Trump won Florida by 1.2 points, Pennsylvania and Wisconsin by 0.7, and Michigan by 0.3, a total of 75 electoral votes. Flip those and Hillary wins 307-231. Flip Florida and any one of the other three and she wins narrowly. Dude.
Nah, I’m kidding. I don’t buy this conclusion for a second, even not having read the study. It smells like the flip side of Trump’s theory that illegal immigrants somehow duped election officials by the millions and cast enough illegal votes to give Hillary a popular-vote win. In both cases you’ve got nefarious actors, either ineligible voters or (presumably) the darned Russkies, somehow manipulating the final numbers. Although, in fairness to the authors, they’re not alleging that any illegal votes were cast in the “Twitter bot” hypothesis. It seems to be purely a matter of bots influencing voters in key areas somehow.
Their rough calculations suggest bots added 1.76 percentage point to the pro-“leave” vote share as Britain weighed whether to remain in the European Union, and may explain 3.23 percentage points of the actual vote for Trump in the U.S. presidential race…
According to the study, bots tended to influence people most when their message backed up their prior opinion. For instance, Trump supporters tended to react to messages spread by pro-Trump bots. And information reverberated quickly: it was generally disseminated and absorbed among Twitter users in 50 to 70 minutes…
The authors identified bots by their unusually large number of tweets, whether they tweeted the middle of the night, and whether they re-posted identical messages, among other criteria.
To figure out how tweeting influenced votes, the study authors looked at the share of pro-leave or pro-Trump tweets by geography to check how closely votes were correlated with Twitter activity. They then figured out how much the accounts they defined as bots added to the volume of tweets advocating Brexit or Trump, and extrapolated from there.
There’s a lot riding on that “may.” The idea seems to be that correlation does (or “may”) mean causation when it comes to the volume of bots in a given region and the number of votes Trump got there. If he performed above expectations consistently in areas where there was heavy bot activity, the theory goes, then maybe it was the bots that were influencing people. But … how? The idea that thought-farts like tweets are meaningfully influencing anyone, let alone influencing how they’re likely to vote for president, is inherently ridiculous. Bot tweets are usually obviously propagandistic too, blunting their persuasive effect. And Twitter’s reach is small by social-media standards. Facebook has many, many more users. Somehow tweets were potentially worth four to five million votes?
If there’s a “bot effect” in the data, I wonder if the causation didn’t run the other way. Maybe whoever was creating these bots had a means of detecting real, organic grassroots interest in Trump on Twitter by voters in particular regions, then logically chose to target those areas by ramping up the bots in that area to reach a more receptive audience. It might not be that the bots were driving voter interest, in other words, but that voter interest was driving the bots. Although that would be an interesting finding too, as it would mean that the people responsible for the bots were picking up clues that Trump was going to overperform in certain regions that even some pollsters hadn’t detected.
It could even be that the bot puppeteers had a better sense of the real electorate than, er, Hillary Clinton’s campaign did. Digest this quote, from the new book “Chasing Hillary” about the Clinton campaign, noting the state of play on Team Hillary around a week before the big vote:
“Robby” is of course Clinton campaign manager Robby Mook, the same guy who thought the thing to do after she underperformed and lost the Michigan primary to Bernie Sanders was to *not campaign too much* in the Rust Belt in the general election. If the bots were worth three points to Trump, how many points was Robby worth to him? Twenty?