How Twitter Predicted the Iowa Caucus Results

Looking at the chatter on Twitter from users in Iowa was a better indicator of success at Monday’s caucuses than data from other social networks.

In the week leading up to Iowa, the volume of Twitter conversation in Iowa about Ted Cruz increased, pulling him narrowly ahead of front-runner Donald Trump. Data from Facebook and Google did not show the same results.

“How Iowans were tweeting about both GOP and Democratic Presidential candidates in the week leading into the Caucuses provided an insightful perspective—in the end, Iowans Caucused like they tweeted,” Nick Pacilio, who handles communications for Twitter’s government and elections team, told TIME.

Among Republicans, Cruz won Iowa with 28% of the vote with 28% and Trump and Rubio followed, with 24% and 23% of the vote, respectively. But before a victory at the Iowa caucuses Monday there was a shift in momentum of online chatter about the candidates that predicted Ted Cruz to win, according to data from Twitter.

Twitter conversation in Iowa from Jan. 24-31 showed Cruz and Marco Rubio rising while Trump’s large lead shrunk. Cruz’s share of the total Twitter conversation in Iowa rose to 31%, narrowly besting Trump’s share of the conversation at 30%.

Results for the Democrats on Twitter mirrored the actual results of what was a tight race, but a victory for Hillary Clinton. Conversation around Sanders gained 9%, while Clinton saw a 4% drop on Twitter during the weeks leading up to Iowa.

“Momentum shifts in Twitter conversation in the weeks leading into the Caucuses favored both Cruz and Rubio over Trump on the GOP side, and Sanders on the Democratic side, leaving Clinton with only a slight advantage,” Pacilio told TIME.

Other social networks showed a different picture from the results at the end of the night. Trump led Cruz in the number of Google searches about the candidates and in the total Facebook conversation on Monday. Sanders led Clinton in the share of the conversation around Democrats on Facebook, 73% to 23%—far from predictive of the tight race between the two candidates in Iowa.