouldn’t overread this. Harris, too, has had a couple of surprisingly weak polls in Minnesota and New Mexico (another Southern state—interesting). All candidates get weak polls somewhere, sometimes. Meanwhile, Trump is holding up fine in Nebraska and Indiana. However, this shows that Selzer’s poll, while extreme, isn’t totally without corroboration.
Sean Trende has a good, balanced take on how we should interpret the Selzer result, which goes into much more detail.
What do I get out of all of this? To a first approximation, nothing. Looking at outlier polls is a very dicey proposition to begin with. We have evidence on both sides about the national environment, but nothing decisive. We should treat this like we’re treating everything else: null. No signal.
A Very Speculative Synthesis
That was a heck of a lot of work to get to three NULLs, but three NULLs is still useful. Since the signals are absent or mixed, that might be a signal that the herd is right, that the polls are accurate, and that this election really is going to be an absolute nailbiter!
Also, now that we’ve done all that work, let me tell you something that’s whispering at my heart, even though I don’t think you should put any weight on it:
The swing-state outlier polls mostly suggested surprising Republican strength in the Sun Belt, where there’s a lot of racial diversity. This matches poll data showing Republicans gaining with Blacks and (especially) Latinos this cycle. That’s also something we might be seeing in the Nevada early vote data, where Republicans leapt out to a large and unprecedented early lead—even though Jon Ralston doubts it will hold up in the end.
The Selzer poll, on the other hand, suggests Democratic strength among Whites. (Iowa is super-White, and the crosstabs on her poll suggest especially strong White support.) The Rust Belt is full of White people.
These different pieces of data could, maybe, be painting a picture where Republicans overperform in the South and Democrats overperform in the North. This is consistent with the mid-October NYT/Siena polls. It also fits with what we see in the other outlier polls, which are much bluer in the Rust Belt than in the Sun Belt.
If that happens, Harris really very probably wins the electoral college 270-268, unless Trump (miraculously) comes back to win New Mexico or Hawaii or Nebraska’s 2nd District and (barely) holds on to Iowa.19 Here’s a map:
Either candidate could easily win the popular vote in this scenario, but Harris would win the White House.
If you held a gun to my head and demanded I predict the election, I would get very worried about whether telling you constituted cooperation in evil, and you’d probably eventually get annoyed and just shoot me.20 However, if you asked politely for my best prediction, I would tell you that the modal outcome is this map, with Harris winning 270-268. The modal outcome is not very likely (less than 4% likely, per Silver Bulletin’s “128 paths” feature), but, after doing all the haruspicy I can, I believe this map is more likely than any other specific map.
It’s a very, very close race.
Maybe.
What I’m Watching Election Night
Now that we’ve looked at every possibility from every angle,21 it’s time to talk about how to actually track the election tomorrow night.
Stop me if you’ve heard this one before:
The day of the Brexit vote, the very first constituency to report results, Newcastle-upon-Tyne, reported a defeat for Brexit, 51% Leave to 49% Remain. The defeat for Brexit there was expected. However, the margin was wrong: experts had expected Brexit to fail in Newcastle by 12 points, not 2 points. Newcastle had been a “safe state” for Brexit opponents that turned into a very narrow win. I was over at my parents’ for dinner, just buzzing by the computer for a quick results check, didn’t even bother sitting down—and I remember seeing that result, sitting down heavily, and thinking, “Welp. The U.K. just left the E.U.”
The rest of the night went the same. Areas that were expected to vote for Brexit by a narrow margin voted for it by a large margin; areas that were expected to oppose it by narrow margins ended up supporting it by narrow margins. There were a few places where the anti-Brexit “Bremain” vote did better than expected, as there always are… but not many. Networks finally called the race for Brexit hours after the result was clear.
This happened again in the 2016 presidential election. I told my family at around 8:15 PM (Central Time) that Trump was probably going to be the President-Elect, because I could see that all the states Trump needed were getting called for him (or were on the cusp of it), while all the states Clinton needed were tossups… and leaning red.
Once again, in 2020, I was closely tracking a random House race in suburban Kentucky and realized from looking at that one race that, holy cow, the polls actually had missed that year, and missed big. I buckled in for a long, long night and finally went to bed after I saw the Milwaukee results put Biden over the top in Wisconsin.
In 2022, I did more of the same.
Here are my secrets: Good Data and Top Men.
Good Data
You can often tell who’s doing well based on early race calls. If North Carolina is called quickly for Trump, that’s a bad sign for Harris. If it’s called for Harris, it’s a much worse sign for Trump. If Minnesota is too-close-to-call for a long time after Georgia has been called for Trump, hang it up for Harris.
Most network decision desks are very, very cautious about race calls. They all remember the 2000 election, when they called Florida for Gore, then uncalled it, then called it for Bush, then uncalled it again, then left it uncalled for a month. After that fiasco, network decision desks were banned from calling races until their certainty went above (I believe) 99.5%. But most of us can tell, with very high confidence, where a race is heading a long time before that.
