The Case for a Neutral 2022

[GALeftist]

I would like to preface this post by saying I do not think this is the modal outcome. I think Republicans will probably end up with a fairly comfortable majority in the House and ~51 seats in the Senate. However, I do think that the possibility of a bluer year than that is being dismissed out of hand recently, which confuses me given that the things that made people think it was plausible a couple months ago honestly haven't changed much.

1. Special Results

This is the big one; post-Dobbs, every single special election to the House had Democrats overperform not only expectations but also Biden's margins by an average of 4.75 points ignoring Alaska. Remember, with a median seat of Biden+2ish, theoretically Republicans need a rightward shift of at least 2.5 points from 2020 for a bare majority, so even if special election turnout or whatever is to blame for some of that, Rs are in trouble unless it accounted for like 7 points versus a general election. This changed people's outlook on the midterms for a while, because historically specials have been decently correlated with generals, but that seems to have evaporated by now. Obviously, the nature of special elections is that there aren't going to be any right before the election, so maybe whatever caused that shift has evaporated by now, but people seem to be just taking that for granted without much evidence.

2. Washington primaries

According to my data, the WA-SEN 2022 primary was D+13.7, while the Washington congressional primaries in 2020 were D+13.6. Historically, these margins have been quite predictive for Washington and fairly predictive nationally. Again, I remember after Rs underperformed expectations in the primary everyone moved this race to Safe D, but now that some time has passed people are talking about it like it's a sleeper again. Don't know why this is. And, again, remember that Rs don't just need to match 2020 to win the House, they need to do better than Trump by about 2.5 points of margin at least.

3. There's no law saying polling error can only benefit Republicans

I think people are really erring in restricting their analysis to 2016-2020 (so essentially n=3). Like, depending on how likely a systemic polling error is, you could plausibly end up with two cycles that underestimate Republicans and one that's pretty good just by pure random chance even if polling errors in both directions are equally possible. It's not like it's never happened before, either; all the polling for the House specials underestimated Democrats. Not sure what the mechanism there was, but you can come up with a number of plausible explanations. Maybe pollsters are overcorrecting for 2020 (I'd argue that this is almost certainly true in some cases like D4P). Maybe nonresponse bias doesn't actually select for Democrats, it just selects for some demographic of people who were very likely to be Biden voters and very likely to swing right in 2022 relative to the national population. Again, I don't think these are true, but I also think it's silly to dismiss them out of hand before the election.

The main counterpoints to all this, if I had to guess, would be 1. midterm fundamentals and 2. polling has shifted back towards Republicans recently. I think these are fine explanations for why we could be looking at a good year for Rs, but the fact is that 1 was always true and 2 has some major caveats (most importantly that there have been so few polls this year compared to other years), so I think it's weird how much the perception of the cycle has changed recently. Mine really hasn't since Ryan won NY-19.

[MT Treasurer]

so even if special election turnout or whatever is to blame for some of that, Rs are in trouble unless it accounted for like 7 points versus a general election.

...The thing is it likely did account for that difference, though?

In 2017/2018, virtually all the special elections except GA-06 showed far more substantial swings away from the GOP since 2016 than would actually materialize in November 2018. The November 2016 -> November 2018 swing wasn’t nearly as pronounced as the November 2016 -> special 2017/2018 swing. Why do you think that was?

If we apply the same 7-point shift to this cycle's special elections, you’d expect the specials to basically resemble or match the 2020 results even in a R+3-4 year, and, well, that’s what happened in most of them.

[Associate Justice PiT]

     The thing I would note in regards to point 3 is that this factor has an n>3. The trend whereby polls overstate liberal/left-wing performance is traditionally called the "Shy Tory Effect", getting its name from when it was first seen in Britain in the 1990s. It also happened recently in Brazil, with polls greatly underestimating the strength of Bolsonaro and his allies.

     Granted, nobody knows what exactly causes the Shy Tory Effect, and so a reversal is in principle possible based on obscure factors. But I think the body of evidence supporting a polling error that benefits the GOP is stronger than you realize. There's definitely also been neutral elections, but I don't know that it is historically supportable to say that a polling error benefiting Democrats is as likely as one benefiting Republicans.

