top of page
Writer's pictureJeff Burdick

Why most 2022 midterm forecasts collapsed so predictably

The job description of prognosticator is pretty straight forward, requiring just a single qualifier: To predict. Correctly.


So what does it say that for most of the last two years the entire prickle of U.S. political prognosticators so confidently – but wrongly – predicted the GOP would dominate the 2022 midterm elections? The only question among them seemed to be whether it would be a red wave or red tsunami. Meanwhile over that same 1½ years, this publication has been pointing out just how thin and fundamentally flawed all this Beltway group think was.


If the GOP fails to win either house of Congress, it will be just another high-profile failure by the national prognosticators and pontificators. Recall they predicted Hillary Clinton would cruise into the White House in 2016. Then supposedly after adjusting their polling models, they forecast an easy double-digit Biden victory in 2020. (His final national vote margin was 4.4%, and he narrowly won the electoral college by a combined 43,000 votes in Arizona, Georgia and Wisconsin.) They also confidently projected 2020 House gains for the Democrats, who instead lost 11 seats and almost the majority.


That’s three consecutive high-profile failed predictions. Given all this, you’d think every prognostication web site and article should feature a prominent qualifier stating “Poor past performance is no guarantee of future accuracy.” Instead, here is a sampling of their analytical errors so far this cycle:


  • On April 1, after completion of just one state primary, Amy Walters of The Cook Political Report put a fork in the midterms thusly: “Every metric we use to analyze the political environment — the president’s approval rating, the mood of the electorate, the enthusiasm gap — all point to huge gains for the GOP this fall.” (Emphasis mine.) But now Walter acknowledges the success of so many Trump-endorsed extremists in GOP primaries has probably lost the GOP the U.S. Senate and maybe even U.S. House control.

  • On July 26, Nate Silver’s FiveThirtyEight predicted a just 16 in 100 chance for Democrats to hold the House. “Things do not look good for Democrats in the House. Even if Democrats were to win all the races currently designated as toss-ups [14], plus hold onto all the seats they’re favored to win [198], they would still wind up short.” However the site recently conceded the expected GOP “enthusiasm advantage” more than reversed after the Dobbs decision. This means the site’s designation of toss-up, leaning and likely districts need a thorough re-do. Yet, it still gives Dems just a 26% chance in the House.

  • Even as late as Aug. 12, Steve Shephard of Politico.com didn’t foresee any dynamics that could change “national headwinds” that leave “Republicans well positioned to wrest back control of the House, and quite possibly the Senate.” However, Monday the site changed its Senate prediction to from “lean” Republican to “remains a toss-up.” Remains a toss-up? Is Politico trying to hide its previous predictions favoring GOP winning back the Senate?

Other bellwether outlets and rating services peddled similar verisimilitudes with the same great assurance. These include CNN, Fox News, Sabato’s Crystal Ball, the New York Times and Washington Post. The rest of the political industry in turn parroted these predictions just as confidently. But then August 2022 happened.


That’s when beginning on Aug. 2 Kansas voters rejected by an astounding 18 points a referendum to amend the state constitution to allow for a banning of potentially all abortions. And despite all the talk of Biden’s low approval numbers being an intractable anchor for Democrats, the albatross necktie now appears firmly on the Republicans with special-election Republican landslides evaporating in Minnesota and western New York. And “likely” red districts in New York’s Catskills and Alaska’s at-large seat actually going for the Democrats.


At the same time, there was also Sen. Mitch McConnell admitting poor “candidate quality” among GOP Senate nominees made it less likely the GOP will win back that chamber. This was a direct reference to Trump’s radical candidates winning so many primaries. Across the country many key Republican candidates were also shockingly behind in fundraising and not on TV as their Democratic opponents were hammering home a turnout motivating post-Dobbs message. And none of this was unpredictable, unless of course you are a professional political prognosticator.


SactoPolitico’s Analysis


Predictably, most bellwether outlets are presently revising their predictions, but changing your predictions after your predictions have proven misguided doesn’t count as a new prediction. It represents the running of a correction. (Remember that job description: to predict, correctly.)


But let’s be clear about what SactoPolitico has been correct about so far. My prediction wasn’t a Democratic rout instead of a Republican rout. That would have been foolhardy. No, my analysis merely showed a Republican rout seemed highly unlikely and the midterms would be very competitive. Plus, if one had to chose whose cards to play, I consistently said I’d rather play the Democratic hand. The surge in inflation was one dynamic that helped the GOP hand, but pretty much everything else has helped the Dems.


