Election models megathread

[Fmr. Gov. NickG]

These models are (and should be) mostly based on polls.  And a lot of races have very few polls at this point.  Are people suggesting that 538 should just arbitrarily change their results based on intuition in these case?

[Fmr. Gov. NickG]

The senate model is clearly underestimating Republicans due to reliance on polls, the GOP has outperformed what the polls said in the Senate 4 cycles in a row, 2014, 2016, 2018 and 2020, in all 4 cycles the GOP won more senate seats than most projected, no reason to believe the polls have corrected for the bias.

I know that’s true for 3 of the elections, but are you sure about 2018? I thought the results matched the polls in the Senate then - it was only Democrat hopium that was defeated.

The 2018 senate polls were biased, look at states like Indiana, Missouri, Florida etc, even in many senate races democrats won, they won by smaller margins than the polls said.

https://projects.fivethirtyeight.com/2018-midterm-election-forecast/senate/

GOP did better than the median 538 senate forecast.

Much worse than the inaccuracy of their predictions in 2018 was 538’s insistence on presenting everything in bizarre fractions. Jackie Rosen was a “4 in 7” favorite to win while Beto O’Rourke was a “2 in 9” underdog.

[Fmr. Gov. NickG]

https://twitter.com/DecisionDeskHQ/status/1549763196084310028

I can't fathom how they have Florida SEN as more competitive than Wisconsin SEC.  Ron Johnson has 96% probability of winning while Mark Kelly and CCM are solid favorites to hold their seats?  Seems pretty inconsistent.

[Fmr. Gov. NickG]

Why is it that switching from the Deluxe to Classic model on 538 dramatically improves Dem’s chances of holding the Senate, but has absolutely no effect on their chances of holding the House?

[Fmr. Gov. NickG]

Why is it that switching from the Deluxe to Classic model on 538 dramatically improves Dem’s chances of holding the Senate, but has absolutely no effect on their chances of holding the House?

Different polling results.

But the polls included in both versions of the model are the same, right?

[Fmr. Gov. NickG]

https://twitter.com/NateSilver538/status/1558095352485511170

I'm very confused about how the House model can think the GCB is slightly over 2020t R + 4 (about a 9 point swing from 2020) yet the Senate model can in good faith rate PA-Sen and AZ-Sen as Lean D. Sure, Oz and Masters have their issues but are they significant to matter that much?

And the Governors forecast almost seems to indicate a D wave, with MI-Gov being safe D, MN, WI, and PA being likely D, and KS being tossup.

Even just look at a state by state level in the House. AZ-Sen is lean D yet AZ-01 and AZ-06 are safe R, districts that Kelly will prolly have to win or at least come close in to win statewide.

I have a feeling that in reality their 3 models are not very well correlated at all.

I do think the 538 model tends to overvalue incumbency across the board, which somewhat explains AZ.  Like I definitely don’t think Dems are more likely to win NC than WI. 

But candidate quality can easily create a >5 point swing in individual race; we see this in multiple races in every cycle.  Just in Pennsylvania alone, Bob Casey would have won in a R+4 environment in both 2012 and 2018 if you apply a uniform swing from the House PV.

[Fmr. Gov. NickG]

The Map isn't there for Democrats to retain the House.

I will be watching the Virginia 2nd Congressional Race on Election Night between Rep. Elaine Luria and her Republican Opponent Jen Kiggans. This District is a total Swing District similar to NY-19. If Kiggans wins Republicans Chances of retaking the House jumps to 95 %.

