Skip to content

“New research suggests Sanders would drive swing voters to Trump — and need a youth turnout miracle to compensate.”

Political scientists David Broockman and Joshua Kalla write:

Decades of evidence from academic studies suggests that more moderate nominees tend to perform better in general elections than more ideologically extreme nominees. For example, [political scientist Alan Abramowitz found that] Democratic US House candidates who supported Medicare-for-all fared approximately 2.2 percentage points worse [on average] in the 2018 midterms than candidates in similar districts who did not. . . .

This makes sense to me. Just to clarify that last point: they’re not claiming that the act of supporting Medicare-for-all caused an average 2 percentage point drop in the vote; rather, they’re saying that candidates with that political position received on average 2 percentage points less of the vote, compared to candidates in similar districts who did not. The candidates who supported Medicare-for-all were, I assume, more left-wing on average than other issues as well. In this case, I’d take the position on Medicare-for-all is an (imperfect) measure of left-right ideology.

From that perspective, the 2 percentage point difference makes sense and is consistent with what we’ve found before on the small average electoral benefit for moderation (see here, for example) and the graph shown above, which appears in Red State Blue State.

2 percentage points of the vote is not much—and for the presidential race I’d expect a smaller effect, maybe 1 percentage point—but in some recent elections such as 2000 and 2016, a shift in vote of 1 percentage point could be enough to swing the election, so I agree with Broockman and Kalla that this is worth talking about. For example, after the 2016 election, I speculated that “Sanders’s ideological extremism could’ve cost the Democrats a percentage or two of the vote. . . . [but] Sanders could’ve won a million or two votes less than Clinton, and still won the election. . . .”

Broockman and Kalla’s new work on this topic focuses on an analysis of recent trial-heat polls. They write:

But early polling testing how Democratic nominees would fare against Trump suggests a different conclusion: Bernie Sanders, the most left-wing candidate in the Democratic primary, polls as well against Trump as his more moderate competitors in surveys. . . . Why does Sanders look similarly electable to leading moderates in polls against Trump? We fielded a 40,000-person survey in early 2020 that helps us look into this question with more precision. . . .

They conclude that, in these polls, Sanders does about 1 percentage point worse on average among existing voters compared to more centrist alternatives (Buttigieg, Biden, or Bloomberg; they didn’t check Klobuchar), but that, in the poll, “the large number of young people who say they will only vote if Sanders is nominated is just enough to offset the voters Sanders loses to Trump in the rest of the electorate.”

Broockman and Kalla conclude:

Democrats should not be very reassured by early polls that find Sanders faring as well against Trump as the more moderate candidates: These numbers may only look decent for Sanders because they assume he will inspire a youth turnout miracle. . . . The gamble Democrats supporting Sanders based on his early polls against Trump must be ready to make is that, despite the evidence to the contrary, the lowest-participating segment of the electorate will turn out at remarkably high rates because Sanders is nominated.

Mostly this makes sense, except that I think they’re overstating the case by framing things a bit too deterministically. The benefits of moderation only appear on average. I think it would be more accurate to say that Democrats supporting Sanders are also hoping that he will perform better among existing voters than is shown in these polls. But I agree with their general point that, if you want to base your analysis on these polls, it makes sense to try to untangle support among regular voters from anticipated changes in turnout.


  1. Zhou Fang says:

    I’m not personally a fan of Sanders but I do wish Broockman and Kalla did not bury in their Vox writeup the effect they estimated:

    “However, when we do not make these assumptions and instead (1) weight our data to the 2016 electorate (downweighting young people) and (2) disregard self-reported turnout intentions by imputing votes to partisans who currently refuse to express
    a preference, the more moderate candidates win more votes against Trump than Sanders does by a statistically significant ≈2 percentage points.”

    I.e. the correction makes about a 1% difference. Enh.

    • Andrew says:


      Yes, in my post I said, “for the presidential race I’d expect a smaller effect, maybe 1 percentage point—but in some recent elections such as 2000 and 2016, a shift in vote of 1 percentage point could be enough to swing the election…”

      • Zhou Fang says:

        There’s just so many unknowns. I don’t have the time to try and build a model of “probability of victory conditional on the February head to head poll margin”, but I can’t imagine it can really be that large an influence, that 1%.