My counter-analysis began in March 2021 with a piece titled “Are midterm losses inevitable for Democrats? Maybe not.” This was followed two months later by what proved my unique and prescient analytic take on GOP gerrymandering titled “Despite Census news, GOP House gains not guaranteed.” These were then followed by some regular updates to see how our midterm analysis was holding up, and each time the many flaws in the Beltway consensus became ever clearer.


It’s important to note it wasn’t that SactoPolitico analyzed the same facts and came to a different conclusion. No, I simply refused to base our prediction on surface-level “trends” and group-think assumptions. I bore deeper and discovered fascinating countervailing factors at play — all of which you would think would be standard operating procedure for the so-called professional prognosticators. Let’s examine some of these factors that the other outlets lazily failed to examine or suss out.


Historical Trend vs. Rule


The major errors in the consensus view were founded largely on two big analytic whiffs. The first involved a fascinating historical trend long known within the Beltway but not outside it. This is the indisputable fact that the party in the White House generally loses midterm House seats. Going back 150 years to the Grant Administration, this has happened 89% of the time, or 34 of the last 38 midterms. That’s a pretty strong trend, right?


However, pretty much every outlet stopped there and locked it in stone as an almost invariable “rule.” But at SactoPolitico, before determining how heavily to weight this trend, I did what any self-respecting forecaster should always do. I scrutinized when the trend did not apply – specifically those four midterm exceptions to see if they might tell us anything of value. And not surprising, they did. It turns out the four exceptions were not randomly spread over time. Two occurred in just the last six midterms (1998 and 2002). Cause for pause? Without a doubt. Further, both were influenced by unprecedented political events. In 1998, the GOP began its march to impeach a popular President Bill Clinton for lying about personal behavior, which backfired on the GOP in that year’s midterms. Then in 2002, President George W. Bush rode post-9/11 fears to gain eight House seats. So might the historic Jan. 6, 2021 Capitol insurrection and continued Republican support for the stolen election Big Lie spur one more exception to the midterm rule? Let alone an expected conservative Supreme Court ruling against Roe v. Wade? Clearly, that possibility could not be ruled out.

Potentially diminishing the historic midterm trend further, my independent analysis also found that a different historical House trend was broken in 2020. Going back 90 years, nearly 75% of the time the party that wins the White House also adds House seats that election. But this didn’t happen for Biden, with the GOP netting 11 seats in 2020 by flipping many narrowly held districts. So might this mean the GOP already got its midterm bounce two years early? Again, one could not be certain, but it was another tantalizing countervailing factor any self-respecting political prognosticator should have considered.

GOP redistricting gains evaporate


Another major basket the national prognosticators put most their eggs in was a huge presumed advantage the GOP would reap from the once-a-decade reapportionment of House districts among states based on the 2020 Census and the related redrawing of House district lines. However SactoPolitico’s state-by-state analysis found both conclusions highly flawed for very clear reasons.


In terms of gerrymandering, the GOP has a well-earned reputation for very successfully and shamelessly leveraging gerrymandering to its limits. So many prognosticators presumed the same would occur again this cycle. Many even went to so far as to predict that GOP gerrymandering gains would give them unbeatable House control. However, the GOP’s past successes have also left little new juice to squeeze from this tactic, and in most every state I examined, I found the only way for Republicans to expand their gerrymandering fortunes was by making a lot of safe incumbent seats less safe.


And there was the key snag. Simple human nature means all incumbents want less electoral risk, not more — the good of their party be damned. Plus, incumbent members of Congress have a lot of sway in their state parties on redistricting. So even if it made tactical sense to the party to spread Republican voters more widely to compete for more seats, few incumbents are eager to see their +12 or +9 GOP advantages in their district dilute to just +8 and +5. And given that the Republican share of voter registrations has been mostly shrinking nationwide (as the independent category has grown), the risk to GOP incumbents this cycle is even more acute.


So as I expected, most states with Republican-controlled redistricting committees prioritized protecting incumbents over expanding the competitive map. This included a need to protect those net 11 new House seats won narrowly in 2020. Thus a major net GOP redistricting advantage never materialized. Even the gurus at Cook Political Report – who were once quite bullish on GOP redistricting fortunes – conceded by January that “redistricting looks like a wash.” Second, all those predicted GOP reapportionment advantages also disappeared. It was true more red states added House seats after the 2020 Census, and more Democratic states lost seats. However, simplistic assumptions that the loss of a seat in blue states like Illinois and California would inevitably steal a seat from the Democratic column never transpired for very predictable reasons (as covered in this piece).


