Some Observations on Halderman’s Concerns About Election Hacking

Word has been going around that some computer scientists have urged Clinton to challenge the election results because of possible hacking-related voter fraud in key states:

Hillary Clinton’s campaign is being urged by a number of top computer scientists to call for a recount of vote totals in Wisconsin, Michigan and Pennsylvania, according to a source with knowledge of the request.

[…]

The scientists, among them J. Alex Halderman, the director of the University of Michigan Center for Computer Security and Society, told the Clinton campaign they believe there is a questionable trend of Clinton performing worse in counties that relied on electronic voting machines compared to paper ballots and optical scanners, according to the source.

The group informed [the Clinton campaign] that Clinton received 7% fewer votes in counties that relied on electronic voting machines, which the group said could have been hacked.

Their group told Podesta and Elias that while they had not found any evidence of hacking, the pattern needs to be looked at by an independent review.

I have a few observations:

First, as of the time I’m writing this, nobody has found actual smoking gun proof that any election machines were hacked, let alone that a significant number were.

Second, Halderman is a legitimate computer scientist and an expert on computer security. That doesn’t necessarily mean that he’s not also some kind of political hack, but in that absence of evidence of dishonesty, we should probably make a rebuttable presumption that he knows what he’s talking about.

Third, Halderman’s statistical observation, as described here, is pretty weak. It’s based on a correlation showing that Clinton received fewer votes (compared to polls) in counties that used electronic voting machines. Like all correlations, whether it has found something meaningful depends on whether co-factors have been eliminated.

In this case, the problem is that use of electronic voting machines was not randomly assigned to counties. This makes it likely that both the decision to use electronic voting machines and the decision to vote for Trump are somehow related to a third factor. For example, Nate Silver has argued that the effect goes away when you control for race and education. This suggests that (I’m just guessing) perhaps affluent well-educated white people are more likely to lie and say they didn’t vote for Trump, and affluent counties are more likely to spring for electronic voting machines. You’d want to rule out things like that before declaring that the election had been fixed.

Fourth, the preceding does not mean that Halderman is (perhaps dishonestly) leaping to conclusions. Rather, this is how scientific investigations work. You begin by doing a quick and inexpensive investigation to see if it looks there might be something interesting going on. In this case, if they’d found no correlation whatsoever between electronic voting and a deviation from the polling data, they could have pretty much ruled out hacking and moved on to investigating something else. However, because the quick statistical analysis couldn’t rule out some kind of problem, the next step is to investigate further, perhaps by using more advanced statistics, or by examining the paper trail from the vote.

Fifth, Professor Halderman says pretty much the same thing in his post on the subject. He makes it quite clear that his best guess is that the election has not been tampered with. However, given the well-known security vulnerabilities in many of our electronic voting systems — Halderman’s team has hacked real voting machines in the lab — he thinks it would be nice to examine the paper trail, just to make sure. But random college professors have no legal standing to get this done. The legal request has to come from one of the candidates, and as a practical matter, only the losing candidate has an interest in a recount. That is pretty much all that Halderman is asking for: That someone in a position to do so asks to take a look at the evidence.

It’s worth noting that computer scientists as a group are unusually skeptical about the security of electronic voting machines. (I have two degrees in computer science, and I’m skeptical.) Before the first electronic voting machines appeared, computer scientists had spent decades researching how to build secure voting systems, yet it’s clear that few real-world electronic voting machines are based on that research. I’ve also heard that voting machines are not designed and operated in keeping with modern security practices. For all those reasons, a lot of computer scientists think it’s a good idea to stick to paper ballots.

Leave a reply

css.php