It may be tempting, as you're doom-scrolling through swing state results trickling in at a glacial pace, to turn to exit polls. Don't do it. Or, if you do, read the rest of this post first and approach this data with all due caution.
Exit polls seem like a reliable source of information. After all, they're not advance surveys of self-proclaimed "likely" voters who may never actually vote. Traditionally, we think of exit polls as in-person, post-vote interviews whose findings are weighted to reflect local voter demographics.
The problem is that even with that adjustment — and sometimes exciting bits of data get out before any adjustment is made — exit polling at best offers an impressionist portrait of the vote. In 2016, for example, "exit polls estimated that the white working class cast a total of 34 percent" of ballots. Later, more accurate research put it at 44 percent, a significant difference. Likewise, the most famous number from the last presidential race, that 81 percent of white evangelicals voted for President Trump, wasn't exactly right. Better, later research put it about five points lower.
This year's exit polls deserve extra skepticism. One reason is the unusually high proportion of early and mail-in voting. Though the main exit polling firm, Edison Research, conducted phone surveys of mail-in voters and sent pollsters to early voting stations to account for this shift, it's difficult to say how accurate this methodology will be, particularly for fine demographic detail.
Another factor is voter behavior. Voters are less willing to participate in exit polls than they used to be, and there's no guarantee participants tell the truth. They might lie because loved ones can overhear them, because they find it funny, or because they think honesty will produce social blowback.
That brings me to a few titillating exit polls making the rounds as I write. One data set shows a dramatic swing of Hispanic voters away from the Democratic Party since 2016. Similarly, another finds "Trump did better in 2020 with every race and gender except white men." These are major shifts, if they hold. Some early vote counts as well as the very large margins in the Hispanic vote numbers suggest that swing is real. But the second data set is all moves of 5 percent or less. It might prove entirely right — or every number on the list may be dead wrong. We simply have to wait and see.