If you’re comfortable with race callers who are willing to take a few more risks in the interest of speed, follow Dave Wasserman (@Redistrict on Twitter) or Decision Desk HQ, among others. (UPDATE: Tragically, I am told that Wasserman has been snapped up by the NBC decision desk this cycle, so he will be unavailable.) The Fox News decision desk is also fairly aggressive about race calls—in fact, probably a little too aggressive. Their 2020 call of Arizona ended up correct, but was very difficult to defend based on the data they had. Still, if you want early signs, you want aggressive predictions.
A second method for figuring out where the race is heading is the classic “benchmark” method. As voting districts turn in their votes, compare the final tally to the polling average (or, less accurately, the previous election result). If the actual results are consistently bluer than the benchmarks, that’s evidence of a polling error missing the Democrats. You can do this by state, by congressional district, by county, or by precinct, and I’ve explained this enough times before that I won’t bother doing it again. By state is the easiest but slowest method; by precinct is the hardest but fastest method.
The New York Times’s famous Needle basically does this benchmarking in real time, which is amazing, but it is backed by a data engineering team of hundreds… and that data engineering team just went on strike today. It is therefore very unclear whether we will have a Needle tomorrow night. Instead, you can do it the old-fashioned way: with a spreadsheet and paying close attention to reported vote totals in districts with >95% of the vote in. (NEVER try to benchmark voting districts with less than 95% of the vote in. Early/absentee vote will mess you up.) I was working on a messy spreadsheet to help you do just that, but Adam Carlson (@admcrlsn) put up quite a good spreadsheet today, with benchmark counties in Georgia, Kentucky, and Indiana—quick-counting states all. (I would have liked to see Florida here, the quickest state of all, but you can’t have everything unless you’re willing to do it yourself.) Get it here:
2024 Election Night Tracking Spreadsheet (Adam Carlson): https://docs.google.com/spreadsheets/d/1zGy3bfDy7GgGvEmuctExLhhYqVt8bEuMagpXEldQf6k/edit?usp=sharing
(The interesting stuff isn’t in the first tab, but in the bellwether and benchmark tabs.) You can make a copy of this, stick it on your hard drive, and fill it out yourself. Or you can probably just watch as Carlson fills it out himself over the course of the night.
Top Men
At this point, I’ve off-loaded a good portion of my brain to Elections Twitter. I can scroll my Twitter list of “ElectionData” for hours, and those people collectively think much better, faster, and clearer than I do (although they lean very left). I will certainly spend much of Election Night following this list, and I suggest you do, too:
ElectionData Twitter List, by James Heaney (it me)
https://twitter.com/i/lists/1590081544248516609
I value Nate Silver above all others, but Silver Bulletin is doing live chat instead of a live blog tomorrow, and I can’t navigate Substack’s live chat at all. I’ll probably skip this.
My default channel, then, is likely CNN, where Harry Enten is a very good elections analyst. I might also have Fox on, for their aggressive decision desk—and for a little access to conservative thinking in the overwhelmingly progressive world of polling nerds.
I might tweet (@BCSWowbagger), but I doubt I will have much to say that others won’t already be saying better. Also, I will be late to the party: I am election-judging until polls close this year, and will therefore not be home until probably 8:45 or 9:00 Central. Also, for unrelated reasons, I have promised to read my youngest her bedtime story for an extra-long time. By the time I finally log into my command center, it might already be over, or trending strongly one way.
Hopefully, though, you are now armed with the tools to see that coming, even without me.
Whoever wins, I hereby commit to re-watching Gods of Egypt when the electoral college votes. Last time, I watched it, but I never told you how it was.22 Now I do movie reviews, so, I don’t know, maybe something just a little bit good can come out of the moral suicide we are about to attempt. Probably not.
Good luck to both teams, let’s hope nobody tries to steal the election this time, remember to stay sane by reading something funny tomorrow, and may God continue to bless America… restore His blessings to America… have mercy on our souls!
Update 5 November 2024 1:42 AM: This originally said “33” senators. That is the regular order of things. However, my friend Daniel Pareja (you know him from “Give the Rich More Money!”) reminded me in a personal message that it’s actually 34, because Ben Sasse’s former seat in Nebraska is up for a special election to confirm Pete Ricketts in the seat until the end of the original term (in 2026). Ricketts leads by 18 points, unlike his Nebraska Republican colleague Deb Fischer. Daniel further adds:
Also, California is voting twice, once to fill the seat vacated by the death of Dianne Feinstein and retirement of Laphonza Butler until the next Congress is sworn in come January 2025, and also to elect a Senator to serve for six years from 2025 to 2030. Both races have Adam Schiff and Steve Garvey as candidates. This sort of thing has happened before.
