Election Forecast, 2022
State-by-state and overall
Welcome to my substack. It seems appropriate that my first post here is an election forecast. I will be posting about, mostly, political and sports data. If those things are of interest to you, then push this little button here.
Time limitations being what they are, I haven’t had as much time to do House-level races like I did in 2020 (sorry) but I feel pretty confident in my Senate numbers. Without going on too much of a rant, what I’m seeing in the Senate is: forecasters (you know the ones) still misunderstanding the importance of poll margin vs. poll average. Just because an election is “close” (e.g. 49-46, “+3”) doesn’t mean the chance of the “underdog” winning is as good as if the poll average were 46-43, “+3”.
If you follow me on Twitter, chances are you understand why. If not, I’ll definitely be talking about it elsewhere. Maybe even in a book.
As for the forecast implications, does their wrongness (not understanding how polls work) have a bias? Well, surprisingly, no. But it does make a difference in the tails - that is, the chances either party sees a large gain (+3 seats or better) in the Senate.
Because of their incorrect definition of poll error, they tend to underestimate the favorite’s chances when that person’s poll average is very high. Combined with some more complicated stats such as state-to-state correlation, this leads them to overestimate the likelihood of unlikely scenarios.
To put it simply, if you’re modeling win probability from a poll average of 51-39, and also from one of 51-45, while you might conclude the candidate with the larger “poll margin” is “more favored” it’s not as big of a difference as you might think. Both are over 99% historically. In fact, the difference in win probability of 49-46 and 47-43, though it feels small, is much greater. Why?
In order for the candidate polling at 51% to lose, they have to underperform their poll average by well over 1% (source). Just because it hasn’t happened this century doesn’t mean it can’t happen, but for a site that publishes junk historical averages to find meaningless correlations to not take note of this very basic (and statistically valid) data point is problematic. Here’s an old chart from my 2020 analysis that illustrates my point.
Candidates almost never underperform their poll number, especially not by more than 1%, and candidates polling at or near 50 win. The good people at 538 said Trump was 11% to win New Hampshire and 10% to win in Maine. No.
When a candidate is polling at or near 50%, what you should be factoring in (for purposes of win probability) is the probability that candidate underperforms their poll number. It doesn’t seem they do that, at least not much.
Long story short
There are a few races where I think 538 (and their, unfortunately, herding counterparts) have overestimated Democrat’s chances, and a few where they are underestimating them.
Here are my state-by-state forecasts, as well as projected outcome and range of outcomes. Just for fun I have projected what I think the “finish line” will be (i.e. 50% vote share) as well.
I’ve ordered them here by least to most competitive.
Florida
Dems were always a longshot here. I had Demings around 12% at one point but the polls have closed strong for Rs. Rubio is polling consistently over 50, with Desantis even stronger. There’s no silver lining to be had here, but if Dems get this race within 6 or 7, it could signal a high turnout, good night.
My Forecast: Rubio 98% (54.5-44.5)
Finish Line (proj): 4.9M votes
538 Forecast: Rubio 95% (53.8-44.3)
The difference: 95% and 98% seem close enough, but it’s 1/20 vs 1/50. No candidate with a polling average at or over 50% has lost this century, probably longer.
Washington
More undecideds in this election than some others, wouldn’t be surprised at a Dem blowout or D+7. But somewhat like Florida, if it’s in the +7 range it might signal a good night for Rs. But considering the time difference from WA to FL, if it’s going to be a good night for Rs, we’ll probably already know.
My Forecast: Murray 94% (54-46)
Finish Line (proj): 1.7M votes
538 Forecast: Murray 91% (54.3-45.7)
Colorado
The relative lack of polling here makes things somewhat uncertain, but what we do have shows Bennet at or over the finish line. The only reason my forecast isn’t more aggressive (i.e. FL) is the relative lack of polling.
My Forecast: Bennet 96% (53-46)
Finish Line (proj): 1.45M votes
538 Forecast: Bennet 92% (52.6-44.3)
North Carolina
This one is disappointing for me because I had it as a tossup early. I thought it’d be close. Through September the high-undecided Georgia-2020 energy seemed to signal well for Dems. There were other factors that made me list NC my top “sleeper” pickup for Ds. But polls solidified for Rs and this seat went from stretch pickup to minor miracle pickup.
