No one saw it coming. The public polls, the experts, and the pundits: just about everybody got it wrong. They were wrong-footed because they didn’t understand who was going to turn out and vote last Tuesday.
Except for Cambridge Analytica, the data company at the heart of the Trump campaign, working in collaboration with the RNC and Brad Parscale.
The firm knew that Mr. Trump had a very solid shot at winning, because it saw trends that no else did, and it knew how to interpret them correctly.
“The outcome was very difficult to predict, and we didn’t get every state right, but we saw the trends that meant we were quietly confident.”
The team’s internal data saw what was going to happen in states like Pennsylvania and Ohio, because it understood who was going to vote on Election Day.
Uniquely, Cambridge Analytica understood how Trump supporters were different from traditional Republican voters such as those who voted for Mitt Romney in 2012. It knew who those Trump supporters were and the issues they cared about.
Ahead of the election, Cambridge Analytica’s internal data showed the race tightening because its data scientists had seen previously hidden trends in voter sampling, and new trends in absentee ballots and early voting, particularly in rural areas. “This led us to predict a significant lift for Mr. Trump in the industrial Midwest,” says Lead Data Scientist Dr. David Wilkinson.
Among the trends: an increase in the rural vote, a drop in African-American turnout, and a one to three point boost from voters who had hitherto not declared themselves for Trump: the so-called “hidden” vote.
“Their analysis was based not on punditry or the art of politics, but on data science and a rigorously scientific approach to research and polling,” says Cambridge Analytica CEO Alexander Nix.
“This is not something that political intuition would tell you,” says Oczkowski, “but our models predicted most of these states correctly. What also gave us an advantage over polling companies is how quickly we were able to react by updating models to take into account where the demographic was shifting.”