VA-02 was the median House seat on 2020 Pres numbers so yeah def a very important one for both sides (since it seems unlikely Kiggans or Luria would significantly defy traditionally partisanhip and run away with it). It seems like most paths of least resistance to a Dem majority will require them to win a few Trump districts such as IA-03, MI-07, PA-08, or MI-10 since Rs have seats like PA-01, CA-40, NE-02, and maybe NJ-07 pretty locked down (at least enough to be to the right of the tipping point)

For reference, he's a list of seats ranked by 2020 Pres results:

205 - WA08 - B + 6.67
206 - NV03 - B + 6.64
207 - NE02 - B + 6.32
208 - CA45 - B + 6.18
209 - NH01 - B + 5.93
210 - NM02 - B + 5.88
211 - PA17 - B + 5.83
212 - PA01 - B + 4.64
213 - NY19 - B + 4.62
214 - CO08 - B + 4.55
215 - KS03 - B + 4.45
216 - NJ07 - B + 3.66
217 - OH13 - B + 2.82
218 - VA02 - B + 2.05
219 - MI08 - B + 2.03
220 - CA40 - B + 1.87
221 - NC13 - B + 1.69
222 - AZ01 - B + 1.48
223 - MI07 - B + 0.95
224 - PA07 - B + 0.62
225 - NY01 - B + 0.21
226 - AZ06 - B + 0.07
227 - FL27 - T + 0.32
228 - IA03 - T + 0.34
229 - MI10 - T + 0.98
230 - CA41 - T + 1.09

One thing that's interesting is that seat 211 (PA-17) is nearly Biden + 6 which is pretty blue and to the left of the nation but after that the seats drop off for Dems very quickly and by seat 218 you're at Biden + 2. However, after that the list slows down again and you have quite a lot of seats that were extremely close in the 2020 pres race.

We could also have a systematic Polling Error. In 2016, 2018 and 2020 every single time Pollsters undercounted the Shy Trump Vote.

As far as the Senate goes yes Models are showing Democrats with a good lead BUT when you look at WI, GA, NV there are all within the MOE. They can easily go the other way.
If you look at 2014 for example Races in IA, CO, NC started to swing in the Final 2 Weeks of the Campaign.

In 2018, the final generic ballot average on 538 was D+8.6.  And the actual national House vote was….D+8.6.

https://projects.fivethirtyeight.com/polls/generic-ballot/2018/
https://en.wikipedia.org/wiki/2018_United_States_House_of_Representatives_elections

I think it’s more likely that if we see a systemic polling error this year, it will be underestimating Dem support.  The lower-propensity voters that polls are less likely to catch this time around will be younger women motivated by Dobbs.

[Fmr. Gov. NickG]

It seems there is a pretty odd reason why 538's House projections has not budged at all in the last few weeks despite Dem's improving poll numbers.

Over the last two weeks, 538's projected House popular vote -has- improved for Dems, from R+4.1 to R+3.4 today.

However, at the same time, their projected popular vote threshold at which Dems are favored to retain control has also fallen, from R+0.6 to R+0.1 today.  

I can't really figure out why this second quantity would change, unless 538 somehow decided it was absurd that they were showing a slight seats/votes bias in favor of Dems, and changed their model in response.

[Fmr. Gov. NickG]

FiveThirtyEight: Democrats up to 71% in the senate, Republicans down to 73% in the House. You love to see the trendline!

And now they are both 72!

[Fmr. Gov. NickG]

Also, why the hell does the RCP model have the GOP winning Wisconsin and Pennsylvania when their polling averages have Fetterman and Barnes ~5 points up in each?

Because RCP is not a neutral site. It had an R-bias for quite some time, especially with the duration of including certain pollsters in polling averages.

In 2000, RCP had California as only Lean Gore before election day.

They had much bigger problems than that. They were projecting like a 400EV Bush win though tbf that was 20 years ago

Also in 2000 Cali was only like a D+10 state or so

The 2000 election really did shift substantially toward Gore in the last weekend.  This was correctly picked up by several polls right before the election, but like two weeks before election day, the solid consensus was that GWB would win by ~5 points.

[Fmr. Gov. NickG]

Has anyone else noticed that the 538 model seems totally broken with respect to the Georgia run-off.  Look at the predicted distribution of vote shares:



It's not unusual for their to be outliers where one candidate wins by like 20% in a couple simulations of a race predicted to be close; there's always the slight possibility of some huge scandal coming up, etc.  
But in the vast majority of those cases, the race shouldn't go to a run-off.  The distribution of vote share in races that go to a run-off should certainly be narrower than races that don't go to a run-off.  And yet the 538 model seems to show the opposite.  

