In 2012, while watching the Ohio returns, Foley wondered what effect votes counted after Election Day might have. He found something astonishing. Looking at five battleground states, Foley discovered that from 1960 to 2000, there’d always been some change between the Election Night tally and the final results, usually in the hundreds or thousands of votes, and sometimes favoring either party. Starting in 2004, the size of the shifts had become reliably Democratic and significantly larger—nearly 80,000 votes in Virginia in 2008. Foley christened this effect the “blue shift.”

The blue shift remains little studied and poorly understood. In a 2015 paper, Foley and the MIT political scientist Charles Stewart III found evidence that the blue shift was correlated with the number of provisional ballots cast. A California Institute of Technology paper this year studying the blue shift in Orange County, California, found that many provisional voters are younger, more likely to be nonwhite, and more transient, all populations that tend to vote for Democrats. Foley and Stewart also found no strong correlation between mail-in or absentee ballots and a blue shift. These votes have not historically had a strong partisan leaning.

“I think an honest assessment of this is that we’re still learning,” Foley said. “While we have made some progress as a field, I don’t myself feel confident that we’ve really pinned down causality.”