There was a period about a month ago when the outcome of the effort to recall California Gov. Gavin Newsom (D) looked uncertain. There hadn’t been a lot of polls, but some showed opposition to the recall was only slightly ahead of support for it, an unexpected finding in a heavily Democratic state.

More worrisome to Newsom supporters, a poll from SurveyUSA showed the recall passing — meaning that Newsom was likely to be replaced. The leading contender to do so was Larry Elder, a right-wing Republican who echoed much of Donald Trump’s rhetoric. Doomsday loomed.

On the graph below and the one that follows, the arrow begins at the point where a poll showed support for the recall and ends where it showed opposition. So a short, upward-pointing arrow indicates a poll where there was slightly more opposition to the recall than support. A long, downward-pointing one shows significantly more support for the recall than opposition.

For Newsom’s campaign, though, this was enormously useful. Candidates and campaigns never want voters to think that the result is in the bag; if they do, they’re less inclined to vote. Newsom and his allies needed people to come out to the polls in huge numbers to ensure that he could survive the recall and, ideally in their minds, establish a mandate for his approach to the job. So they played up the idea that the recall was close.

Searches for “recall” surged in California in the first three weeks of August. The state was paying attention.

Pollsters began conducting more surveys in the state, and the results increasingly looked good for Newsom. Now, with the pool of respondents more often shifted from registered voters to likely voters — that is, from people who could vote to people who probably would vote — opposition to the recall began to increase. SurveyUSA showed the recall failing in a second survey (as it somewhat awkwardly acknowledged some reasons that first poll might have erred).

By the time Election Day arrived Tuesday, FiveThirtyEight’s polling average had it losing by nearly 16 points.

There was a spate of announcements from pundits that a margin that wide would even survive the sort of polling misses seen in the presidential races in 2016 and 2020. In other words, even if polls again understated support for the right’s position, the recall would fail.

The polls were off the mark — but in the other direction. As of writing, the recall was losing by a nearly 28-point margin, wider than any poll conducted over the course of the recall campaign. That is likely to narrow, given that California takes an average of 13 centuries to complete its vote counting in any election cycle, but there’s little chance the result will land where the polling average had it.

This is an intriguing result. In the wake of both of the past two presidential contests, there were robust efforts to figure out why the polls were so wrong, how in 2016 pollsters failed to predict Trump’s victory and how, in 2020, they failed to recognize how close the result would be. Outside observers seized on the 2016 miss (often ignoring the 3-in-10 chance of a Trump victory that forecasts offered) in part because a Trump loss was expected and a Trump win resulted. The 2020 response was in part a function of observers being wary of what had happened in 2016. And in each case pollsters went back and assessed their own work, finding places for improvement that would better capture the non-college-educated White vote that was central to 2016 and exploring in depth the ways in which 2020 polls missed the mark.

That latter analysis was inconclusive, offering up a number of possible ways in which their research missed Trump support. One was that Republicans who don’t answer polls — perhaps because of the emphasis Trump had placed on rejecting them — simply vote more heavily for Republicans. This makes some sense: If you agree with Trump that polling is to be shunned, you’re probably going to vote for Trump for president. But this is also largely theoretical, since it’s hard to describe the ideologies of people who don’t respond to polls — because they don’t respond to polls.

So consider what happened in four sets of polls: the 2016 and 2020 presidential votes, the 2021 California recall and the 2014 and 2018 midterm elections. In 2016 and 2020, with Trump on the ballot, polls understated Republican support. In the pre-Trump era of 2014, polls also missed a surge in Republican support. In 2018, though, with Trump not on the ballot and Democrats energized against him, the polls were largely accurate. In the California recall, the polls at this point seem to have again understated Democratic support.

One consistent aspect seems to be energy. In 2014, 2016 and 2020, polls didn’t capture Republican energy. (For example, it is possible according to the post-election analysis that the polling miss in 2020 may in part have been a function of new voters in November.) In 2021, they seem to have underestimated energy on the left.

It’s also possible that they overestimated energy on the right, perhaps overcorrecting. Exit polling suggests that about 4.3 million Whites without college degrees voted in the 2018 gubernatorial race that first brought Newsom to office. About 3.5 million Whites with degrees did as well. In the recall, the number of non-college-educated Whites fell by 2 million (so far and according to exit polling, which should be taken with a grain of salt) while the decline among college-educated Whites was only about 800,000. Since Whites without college degrees are the only race-slash-educational group that backed the recall, that seems significant. It’s also possible that Whites without degrees who were ambivalent about Newsom stayed home, since a majority of that group voted for him three years ago.

A core question in the wake of Trump’s presidency is whether Republicans will be able to maintain the energy that he brought to the party. Elder tried, embracing Trumpian positions and tactics, but Republicans made up only slightly more of the electorate Tuesday than they did in 2018, despite this being an off-year election. They also didn’t vote much more heavily against Newsom.

It is possible that 2016 and 2020 were Trump-specific flukes in the polling. It is possible, too, that that flukiness will reappear in 2024 without affecting races in the interim. But it is not the case that the polling in California this year was obviously undercounting Republican support. If anything, it overstated Republican support at a key moment, making it easier for Newsom to survive the recall effort.

Emily Guskin and Lenny Bronner contributed to this report.