[Sestak]

     The thing I would note in regards to point 3 is that this factor has an n>3. The trend whereby polls overstate liberal/left-wing performance is traditionally called the "Shy Tory Effect", getting its name from when it was first seen in Britain in the 1990s. It also happened recently in Brazil, with polls greatly underestimating the strength of Bolsonaro and his allies.

     Granted, nobody knows what exactly causes the Shy Tory Effect, and so a reversal is in principle possible based on obscure factors. But I think the body of evidence supporting a polling error that benefits the GOP is stronger than you realize. There's definitely also been neutral elections, but I don't know that it is historically supportable to say that a polling error benefiting Democrats is as likely as one benefiting Republicans.

The one thing here is that (unlike in 2020) everyone already seems to be pricing in that sort of polling error. To use one example, if Oz loses (which I don't expect), the surprise will not be that "there was a polling error that benefitted the Democrats", the surprise will be "there was no polling error". The "general expectation" already is a polling error leading to an Oz win by a couple points; Oz outperforming expectations would be a case where the GOP-favored error not only existed but was fairly large.

It may still be the less likely option but Democrats matching/beating expectations becomes a much more viable possibility when that can be done even without beating (and perhaps even while underperforming) public polling.

[TheDeadFlagBlues]

     The thing I would note in regards to point 3 is that this factor has an n>3. The trend whereby polls overstate liberal/left-wing performance is traditionally called the "Shy Tory Effect", getting its name from when it was first seen in Britain in the 1990s. It also happened recently in Brazil, with polls greatly underestimating the strength of Bolsonaro and his allies.

     Granted, nobody knows what exactly causes the Shy Tory Effect, and so a reversal is in principle possible based on obscure factors. But I think the body of evidence supporting a polling error that benefits the GOP is stronger than you realize. There's definitely also been neutral elections, but I don't know that it is historically supportable to say that a polling error benefiting Democrats is as likely as one benefiting Republicans.

What has happened in other countries is totally irrelevant to the US. It doesn't matter that sometimes the right benefits from polling error abroad. We also know that the left sometimes benefits as well - this happened in Chile last year, so what?

In the end, empirical evidence from the US shows that polling error isn't systematically associated with the right or left over the past 40-50 years. Sometimes Democrats are underestimated, sometimes Republicans. We do know that in some states, Democrats are almost always underestimated and this is true for the GOP as well but, nationally, we really are relying on a few elections that happened recently.

To be clear, I think Democrats will be crushed on Tuesday and that GOP will benefit from polling error but that's more about non-response bias related to "low social trust" voters and not about "shy Tory" - latter is very different phenomenon. A certain type of working class voter won't respond to surveys - not the same thing as someone lying about their preference.

[RI]

Even an R+1 environment (which the WA primary suggested) results in the Republicans winning a solid margin in the House and most of the toss-up Senate seats. That's a 5.5-point swing from 2020.

[Senator Incitatus]

     The thing I would note in regards to point 3 is that this factor has an n>3.

It's also n>3 because it didn't happen in "three elections," but in the most recent dozens or hundreds of elections. Just because they occurred on the same three days doesn't make them three events.

[TheDeadFlagBlues]

Even an R+1 environment (which the WA primary suggested) results in the Republicans winning a solid margin in the House and most of the toss-up Senate seats. That's a 5.5-point swing from 2020.

If it's a R+1 environment, Democrats will probably hold all of their Senate seats and flip PA.

[Kahane's Grave Is A Gender-Neutral Bathroom]

Is it possible the GOP firms are pushing the averages so they can claim fraud with something like 'even the fake D polls had us up!' or something?

[GALeftist]

so even if special election turnout or whatever is to blame for some of that, Rs are in trouble unless it accounted for like 7 points versus a general election.

...The thing is it likely did account for that difference, though?

In 2017/2018, virtually all the special elections except GA-06 showed far more substantial swings away from the GOP since 2016 than would actually materialize in November 2018. The November 2016 -> November 2018 swing wasn’t nearly as pronounced as the November 2016 -> special 2017/2018 swing. Why do you think that was?