        Let’s just say I prefer the window title of “Can Bernie Sanders beat Trump? The data is complicated” to the textual headline of “Bernie Sanders looks electable in surveys — but it could be a mirage”.

        • Andrew says:


          I agree with your preferred headline. The larger point is that we have prior reason to think that a more extreme candidate could be expected to get something like 1 or 2 percentage points less of the vote, compared to a moderate candidate, and these polls are consistent with that expectation. But there’s lots of uncertainty, so the “Sanders or Warren will do worse” expectation is just an expectation, not a known effect.

          • Ian Fellows says:

            idk, it seems almost like statistical malpractice to me. Saying that a candidate doesn’t perform as well when we impute the voting patterns observed during a different election with a candidate that had a vastly different ideological viewpoint/coalition strikes me as misleading. This is especially true since the actual expressed voting preferences of the individuals surveyed are explicitly thrown out.

            You could have also argued in 2016 that trumps electability relied on a decrease in black voting turnout. re-weighting the polls (remember that whole thing) to the 2012 levels, Hillary would clearly win. After all, what difference is there between Hillary Clinton and Barack Obama? I certainly can’t think of anything.

            • Andrew says:


              Just based on the traditional fundamentals, Clinton in 2016 had the advantage of Obama in 2012 of a stronger economy but the disadvantage of not being an incumbent. Clinton did win the popular vote and only lost the electoral vote through an unusual circumstance regarding geographic distribution of votes. Both Clinton and Obama are center-left. The ideological positioning model suggests that either Clinton or Obama would be expected on average to do better than a more left candidate such as Warren or Sanders, and they’d be expected on average to do a bit worse than a more centrist candidate.

              One of the difficulties of applying this model to the 2020 election is that, although there are two far left candidates running, there are no centrist Democrats remaining in the race. (I’d label Biden as center-left, comparable to Obama and Hillary Clinton.) So I’d expect any effects of ideology (for example, comparing Sanders to Biden, Buttigieg, or Klobuchar) to be small. I did not think about that when writing the above post.

              • Ian Fellows says:


                I think you are right in your thinking about the fundamentals. I think that we should be very skeptical of “unskewing” polls though. If a pollster in 2016 just assumed that African Americans would turn out for Clinton at the same rates that they turned out for Obama ignoring what they said they were going to do, I’d have taken their Statistician card away from them.

                To me, there doesn’t seem to be a big conflict between a positioning model of behavior and a real Sanders advantage. Under a left-right ideological model with a constant electorate the candidate closest to the median voter will tend to win. If voters only turn out for candidates within X ideological points of their position and the left-right distribution of the population is bimodal, then the extreme candidate will tend to beat the moderate. Isn’t this all just the age old tradeoff between inspiring your base to vote or broadening it?

                Saying that Sanders would inspire the base and a centrist would broaden it is pretty obvious. Assuming that he wouldn’t inspire the base in order to suggest that a centrist would perform better is, at best, questionable.

              • Zhou Fang says:

                The primary issue with poll “unskewing” is the imposition of (often very wild) assumptions on to a whole system of weighted polling that the unskewer doesn’t understand. However, a reasonable(-ish) assumption for turnout patterns is that they stay fairly consistent from election to election, so while we probably don’t believe things will be exactly the same, it is probably not illogical to want to incorporate them into the priors and suppose that a best estimate for the next election will be somewhere between what happened last time, and what current polls indicate.

                There’s a basis for sensitivity checking even if you don’t necessarily accept the youth turnout assumption as entirely true. That said, my takehome message from this work is that the turnout assumptions aren’t really massively significant from a practical election prediction POV.

                The ‘extremist penalty’ argument is mainly about extremists spurring opponents to turn out.

        • jim says:

          Also, headlines like:

          “we might know something but we’re not sure”

          Don’t sell clicks, newspapers etc.

          I wonder if the press’s focus on hyping science and research is a natural result of overall improvements in society – i.e. as body counts from crime and wars have fallen, there’s just not that much to report anymore. Research, with it’s many unknowns, presents a great opportunity for imaginative headlines.

        • Z says:

          > I prefer the window title of “Can Bernie Sanders beat Trump? The data is complicated”…

          I wouldn’t phrase it as “Can he”, since I think we know there’s at least a reasonable chance that he can. I would phrase it as: “Is Bernie less electable than other Democratic candidates? Probably at least a little”. And the obvious caveat applies that the probability statement is only a statement about how candidates who are like Bernie in certain *but not all* ways have performed *on average* in the past.