An example of why can be found in Texas, which was awarded two more House seats for this cycle after adding four after the 2010 census. But as Texas' 2012 experience shows, it’s no gimmee that those two extra 2022 seats will net the Texas GOP two more GOP members to the state delegation. To do so would require significantly spreading an already thinning Republican registration advantage among two more Congressional districts. This was attempted and failed 10 years ago. That’s when the addition of four seats to the Texas delegation actually resulted in the Republican advantage in Texas’ House delegation dropping from +14 to +12.

Dobbs decision


Nowhere has Beltway group think been more head-scratching than in not anticipating the huge impact of a repeal of Roe v. Wade. And the Supreme Court’s Dobbs decision didn’t come out of left field. Court watchers have been warning since last year that the Alabama and Dobbs cases could both lead to a major or complete rollback of Roe. Yet even when the Supreme Court’s draft ruling leaked in May, the Beltway handicappers were still assuredly saying the issue was not a political game changer. When the Dobbs decision was handed down two months later, the political high priests again said, “Won’t be a factor.”


But how wrong they were.


Just how major a failure this was cannot be underemphasized. Just consider how so many prognosticators actually believed the chaotic withdrawal of America from Afghanistan in 2021 would be of greater consequence to the midterm outcome than the repealing a nearly 50-year-old right of women’s to control of their own medical decisions.


One can’t even call this a failure of imagination. The game-changing issue was there for more than a year for everyone to see. SactoPolitico saw it and cautioned about it way back in January: “What if the Supreme Court overturns Roe v. Wade entirely or mostly over the objections of at least 60% of Americans? How many of those valued suburban female voters do you think the GOP will lose?”


That Referendum on Biden


The next most heavily cited midterm disadvantage for Democrats has long been Biden’s low approval ratings. But once again, prognosticators seemed to take this disadvantage as a settled matter of fact with no accounting for any variant evidence. After all, one only had to look at Trump’s low approval ratings in 2020. For most of that year, Trump was under 40% too. Yet his party still gained House seats in 2020 despite that election being even more of a referendum on Trump. From my analysis, Biden’s approval ratings contained a lot of interpretive uncertainty. First as compared to Trump’s approval numbers, Biden’s ratings seemed comparatively too low because Democrats are more likely to express unfavorable attitudes toward their party’s presidents than Republicans are. So although Biden’s low approval numbers weren’t a good thing for Democrats, it wasn’t possible to confidently correlate them directly to any change in voter turnout or to independent voter decisions in swing districts. Related to this was the “generic ballot” argument. Because 435 House districts are too many for any outlet to poll with any accuracy, some pollsters prefer to give extra weight to voters’ preference for a generic Democrat versus a generic Republican to represent them in Congress. But this doesn’t really improve forecasting accuracy much because even when you isolate the generic-ballot question to just voters in swing districts and swing states, those voters don’t ultimately choose between two generic candidates. Instead they chose between candidates with very specific positions, personalities and campaign characteristics.


And this is where the success of all those Trump-endorsed extremists threw a wrench into any value for this “metric.” Trump-endorsed Republicans aren’t considered “generic” Republicans, and this requires prognosticators to roll up their sleeves and study those 30-40 swing districts far more closely.


The 50-50 Myth


News consumers also hear regularly that America is a 50-50 nation politically. Fifty percent lean Democratic, and 50% favor or lean toward the Republicans. But any political demographer and campaign pollster will tell you this is not how they view the electorate. That’s because generally anywhere from 15%-30% of voters in any state or House district are unaffiliated with either party, and this category keeps expanding every cycle.


So right there, throw any 50-50 conception out the window. Each party needs to turn out their solid supporters, but all close elections are determined by how these unaffiliated voters swing. Any prediction that doesn’t attempt to isolate and study these voters’ attitudes will always be superficial at best.


Plus this cycle, the GOP had to deal with the scorched-earth primary tactics by most Trump-endorsed extremist candidates. Where Trump’s candidate won, this risks losing not just more unaffiliated voters but also moderate GOP voters (the so-called unpatriotic RINOs) to the Democratic candidate. Or some moderate just don’t turn out.


What happens in November remains to be seen, but these dynamics were clearly seen in the rank-choice results of Alaska’s special election to fill its sole House seat. Alaska has nearly twice as many registered Republicans than Democrats, but 44% of all Alaskan voters aren’t affiliated with either national party. After the first round of voting, the Democrat Mary Peltola was the top vote getter with 40.2%, but her two Republican opponents (Sarah Palin and Nick Begich) combined for nearly 60% of the vote. But in later rounds, nearly 30% of Begich’s voters listed Peltola as preferred to Palin, which ultimately gave Peltola a 3-point victory over the far more extreme Trump-endorsed Palin.


