We all know what happened in the 2016 U.S. election for president: The opinion polls of voters immediately before the election showed that Hillary Clinton would be the likely winner of the election. In fact, the Trump campaign had not even prepared an acceptance speech for Trump because the odds of winning were so poor. However, the election outcome was a surprise because Trump got just the right number votes in the right places to get enough votes in the Electoral College to squeeze by even though the cumulative vote total gave Clinton about 3 million more votes.
I found this event very surprising and disturbing. When I took statistics in college I was required to do a term research project. The project I picked was to use statistical tools to measure the accuracy of the polling in U.S. presidential elections and measure the accuracy of the polling compared to the actual results. I looked at elections going back about 20 years and polls conducted at different times before the election. Not surprisingly, polls conducted well before the election (6 months or more) were less accurate than polls immediately before the election. But what was surprising was that the polls immediately before the election were super accurate: the actual results were accurate within a fraction of 1% of the actual count for polls conducted one week of the election. I wrote the term paper in 1974 (remember Watergate?!) and I have no reason to think that current polling techniques are any worse than they were 45 years ago. In the presidential elections from 1976 to 2012 most polls accurately predicted the outcomes of those elections.
Of course we all carefully watched the 2018 mid-term elections. I was very interested to see whether the election outcomes would be consistent with pre-election polling data. In the 2018 election the results were very consistent with the polling. The only “surprises” were in the races where the difference between the candidates was within the margin of error for the poll. In other words, the election results in 2018 matched the polling just like every election from 2014 and before. The outlier is the 2016 election.
So why the different outcome in 2018, especially since the Russian government’s disinformation and cyber attack campaign is still in full swing? What has changed? Perhaps the answer is in this news article in the Washington Post today: U.S Cyber Command Disrupted Internet Access of Russian Troll Factory
The Washington Post article states that the U.S. took action to block the Internet access of key Russian government and military cyber operations during the day of the election and “a day or so afterward as the votes were tallied,” according to Sen. Mike Rounds (R-S.D.). Few technical details were released but apparently the U.S. also sent messages through e-mail and pop-ups to the Russians to let them know that there is a new sheriff in town and the U.S. was aware of their actions. What is interesting is the fact that Senator Rounds mentioned that it was necessary to block the Russians after the election WHILE THE VOTES ARE COUNTED. This may tell us about what really happened in 2016: it strongly suggests more than a disinformation campaign — it suggests direct hacking of the vote totals in 2016 in a few key states. We do not have direct evidence to support this claim, but it is the only theory that fits the public facts that we know as of today.