2022 Elections: A Post-Mortem (Part 1)

I first previewed 2022 United States midterm elections, which ended on November 8, 2022, in three June 2021 essays. First, I updated my single-variable model of elections for the United States House of Representatives (“House”). In the second and third essays, I took a “wicked early” look at, respectively, the 36 elections for governor and 35 elections for United States Senate (“Senate”). As of June 2021, it seemed very likely Democrats would lose their House majority, while they had a decent chance to retain their Senate majority and do no worse than “hold serve” in gubernatorial elections.

The latter two essays focused on the “fundamentals”: national partisan environment, state partisan lean and incumbency. I measured national partisan environment using generic ballot polls;[1] in June 2021, Democrats led by 3.6 percentage points (“points”), though by Election Day (“ED”) 2022, Republicans led by 0.2 points with approximately 9.0% still undecided. When estimating the probability Democrats would retain their House majority, I used my regularly-updated weighted-adjusted polling average (“WAPA”) of the generic ballot polls. Meanwhile, I measured state partisan lean with 3W-RDM.

On November 7, 2022, I published my final projections of who would control the House, as well as how many Senate seats and governor’s mansions each party would control. Rather than “predict” individual races, I presented Democratic win probabilities for overall House control and for each of the 71 Senate and gubernatorial elections.

So, now that two weeks have passed since Election Day, the question is:

How did I do?

***********

The Big Picture. Unless stated otherwise, 2022 election data come from MSNBC.

My House model consistently gave Republicans the edge to capture the minimum five seats required for a majority, regardless of whether I used all polls conducted since January 1, 2021 or only those conducted entirely on or after September 5, 2022 (Labor Day). After Labor Day, Democratic-House-majority probability ranged between 20.3% (September 5, all polls) and 39.6% (October 3, post-Labor-Day-only). At 6:30 pm EST on November 16, NBC News called Republican Mike Garcia the winner in California’s 27th district, giving Republicans the 218 seats needed for a majority.

Republican majority in House? Check.

My Senate projections consistently had Democrats favored to hold at least 50 seats, with Democratic Vice President Kamala Harris breaking the tie, again irrespective of poll set. Democratic-majority probability ranged between 65.1% (November 7, post-Labor-Day-only) and 88.5% (October 10, all polls). At 9:17 pm EST on November 12, NBC News called Democratic incumbent Catherine Cortez Masto the winner of the Nevada Senate race, giving Democrats 50 seats.

Democratic majority in Senate? Check.

My gubernatorial projections consistently showed Democrats more likely to add governor’s mansions than to lose them, again irrespective of poll set. The probability Democrats add governors ranged between 50.5% (November 7, all polls) and 59.9% (September 26, all polls). At 8:48 pm on November 14, NBC News called Democrat Katie Hobbs the winner of the Arizona gubernatorial election, giving Democrat a net gain of two governor’s mansions with one election (Alaska governor) still to be called as of this writing.

Democrats add governors? Check.

Thus, this concluding paragraph from my final pre-election preview appears prescient, depending on the final national House vote tallies.

If these polling averages are unbiased, Senate control may well depend upon mail ballots in Nevada and a runoff in Georgia. The Senate will land somewhere between 49 and 51 Democrats. And Democrats will likely land between a net gain of 0 and 2 governors. The House appears likely to go Republican, though a pro-Democratic polling error of just 2.0 points makes that at least a 50-50 proposition. And, as of now, the political environment is anywhere from Democrats +2 to Republicans +2.

To be fair, these were relatively easy projections, given fundamentals and polling. The details, to which we turn next, are always more challenging.

***********

The House. Going into Election Day, I pegged Democrats as modest underdogs to maintain House control (20-25%), which proved to be correct. I also estimated Democrats would lose between 10.9 and 12.7 seats. Since Labor Day, my Democratic seat loss estimate ranged between 6.6 (October 10, post-Labor-Day polls) and 12.7 (twice). In essence, I “projected” Democrats would lose between six and 13 seats.