Every individual special election has its peculiarities, and this Alaska race is no different. But the risk of Trump-endorsed candidates losing independent and moderate Republican voters should have always been a real factor for prognosticators to consider. However what we regularly heard (until recently) was how Republicans had an at least 6-8% enthusiasm advantage that would turn even districts that went strongly for Biden in 2020 into GOP midterm flip possibilities. The current dynamic shows the direct opposite at play with Democrats expanding the competitive map and not the Republicans – something that again should have been contemplated by any expert forecaster.


District-Level Polling


In all these ways, most prognosticators erred by consistently overweighting Democratic negatives and vastly ignoring or underbaking GOP disadvantages. Hurting the national prognosticators further is the absence of any good independent district-level polling or even much strong local reporting on these swing races.


Consider last weekend, USA Today published a list of 12 key House races to watch. Half of these districts included parts of major metropolitan areas like Las Vegas, Cincinnati and Denver, but no independent polling existed by any local media outlets, as few budget money anymore for local polling.


Sometimes, individual campaigns and the national parties will share their polling when they think it favorable to their side. But they usually don’t share all the details on methodology or any details that don’t match their preferred spin. So this data is often not terribly useful for informing forecasts about those key swing races.


Also missing is strong political campaign coverage by local media outlets. This used to be a local media staple with any local reporter loving a good contested race. But nowadays getting good, timely political reporting of any kind has become rarer and rarer outside the largest media markets. And with little savvy local political insight to guide it, most national predictors fall back on highly flawed national “metrics” and assumptions.


Take California’s political capitol of Sacramento. One of the possibly pivotal House races involves the newly redesigned CA-03 that sprawls across 10 mostly rural counties, but whose population centers on the northeast suburbs of Sacramento. Trump would have won there in 2020 by just 1.9 points, yet, the major local Sacramento newspaper, radio and TV stations have covered little of the race. Maybe that changes post-Labor Day, but with mail-in ballots going out next month, there is little to go on in this race.


In many ways, this seat is a microcosm of the dynamics that will determine who ultimately controls the U.S. House. The open seat pits the Trump-endorsed Republican State Assemblyman Kevin Kiley against the Navy veteran moderate Democrat Dr. Kermit Jones. More than $3 million has been spent on the race so far. The Democrat has a decent ground game, has focused on district-level issues, and has hammered away at the Trump connection and abortion. The Republican has avoided most district issues except inflation and gas prices and has tried to make the race a referendum not on Biden but California Gov. Gavin Newsom.


All indications are this will be a closely run race. But if the Dobbs effect and anti-Trump sentiments seen in special elections prevail, the Democrat is positioned to benefit. However, all the big ratings services list the seat as “likely” to go Republican. FiveThirtyEight even predicts the Republican to win by an astounding 8 points. Such a prediction is not based on any polling, not on any historical data, not on any local reporting, just on very broad – and outdated – assumptions that Republicans still enjoy a big “enthusiasm advantage.”


I have reviewed the ratings of other races in other states that feature many of the same suspect analysis. This has led this publication to believe there will be a lot of “astonished” forecasters on election night who are surprised at how competitive many “likely” Republican seats turned out to be.


Ghost of 1948


This doesn’t mean the Democrats are a lock to hold the House or the Senate. Remember our forecast was (and continues to be) this will likely be a very competitive midterm, but if we had out choice of hands to play, we’d still take the Democrats’ cards.


Even though many outlets are now softening their references to red waves and tsunamis, I still continue to be an outlier with this election forecast. Many have been doubling down on their bad bets for so long, it is hard for them to switch to a more tempered “wait and see” attitude.


For awhile there, I was actually tracking how similar the national forecasters’ 2022 predictions were to that most famous of political prediction flubs: President Harry Truman’s 1948 re-election victory. Political junkies remember the famous Chicago Tribune “Dewey defeats Truman” front page, but the Tribune was far from alone. Four weeks before, Newsweek featured predictions from 50 political experts, and unanimously, all 50 forecast an easy victory for New York Gov. Thomas Dewey.


In fact, the drumbeat against Truman was so strong that a month earlier the survey firm Roper even announced it would not publish any more polls on what it considered an all-but-decided presidential contest.


The main difference between 1948 and 2022 is all those August special elections this year provided a real wake-up call for most forecasters. None appear so far to understand why their forecasts were so predictably wrong for so long, but as a final prediction, I won’t be surprised if most scurry back by election day to something closer to “toss-up” territory.


Comments


bottom of page