It’s obviously true that many Americans weren’t expecting Donald Trump to win the presidency in 2016. He trailed in national polling nearly every day of that year and trailed in enough state polling that an electoral-vote majority seemed out of reach. As the race tightened shortly before Election Day, forecasts of the race still showed Hillary Clinton with a strong advantage. FiveThirtyEight’s model showed one of the closest contests at the end of the race, but even that gave Clinton a 7-in-10 chance of winning.

Or, basically, a 2-in-3 chance. Which, put that way, doesn’t seem that solid. It’s about the same odds as getting something other than a one or a two when you roll the dice — decent, but not a sure thing. So, largely thanks to polls in a few states that vastly overestimated Clinton’s chances, the results of the election were different from what many Americans expected.

Trump, of course, claimed not to be surprised (though reporting has repeatedly suggested that he and his team very much were). He brushed off his low poll numbers by claiming that there was a big pool of pro-Trump voters who simply refused to tell pollsters they liked him, for whatever reason. He had a silent majority, he would say, a conveniently unmeasurable universe of support that would carry him to victory. Then, lo and behold, he was actually victorious.

All of this provides key context for Trump’s 2020 reelection campaign. Politicians, always interested in controlling how people view polls, have embraced the apparent uncertainty that emerged in 2016 as evidence for whatever their position happens to be.

For Democrats, it’s about turnout: You need to go vote because 2016 showed that you can’t rely on polling. For Republicans, it’s something similar: Don’t discount Trump/me; after all, remember what happened in 2016. For Trump, the 2016 results have allowed him to continually argue that he does have that silent majority at his back, even though there’s no evidence at all that he actually does. As long as people believe his claim, he gets the veneer of a mandate from claiming that polls simply don’t see what’s actually there.

And people believe his claim, as new polling from Pennsylvania shows. Monmouth University on Wednesday released a poll showing former vice president Joe Biden leading Trump by double digits in the pivotal state. Yet Pennsylvanians are as likely to think Trump will win the state as they are to think that Biden will.

Why? In part, clearly, because so many of them also report having been surprised by what happened in 2016.

The Monmouth poll, though, put a fine point on it. Asked if they believed that there actually did exist a secret pool of Trump voters, people who “who support Donald Trump but wont tell anyone about it,” a majority said there was.

Republicans were more likely than respondents overall to say that such voters existed. By contrast, voters in counties that Trump won easily in 2016 were less likely to say such voters existed, presumably because they were more likely to see vocal Trump supporters around them.

No one really thinks there are similarly secret Biden voters, according to Monmouth. It’s just this idea that there’s this pro-Trump undertow that’s captured Pennsylvanians’ imaginations.

What’s remarkable about this finding isn’t just how Trump’s portrayal of his support has been accepted. It’s also the idea that polling missed something significant in Pennsylvania four years ago, leading to a staggering surprise.

In reality, that’s not the case. Unlike in Michigan and, to a larger degree, Wisconsin, Pennsylvania polling in 2016 wasn’t that far off the mark. The final RealClearPolitics average of polling in the state gave Clinton a two-percentage-point lead — a lead, in other words, that’s far from certain. She lost the state by less than one point, meaning that the polling average was off by a bit under three points. It was off by less than the national polling average was in 2012, by comparison, but since that error didn’t change the expected outcome (Barack Obama won by more than expected), it’s not seen as being as incorrect.

The difference between the average and the reality in Pennsylvania was about 1.7 points larger than same difference at the national level — a miss, but not a mind-boggling one. As FiveThirtyEight put it right before the election, Trump was only “a normal polling error behind Clinton.” That error showed up.

Note that Biden’s position in Monmouth’s poll is far better than Clinton’s position was in the late-campaign average. Biden is up by 13 points in the new poll, a bigger margin than Clinton enjoyed at any point in the RealClearPolitics average in 2016. Swing Monmouth three points away from Biden … and he’s still winning by 10 points. Yet Pennsylvanians still think that Trump will steal the election away thanks to a massive pool of voters who for inexplicable reasons won’t even tell pollsters that they like Trump.

Rep. Mike Kelly (R-Pa.) tried to bolster that sense in a conference call this week. He argued that Pennsylvania would be close because pollsters missed a surge of new voters who turned out in 2016. Such voters are often considered less likely to vote by pollsters for the simple reason that they tend to vote less frequently.

Trump won, he said, “because there were a lot of people who came out to vote for the first time because they believed in the message and they believed in the man.” Not really secret voters, but unmeasured ones who will swing the result.

A review of who those new voters were shows some interesting patterns. Looking at partisan and age groupings among voters who registered in 2016, 2018 and 2020 (and who are still registered in the state), we see immediate differences between 2016 and 2018.

Among those who registered in 2016, those who actually voted that year were more heavily Republican — bolstering Kelly’s argument. Those who voted in 2016 and 2018 were more likely to be Democrats.

The density of Democrats was heavier among those who registered in 2018, and the density of Democrats among actual voters was heavier still. So far in 2020, new registrations have looked generally the way they did in 2016, though slightly more heavily Republican and less heavily Democratic. (The total number of registrants this year is also now about a quarter of what was seen through Election Day four years ago.)

But, then, we’re talking about 56,000 newly registered Republicans this year — in a state where more than 6 million votes were cast four years ago. The underlying sentiment may have some validity, but the effect compared with what the poll shows statewide is de minimis.

It’s unlikely that Biden holds a 10-plus-point lead until Election Day, and the likelihood of a narrowing race may itself power some of the sentiment that Trump will win despite how people feel in the moment. After all, Clinton had a big lead four years ago that eventually withered enough that a quiet Trump surge could flip the state.

But that’s just another facet of what we’ll call ’16 Syndrome: the idea that since polling only captures an incomplete picture of the current moment, something weird can and will happen. That’s not wrong, really. It’s just that there’s a reason we do polling: Surprises aside, it has a very good track record.