Sen. Bernie Sanders (I-Vt.) delivers remarks in the Wells Fargo Center on the first day of the 2016 Democratic National Convention in Philadelphia on July 25, 2016. (Shawn Thew/European Pressphoto Agency)

On Tuesday night, Sen. Bernie Sanders (I-Vt.) was in Detroit to host a town hall meeting with Rep. John Conyers Jr. (D-Mich.). It felt, writes The Washington Post’s Dave Weigel, like a “campaign rally.” Indeed, Sanders is still being discussed as a front-runner for the Democratic nomination in 2020.

New data is shedding light, however, on Sanders’s role in the last election — and on how many Sanders voters ended up supporting Trump. It’s a question many in the party will be asking about a candidate who may want to compete again for the Democratic nomination.

How many Sanders voters voted for Donald Trump?

Two surveys estimate that 12 percent of Sanders voters voted for Trump. A third survey suggests it was 6 percent.

First, the political scientist Brian Schaffner analyzed the Cooperative Congressional Election Study, which was conducted by YouGov and interviewed 64,600 Americans in October-November 2016. In that survey, Schaffner found that 12 percent of people who voted in the primary and reported voting for Sanders also voted in November and reported voting for Trump.

Schaffner examined only voters whose turnout in the primary and general election could be validated using voter file data. This excludes people who said they voted but actually did not — although it also excludes people who voted in caucuses or party-run primaries, for which validated turnout data are not as readily available.

Second, the same 12 percent figure emerges in the 2016 VOTER Survey, which was also conducted by YouGov and overseen by the Democracy Fund Voter Study Group (of which I am research director). In 2016, this survey interviewed 8,000 respondents in July, when they were asked their primary vote preference, and then in December, when they were asked their general election preference. This has the advantage of measuring primary preference closer to the primaries themselves and then tracking people over time. But their turnout in both elections has not been validated as of yet.

The third survey is the RAND Presidential Election Panel Survey, which interviewed the same group of about 3,000 Americans six times during the campaign. Again, this survey has the advantage of tracking voters over time, but nobody’s turnout has been validated. Among voters who reported supporting Sanders as of March 2016, 6 percent then reported voting for Trump in November.

There is no way to know whether 12 percent or 6 percent or some other estimate is The Truth, and there are enough differences among these surveys that we cannot easily pinpoint why the numbers differ. So we should take these estimates with some caution.

On the second night of the Democratic National Convention in Philadelphia, delegates supporting Sen. Bernie Sanders (D-Vt.) walked out in protest, despite Sanders's calls for unity within the party. (Peter Stevenson,Adriana Usero,Dalton Bennett,David Bruns,Jayne Orenstein,Alice Li/The Washington Post)

How does this compare to other elections?

It’s a perennial question whether supporters of losing primary candidates will vote for their party’s nominee in the general election. So let’s compare the Democratic primary with the Republican primary. In the VOTER Survey, only 3 percent of those supporting Texas Sen. Ted Cruz reported voting for Hillary Clinton, as did 10 percent of Florida Sen. Marco Rubio’s supporters and 32 percent of Ohio Gov. John Kasich’s supporters. So Sanders supporters were about as likely to vote for Trump as Rubio’s supporters were to vote for Clinton, and far less likely than Kasich supporters were to vote for Clinton.

Another useful comparison is to 2008, when the question was whether Clinton supporters would vote for Barack Obama or John McCain (R-Ariz.) Based on data from the 2008 Cooperative Campaign Analysis Project, a YouGov survey that also interviewed respondents multiple times during the campaign, 24 percent of people who supported Clinton in the primary as of March 2008 then reported voting for McCain in the general election.

An analysis of a different 2008 survey by the political scientists Michael Henderson, Sunshine Hillygus and Trevor Thompson produced a similar estimate: 25 percent. (Unsurprisingly, Clinton voters who supported McCain were more likely to have negative views of African Americans, relative to those who supported Obama.)

Thus, the 6 percent or 12 percent of Sanders supporters who may have supported Trump does not look especially large in comparison with these other examples.

Was this enough to cost Clinton support in key states?

This is a huge hypothetical, of course. Clinton’s losses in the Rust Belt, which cost her an electoral college majority, can be attributed to many factors. And a lot depends on the exact number of Sanders supporters who did not vote for her.

Schaffner generated some state-level estimates, which G. Elliott Morris quickly noted were large enough to exceed Trump’s margin of victory in Wisconsin, Michigan, and Pennsylvania.

Even if we assume that the overall percentage of Sanders supporters who voted for Trump was 6 percent and not 12 percent, and assume therefore that we can cut every state estimate in half, the estimated number of Sanders-Trump voters would still exceed Trump’s margin of victory.

But again, attach a lot of caveats to that analysis.

What kinds of Sanders voters supported Trump?

Perhaps the most important feature of Sanders-Trump voters is this: They weren’t really Democrats to begin with.

Of course, we know that many Sanders voters did not readily identify with the Democratic Party as of 2016, and Schaffner found that Sanders-Trump voters were even less likely to identify as Democrats. Sanders-Trump voters didn’t much approve of Obama either.

In fact, this was true well before 2016. In the VOTER Survey, we know how Sanders-Trump voters voted in 2012, based on an earlier interview in November 2012. Only 35 percent of them reported voting for Obama, compared with 95 percent of Sanders-Clinton voters. In other words, Sanders-Trump voters were predisposed to support Republicans in presidential general elections well before Trump’s candidacy.

Schaffner found that what distinguished Sanders-Trump voters from Sanders-Clinton voters wasn’t their attitudes about trade, but their attitudes about race. When asked whether whites are advantaged, Sanders-Trump voters were much more likely to disagree than were Sanders-Clinton voters.

The same thing is true in the VOTER survey — and, again, this was evident well before 2016. When originally interviewed in December 2011, Sanders-Trump voters were actually more likely to favor increasing trade (59 percent said so) than Sanders-Clinton voters (48 percent said so).

And, again, the bigger cleavage involved race. For example, in December 2011, 75 percent of Sanders-Trump voters agreed with this statement: “If blacks would only try harder they could be just as well off as whites.” Only 16 percent of Sanders-Clinton voters agreed.

Similarly, when how they felt about whites and blacks on a 0-100 scale, Sanders-Trump voters rated blacks 9 points lower than Sanders-Clinton voters. But Sanders-Trump voters rated whites 8 points higher.

The same thing was true for other minority groups. Compared with Sanders-Clinton voters, Sanders-Trump voters rated Latinos 11 points less favorably, Muslims 20 points less favorably, and gays and lesbians 31 points less favorably.

In short, it may be hard to know exactly how many Sanders-Trump voters there were, or whether they really cost Clinton the election. But it doesn’t appear that many of them were predisposed to support Clinton in the first place.