Here at FiveThirtyEight, we’ve been skeptical of claims of irregularities in the presidential election. As we pointed out last week, there are no obvious statistical anomalies in the results in swing states based on the type of voting technology that each county employed. Instead, demographic differences, particularly the education levels of voters, explain the shifts in the vote between 2012 and 2016 fairly well.
But that doesn’t mean I take some sort of philosophical stance against a recount or an audit of elections returns, or that other people at FiveThirtyEight do. Such efforts might make sense, with a couple of provisos.
The first proviso: Let’s not call it a “recount,” because that’s not really what it is. It’s not as though merely counting the ballots a second or third time is likely to change the results enough to overturn the outcome in three states. An apparent win by a few dozen or a few hundred votes might be reversed by an ordinary recount. But Donald Trump’s margins, as of this writing, are roughly 11,000 votes in Michigan, 23,000 votes in Wisconsin and 68,000 votes in Pennsylvania. There’s no precedent for a recount overturning margins like those or anything close to them. Instead, the question is whether there was a massive, systematic effort to manipulate the results of the election.
So what we’re talking about is more like an audit or an investigation. An investigation that would look for signs of deliberate and widespread fraud, such as voting machines’ having been hacked, whole batches of ballots’ intentionally having been disregarded, illegal coordination between elections officials and the campaigns, and so on. Such findings would probably depend on physical evidence as much or more than they do statistical evidence. In that sense, there’s no particular reason to confine the investigation to Wisconsin, Michigan or Pennsylvania, the states that Hillary Clinton lost (somewhat) narrowly. If the idea is to identify some sort of smoking gun indicating massive fraud perpetrated by the Trump campaign — or by the Clinton campaign, or by the Russian government — it might be in a state Clinton won, such as New Hampshire or Minnesota. Or for that matter, it might be in a state Trump won fairly easily, like Ohio or Iowa.
A second “condition” is that the burden of proof for claims of a fixed election ought to be high. That’s because there’s enough evidence for there to be a clear presumption against theories of massive vote-rigging:
- Many individuals and organizations have already checked for signs of irregularities. The journalists at ProPublica, for example, had more than 1,100 people monitoring the vote on Election Day and found no major irregularities. The campaigns also employ their own election monitors, lawyers and statisticians. The Clinton campaign, in particular, “had not uncovered any actionable evidence of hacking or outside attempts to alter the voting technology,” according to their general counsel, Marc Elias, although they’ll participate in “recount” efforts brought about by the Green Party’s Jill Stein.
- It’s awfully hard to rig an election because of the degree of coordination required across dozens of localities in dozens of states. The decentralized nature of U.S. presidential elections — whatever other problems it might cause — is thus a partial check against widespread election hacking.
- Whether or not the election outcome should have been foreseeable — and we’re of the view that people ignored clear signs of trouble for Clinton in the polls — the results make a lot of sense after the fact based on demographic trends. Specifically, the polls underestimated Trump’s support among white voters without college degrees, but were fairly accurate otherwise. Importantly, Trump’s overperformance occurred not only in swing states but also in states that he won easily, such as North Dakota. And he significantly improved on Mitt Romney’s performance in regions of states that otherwise weren’t competitive, such as upstate New York. If someone hacked the election, they did a clever job of covering their tracks by producing consistent-seeming demographic swings across a number of competitive and noncompetitive states.
On the flip side, when you’re examining results in thousands of precincts, it’s easy to detect results that look like funny business even when they have a perfectly innocent explanation. If hundreds of researchers are performing hundreds of statistical tests on hundreds of results, there are going to be a lot of false positives, including some that researchers claim have an extraordinarily high degree of statistical significance. (See also: our reporting on “p-hacking” and the scourge of false positives in the scientific community.) It’s also easy to uncover actual but isolated irregularities, such as malfunctioning voting machines, which nonetheless aren’t part of a broader conspiracy to rig the results.
It’s easy for smart people to be deceived by these claims, especially if they’re motivated to see a particular result. And it’s easy for journalists to spread misinformation by highlighting the claims, without providing sufficient scrutiny of them. In 2004, left-leaning sites made a cottage industry of claims that George W. Bush had stolen the election from John Kerry in Ohio, despite a lack of evidence and a wider margin (more than 100,000 votes) than the ones that separate Clinton and Trump in Wisconsin and other states now.
In many ways, undertaking an audit of the election results is tantamount to performing a test for a rare but potentially fatal disease. You want to weigh the probability of successfully detecting an anomaly against the invasiveness of the procedure and the chance of a false positive result. Oftentimes, the risk outweighs the reward. For instance, many experts warn against mammograms for women in their 40s because the underlying risk of breast cancer is low for women of that age and the rate of false positive tests is high, causing undue stress for the patients and subjecting them to further tests and operations that might be harmful.
What are the costs of an election audit? Running them will cost several million dollars, but that’s fairly trivial in an era of billion-dollar campaigns. Instead, since these audits aren’t routine — although maybe they should be — the cost is mostly that they could undermine the perceived legitimacy of the election and the longstanding norm toward uncontentious transitions of power from one president toward the next. Which might be more persuasive … if Trump hadn’t spent the weekend peddling a conspiracy about how he thought the results were rigged in Clinton’s favor because millions of people had voted illegally.
So, case closed, right? Trump, Stein, Clinton-backing Democrats — everyone’s talking about irregularities. So let’s get on with the recount … er, the audit? Actually, I think people still ought to be careful what they wish for. Remember, we’ll be in a signal-poor environment. The audit will kick up a lot of dust, but it’s unlikely that there will be anything there, and even harder to prove anything. Trump is a masterful troll, and trolls — and aspiring authoritarians — can thrive in environments where there’s a lot of confusing information. It’s also worth noting that Stein’s motivations for financing the recounts are ambiguous and she might be prone toward making sensational claims. People’s BS detectors ought to be set to extra high, even as compared to their already-high 2016 levels. (If audits were automatic — a small share of ballots are checked after every election — they might work to build confidence in our elections rather than potentially undermining it.)
Ultimately, though, I’m in the information business. An audit very probably won’t detect a conspiracy, but it will reveal information about our voting systems. FiveThirtyEight and most other American news organizations are founded on the premise that more information is better, even if it risks being misinterpreted. I’ve never questioned that premise more than I have over the course of this election. But over the next four years, we’re all going to have to get used to an environment in which nuggets of insight come buried in mounds of misinformation. An audit is as good a place as any to start.