It also shows a suspiciously small number of simulations where the race actually -is- very close.  Unlike every other close race where the results are approximately normally distribute with a mean close to 0, here the distribution appears bimodal with a local minimum around 0.  

Finally, the estimate for the Libertarian vote share varies from almost 5% all the way down to 0.2%.  Should the estimate be substantially more confident than that?

[Fmr. Gov. NickG]

Dems went from 65 to 67 in 538 Senate and 29 to 30 in the House? Any ideas what caused this? No polling that came out today was divergent enough or high rated enough to cause such a change. Maybe their fundamentals model incorporated some news.

Its showing 66 and 31 right now, so a slightly smaller change in the Senate but bigger chance in the House.

The generic ballot has ticked up a couple tenths for the Dems the past couple days.

Also, the 538 model assumes the environment will get progressively better for Republicans as the election approaches.  So every day that passes with no real change actually improves Dem chances.

This is an aspect of their model that has already proven fundamentally wrong, and yet I don't think they will get much grief for it, because it presumably won't really affect their final prediction.

[Fmr. Gov. NickG]

Unfortunately Nate "Plastic" Silver has become a complete partisan hack.

My impression is that the Twitter consensus has come to believe Nate Silver is a Republican partisan hack due to some of his positions on covid.

[Fmr. Gov. NickG]

538 has Democrats' odds of holding the Senate at 63%, which is their lowest number since August 23rd.

I would guess that most of the movement in all of 538's models today is due to changes in the generic ballot.

[Fmr. Gov. NickG]

I really wish 538’s models had options for “only include Trafalgar polls” and “exclude all Trafalgar polls”.

[Fmr. Gov. NickG]

I'm really confused about how 538's model is working.  Siena releases 4 polls this morning showing Dems doing better than expected in every race.  Siena is an A+ rated pollster on 538, in fact the only A or A+ rater pollster polling any of these races as far as I can tell.

And these polls don't move the needle the slightest bit today any of these races.  What is 538 basing these predictions on if not these sort of polls?  (I know there's generic ballots and fundamentals and stuff, but shouldn't they be relying on this at least a little?)

[Fmr. Gov. NickG]

Did they? It's back up to 51-49 Dems.  

It was 50-50 a minute ago.

I truly cannot comprehend this.  Hasn’t every single Senate poll today been good for Dems? (I guess except for the high school “poll” in WI)

[Fmr. Gov. NickG]

Does anyone know what "adjusted poll average" means?

It's the RCP average plus the average poll underestimation.

But why are NV and AZ projected as GOP pick-ups when Dems lead the adjusted average?

[Fmr. Gov. NickG]

Does anyone know what "adjusted poll average" means?

It's the RCP average plus the average poll underestimation.

But why are NV and AZ projected as GOP pick-ups when Dems lead the adjusted average?

I don't understand what "average poll underestimation" means.

In NH, Biden was +10.4 in the RCP average but won by 7.3, so a 3.1% underestimation.  Same with 2016; Hillary was up 3.3 but won by like 0.3, so a 3% understimation.

[Fmr. Gov. NickG]

For the Senate, suddenly 538 moved a full 3 points in the last hour or two, to 58-42 odds that the Pubs get 51 or more seats. It appears to be primarily driven by Oz's odds of winning by a substantially higher percentage, and Walker in GA, even as Laxalt fades a bit - the spirit of Ralston effect.

https://projects.fivethirtyeight.com/2022-election-forecast/senate/?cid=rrpromo

Yeah that’s super weird.  Fetterman just dropped 10 points based on I have no idea what.

Edit: Actually I have a pretty good idea what it is….Fetterman actually rose slightly in the “Classic” model; he only dropped in Deluxe.  

So it must be all based on Sabato picking Oz in his final predictions.  Also explains the bump for Walker, Kelly, and CCM.