  2. A. Donda says:

    “Sanders’s ideological extremism” – the fact that this phrase is nonchalantly used, and without making a joke either, only shows how f***ed up your countries’ politics are.

    • Adede says:

      Yep. What about Trump’s ideological extremism driving people to Sanders?

      • More Anonymous says:

        Adede, note that Republicans and Independents on average view Trump as somewhere between “slightly conservative” and “conservative”, not as “extremely conservative”. For example, in Figure A1 of the paper Andrew’s post is about, you can see that Republicans view Trump as being as conservative as they think Bloomberg and Biden are liberal.

        Some Independents might be driven to Sanders by views that Trump is ideologically extreme… but there seems to be a lot of confusion among Independents, with many even viewing Sanders and Warren as conservative and Trump as liberal-ish. Although I guess that could be an artifact of oversmoothing in Figure A1, I can’t tell.

        I have a lot of problems with the paper under discussion, but Figure A1 is very useful!

      • Bill Spight says:

        Trump’s extremism is not on the left-right dimension, but on the authoritarian dimension. Trump is extremely authoritarian, not only by asserting dominance whenever he can and demanding personal loyalty, but by loving him some dictators, like Putin and Kim. It is in fact his authoritarianism that makes him a danger to the republic.

        • jim says:

          +1. Every word spot on.

        • jim says:

          So yes, your comment points out that there are multiple dimensions to ‘extreme’ – not just the commonly-touted “left-right” axis.

          So what’s happening is people are confusing these different dimensions of ‘extreme.’ They’re inadvertently lumping them together – n-dimensional Simpson’s Paradox – and thus misunderstanding the underlying data.

        • More Anonymous says:

          Bill and Jim — I don’t disagree with you, but how does the authoritarian aspect of Trump affect the electability of Sanders vs. the electability of the other Democratic candidates?

          Curious if you see a connection that I’m missing.

          • jim says:

            Good question.

            Andrew has presented evidence that the electorate is polarized, and suggested that the “swing” vote is only a couple of percentage points. He’s arguing that the electability of Sanders relies on this swing vote.

            Bill’s point and my broader argument is that the “electorate polarization” meme is misrepresenting both the polarization and the swing vote as being along a simple “left-right” scale, with “extremism” along that scale being the key factor in electability.

            Here’s why:
            1. Trump’s brand of extremism doesn’t lie on that scale. So the “left-right” scale doesn’t measure how and why people respond to trump.

            2. Trumps positions are scattered across that scale – mostly weighted right but with some positions that don’t fit in either camp, like ending wars.

            3. The positions that constitute ‘extreme’ have changed even since 2016. Overall, the nation seems to have moved right somewhat, with some Trump positions that were previously viewed as extreme even in his party now being roughly centrist – like his position on China. Sanders has, by virtue of staying in the same place since 2016, moved left.

            so I’m saying on average the pundits are estimating Sander’s chances higher than they really are because they’re misunderstanding behavior and scale of the swing vote.

            I’m not predicting the election, mind you; I’d be just as surprised if Trump won this year as I was in 2016. But I think Sanders’ electability is overrated.

            • More Anonymous says:

              Thanks for the explanation! I see your point. It would definitely be useful to have survey results on how often people intend to switch vote to the Democrat or turn out for the Democrat because of Trumps authoritarian tendencies, rather than because of his ideological location on a liberal-conservative scale.

  3. Jane says:

    Where does Trump’s victory fit in the “Decades of evidence from academic studies suggests that more moderate nominees tend to perform better in general elections than more ideologically extreme nominees.” Or does that research only apply to Democrats?

    • Andrew says:


      The estimated effects are all small—1 or 2 percentage points of the vote—and only on average. Also, Trump does have certain far-right positions, but overall on issues he’s on the right, at around the same place as the Republican party. In his campaign he took some far-right positions but also some more centrist positions as well. And Trump did lose the popular vote. I don’t think his performance in the 2016 general election contradicts the evidence on moderation and extremism.

      • Dale Lehman says:

        But it does suggest to me that the moderation-extremism dimension is too narrow. In some respects, Trump is as extreme as you can get. In traditional conservative-liberal views (perhaps economic), not so much. So, I think the real question now is whether moderation has an advantage over extremism in other dimensions than those traditional ones. It appears to me (caveat: confirmation bias and other biases present) that only extreme personalities can get elected now.