My Forecast: Budd 88% (52-46)
Finish Line (proj): 2.4M votes
538 Forecast: Budd 80% (51.4-46.8)
The difference: Budd’s poll average is at or around 49. It would be really, really hard for Beasley to overcome that. 12% is basically the chances of guessing the roll of a dice, so it’s not impossible. But if you’re hoping for a good D night, you have to consider this size of upset is also possible in places like New Hampshire (more on them soon).
Wisconsin
Here’s the first example of a race that’s not as close as it looks if you only look at the margin. Yes, Dems can get lucky here (see my post about dice rolls above) but Johnson is consistently over 49 - hard to beat.
My Forecast: Johnson 85% (51.5-48)
Finish Line (proj): 1.5M votes
538 Forecast: Johnson 78% (52.1-47.9)
Ohio
A few late polls moved me from Ryan ~1/3 to Ryan ~1/4. Undecideds seem to be coming home for Rs and Ryan’s “Trump-sized” chance (high undecided persuadables) shrunk a bit. But if Ds have a REALLY GOOD night (opposed to a medium good night) OH will be the first indicator. Early vote is counted here first so beware the blue mirage.
Bonus content because OH is my home state and I live in a gerrymandered hellscape.
Keep an eye on OH-1 (Cincinnati suburbs), and OH-13 (Akron & suburbs), and if Ryan can “run up” his vote totals in the big cities. Ryan’s path to victory relies on that, plus flipping (or making competitive) the Obama-Trump districts in Northeast Ohio, which happen to be his home territory. The relative weakness of Nan Whaley as a governor candidate might prove helpful if she can pull Montgomery County, where Dayton is located and where she was Mayor, which is one of (and by far the largest) of Ohio’s EIGHT Obama-Trump counties.
It might sound like a tall task - and it’s not a small one - remember that House races are horribly under-polled and I don’t think 538’s “uncertainty” is high enough in what I believe will be competitive races.
My Forecast: Vance 77% (51.5-48)
Finish Line (proj): 2.5M votes
538 Forecast: Vance 85% (52.9-47.1)
The difference: I think Ryan does well enough in a few important places to make things interesting. If Ds get 52, Ohio might be the 52nd.
New Hampshire
If you know me, you know I’m not shy about my partisan leanings and political preferences. However, I’m very up front about my bias because I try not to let my bias influence my work.
When I first released my forecasts, I was way to their left, then falling victim to hype and small samples, they moved way to my left.
I’m to the right of 538 on both NC and WI.
I’m to their left on NH and a few others. This may be the first appearance of what is becoming the notorious partisan (R) poll flood. Now, you may be of the mind that you should throw out “bad” polls. Not exactly. Who defines “bad?” Are all partisan polls bad? Or just ones perceived as “outliers?” Don’t let bias influence your work.
Even “bad” polls can give you valuable insight. But what’s happening at 538 and, by extension, basically everywhere else is they’re essentially counting these partisan polls individually and wrongly rewarding them for being more recent.
Well, Nate Silver himself addressed this issue by saying, effectively:
Their averages adjust for bias. (Good thing we’ve already covered he doesn’t know what a poll error is).
Dems could do the same thing but they don’t so that means…something
This is not to say Rs can’t get lucky here. See my posts about Wisconsin and North Carolina and guessing the roll of a dice. But I don’t think New Hampshire is as “in play” as the “poll margin” might suggest.
My Forecast: Hassan 85% (51.5-47)
Finish Line (proj): 375K votes
538 Forecast: Hassan 73% (51-47.3)
Arizona
Another margin vs average issue. This race is one where the “toss up” narrative is too strong in my opinion. Again, yes, Rs can win here and it wouldn’t be the upset of the century, but it’s closer to dice roll than coin flip.