If we apply the same 7-point shift to this cycle's special elections, you’d expect the specials to basically resemble or match the 2020 results even in a R+3-4 year, and, well, that’s what happened in most of them.

I see this talking point all the time, and it's appealing because the particularly unusually close races tend to stick in our mind, but the fact is that it doesn't seem to be supported by the data.

GA-06 special was R+3.6, GA-06 general was D+1.0, error of R+4.6
KS-04 special was R+6.2, KS-04 general was R+18.8, error of D+12.6
MT-AL special was R+5.5, MT-AL general was R+4.6, error of R+0.9
SC-05 special was R+3.1, SC-05 general was R+15.5, error of D+12.4
AZ-08 special was R+4.7, AZ-08 general was R+11, error of D+6.3
TX-28 special was R+20.7, TX-28 general was R+23.7, error of D+3
OH-12 special was R+0.8, OH-12 general was R+4.2, error of D+3.4

That's an average error of D+4.6, not particularly close to the 7ish points you'd expect Rs to need to win the House. Only KS-04 and SC-05 reached that magic number, in fact.

(CA-34 was D vs. D, UT-03 had a third party in the special but not the general, PA-18 was redistricted; therefore, all were omitted from this analysis.)

    The thing I would note in regards to point 3 is that this factor has an n>3. The trend whereby polls overstate liberal/left-wing performance is traditionally called the "Shy Tory Effect", getting its name from when it was first seen in Britain in the 1990s. It also happened recently in Brazil, with polls greatly underestimating the strength of Bolsonaro and his allies.

Sure, but it doesn't happen in every election. Didn't happen in the specials, didn't happen in 2018, didn't happen in 2012. Didn't even happen in the UK in 2017 when Labour greatly outperformed its polls and forced the Tories into C+S with the DUP. Am I to conclude there is a "Shy Corbynista" effect? Why in 2017 and not 2019? For that matter, how can I tell when Tories are being shy, as it doesn't seem they were very shy in 2017? The fact is, until a consistent pattern emerges in the United States, it is not safe to assume that 2020 polling error is generalizable, and such a pattern has not yet emerged.

It's also n>3 because it didn't happen in "three elections," but in the most recent dozens or hundreds of elections. Just because they occurred on the same three days doesn't make them three events.

Unless you think that it's a coincidence that polls were pretty good almost everywhere in 2018 and off in the same direction by similar magnitudes in 2016 and 2020, this is transparently false. We aren't talking about individual races here, we are talking about systemic polling error, and in general those errors emerge in cycles, not specific races.

[Alben Barkley]

Is it possible the GOP firms are pushing the averages so they can claim fraud with something like 'even the fake D polls had us up!' or something?

It's certainly possible, but trustworthy polls like Selzer are showing a significant shift right over the past month as well.

Even an R+1 environment (which the WA primary suggested) results in the Republicans winning a solid margin in the House and most of the toss-up Senate seats. That's a 5.5-point swing from 2020.

If it's a R+1 environment, Democrats will probably hold all of their Senate seats and flip PA.

How so? I mean I can see how it's possible but the swing states still generally lean R, so all else equal would probably narrowly stay with the GOP in that kind of environment. You might get some exceptions in places like AZ thanks to "candidate quality," but it would be a mistake to assume Dems will just automatically win these seats in an R +1 environment.

[GALeftist]

Is it possible the GOP firms are pushing the averages so they can claim fraud with something like 'even the fake D polls had us up!' or something?

It's certainly possible, but trustworthy polls like Selzer are showing a significant shift right over the past month as well.

Maybe, but Reynolds's margin remained unchanged. I am more interested to see the GCB than anything; that should tell us if it was just closet Grassley supporters coming home or a larger rightward shift.

[Southern Senator North Carolina Yankee]

Is it possible the GOP firms are pushing the averages so they can claim fraud with something like 'even the fake D polls had us up!' or something?

It's certainly possible, but trustworthy polls like Selzer are showing a significant shift right over the past month as well.