    • jim says:

      that’s a good question: what does ‘extreme’ mean?

      The problem “extreme” is often a moving target. In some cases it represents natural endpoints of a spectrum – like pro-life and pro-choice. In others, it’s apparently a difference from the status quo (where the median of the political spectrum sits). And even on a natural spectrum, often only one end is considered “extreme”.

      Is Trump ‘extreme’ because – unlike almost anyone in either party – he seeks *less* U.S. military intervention overseas? On the basis of the political spectrum, this would be ‘extreme’ since it’s far distant from most American politicians. On a foreign policy basis, however, it’s a moderate or a “dove” position. Or does “moderate” now imply the status quo (long-term low-level intervention) as opposed to the natural “extremes” of the spectrum (all out war vs. no war).

      The China question is interesting too. Trump’s anti-china position was initially viewed as extreme, but it has become more central to both parties since 2016, with democrats advocating against Huawei’s 5G system. It seems that the Bush/Obama position of engagement was quietly tolerated but resented by many and once Trump went the other way, those people came out of the woodwork in support of the anti-china position.

      Also, many of the issues that Democrats advocate today would have been wildly extreme in their own party during Bill Clinton’s term.

      So while it seems like a sound observation that in recent history the swing is around 2%, that’s doesn’t imply the same people swinging on the same issues, because what’s “extreme” is variable. It has changed dramatically since Bill Clinton’s presidency and significantly even since 2016, and is changing even now.

      • Martha (Smith) says:

        “So while it seems like a sound observation that in recent history the swing is around 2%, that’s doesn’t imply the same people swinging on the same issues, because what’s “extreme” is variable. It has changed dramatically since Bill Clinton’s presidency and significantly even since 2016, and is changing even now.”


  4. Martha (Smith) says:

    So (with Super Tuesday coming up), which Democratic candidate is most likely to beat Trump in November?

  5. Ethan S. says:

    It might be good to reevaluate some of these arguments now that we have the results from Super Tuesday. In particular, it looks like Sanders was not able to significantly increase turnout among the youth in the primary. A primary is not a general election, but this casts some doubt on whether or not it’s possible for Sanders to significantly increase turnout in the general.

    • Martha (Smith) says:

      Taking conditional probabilities into account, it might be possible for Sanders to get significantly increased turnout in the general election, if he gets the nomination: I am guessing that there are significant numbers of people for whom getting Trump out of office is the highest priority. If Sanders does get the nomination, such people are likely to not just vote for Sanders, but to work to get out the vote for him, even if he was not their first choice in the primary. Also, there are probably a lot of people who voted for Biden even though they would prefer Sanders, on the assumption that Biden would be more likely to beat Trump in the general election.

  6. KD says:

    This article argues (convincingly imho) that there are a number of serious problems with the paper:

    One problem is that their Figure 8 switches between denominators. They argue Sanders needs a 11 percentage point increase in turnout among young non-Republican voters. But in the figure, they show changes in turnout among *all* young voters (which is incorrect because around a third of young people are Republicans). Correcting this and other errors implies Sanders would only need a 6pt increase, not an 11pt increase.

    The more fundamental problem is that the 11pt increase that Sanders ‘needs’ is the counterfactual difference between turnout among the young if Sanders is the candidate, and turnout among the young if a moderate like Biden is the candidate. This does not imply an 11pt increase in turnout between 2016 and 2020. To take an extreme example, it might be that youth turnout would fall 11pts relative to 2016 if Biden was the nominee and would stay the same as 2016 if Sanders was the nominee.

    Even according to the authors’ own analysis, when they reweight to 2016 turnout, Sanders leads Trump by 0.8pts while Biden leads with 0.2pts.

  7. John Lashlee says:

    Thanks, I came here to post this. Bruenig is even saying that Brockman and Kalla are being unresponsive, which sucks.

  8. Kaiser says:

    That last paragraph is gold. The data pundits are falling into the trap of using average predictions to predict a single election. Being mathematically eliminated from a round-robin competition is one thing but the simulation models can only statistically “eliminate” someone. Just one narrow path is enough. Trump proved that in 2016.

Leave a Reply