Did YouTube Attention Predict How Counties Voted in 2016? Evidence from the Trump and Clinton Campaigns

Do the political videos people watch online tell us how their area will vote? I test this with a record, by county and by day, of YouTube views of Donald Trump and Hillary Clinton content from 1 January to 15 October 2016, covering 2,184 US counties in 49 states. Counties where Trump took a larger share of candidate views voted more Republican, but nearly all of that is geography and past voting. Within a state, holding demographics and the 2012 vote equal, a 10 point higher Trump share goes with a 0.53 point larger Republican swing from 2012 to 2016 (95% CI 0.33 to 0.74). The 2008 to 2012 swing shows no such relation. Comparing neighbouring counties gives smaller but still positive estimates: 0.23 for neighbours in the same state, and 0.33 to 0.40 for neighbours across a state line once each border is held fixed. Allowing for noise in the view share raises the baseline to about 0.6. Attention measured before Trump was formally nominated predicts the swing about half as strongly as general election attention, which fits a stable county orientation better than exposure to the campaign. The swing comes mainly from lost Democratic votes, with a smaller Republican gain and no change in total votes. Coarse Google Trends rankings from 2015 do not predict the swing, and the YouTube share does not improve prediction for counties in states the model has not seen. In short, YouTube views are a small local sign of where a county was already heading. They do not predict votes beyond that, and they are not evidence that videos changed anyone's vote. Preprint, not peer reviewed. The author writes in a personal capacity; views are the author's own.