As of this writing, only three House seats are uncalled (California’s 3rd, 13th and 22nd districts),[2] though Republicans have multi-thousand vote leads in the 3rd and 22nd. It is thus likely Republicans will end the 2022 elections with 221 or 222 House seats, a net gain of eight or nine relative to just after the 2020 elections – well within the 6-13 seat range overall and slightly better than the 11-13 range as of Election Day.

Curiously, these projections were based upon Democrats losing all House votes cast nationwide by between 0.2 and 0.6 points (assuming undecideds broke evenly between Democrats and Republicans). As of this writing, though, Democrats trail by 3.3 points (50.9% to 47.6%), albeit with a few million mostly Democratic votes still to be counted in California. To be fair, given 1.5% of the vote going to neither major party (which I also estimated correctly), this is the anticipated margin had the final remaining undecideds broken 2-1 Republican; NBC exit polling suggest the 12% who decided in the last week for whom to vote in their House district split closer to 52.5% Republican, 44.0% Democratic, 3.5% Other, possibly putting the final margin at R+1.3. Still, in an R+3.3 environment, Democrats would be expected to lose 21.3 seats, albeit with a 95% “prediction interval” of 0.3 to 43.0 seats, which just barely encompasses the actual loss of 8-9 seats.

Bottom line: Republicans narrowly regained House control, as projected, though Democrats lost fewer seats than my model estimated.

***********

The Senate. From the first Senate distribution, published on September 26, Democrats were anticipated to end the 2022 election with between 49 and 52 seats, with the average/median number of seats rising and falling over time. With the Georgia Senate election headed to a December 6 runoff between Democratic incumbent Raphael Warnock and Republican Herschel Walker, Democrats will end the 2022 elections with 50 or 51 seats, right in the middle of my projections. And the only Senate election (with Georgia pending) I “called” incorrectly was in Nevada, where I gave Cortez Masto a 43.7% chance to win (Tilt Republican).

Digging further, meanwhile, Table 1 compares ED Democratic Senate margins (as of 11:00 pm EST, November 20, 2022) to final estimates based on 1) fundamentals (3W-RDM plus generic ballot estimate plus incumbency adjustment[3]), 2) polling (WAPA) and 3) my final projection (weighted combination of fundamentals and polling, with average polling weigh 86%[4]). I exclude Alaska, where the winner will be one of two Republicans, incumbent Lisa Murkowski or Kelly Tshibaka. In Utah, Independent Evan McMullin was effectively the Democratic nominee. No public polls were conducted of the Senate elections in Hawaii, Idaho and North Dakota.

Table 1: Election Day Democratic Senate margins compared to fundamentals, polling and projection estimates, sorted by state partisan lean

State3W-RDMED MarginED Margin minus…
   FundamentalsPollingProjected
Hawaii29.045.210.8n/a10.8
Vermont28.940.511.815.114.7
Maryland26.231.0-0.68.05.4
California24.922.8-3.82.61.9
New York20.213.5-12.1-2.7-3.2
Connecticut13.915.0-4.30.3-0.2
Washington13.714.6-4.58.17.2
Illinois13.314.5-4.2-0.2-0.6
Oregon10.115.0-0.5-3.5-3.3
Colorado5.714.63.56.56.2
New Hampshire1.29.12.65.45.1
Nevada-0.50.9-3.91.81.1
Pennsylvania-2.34.77.22.63.0
Wisconsin-2.41.06.94.04.2
Florida-5.5-16.4-7.4-9.1-8.9
North Carolina-5.8-3.22.80.40.6
Arizona-6.14.99.41.62.5
Georgia-6.50.95.80.61.1
Ohio-9.8-6.53.5-3.1-2.4
Iowa-9.8-12.11.2-2.4-2.2
South Carolina-15.9-25.9-6.5-7.4-7.1
Missouri-19.0-13.35.9-1.0-0.8
Indiana-19.6-20.72.4-11.0-8.9
Kansas-21.3-23.21.6-4.0-3.4
Louisiana-22.3-23.22.60.20.6
Utah-27.6-10.520.6-1.22.4
Alabama-29.2-35.7-6.3-4.6-4.7
South Dakota-29.6-43.4-10.3-17.2-16.3
Kentucky-30.3-23.610.2-8.13.0
Arkansas-30.3-34.8-1.0-11.9-9.7
Idaho-34.8-32.06.3n/a6.3
North Dakota-35.4-31.47.5n/a7.5
Oklahoma-37.8-32.29.1-11.7-8.1
Oklahoma SE-37.8-26.614.7-10.6-6.2
Average  2.1-1.7-0.4
Median  2.6-1.00.6
Average – Abs Value  6.05.44.8
Median – Abs Value  4.54.03.4

