Has anyone done deep data analysis on the last election results?
February 22, 2017 6:10 PM
Specifically, has anyone done or tried to do an analysis of the latest election results data to determine if there's any correlation between polling places Trump narrowly won in swing states (especially those with Republican governors) and electronic voting systems that don't produce a verifiable paper trail? Or anything similar, looking at discrepancies between exit polls and reported results?
A few years back, when Bush narrowly eked out his first victory over Al Gore, there was a massive push of popular concern about manipulation of electronic voting systems that don't produce an auditable paper trail. There were isolated cases in which poll hacking was suspected, but because the system's counts weren't fully accountable and verifiable by design, there was no way to tell with a high degree of certainty if the machines had been hacked or not.
Security experts investigated these systems and found they were vulnerable to a variety of security exploits that even moderately skilled hackers could exploit to potentially manipulate election results. The public attention moved on from the issue, but apart from reforms at the local levels in some specific (mostly Democratic) localities, no system wide measures were ever put in place to correct for these security gaps and the dangerous lack of accountability as far as I know.
Since Trump eked out such a narrow, precisely targeted victory that relied on gaming the electoral college system, even minor swings of a few thousand votes in certain key districts could potentially make the difference between a major loss in both the electoral college and popular vote and a major electoral college win. If Russia intended to swing the election for Trump, it's not hard to imagine Russian intelligence services would certainly have had plenty of technically sophisticated hackers to throw at the problem.
I'd like to know if anyone has even attempted a serious analysis to rule out the possibility our election systems may have been compromised through a coordinated intelligence operation to carry out such targeted, precise attacks, or if there might be any publicly accessible, aggregated data sets that could be mined to look for such correlations.
A few years back, when Bush narrowly eked out his first victory over Al Gore, there was a massive push of popular concern about manipulation of electronic voting systems that don't produce an auditable paper trail. There were isolated cases in which poll hacking was suspected, but because the system's counts weren't fully accountable and verifiable by design, there was no way to tell with a high degree of certainty if the machines had been hacked or not.
Security experts investigated these systems and found they were vulnerable to a variety of security exploits that even moderately skilled hackers could exploit to potentially manipulate election results. The public attention moved on from the issue, but apart from reforms at the local levels in some specific (mostly Democratic) localities, no system wide measures were ever put in place to correct for these security gaps and the dangerous lack of accountability as far as I know.
Since Trump eked out such a narrow, precisely targeted victory that relied on gaming the electoral college system, even minor swings of a few thousand votes in certain key districts could potentially make the difference between a major loss in both the electoral college and popular vote and a major electoral college win. If Russia intended to swing the election for Trump, it's not hard to imagine Russian intelligence services would certainly have had plenty of technically sophisticated hackers to throw at the problem.
I'd like to know if anyone has even attempted a serious analysis to rule out the possibility our election systems may have been compromised through a coordinated intelligence operation to carry out such targeted, precise attacks, or if there might be any publicly accessible, aggregated data sets that could be mined to look for such correlations.
This thread is closed to new comments.
You are probably aware that some analyses got press soon after the election, but if not you can get interpretations that "persuasive evidence" was found or Scientific American saying it'a bad analysis.
I'm not neutral in which take is correct. Scientific American is right. And I think that ties into why you likely won't see more studies, certainly not ones taken seriously by people with statistical reputations to protect*. This strikes me as the sort of analysis that will have every single problem that leads to the "irreproducibility crisis" in science. Take a noisy data set, have trillions of possible ways to parse it, and look for very small anomalies, have no way to repeat the experiment, and you will be guaranteed to find outliers. There certainly is no way to "rule out" or "rule in" problems if you are considering hacking at the level of tens of thousands of votes. Andrew Gelman's colorful metaphor for this sort of thing is trying to weigh a feather by giving it to a kangaroo bouncing on a bathroom scale.
If the hypothesis was that there was heavy manipulation--I dunno the number but hundreds of thousands, maybe millions?--of votes were hacked in key states you might have a fighting chance of detecting it. But you'd need more traditional methods than data mining to test a hypothesis of small manipulations though. (FWIW I personally wish we'd seen recounts.)
*Another prediction: The mainstream press will use these criticisms to ignore such studies, which is the correct action but for the wrong reason--they publicize lots of crappy stuff when it gets headlines and clearly can't distinguish good studies from bad. But they'll view this as universally off limits because "afflicting the comfortable" is no longer part of their ethos.
No I'm not constantly seething about the middlebrow pandering from the NYT or NPR over the last several years why do you ask?
posted by mark k at 10:10 PM on February 23, 2017