Originally published at the Judy Ford Wason Center for Public Policy.

In July of 2018, my innovative forecasting model raised eyebrows by predicting some four months before the midterm election that Democrats would pick up 42 seats in the House of Representatives. In hindsight, that may not seem such a bold prediction, but when my forecast was released, election Twitter was still having a robust debate as to whether the Blue Wave would be large enough for Democrats to pick up the 23 seats they needed to take control of the House of Representatives and return the Speaker’s gavel to Nancy Pelosi.

Based on its 2018 performance, my model, and the theory that structures it, seem well poised to tackle the 2020 presidential election – 16 months out. I’ll serve up that result below, but first let’s set the table by reviewing my model’s 2018 forecasting success.

Not only did I predict that they would gain nearly double the seats they needed, but I also identified a specific list of Republican seats Democrats would flip, including some, such as Virginia CD7, that were listed as “Lean Republican” by the majority of race raters at the time. At a time when other analysts coded even the most competitive House races for Democrats as Lean or Tilt Democrat, I identified 13 Republican-held districts as “Will Flips,” 12 as “Likely to Flip,” and 6 as “Lean Democrat.” I also identified a large list of “Toss Ups,” from which I would later identify the remaining “flippers.” In addition, I identified some “long-shot toss-up” districts that could be viable flips under some turnout scenarios. Of the original 25 districts I identified as definitely or highly likely to flip, all but one, Colorado CD3, did so, possibly because the party failed to invest in their nominee there.

2018 House Predictions

The post-election diagnostics of my forecasting model, which departs significantly from the approaches used in conventional election forecasting models, such as those used by FiveThirtyEight, reveal just how powerful my model was at identifying the House districts and Senate races capable of producing Blue Wave effects powered by Trump backlash in the electorate. Indeed, the places I went astray in my final, “handicapped” predictions are races where I ignored the clear signals of my model, such as Georgia’s 6th congressional district, which my model was quite clear about flipping, and Kentucky’s 6th, which my model was quite clear couldn’t. Still, in other races, my manual handicapping was necessary, and correct, because despite its overall accuracy, my model underpredicts the Democrats’ two-party vote share in Utah’s 4th district.

Predictors of 2018 Democratic Party Vote Share

2018 Model

Looking ahead to the 2020 Electoral College map, my model delivers on two of the most critical elements of election forecasting: lead time and parsimony, that is, simplicity. It’s probably not lost on you, dear reader, that I am offering a forecast not for the presidential primary election, itself still in its infancy, but for the November 2020 general election that is some 16 months away. And I am offering a forecast free from all the trappings you are used to. There are no poll aggregators, no daily or weekly updates, no simple versus deluxe versions. Right now, there is not even a nominee! By and large, I don’t expect that the specific nominee the Democratic electorate chooses will matter all that much unless it ends up being a disruptor like Bernie Sanders.

Indeed, the only massive restructuring I might have to make to this forecast involves a significant upheaval like the entrance of a well-funded Independent candidate such as Howard Schultz into the general election, which our national survey in March shows would likely to pull 5 votes away from the Democrats’ nominee for each one vote it would pull away from Trump. Other potential significant disruptions might be a ground war with Iran, an economic recession, or a terrorist attack on the scale of 9/11. Otherwise, the country’s hyperpartisan and polarized environment has largely set the conditions of the 2020 election in stone. As unpopular as Donald Trump is today, and no matter how badly he trails his potential Democratic rivals in head-to-head polls right now, on Election Day Donald Trump will earn the vote of somewhere around 90% of self-identified Republicans. And as 2018 demonstrated, Republicans will increase their turnout rate over 2016. This, combined with a floor for Trump among Independents of around 38% (because of right-leaning Independents) and an infusion of cash that will dwarf his 2016 efforts, Trump has a floor that is at least theoretically competitive for reelection and will force Democrats to compete hard to win the presidency. The polarized era doesn’t produce Reagan Era Electoral College landslide maps.

Before revealing what my model has to say about 2020, I note one very important point of methodology. To construct predicted two-party vote shares for the Democratic Party’s nominee in each state, I use the best turnout estimate available for each state in 2018 for the Democrats. This is important because it allows me to capture the turnout surge we also saw among Republicans in 2018. Although I predicted an enormous surge in turnout among Democrats and Democrat-leaning Independents, the size of the corresponding surge among Republicans surprised me somewhat. I predicted the surge of Democratic turnout via negative partisanship, activated by the tangible threat of living under a unified government controlled entirely by Donald Trump.

What I did not anticipate was that, at least among Republicans, a threat response can be artificially generated at a mass scale and at a time when a party’s voters should be placated. Despite controlling the White House and both chambers of Congress in 2018, turnout surged nearly as much among Republicans, leading to the highest overall midterm turnout rates we have seen since 1914. Overall turnout ended up at a whopping 50.4%, tempting many analysts afterward to conduct comparisons between 2016 and 2018, a presidential-to-midterm comparison that is usually apparently absurd. Trump and the RNC accomplished this by running a base-centric mobilization campaign focused largely on stoking fear of immigration; a strategy they will replicate for 2020 while adding socialism into the mix.

Because my 2020 model relies on the 2018 vote to estimate the 2020 vote, it is naturally designed to account for this unexpected bipartisan turnout surge. As such, my expectation is the 2020 model will be better than the 2018 model, which was built with Virginia’s one-sided Democratic turnout surge as a turnout guide.

