The Irony of Hillary Clinton’s Data Analytics

Barack Obama’s campaign adopted data but Hillary Clinton’s campaign has been molded by data from birth. Politico has the remarkable story:

Staff in Clinton’s analytics department sit under a sign that hangs from the ceiling with the words “statistically significant” printed on it. And overnight, in some of the few hours that headquarters isn’t whirring with activity, the team’s computers run 400,000 simulations of the fall campaign in what amounts to a massive stress-test of the possibilities on Nov. 8.

…“I have never seen a campaign that’s more driven by the analytics,” [one] strategist said. It’s not as if Kriegel’s data has ever turned around Clinton’s campaign plane; it’s that her plane almost never takes off without Kriegel’s data charting its path in the first place.

…Among the pioneering areas Kriegel’s analytics team has studied, according to people familiar with the operation, is gauging not just whom to talk to, how to talk to them and what to say — but when to say it. Is the best time to contact a voter, say, 90 days before the election? 60 days? One week? The night before? It is a question Obama’s team realized was crucial to mobilizing voters in 2012 but had never been truly analyzed. With a full calendar of competitive primaries, Kreigel and his team had plenty of chances to run rigorous, control-group experiments to ferret out answers to such questions earlier this year.

Here is one fascinating bit on the algorithms that were used to estimate delegate flippability in the primary:

First, the campaign ranked every congressional district by the probability that campaigning there could “flip” a delegate into Clinton’s column. Because every district has a different number of delegates allocated proportionally (in Ohio, for instance, 12 districts had 4 delegates each while one had 17), this involved polling and modeling Clinton’s expected support level, gauging the persuadability of voters in a particular area and then seeing how close Clinton was to a threshold that would tip another delegate in her direction. (At the most basic level, for instance, districts with an even number of delegates, say 4, are far less favorable terrain, as she and Bernie Sanders were likely split them 2-2 unless one of them achieved 75 percent of the vote.)

That so-called “flippability score” was then layered atop which media markets covered which seats. If a media market touched multiple districts with high “flippability” scores, it shot up the rankings. Then the algorithm took in pricing information, and what television programs it predicted the most “flippable” voters would be watching, to determine what to buy.

The irony? More questions are being asked, more data is being collected and more randomized experiments are being run in the effort to win the presidency than will ever be used to choose policy by the presidency. Sad.

Comments

Comments for this post are closed