While we wait for the voting to finish in Virginia there are plenty of pundits staring obsessively at the final polling out of the commonwealth. Depending who you ask, it looks like it’s going to be close, with most likely a less than five-point margin for the winner. Many of the polls are showing Ralph Northam with a thin lead, somewhere in the range of the margin of error. So why am I seeing people predicting that Gillespie will pull it out at the last minute? It’s still all about 2016, people.

Personally, I have calculated that Gillespie will win by three points using a highly scientific survey of how many pigeons were perched on the phone lines outside my house after breakfast. This is a tried and true method, but others prefer a tedious process involving asking large numbers of poll respondents what they think and extrapolating from there. But should we have faith in those numbers?

For one bit of catnip to pass the time, take a look at this article by Michael Tomasky at the Daily Beast. He’s quite concerned that the Democrats may be counting their chickens before they’ve hatched just as they did last November. And the reason is the seeming inability of the polling community to lock down accurate exit polls for the dreaded “white working class” vote. His concerns are substantiated by this report which shows that for at least a few election cycles now, the percentage of non-college educated whites showing up at the polls has been seriously undercounted in exit polling and perhaps in early analysis as well.

The exit polls, they found, dramatically understated the percent of the total vote cast by non-college whites—that is, the category that best correlates to the famous white working class. Exit polls had non-college whites casting 34 percent of all 2016 ballots, and college-educated whites casting 37 percent. The actual numbers, according to the new study? College-educated whites were in fact just 29 percent of the total vote, and non-college whites were a whopping 45 percent of the vote.

I know what you’re thinking: There was a massive Trumpian white working class surge that the exit polls missed. But even that isn’t really right. Because they studied the 2012 results too, and found that almost exactly the same thing happened then. The 2012 exit polls had both groups of white voters at about 36 percent. But Teixeira, Halpin, and Griffin found that college-educated whites accounted for 28 percent of the 2012 vote, and non-college whites 45 percent. So apparently, exit polls just consistently undercount the white working class.

So in 2016 the working class white vote was predicted by the exit polls to account for 34% of the ballots, in reality it was 45%. Nearly the same margin of error applied in 2012. I think the main takeaway here isn’t how they got it so wrong, but the simple fact that it seems to be happening. It’s not enough to tip an election in a state where someone has a decisive lead, but in a race that’s hovering around the margin of error? That’s the difference between winning and, well… Hillary Clinton.

It probably also accounted for the final numbers in Pennsylvania. But let’s not pretend that it’s just the exit polls. Turnout predictions attempt to capture demographic groups which reliable lean toparty one part or the other. And the total voter turnout is depressingly low compared to the number of registered voters. That means that if one particular group overperforms in a particular cycle, the final totals can really shift.

Think about 2008 and, to a slightly lesser extent, 2012. Black voter turnout was well outside the normal ranges with Barack Obama on the ticket. Part of that was anticipated by pollsters, but the surge was even larger than they had predicted. Unfortunately for Hillary, in 2016 they didn’t scale those predictions back down all the way to their previous levels. At the same time, the aforementioned non-college educated white vote in places like Pennsylvania came out of the woodwork. A state which Donald Trump was supposed to lose actually went to him outside the margin of error.

So how well is everyone counting those two demographics in Virginia this month? I suppose we’ll know in about eight hours. And if Northam somehow manages to lose, be sure to stop by and drop off some birdseed for my pigeons.