While I’m here, I may as well add that Daniel also reminded me that it’s not just the 435 voting members of the House who are standing for election:
Even more minor fact check: There are also the six non-voting members of the House from the various territories of the United States (who do nonetheless vote in committee, just not on the floor), all of whom are facing re-election this year. (Puerto Rico's Resident Commissioner serves a four-year term.) Currently three of these caucus with the Democratic Party (from the District of Columbia, the United States Virgin Islands, and the Northern Mariana Islands) and three with the Republican Party (from American Samoa, Guam, and Puerto Rico).
Was it really just yesterday? Man, long day.
There’s another winner from this election: whether Trump wins or loses, Vance has, with one debate, given himself an enviable chance at the 2028 Republican nomination.
As we will see later on, it is right and just that WaPo published that outlier. It is very important that pollsters don’t bury their apparent outliers, even though they usually do turn out to be outliers. That kind of thing is why WaPo remains a widely-respected and usually pretty accurate pollster.
You can get a higher confidence level in your margin of error, but this makes your poll even less useful.
For example, a poll might have 95% confidence—a 1-in-20 chance of producing an outlier—with a margin of error of +/-3. If you adjust that to 99% confidence, you’ll reduce the odds of an outlier to 1-in-100… but your margin of error will get much larger (around +/-5%).
Arizona, Nevada, Pennsylvania, Michigan, and Wisconsin all feature contested Senate races where Democrats have clear but surmountable polling leads.
Republicans have clear but surmountable polling leads in Senate elections in Nebraska, Texas, and Florida, where they are defending their seats.
Now that Nate Silver has moved to Silver Bulletin, I don’t really trust the 538 House model this year. However, since Silver Bulletin isn’t doing a House model, I must look at the graphs I can look at.
This chart shows how easily Dems can win control after a Republican popular vote win:
Collins, Murkowski, and McConnell are all pretty much guaranteed to sustain it, and I will bet that John Thune and Bill Cassidy would, too.
2014, 2016, and 2020. Yes, I am counting midterms.
A bit less than that. Around 3.5%, to be precise. This is based on 5,868 home runs in the 2023 season after 164,418 at-bats.
That “maybe” is doing a lot of work. 1876 was pretty dang close, but, of course, we didn’t have polls back then to tell us whether it was going to be close.
I know I am grossly overestimating how many people have any particular electoral maps burned into their heads. XKCD 2501 strikes again:
Alt text: “How could anyone consider themselves a well-rounded adult without a basic understanding of silicate geochemistry? Silicates are everywhere! It's hard to throw a rock without throwing one!”
Does that mean we should still be betting against last week’s Trumpy Vibe, since maybe that did corrupt the polls? Maybe, but the signal is not strong enough for me to bank anything on it.
As far as I’m aware, so going back to 2016.
I defined “not herding” as the pollster reporting topline results that were at least 3 points outside the polling consensus. (I used today’s polling consensus, November 4, for the published version of the spreadsheet.)
Also, I used only polls with a weight of 0.5 or higher, and, if a firm had multiple outlier polls in the same state, I used only the latest.
Silver Bulletin says Iowa’s movement is 84% correlated with Wisconsin, 70% with Michigan, 69% (nice) with Pennsylvania, and 57% with North Carolina.
Admittedly, Arkansas is more Southern than Rust Belt.
I just saw this while doing final edits on the article.
Surprisingly, Silver Bulletin says right now (in its 128 Paths to Victory chart), that Trump actually has a 17% chance of pulling out a win if he wins the whole Sun Belt and Harris the entire Rust Belt. I don’t know which specific states Silver Bulletin has in mind as possible targets. New Mexico would clearly be very difficult, even after Trump runs the table in all surrounding states. Perhaps the Silver Bulletin model thinks Trump could still pick off New Hampshire?
In “The polls are close, but that doesn’t mean the results will be,” the team also considers a situation where there’s a pro-Trump polling error of 3.1 points in the South and West, cancelled out by a pro-Harris polling error of 3.1 points in the Midwest and Northeast. To my eyes, that looks like a near-guaranteed Harris win, 270-268, but Silver Bulletin gives Trump a 27% chance at a comeback in that scenario. Again, I’m not sure what they have in mind here. I trust Silver and McKown-Dawson, but I’m having a hard time envisioning this map. Perhaps they simply think Trump still has at least an outside chance at picking off one of the Blue Wall states even with a polling error favoring Harris in that region.
I’m pretty sure it doesn’t, but I’m trolling here.
This is actually not even close to true. We could go for another 20,000 words about plausible outcomes. But this is the best overview we can have in 6,000 words!
No, the throwaway line in “It’s Okay To Feel Okay” doesn’t count.
A fitting song from a classic 80s film for the evil mobsters that will be running the Executive Branch come January (the Presidential Election is basically a choice between an evil MAGA mobster and an evil limousine liberal mobster, IMO):
https://youtu.be/Olgn9sXNdl0?feature=shared