Overall, my final projections – based upon all polls released after November 1, 2021 – were extremely close, deviating from the final margin by an average 0.4 points in the Republican direction, with a median of 0.6 points in the Democratic direction. My conclusion that polls underestimated Democratic strength by 1-3 points appears not to be correct, though, as polling averages underestimated Republican strength by 1.7 points, on average, with a median deviation of 1.0 points. These relatively small misses could easily be explained by a late break of undecided voters for Republicans. Finally, fundamentals – even using the more generous generic ballot poll estimate of R+0.2 – underestimated Democratic margins by an average 2.1 points, with a median of 2.6 points; these values rise to 5.4 and 5.9 with the current Republican lead of 3.3 points in the total House vote.

The story is similar when looking at how much ED margin differed from estimates, regardless of direction. The average absolute value of the differences was 6.0, 5.4 and 4.8 for fundamentals, polling and projections, respectively, while medians were closer at 4.5, 4.0 and 3.4. These are good values historically, and they justify my decision not to make all polling weights equal; projections – each with their own polling weight – were the most accurate of the three estimates.

Nonetheless, these low mean and median deviations mask a fundamental divide in polling accuracy. Democratic Senate nominees overperformed WAPA by an average of 3.2 points in the 17 Democratic (Hawaii to New Hampshire), swing (Nevada, Pennsylvania, Wisconsin) and slightly Republican states (North Carolina, Arizona, Georgia). By contrast, in Florida, where Democrat Val Demings underperformed WAPA by 9.1 points, and the 16 most Republican states, Democratic Senate nominees underperformed by a mean 6.9 points, continuing a multi-cycle of polling missing Republican voters in the most Republican states. And while Democratic overperformance in the former set of states could be attributed to undecided voters breaking heavily Democratic, this is – at best – half the explanation in the most Republican states. Democratic overperformance is more likely to be evidence of Republicans trying to psyche out Democratic voters by “flooding the zone” with garbage and partisan polling, especially in the Democratic-leaning states of Colorado, New Hampshire and Washington, where Democratic incumbents Michael Bennet, Maggie Hassan and Patty Murray overperformed WAPA by a mean 6.7 points. Even in the six swing/slightly Republican states, Democratic nominees overperformed by 1.8 points, using Warnock’s 0.9-point ED margin over Walker. Put another way, in the 11 Senate elections outside of Florida receiving the most attention – the above nine plus Iowa and Ohio – Democrats overperformed by an average of 2.3 points, squarely in the middle of my 1-to-3-point estimate. Notably, these 11 Senate elections were polled a mean 39.2 times, while the other 22 elections (excluding 44 Florida polls) were only polled a mean 5.7 times.

Mean Democratic overperformance in the Democratic/swing/slightly Republican states is a similar 3.3 points for projections, with Republican underperformance in Florida and the most Republican states dropping to 3.5 points. Because these values include fundamentals, they theoretically include latent Democratic votes in the more Democratic states and Republican votes in the more Republican states, effectively disproving the “undecideds broke heavily in a partisan direction” hypothesis. Finally, Democratic Senate nominees overperformed fundamentals more in the 17 Republican states (3.2) than in the 17 Democratic/swing/slight Republican states (1.6).

The correlation between state partisan lean and polling over/underperformance is a very high 0.85, dropping only slightly to 0.65 for projections; there is essentially no correlation (0.04) with fundamentals.

Bottom line: As projected, Democrats will have 50 or 51 Senate seats following Election Day. And while I only “called” the Nevada Senate election wrong, the key takeaways are:

  1. Projections were extremely accurate, on average, followed by polling.
  2. Allowing polling weight to vary by state was the correct choice.
  3. Democrats overperformed projections and polls in Democratic/swing states, while underperforming in Republican states, including Florida.
  4. Democrats overperformed in the 11 most-polled states other than Florida by an average 2.3 points, in line with my pre-election conclusions.

