Using Facebook Data to Predict the 2016 U.S. Presidential Election

June 2021
PLOS One

We use 19 billion likes on the Facebook posts of U.S. fan pages from Jan 2015 to Nov 2016 to measure the dynamic ideological positions for politicians, media and users at the state and local level. We then use these measures to derive support rates for 2016 presidential candidates in all 50 states, to predict the election, and to compare them with state polls and actual vote shares. We find the following: (1) Under the Hotelling model assumption, support rates calculated from FB predict that Trump will win the electoral college votes by a large margin, in fact, 293 votes. (2) The trends between state-level Facebook support rates and state level polls are similar and pass all of the co-integration tests. Popular vote predictions also perform well. (3) Compared with actual vote shares, polls often overestimate Clinton’s support in right-leaning states, while Facebook often overestimates Trump’s support in these states. Finally, we discuss several limitations of this method. Overall, we provide a new method to forecast elections with low cost and almost in real time.

Spotlights
  • Use 19 billion likes to measure dynamic ideological positions of users and fan pages.
  • Guess user’s geolocation by likes and measure state level support rates for candidates.
  • Assume that users would support candidates with closer ideology.
  • FB support rates predict election outcome well, and share similar trends with polls.
  • Polls systematically overestimate Clinton’s support in right-leaning states.