Maybe, but Reynolds's margin remained unchanged. I am more interested to see the GCB than anything; that should tell us if it was just closet Grassley supporters coming home or a larger rightward shift.

I see a tagline when I search Iowa GCB R's lead in three of.." but when I click I get paywalled by the DMR site.

[GALeftist]

Even an R+1 environment (which the WA primary suggested) results in the Republicans winning a solid margin in the House and most of the toss-up Senate seats. That's a 5.5-point swing from 2020.

Sorry for the double post but where are you getting this estimate? Here's what I have:

2022 D+14 SEN primary vs. ? actual
2020 D+14 congressional primary vs. D+20 congressional actual
2018 D+24 congressional primary vs. D+28 congressional actual
2016 D+15 congressional primary vs. D+11 congressional actual

That is to say, this seems to suggest that 2022 is similar to 2020 or perhaps 2016, D+3 or R+1 on the GCB respectively. (And before someone mentions it, yes, I used SEN data; there was a lot of crossover by Democrats to try to save Newhouse and Beutler, so I think SEN is a better measure.) This is actually kind of what I mean. At the time of the Washington primaries, everyone accepted that they were really good for Dems relative to expectations, but now everyone seems to have retconned it. Really weird.

[ProgressiveModerate]

Even an R+1 environment (which the WA primary suggested) results in the Republicans winning a solid margin in the House and most of the toss-up Senate seats. That's a 5.5-point swing from 2020.

Sorry for the double post but where are you getting this estimate? Here's what I have:

2022 D+14 SEN primary vs. ? actual
2020 D+14 congressional primary vs. D+20 congressional actual
2018 D+24 congressional primary vs. D+28 congressional actual
2016 D+15 congressional primary vs. D+11 congressional actual

That is to say, this seems to suggest that 2022 is similar to 2020 or perhaps 2016, D+3 or R+1 on the GCB respectively. (And before someone mentions it, yes, I used SEN data; there was a lot of crossover by Democrats to try to save Newhouse and Beutler, so I think SEN is a better measure.) This is actually kind of what I mean. At the time of the Washington primaries, everyone accepted that they were really good for Dems relative to expectations, but now everyone seems to have retconned it. Really weird.

The issue is WA is not necessarily a very good barometer to the nation. Given the current coalitions, Washington seems like one of the places to be least affected by an R wave along with much of the Northeast, Colorado, and racially polarized parts of the southeast.

[Amenhotep Bakari-Sellers]

We already know it's a 303 map

[ProgressiveModerate]

Even an R+1 environment (which the WA primary suggested) results in the Republicans winning a solid margin in the House and most of the toss-up Senate seats. That's a 5.5-point swing from 2020.

If it's a R+1 environment, Democrats will probably hold all of their Senate seats and flip PA.

Considering NV/GA/AZ all voted about 3 points to the right of the nation in 2020, I think in an R+1 environment they'd probably lose at least one, even once you factor in "candidate quality" and the dynamics in those specific states.

[Amenhotep Bakari-Sellers]

OK Gov, OH or NC or UT and WI SEN can be won still we Ds have won red states outside of TX and FL, we won OH, MT, WVA SEN AND KS and KY 2018/2019 and AK 2022

The reason why FL/TX are not prone to D Takeovers the Ds haven't won a Gov race since Ann Richards and Laughton Child's Gwen Graham should of been the nominee in 2018 fir GOV

[Sestak]

Is it possible the GOP firms are pushing the averages so they can claim fraud with something like 'even the fake D polls had us up!' or something?

It's certainly possible, but trustworthy polls like Selzer are showing a significant shift right over the past month as well.

Maybe, but Reynolds's margin remained unchanged. I am more interested to see the GCB than anything; that should tell us if it was just closet Grassley supporters coming home or a larger rightward shift.

I see a tagline when I search Iowa GCB R's lead in three of.." but when I click I get paywalled by the DMR site.

Implication here would be that Selzer is showing Axne ahead?

