The incident that served as the best predictor of the 2020 Democratic presidential primary actually happened in 2011.

That October, then-Gov. Rick Perry (R-Tex.) was leading the Republican field as the party tried to identify a candidate who could beat incumbent President Barack Obama. In a Washington Post-ABC News poll, Perry not only led in terms of support but was also seen as the candidate who had the best chance of winning the general election.

Then, during a debate, he tried to name the three government agencies he would eliminate as president. He came up with two, substituting an “oops!” for the third. In our next poll, his support had collapsed — as had the sense that he was the best bet to beat Obama.

Perry’s dive, which we wrote about in May, should have served as a warning to Democrats considering whom they want to nominate to run against President Trump. Centering that choice on electability means having to determine electability — something voters in the past have shown they’re pretty terrible at.

The Post-ABC poll released Wednesday offered a new, perhaps too-late reminder of the same thing. Former vice president Joe Biden’s position at the top of the primary field was linked to perceptions that he was the best bet for defeating Trump. Since January, though — and since Biden fared terribly in the first two nominating contests — Biden’s position has plummeted. His support was cut in half since our last poll as Sen. Bernie Sanders (I-Vt.) and former New York mayor Mike Bloomberg rose.

At the same time, the sense that Biden had the best shot of beating Trump also dropped dramatically.

Those two factors, electability and support, have been tightly correlated in the Democratic primary contest (as they were in the Republican contest in 2012). As perceived electability rose or sank, so did support. Since January, Sanders and Bloomberg have seen increases in both support and perceived electability while Biden has dropped (a lot) on both metrics. Only Sen. Elizabeth Warren (D-Mass.) stands out among the top-tier candidates: Her perceived electability dropped even though her overall support was steady.

Bloomberg has invested heavily in making the case that he’s the most electable candidate against Trump, using his massive wealth and hurricane of ads as evidence that he can tackle the incumbent president. But respondents to our poll in states where those ads have aired — mostly states that vote on Super Tuesday — are about as likely to view Bloomberg as electable as are respondents from other states.

It’s hard to know how to read that particular bit of data. It may, for example, suggest that support for a candidate drives perception of electability to some significant extent, though we tend to think of the causality working in the other direction.

Asked which was more important, electability or picking a candidate based on issues, most primary voters in our poll said that electability was the priority. That varied widely by demographic, however.

Independents were more likely than Democrats to say that issues were the priority. Women and white respondents were more likely than men or nonwhites to say that electability was. There’s a massive split by age, with younger voters — read: Sanders voters — being among the only groups to say issues were more important while older voters emphasized electability.

On this one, the younger voters have an advantage. Prioritizing a candidate’s stance on political issues is, at least, something that can be evaluated. Electability, on the other hand, is more subjective, a gut assessment of how a candidate will fare. The sort of thing that makes a Texas governor or former vice president seem like a sure bet, until they aren’t.

What’s particularly noteworthy about the focus on electability is that polling consistently shows that the leading Democrats all fare about as well against Trump as one another. In some polls, one candidate has an advantage over Trump of a few points nationally; in others, a different candidate might. To the extent that electability perceptions are rooted in anything substantive, they’re generally based on subtle differences between candidates in polling months ahead of a general election.

In 2016, Hillary Clinton seemed like the most electable candidate in the general election field up until polls closed on Election Day. Picking a party nominee based on making a better prediction of that outcome this time around seems like a more challenging prospect than voters like to admit. As former Biden supporters can attest.