Using Facebook data to predict the 2016 U.S. presidential election

PLoS One. 2021 Dec 1;16(12):e0253560. doi: 10.1371/journal.pone.0253560. eCollection 2021.

Abstract

We use 19 billion likes on the posts of top 2000 U.S. fan pages on Facebook from 2015 to 2016 to measure the dynamic ideological positions for politicians, news outlets, and users at the national and state levels. 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-level polls and actual vote shares. We find that: (1) Assuming that users vote for candidates closer to their own ideological positions, support rates calculated using Facebook predict that Trump will win the electoral college vote while Clinton will win the popular vote. (2) State-level Facebook support rates track state-level polling averages and pass the cointegration test, showing two time series share similar trends. (3) Compared with actual vote shares, polls generally have smaller margin of errors, but polls also often overestimate Clinton's support in right-leaning states. Overall, we provide a method to forecast elections at low cost, in real time, and based on passively revealed preference and little researcher discretion.

MeSH terms

  • Humans
  • Models, Theoretical*
  • Politics*
  • Predictive Value of Tests
  • Social Media*
  • United States

Grants and funding

The authors received no specific funding for this work.