[Associate Justice PiT]

     The thing I would note in regards to point 3 is that this factor has an n>3. The trend whereby polls overstate liberal/left-wing performance is traditionally called the "Shy Tory Effect", getting its name from when it was first seen in Britain in the 1990s. It also happened recently in Brazil, with polls greatly underestimating the strength of Bolsonaro and his allies.

     Granted, nobody knows what exactly causes the Shy Tory Effect, and so a reversal is in principle possible based on obscure factors. But I think the body of evidence supporting a polling error that benefits the GOP is stronger than you realize. There's definitely also been neutral elections, but I don't know that it is historically supportable to say that a polling error benefiting Democrats is as likely as one benefiting Republicans.

What has happened in other countries is totally irrelevant to the US. It doesn't matter that sometimes the right benefits from polling error abroad. We also know that the left sometimes benefits as well - this happened in Chile last year, so what?

In the end, empirical evidence from the US shows that polling error isn't systematically associated with the right or left over the past 40-50 years. Sometimes Democrats are underestimated, sometimes Republicans. We do know that in some states, Democrats are almost always underestimated and this is true for the GOP as well but, nationally, we really are relying on a few elections that happened recently.

To be clear, I think Democrats will be crushed on Tuesday and that GOP will benefit from polling error but that's more about non-response bias related to "low social trust" voters and not about "shy Tory" - latter is very different phenomenon. A certain type of working class voter won't respond to surveys - not the same thing as someone lying about their preference.

     You do make a valid point that I should not be associating this with what happened to the Tories in 1992 since GOP voters today and the UK Conservative base 30 years ago are very different groups. With that in mind, I think it is somewhat misleading to look at "empirical data...over the past 40-50 years". The GOP base has changed quite a bit over that timeframe, and the most recent shift related to Trump's rise is likely related to the emergence of this polling error.

[Southern Senator North Carolina Yankee]

Is it possible the GOP firms are pushing the averages so they can claim fraud with something like 'even the fake D polls had us up!' or something?

It's certainly possible, but trustworthy polls like Selzer are showing a significant shift right over the past month as well.

Maybe, but Reynolds's margin remained unchanged. I am more interested to see the GCB than anything; that should tell us if it was just closet Grassley supporters coming home or a larger rightward shift.

I see a tagline when I search Iowa GCB R's lead in three of.." but when I click I get paywalled by the DMR site.

Implication here would be that Selzer is showing Axne ahead?

It was for there previous poll, but I couldn't tell that before since it was behind a paywall.

[Vosem]

Even an R+1 environment (which the WA primary suggested) results in the Republicans winning a solid margin in the House and most of the toss-up Senate seats. That's a 5.5-point swing from 2020.

If it's a R+1 environment, Democrats will probably hold all of their Senate seats and flip PA.

That would suggest very strong Democratic trends in a number of states; PA/AZ/GA all should be gone at ~D+3, and while there's lots of evidence that GA has a very strong Democratic trend and might well defy the wave this time around, the evidence of this for AZ or PA is much more lacking. (At D+1 NV should be flipping as well).

It would be modestly surprising but not impossible if Democrats won one of PA/AZ/GA/NV at R+1, but 2+ would suggest a set of very well-run campaigns, and 4/4 would be insane. I don't think there's much reason to think that'll happen; I've said since this summer that an underrated possible outcome is an outright Democratic victory in the GCB (like D+3) which is still accompanied by Republicans sweeping competitive Senate and House races.

[Sbane]

Even an R+1 environment (which the WA primary suggested) results in the Republicans winning a solid margin in the House and most of the toss-up Senate seats. That's a 5.5-point swing from 2020.

I agree that a republican win by 1 point in the NPV equals a tossup/tilt R senate, but it’s not going to be a blowout in the house at those numbers. Republicans probably still win but only about 220-230 seats.

[GALeftist]

I am getting pretty close to accepting my accolades. Certainly the second best midterm performance since 2000.

[Absentee Voting Ghost of Ruin]

I am getting pretty close to accepting my accolades. Certainly the second best midterm performance since 2000.

There are enough key seats still up there that if everything breaks the GO0's way over the next week they could eke out control of both houses, I think kudos are absolutely due.