My Forecast: Kelly 82% (50.5-48)
Finish Line (proj): 2M votes
538 Forecast: Kelly 66% (50.2-48)
Pennsylvania
Now begin the true swing states. This race might be the only race where a little recency bias (e.g. favoring 2 week old polls over 3 week old polls) is justified. That being said, there has only been ONE poll (even partisan) that had Oz over 48%.
Yes, undecideds have upticked late, but here’s the big question:
What do we know about the undecideds?
Considering 538’s poll average had Fetterman over 50% in September, is it reasonable to assume the debate was so bad that Fetterman voters will defect to Oz? Or just that they defected to undecided, while the “Oz undecideds” have come home?
The answer is probably a mix, but Fetterman should still be the favorite here. A polling average over 50%, even if it recedes, isn’t mere noise. My money would be on D voters to come home and Fetterman to flip PA blue, even if it’s by a modest margin.
My Forecast: Fetterman 65% (50.5-48)
Finish Line (proj): 2.9M votes
538 Forecast (the night of 11/6): Fetterman 54% (49.3-48.7)
538 Forecast (the evening of 11/7): Oz 56% (49.4-48.6)
The difference: Their win probability moved 10% on the back of, I shit you not, a 450 Likely Voter sample that’s actually very good for Fetterman, and a Registered Voter sample strongly favoring Oz from a polling firm that they don’t have a link for.
Nevada
This race has been, effectively, a tossup from the beginning. A few quality polls of late have showed Laxalt ahead. In the past week I’ve nudged this race from slightly favoring Ds to slightly favoring Rs.
The early vote in Nevada seems to range from okay to good for Ds depending on how you read it, but early vote is not as reliable of a predictor (in my opinion) as poll data. If you need some hopium, you can nudge my forecast to the left based on the not terrible early vote and make it a coinflip.
My Forecast: Laxalt 60% (49-48.5)
Finish Line (proj): 600K votes
538 Forecast: Laxalt 52% (48.8-48.6)
Georgia
Finally, my good friend, Bluorgia. Analysts slept on Dem turnout there in 2020, and could be doing so again in 2022. Can they avoid a runoff is the better question. Kemp is running strongly ahead of Abrams, but Walker isn’t exactly the ideal candidate for Rs.
Here, I’ve forecasted both win probability and runoff probability. The most likely scenarios are that Warnock wins outright and is the 50th or 51st Dem in the Senate, or it goes to a runoff where Ds need Georgia to maintain 50-50 (in which case I’d have them favored in a runoff) or that Walker wins outright (in which case 50-50 is best case for Ds, more likely he’s the 51st R).
My Forecast: Warnock 40% outright, 60% overall.
Finish Line (proj): 2.2M votes
538 Forecast: Walker 63% (49.9-48.7)
The difference: 538’s recency weighting is especially egregious here giving Trafalgar’s 50 for Walker MUCH MORE WEIGHT than
SurveyUSA’s 49 for Warnock,
Marist’s 49 for Warnock, or
Emerson’s 50 for Warnock.
Trafalgar’s poll was Nov 4th, whereas the three I mentioned began on October 29th, October 31st, and gasp October 28th, respectively. Despite their “A” ratings, November 4th is just much more recent than October 28th (sarcasm font). I think the difference here is pretty indefensible, even if Walker pulls it out, the extreme weight-by-recency methodology is terrible.
Overall Forecast
Deep breath. Dems are roughly 65% to hold the Senate. Not overwhelming favorites, but favorites. If I told you that in 2020 (I did) you’d probably have been overjoyed.
There’s about a 28% chance we end up where we started, 50-50.
I have Dems about 15% to get to 52+ (hold all current seats + PA, + OH or WI or NC).
On the other end, Rs have slightly thicker tails at 49 and 48 because a few things could go well for them: PA (I have them 35%) and NV (60%) would get them to 49. GA, 48.
Meanwhile, our friends at 538 think Ds have a 10% chance of holding 53 or more seats, and Rs have a 20% chance of holding 53 or more. That is, nearly a 1/3 chance that there’s a net gain/loss of 3. See beginning of post: tails.
Hope to hear from you all on and after election night. Thanks for reading, please subscribe!