***********

Governors. From the first gubernatorial distribution, published on September 26, Democrats were anticipated to end the 2022 election controlling 22 or 23 governor’s mansions (a net gain of 0 or 1), with the final projections giving Republicans a slightly better than one-in-four chance to net governor’s mansions. With the near-certainty Republican Mike Dunleavy has been reelected governor of Alaska, Democrats will soon control 24 governor’s mansions, one or two more than expected. And the only gubernatorial election I “called” incorrectly was in Arizona, where I gave Hobbs a 23.1% chance to win (Lean Republican).

Table 2 compares ED Democratic gubernatorial margins (as of 11:00 pm EST, November 20, 2022) to final estimates based on 1) fundamentals, 2) polling and 3) final projection average (average polling weight = 88%) projection. I exclude the three-way race in Alaska because of ranked-choice voting. No public polls were conducted of the gubernatorial elections in Hawaii and Idaho.

Table 2: Election Day Democratic gubernatorial margins compared to fundamentals, polling and projection estimates, sorted by state partisan lean

State3W-RDMED MarginED Margin minus…
   FundamentalsPollingProjected
Hawaii29.026.4-2.4n/a-2.4
Vermont28.9-47.0-61.8-6.6-13.0
Maryland26.231.75.72.52.8
Massachusetts26.128.93.03.13.1
California24.919.2-15.9-1.1-2.7
New York20.25.7-14.3-2.3-3.5
Rhode Island16.619.0-2.65.84.6
Connecticut13.912.9-11.2-1.7-2.3
Illinois13.311.5-12.0-2.4-3.7
Oregon10.13.6-6.32.91.6
New Mexico6.36.4-10.10.10.0
Colorado5.719.33.46.46.2
Maine4.514.3-0.46.65.7
Minnesota1.87.7-4.3-0.6-0.7
New Hampshire1.2-15.5-2.61.00.7
Nevada-0.5-1.4-11.10.4-0.9
Michigan-0.710.61.15.04.6
Pennsylvania-2.314.617.14.05.3
Wisconsin-2.43.4-4.43.53.0
Florida-5.5-19.40.2-9.3-8.1
Arizona-6.10.66.92.52.9
Georgia-6.4-7.56.0-0.6-0.1
Iowa-9.8-18.65.3-1.8-1.7
Ohio-9.8-25.5-1.6-5.4-5.1
Texas-12.0-11.015.1-2.8-1.0
South Carolina-15.9-17.312.7-9.8-3.8
Kansas-21.32.313.4-1.30.4
Nebraska-25.1-23.41.9-17.4-12.7
Tennessee-27.2-32.09.3-6.5-3.3
Alabama-29.2-37.75.6-1.6-1.1
South Dakota-29.6-26.816.9-12.9-9.1
Arkansas-30.3-28.02.5-8.5-6.6
Idaho-34.8-40.28.7n/a8.7
Oklahoma-37.8-13.738.2-9.3-0.9
Wyoming-47.5-58.33.3-4.1-4.0
Average  0.5-1.9-1.0
Median  2.2-1.3-0.9
Average – Abs Value  9.94.53.9
Median – Abs Value  6.03.13.2

My final gubernatorial projections – based upon all polls released after November 1, 2021 – were also extremely close overall, overestimating the final margin by an average and median of 1.0 points and 0.9 points, respectively. My conclusion that polls underestimated Democratic strength by 1-3 points again appears not to be correct, though, as polling averages underestimated Republican strength by 1.9 points, on average, with a median miss of 1.3 points. These relatively small misses could easily be explained by a late break of undecided voters for Republicans. Finally, the fundamentals – even using the more generous generic ballot poll estimate of R+0.2 – underestimated Democratic margins by an average 0.5 points, with a median underestimate of 2.2 points; these values rise to 3.8 and 5.5 with the current Republican lead of 3.3 points on the total House vote.

