Big Data Leads to a Big Win in a Big Election
I am not writing this as a political statement and will refrain from delving into a discussion of politics and the political landscape that has unfolded since last week’s Presidential election. Instead, I wanted to focus on some of the comments by major players from both sides in the days following election night. There are many parts to the narrative explaining the results for this most recent election cycle and undoubtedly there will be much more to come, but one thing that has surfaced as the smoke clears is the importance that great data and analysis had on the outcome of the election.
"Nothing happened on election night that surprised me."
In an interview given by David Axelrod, the campaign manager for the President, he was "not surprised" by the results of the election. Why not? He had great confidence in the internal polling conducted by the campaign, which turned out to be “remarkably close” to actual results in the battleground states. By building on the data they already collected from the 2008 campaign, they were able to construct models for the battleground states with deadly precision. They knew what buttons to push to turn out its base. They segmented their base and devised targeted messages to each group. While Republicans accused the President’s campaign of being about small things, this is how they won. In the end, they had a better understanding of who was likely to vote and, perhaps most importantly, who was not likely to vote. This latter point is key because that was a fatal flaw of Romney’s internal polling efforts. This knowledge coupled with a superior ground game in key districts was critical to the campaign’s success. We can sum it up as:
Data versus gut?
Romney was reportedly shocked at the results of the election. Why is that? Because the analysis conducted by his team led to decisions based on, what turns out to be, a faulty assumption that the Democrat turnout would be closer to +3 points rather than the +6 points that the President’s team modeled. And guess which campaign was correct? While many of the national polls were also coming away with large 6-7 point Democrat advantages in their polling samples, politicos on the Republican side could not fathom the same advantage. It did not pass the smell test for them given the economic climate. Honestly, my gut led me to the same conclusion absent the raw data, particularly given the challenges that many pollsters face in constructing representative samples. While accusing the pollsters of being biased toward one side, it looks like big data won in this cycle. I wonder if this bias against the +6 to +7 turnout crept into the Romney campaign and prevented them from fully vetting their +3 point assumption with data. If so then that was a critical failure and may be the reason why campaign leaders felt they had “let Mitt Romney down.”
Now with all this being said, a look back into the past shows that polls are hit and miss. In 2004, a national poll had Kerry ahead of President Bush in the battleground state Ohio on the eve of the election. They were wrong. Polls are still polls and nothing is certain until election day. But it does appear that the President’s campaign built upon its success in 2008 and achieved a significant competitive advantage over its rival in the use of data. While Team Romney’s data and analysis prompted an expansion of the campaign into traditionally Democrat fortresses, the faulty models gave them a false sense of security and hid underlying problems in the battleground states that Romney needed to win the election. Republicans have reportedly made significant gains since 2008 in the use of data, but I think this election shows that they still have a lot of catching up to do. It's a hard lesson to learn, but one that can help them in future cycles.