Nov 23, 2016 at 2:22 PM
Demographics, Not Hacking, Explain The Election ResultsBy Carl Bialik and Rob Arthur
Filed under 2016 Election
According to a report Tuesday in New York Magazine, a group of computer scientists and election lawyers have approached the Hillary Clinton campaign with evidence they believe suggests the election might have been hacked to make it appear that Donald Trump won the Electoral College when Clinton really did. The hacking claim appears to be based on concerns about tampering with electronic voting machines. We’ve looked into the claim — or at least, our best guess of what’s being claimed based on what has been reported — and statistically, it doesn’t check out.
There’s no clear evidence that the voting method used in a county — by machine or by paper — had an effect on the vote. Anyone making allegations of a possible massive electoral hack should provide proof, and we can’t find any. But it’s not even clear the group of computer scientists and election lawyers are making these claims. (More on this in a moment.)
The New York article reports that a group that includes voting-rights attorney John Bonifaz and computer scientist J. Alex Halderman presented findings last week about Wisconsin, Michigan and Pennsylvania to top Clinton campaign officials to try to persuade them to call for a recount. Exactly what those findings were isn’t clear. The New York article includes just one example, a finding that Clinton did worse in counties in Wisconsin that used electronic voting machines instead of paper ballots.1 It’s not clear what data the group was using to call for a recount in Michigan and Pennsylvania, or if it was looking at data at all: It could have chosen those states because they were the ones besides Wisconsin that Trump won with the smallest margins. Bonifaz, Halderman and the Clinton campaign officials mentioned in the article didn’t respond to requests for comment or more detail about the study.
But in a Medium post on Wednesday, Halderman said the New York article “includes some incorrect numbers” and misrepresented his argument for recounts. He laid out an argument based not on any specific suspicious vote counts but on evidence that voting machines could be hacked, and that using paper ballots as a reference point could help determine if there were hacks. “Examining the physical evidence in these states — even if it finds nothing amiss — will help allay doubt and give voters justified confidence that the results are accurate,” Halderman wrote.
Without a recount, all we can do for now is look for any meaningful difference in the three states named in the New York article between votes in counties that used paper ballots and votes in ones that used machines. That quickly crossed Michigan off the list: The entire state uses paper ballots, which are read by optical scanners.2 So we couldn’t compare results by type of voting in that state. Instead, we checked the six other states with a margin between Clinton and Trump of less than 10 percentage points that use a mix of paper and machine voting: Arizona, Florida, North Carolina, Ohio, Texas and Virginia.
For each county in those states, we looked at Clinton’s vote share and whether it was associated with the type of voting system the county used, based on voting-system data compiled by a nonprofit electoral-reform group called Verified Voting and 2016 vote data from Dave Leip’s U.S. Election Atlas and ABC News.3 It doesn’t make much sense, though, to just look at raw vote counts and how they differed, because we know there are many factors that affect how a county voted, both in those states and everywhere else around the country. So we separated out two of the main factors that we know drove differences in voting results: the share of each county’s population age 25 and older with a college degree, and the share of the county that is non-white.4
We found no apparent correlation5 between voting method and outcome in six of the eight states, and a thin possible link between voting method and results in Wisconsin and Texas. However, the two states showed opposite results: The use of any machine voting in a county was associated with a 5.6-percentage-point reduction in Democratic two-party vote share in Wisconsin but a 2.7-point increase in Texas, both of which were statistically significant.6 Even if we focus only on Wisconsin, the effect disappears when we weight our results by population. More than 75 percent of Wisconsin’s population lives in the 23 most populous counties, which don’t appear to show any evidence for an effect driven by voting systems.7 To have effectively manipulated the statewide vote total, hackers probably would have needed to target some of these larger counties. When we included all counties but weighted the regression by the number of people living in each county, the statistical significance of the opposite effects in Wisconsin and Texas both evaporated.8
Even if the borderline significant result for Wisconsin didn’t vanish when weighting by population, it would be doubtful, for a few reasons. You’re more likely to find a significant result when you make multiple tests, as we did by looking at eight states with and without weighting by population.9 Also, different places in Wisconsin and Texas use different kinds of voting machines; presumably if someone really did figure out how to hack certain machines, we’d see different results depending on which type of machines were used in a county, but we don’t. And Nate Cohn of The New York Times found that when he added another control variable to race and education — density of the population — the effect of paper ballots vanished.
It’s possible nonetheless that the election was hacked, in the sense that anything is possible. (And the best hackers are experts in erasing their tracks.) Maybe hackers knew which control variables we’d look at and manipulated the vote in a way that it would look like it was caused by race, education and population driving different voting preferences. Maybe hackers didn’t manipulate the share of votes in individual counties, but rather the turnout, increasing the number of votes in counties likely to favor one candidate or another. Maybe some irregularities at the county level in early Wisconsin vote-counting are signs of wider problems. Maybe we’d find something if we dug down to the precinct level, or if we looked at other states with mixed voting systems. But at a time when the number of voters without confidence in the accuracy of the vote count is rising, the burden of proof ought to be on people claiming there was electoral fraud. The paradox is that in our current electoral system, without routine audits, seeking proof requires calling for a recount, which in itself can undermine confidence in the vote.
Carl Bialik is FiveThirtyEight’s lead writer for news. @CarlBialik
Rob Arthur is FiveThirtyEight’s baseball columnist and also writes about crime. @No_Little_Plans