So, with no further ado:

Barring a shock to the system, Democrats recapture the presidency. The leaking of the Trump campaign’s internal polling has somewhat softened the blow of this forecast, as that polling reaffirms what my model already knew: Trump’s 2016 path to the White House, which was the political equivalent of getting dealt a Royal Flush in poker, is probably not replicable in 2020 with an agitated Democratic electorate. And that is really bad news for Donald Trump because the Blue Wall of the Midwest was then, and is now, the ONLY viable path for Trump to win the White House.

Why is Trump in so much trouble in the Midwest? First, and probably most important, is the profound misunderstanding by, well, almost everyone, as to how he won Michigan, Wisconsin, and Pennsylvania in the first place. Ask anyone, and they will describe Trump’s 2016 Midwestern triumph as a product of white, working class voters swinging away from the Democrats based on the appeal of Trump’s economic populist messaging. Some will point to survey data of disaffected Obama-to-Trump voters and even Sanders-to-Trump voters as evidence that this populist appeal was the decisive factor. And this is sort of true. In Ohio, Trump managed the rare feat of cracking 50%. Elsewhere, that explanation runs into empirical problems when one digs into the data. Start with the numerical fact that Trump “won” Pennsylvania, Wisconsin, and Michigan with 47.22%, 48.18%, and 47.5% of the vote, respectively, after five times the normal number in those states cast their ballots for an option other than Trump or Clinton. This, combined with the depressed turnout of African Americans (targeted with suppression materials by the Russians) and left-leaning Independents turned off by Clinton (targeted with defection materials by the Russians) allowed Trump to pull off an improbable victory, one that will be hard to replicate in today’s less nitpicky atmosphere. Yet, the media (and the voting public) has turned Trump’s 2016 win into a mythic legend of invincibility. The complacent electorate of 2016, who were convinced Trump would never be president, has been replaced with the terrified electorate of 2020, who are convinced he’s the Terminator and can’t be stopped. Under my model, that distinction is not only important, it is everything.

Trump’s second problem is that along with a turnout surge of Democrats that in many states like Virginia is simply larger than the turnout surges of Republicans because of demographics, he is deeply unpopular among Independents because of all the abnormal, norm-breaking and according to the Mueller Report, even illegal things, he does as president. This has left him with an abysmal approval rating averaging just 34.8% in 2019 among Independents, who largely broke against Republicans in the 2018 midterms as my theory predicted. In a follow-up piece to this forecast, I will show that much of this swing among Independents is actually the product of their own turnout surge, which brought more left-leaning Independents out to the polls by the same negative partisanship mechanisms that moved their partisan counterparts. This is why even the Democrat’s sharp drift to the left as they chase their party’s nomination, following the Republicans down the path of ideological polarization won’t have the impact on the vote choice of Independents Republicans are hoping for in 2018.

At the end of the day, Independents will be asked to weigh what Democrats might do against what Republicans, particularly Trump, are doing; the reverse situation from 2016 when Democrats suffered from the referendum effect among Independents. Even if the Democrat’s nominee is unabashedly liberal, it is not likely Trump can win a referendum among college-educated Independent voters without a dramatic transformation in both tone and style.

Republicans can survive an under-maximized Democratic turnout surge, like the one we saw in 2018 (I’ll be showing this in forthcoming work), but not one that it is combined with the loss of Independent voters and not one without a corresponding Republican turnout surge which can only be accomplished via things likely to further isolate Independent voters and agitate Democrats. 

Does the Democrat’s nominee matter? Sure, to an extent. If the ticket has a woman, a person of color or a Latino, or a female who is also a person of color, Democratic Party turnout will surge more in really important places. If the nominee is Biden he’d be well-advised to consider Democratic voter turnout his number one consideration when drawing his running mate to avoid the critical mistake made by Hillary Clinton in 2016. This is true for any of the white male candidates. If the nominee hails from the progressive wing of the party, it will provoke massive handwringing both within the party and the media that if not controlled could become self-reinforcing. But the Democrats are not complacent like they were in 2016 and I doubt there is any amount of polling or favorable forecasts that will make them so. That fear will play a crucial role in their 2020 victory. We will not see a divided Democratic Party in 2020.

A note on Florida: As it was in 2018, my model is convinced that Florida is going to break in favor of the Democrats. After 2018, I am less convinced, but that is because I know something about Florida that my model does not know: Florida is really, really old. No state hosts more members of the Silent Generation, and Silents are more conservative, more white, more Republican, and thus less hostile to Trump than their generational counterparts. Not only are seniors realigning to the Republican Party, they are also the nation’s most reliable voters. The other issue with Florida is that as white, non-college educated voters, especially older ones, become more Republican, Democrats become increasingly reliant on the turnout of young and/or Latino voters to make up the difference. As of 2018, Democrats still have not cracked the code on getting either young people or Latino voters mobilized (although I will show you that they made gains in 2018 that were critical to their victories). Until they do, states like Florida, Georgia, and Texas remain highly attractive pitfalls.

One more thing: We are in the midst of a long-term, multifaceted coalitional realignment in both parties (among other things) that the academic version of this work tackles. Trump is a product of this realignment and of the hyperpartisanship and polarization that accompanies it. He is not a cause! But he is an accelerant, like kerosene on smoldering coals. The realignment that was puttering along has picked up tremendous speed. My analysis of the House races reveals that the political parties have not entirely adapted to these changes. In 2018, Democrats failed to contest several Republican-held districts with ideal demographics. One of these, Texas CD24, has already drawn a strong challenger. I’ll soon post that material, as well as a deep dive into the role that increased turnout in 2018 played in Orange County, California, where my long-time fans will know I was the very first to predict that Democrats would overrun what had been Reagan Country for decades. As the other Rachel says, watch this space.