The 12% Bernie-to-Trump figure (and 24% Clinton-to-McCain figure)[see note below] comes from Brian Schaffner of the University of Massachusetts, Amherst who based it on extrapolations from the data in the Cooperative Congressional Election Study. Shaffner limited his examination only to those voters whose participation in both the primary and general could be validated. Theoretically, the idea is to weed out people who said they voted in one or other but didn’t. In practice, it cripples the headline conclusions at the end of it. Problem #1: That approach entirely excludes caucus states. Guess who did disproportionately well there? Problem #2: It entirely excludes voters in states where validation info wasn’t readily available. Short version: A large chunk of the U.S. is entirely unrepresented in his figures.

The RAND Corporation did a much better study in which they tracked the same nationwide sample-group, surveying them half a dozen times throughout the campaign process. They found that 6% of Sanders voters subsequently cast their general-election ballots for Trump.

This dovetails with an ABC News/Washington Post poll conducted only days before the election, which found that only 8% of Sanders primary supporters who were participating in the general planned to vote for Trump or had already done so:

That great big blue stack there also helps make the point about the absurdity of efforts to blame Sanders voters for Trump’s victory; the overwhelming majority of Sanders voters voted for Clinton. If Sanders voters hadn’t voted for Clinton, she would have lost badly. Sanders was the energizing hope-and-chance candidate in the race and brought into the process many new people who had never voted or had given up on voting but who came back to support him. Because of Sanders’ presence in the race, Clinton’s final vote-count is padded with x number of votes she wouldn’t have gotten absent Sanders’ presence.

And now that I’ve unnecessarily belabored these points, I’ll note something that usually gets lost in this sort of minutiae: no politician is entitled to the anyone’s vote. If a pol wants someone’s vote, it is incumbent upon that pol to earn that vote. If one doesn’t proceed from the assumption that Clinton was entitled to the votes of people who didn’t want to give it to her, there’s nothing to discuss with it comes to this.

— -

Note: In general, 7–10% of self-identified Democrats will vote for the Republican candidate and vice-versa — that’s the usual level of crossover voting in a modern presidential election. The 24% Clinton-to-McCain figure for 2008 is derived from the same flawed examination as the 12% Sanders-to-Trump. There were, however, a large number of Clinton voters who went on to vote for McCain. CNN reported at the time that

“Exit polling also showed that Democrats who supported Sen. Hillary Clinton during the primaries overwhelming voted for Obama in the general election, 84 percent to 15 percent for McCain.”

That’s a larger number than both the Sanders-to-Trump crossover explained above and the 7–10% crossover typical of any presidential race.