| > no one... thought he would get close to winning The devil is in the timing, there. Relying on pre-election polling is not a safe way to predict an election outcome, and has been badly wrong on quite a lot of occasions. Relying on pre-election punditry is even worse, since it's not only inadvertently biased but actively spun to promote certain outcomes. And relying on anecdote or acquaintance reports... sample bias aside, if you had friends or coworkers who voted for Trump, would they have admitted that openly? Slightly better, we can look at exit polling to see how people claim they voted. At a glance, 2016 exit polling suggested far higher totals for Clinton than we saw, which is suspicious. But exit polling has systematic biases, and simple percentages have been increasingly unhelpful for several successive elections. The Kerry v. Bush exit polls, for instance, skewed towards Kerry in 28 states and Bush in 4. [1] In particular, 2016 exit polls massively oversampled college graduates in an election where a major factor was non-college Democrats breaking right. [2] Fortunately, we get enough demographic data on exit polls and vote counts to correct for some of this stuff in hindsight. It's not perfect, but what we find is that corrected exit data looks far more like the actual vote count than naive exit polls did. [3] If we want to look for evidence of direct vote manipulation, a surprising election outcome isn't sufficient. There are much more specific things we should expect to see. Broadly, I can think of two dichotomies here: digital attacks (probably systematic and likely foreign) vs. physical attacks (probably local and domestic), and widespread versus swing-state manipulation. For digital attacks of any kind, we'd expect to see a discrepancy in outcomes between networked electronic votes and other (non-networked or analog) votes. Fortunately, Jill Stein's recount initiative offered useful data: we have both paper and digital vote counts in Wisconsin and Michigan. The counts don't match precisely, of course, but studies found no evidence of any skew towards one candidate. Vote totals also don't appear to have had meaningful digital/paper discrepancies or departures from voter roll counts, which precludes other attacks like generating extra votes in red-leaning districts. [4] For physical-access hacks, hard evidence would be much less obvious but statistical evidence would be much more obvious, unless we invent a conspiracy capable of traceless, coordinated physical action across the country. This is the sort of thing we did see in 2000, with thousands of votes from specific districts and even specific polling places vanishing. But the unexpected Trump vote counts were widespread and geographically consistent. Trump voters overall mirrored normal Republican demographics, but geographic results and exit poll counts alike show that the biggest changes were a rightward shift among white non-college Democrats, who are heavily overrepresented in the states which unexpectedly broke for Trump. Forging this sort of neat demographic shift ought to be basically impossible, because you couldn't plan the appropriate adjustments until after you had nationwide vote counts in hand. [5] The hard part of manipulating a US election wouldn't be screwing up a few Symantec machines, but doing enough to change the outcomes without leaving extensive evidence of what was done. It's not enough to say that the outcome was a surprise compared to predictions or that Russia has good hackers, we'd at minimum need to see some kind of actual gap between the results and what people put in ballot boxes and told pollsters they'd chosen. I haven't seen anyone make a decent case for that. [1] https://www.vox.com/policy-and-politics/2016/11/8/13563308/e... [2] https://www.nytimes.com/2018/03/29/opinion/2016-exit-polls-e... [3] http://www.pewresearch.org/2018/02/15/commercial-voter-files... [4] https://www.washingtonpost.com/news/monkey-cage/wp/2017/06/0... [5] http://www.pewresearch.org/fact-tank/2016/11/09/behind-trump... |