The big data models got it all wrong.
Traditional polling, online polling and IVR polling, and the big aggregators didn’t see it coming either. Prediction sites like 538 Politics and the New York Times Upshot had the media convinced Hillary was cruising to victory. Even Slate’s innovative Votecastr project, which represents the cutting edge of predictive modeling, couldn’t see a Trump win hiding in the numbers.
{mosads}The conflicts weren’t only in the media. They were in the campaigns too.
Hillary Clinton built a sophisticated analytics and data operation that included many veterans of Obama’s much lauded digital operation.
Trump’s campaign looked different. “I’ve always felt it was overrated,” Trump said back in May of 2016. “Obama got the votes much more so than his data-processing machine. And I think the same is true with me.”
While Trump’s big data gamble didn’t last, it never defined his operation. Late in the campaign Trump did expand his data team but it almost certainly was a far leaner iteration of Clinton’s data bureaucracy.
So in the end, was big data a big flop? Was Obama’s 2008 data Pearl Harbor not what it seemed?
The aggressive day-after predictions that big data is dead are overblown.
Donald Trump did not prove that big data isn’t worth the investment. But what he did prove is that strategy still plays a role in winning campaigns and that more paths to victory exist than the ones that others have trodden.
Trump is not from the world of politicians. His strategy of mastering social media, large rallies, and media wars weren’t neglectful of the science of campaigns. They were strategic bets that invested in Trump’s strengths and capitalized on his celebrity blazing a unique path to victory.
But those candidates who are waiting in the wings should beware of emulating Trump’s unique advantages. Just because it worked for Donald Trump does not mean it will work for everyone. Not every politician is a celebrity and not all are so watchable that NBC wants them heading up their primetime hour.
Just as data analytics didn’t make Hillary Clinton more likable, celebrity and spectacle won’t transform mundane politicians and into powerful communicators. You either have it or you don’t.
Finally, the prediction index and pollsters deserve a break. It’s political ‘science’ for a reason and, as with all science, it has dangerous limitations.
When you follow polls you assume risk. They have margins of error and, ultimately, are based on humans making assumptions about how to weight responses to reveal real trends. Polling aggregators create a false sense of security that averaging polling results together eliminates the underlying risk. A 66% chance of Hillary Clinton winning doesn’t eliminate the possibility that opposite outcome is possible. It’s 34% possible… and, as Donald Trump will tell you, those aren’t impossible odds.
It’s doesn’t make the predictor wrong if we see what we want to see in the results.
Polls have dozens of pages of cross tabs that accompany them and complex methodology sections that communicate some, but not all, of the unique assumptions that pollsters make trying to get it right.
Reading a poll and believing results without acknowledging the underlying assumptions doesn’t mean the poll is wrong, it just means we didn’t know how to read it and we put too much stock in what it’s telling us.
And that’s why big data isn’t dead. Big data wasn’t the centerpiece of Trump’s strategy and it didn’t need to be.
But, for future candidates, big data may be the difference between winning and losing. Data is not a substitute for an innovative strategy. And if 2016 teaches us anything, it’s that data science can’t fix a weak message or compensate for a powerful enthusiasm among voters for change.