One of the biggest flashpoints in the 2020 Democratic presidential primaries was the notion of electability. Candidate after candidate was assessed by primary voters on the basis of who would be most likely to win the general election against Donald Trump, and entire campaigns floundered or surged around their answers to the question.
Critics of the concept frequently argued that it was nebulous, unquantifiable, and selectively picked to hurt specific candidates. Elizabeth Warren’s campaign, in particular, was a particular flashpoint for this issue — she severely underperformed electorally in her 2012 and 2018 Senate runs, leading many to speculate that she was not electable. This was a notion further reinforced by a New York Times poll showing her lagging behind both Biden and Sanders in head to head polling against Donald Trump. In a sharp contrast to the Democratic primary base’s obsession over this concept, however, several political pundits argued that electability was a myth and that candidates could easily break through the conceptions created by it if given the chance.
To answer whether this notion is real, it may be useful to examine whether past electoral overperformance is predictive of future ones. If electability is a thing, one would expect that a candidate’s overperformance in a given cycle generally could and would be replicated to some degree in future ones, even in different environments, whereas the lack of a link would indicate that overperformances are mostly just the product of cycle-specific noise.
Readers may remember that we have two sets of Congressional models to help us assess electoral overperformance: a 2018 Wins-Above-Replacement model (interactive version) and a 2020 one (interactive version). In these, we tried to quantify how well a candidate did electorally in their race relative to the environment, the district specifics, and the resources available to them. To help gauge whether the notion of electability is real, we can examine the overperformances of congressional incumbents that ran in both cycles and see whether there is a clear link between 2018 and 2020 overperformances. Concretely, the question we will try to answer is: Were incumbents who overperformed in 2018 significantly more likely to overperform in 2020, and if so, to what degree?
Below, we show our correlation plot, which visually implies a clear and obvious link between 2018 overperformance and 2020 overperformance. We find that those who overperformed electorally in 2018 were substantially more likely to overperform in 2020 and retained a significant amount of their fundraising-independent electoral strength across cycles, while those who underperformed in 2018 were also more likely to do so in 2020.

The correlation between past and future overperformance is visually evident, and the former does appear to be decently predictive of the latter. In fact, of the 40 multi-term incumbents that overperformed by 5 or more points in 2018, 36 overperformed relative to expectations in 2020. Meanwhile, of the 45 that underperformed by 5 or more points, 38 underperformed relative to expectations in 2020. This is a significant point in favor of the existence of cycle-independent “electability” — there are certain candidates that consistently play better with the general electorate, and this is reflected in election results. While past performance certainly cannot perfectly predict the future, it does seem to serve as some indication of the cycle-independent strength of the candidate.
The question of what exactly makes a candidate electable is a different question altogether and arguably even more difficult to answer — prior research from both us and others in the field at suggests that electoral overperformance is decently correlated with moderation. While the penalty for ideological extremism has significantly declined over time, it still exists and is magnified the farther up the ballot one goes, as recently illustrated in the linked paper by Chris Warshaw and Devin Caughey.
It is not impossible to be on a partisan ideological extreme and overperform in a general election; however, it is significantly more difficult, because base voters generally tend to turn out and vote for whoever the party’s nominee is, and the most votes are thus often gained and lost among the persuadable cohort of swing voters. Moreover, electability is not necessarily static and the perception of moderation can change for candidates on a cycle-by-cycle basis depending on their actions — those who take dangerous positions, like Tom MacArthur when he led the 2017 repeal effort for the Affordable Care Act, can see past residual strength evaporate quickly. But the question of what exactly makes a candidate more or less electable is a different question altogether from whether or not electability is a real concept, and it is one that has been surveyed and covered by us extensively already.
In terms of implications, the next question is whether this tells us anything for 2022. To this, we would say that past underperformances should serve as a red flag for Democratic incumbents in 2022, as heavy spending and strong Republican challengers could easily result in losses even in districts that were somewhat comfortably won by Joe Biden. Examples of extra-vulnerable incumbents are members like Kim Schrier (WA-08) and Mike Levin (CA-49). On the flip side, a candidate like Jared Golden (ME-02) could prove a bit more difficult to dislodge if his above-average performance in 2020 is any indication, even in a district that is significantly less friendly presidentially towards Democrats than Schrier’s.
There aren’t many better predictors of a district’s congressional lean than its presidential margin, and if polarization accelerates, all other factors will continue to diminish in importance. However, dozens of races are still decided by single-digit margins, and factors like electability do still move the needle by enough to make the difference in several seats. And so perhaps the past may not be the worst signal for parties looking to win the future — electability is real, and financially targeting or shoring up consistent underperformers to account for this may be key to deciding control of the House in upcoming years.
I’m a software engineer and a computer scientist (UC Berkeley class of 2019 BA, class of 2020 MS) who has an interest in machine learning, politics, and electoral data. I’m a partner at Split Ticket, handle our Senate races, and make many kinds of electoral models.