The story is similar when looking at how much ED margin differed from estimates, regardless of direction. The average absolute value of the differences was 9.9, 4.5 and 3.9 for fundamentals, polling and projections, respectively, while medians were closer at 6.0, 3.1 and 3.2. The much larger deviation from fundamentals (compared to Senate elections) reflect an astonishing overperformance by Republican incumbent Phil Scott in the very Democratic state of Vermont (-61.8), underperformances by Democratic incumbents in the solidly Democratic states of California (-15.9) and New York (-14.3), and much-stronger-than-expected Democratic performances in the swing state of Pennsylvania (17.1) and the very Republican states of South Carolina (12.7), Kansas (13.4), Texas (15.1), South Dakota (16.9) and, especially, Oklahoma (38.2). In fact, simply removing Vermont and Oklahoma lowers the average of the absolute value of fundamentals deviations to 7.3, though the median only drops to 5.7.

As with Senate elections, however, these low mean and median deviations mask a fundamental divide in polling accuracy. Democratic gubernatorial nominees essentially matched final WAPA (mean=D+0.4) in the 21 Democratic (every state from Hawaii to New Hampshire in Table 2), swing (Nevada, Michigan, Pennsylvania, Wisconsin) and slightly Republican states (Arizona, Georgia), again excepting Florida. In Florida, where Democrat Charlie underperformed WAPA by 9.3 points, and the 13 most Republican states, Democratic gubernatorial nominees underperformed by 6.2 points, nearly identical to the mean Senate deviation in these states. However, there is far less evidence of Republicans trying to manipulate polling averages in key gubernatorial elections, even if Democrats overperformed them in 16 of the most polled states (New York plus every state from Oregon to Texas in Table 2, excluding Florida) by 1.2 points – just inside my 1-to-3-point estimate. Elections for governor in these 16 states were polled a mean 31.2 times, while all other gubernatorial elections (excluding 21 in Oklahoma, 47 in Florida) were polled a mean 5.2 times.

Democratic gubernatorial nominees also matched final projections (R+0.2) in the Democratic/swing/slightly Republican states, while Republican underperformance in the most Republican states drops to 2.6 points. Clouding the issue somewhat, Democrats underperformed fundamentals by an average 5.3 points in the 21 Democratic/swing/slightly Republican states and overperformed fundamentals by 9.4 points in the most Republican states; the correlation between ED margin and fundamentals deviation is only 0.01, while the correlations with polling and projections are much higher, at 0.67 and 0.51, respectively.

Bottom line: As projected, Democrats netted governor’s mansions on Election Day, albeit one-to-two more than expected. And while I only “called” the Arizona gubernatorial election wrong, the key takeaways are:

  1. Projections were extremely accurate, on average, followed by polling averages.
  2. Fundamentals were somewhat less accurate, mostly due to extreme Republican strength in Vermont and Democratic strength in Oklahoma.
  3. Democrats closely matched projections and polls in Democratic/swing states, while again underperforming in Republican states, including Florida.
  4. Democrats overperformed in the 16 most-polled states other than Florida by an average 1.2 points, in line with my pre-election conclusions.

***********

I had planned to conclude with an assessment of individual pollsters, but I will now wait until there is more clarity on the total House vote. Instead, I leave you with this photograph of me as Steve Kornacki – a person for whom I have tremendous respect as a fellow political junkie and data narrator – on Halloween 2022.

Until next time….and if you like what you read on this website, please consider making a donation. Thank you.


[1] Usually a variation of “If the election were held in your district today, would you support the Democratic candidate, the Republican candidate or some other candidate,” though sometimes respondents are asked which party they prefer to control Congress.

[2] I assume Democratic incumbent Mary Peltola (48.7% first choice votes; 87% reporting) will win the Alaska At-large seat when second-choice votes are reallocated on November 23, while Democrat Adam Frisch conceded to Republican House incumbent Lauren Boebert in Colorado’s 3rd district.

[3] 5.5 for Democrats, -3.3 for Republicans

[4] Weight calculated using date of most recent poll, average pollster rating, and how far polling leader was below 50.1%. For details, please see earlier essays in the series.

One thought on “2022 Elections: A Post-Mortem (Part 1)

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s