Why This Astrophysics Whiz Went from Biotech to Politics
Political campaigns and party organizations will use this year's midterm election as a testing ground for technology and data analysis they'll refine for the presidential race in 2016. In their struggle to hold on to the Senate majority and maintain U.S. House seats, Democrats this year are applying sophisticated tools for statewide and local candidates, and in some cases that means working with data firms that specialize in state-specific data services.
FiftyOne Percent is one such firm championing data analytics for Democrats in New Jersey this election cycle. The company aims to take the approaches to modeling voter data that helped turn the 2012 presidential race into the most data-centric in history to down-ballot campaigns. And it's not as simple as merely applying the presidential data models to a smaller, local group of voters.
Sherrie Preische, partner and co-founder of FiftyOne Percent, said national models -- the analysis applied to categorize or "score" voters and their likelihood to support a certain issue or candidate -- are not necessarily transferable to state or local elections.
Ms. Preische came to the four-year-old company in a roundabout fashion. Though she has a history in politics and government -- she ran Rep. Rush Holt's office in New Jersey after campaigning for him in 1998, and later served as executive director of the New Jersey Commission on Science & Technology -- her PhD in astrophysical sciences from Princeton University landed her a business development role at biotech firm Genmab.
"I was looking for another technology to do on my own," said Ms. Preische. "I had this political background as well, and I think going through the 2012 cycle I saw what a big influence data was having on the political scene, and I saw that this was my background, this was my sweet spot." Her partner at FiftyOne Percent, President Mark Matzen, was Mr. Holt's chief of staff from 1999 to 2004.
All Political Data Should Be Local
Using sophisticated data modeling is "still very new for down-ballot campaigns," she said. While national party organizations supply local campaigns with voter scores, she explained, "a national model might be built on, say, a ten-thousand sample size from the whole country or something like that, so you're using national averages across the country to build a support model that works across the country. ... The model might work to describe national averages and national trends, but when you try to apply it to this specific race you're using national averages to describe a small subset of people."
New Jersey voters have their own distinct idiosyncrasies, she said. "There are definitely, as you go down [ballot] in New Jersey, people behave very Democratic at a federal level but as they move down the ballot they behave very Republican in a lot of areas, and you have to account for that. So using scores that are based on support for a presidential candidate for a county race doesn't make sense."
People who might be considered persuadable in a national race may be "unreachable" in a local one, she said. For instance, analysis of gun owners on average nationwide might reflect different beliefs than those of gun owners in central Jersey.
That matters because campaigns employ data modeling and scoring to segment voters and help decide which issues might resonate with them, as well as whether it's worth knocking on their doors or calling them. This information becomes more and more critical as time, money and resources leading up to election day dwindle.
Ad Age chatted with Ms. Preische about how her physics background influences her work in politics.
Ad Age: How does biotech or scientific data in general compare or contrast with political data?
Ms. Preische: Physics research and biotech data can usually be collected in a very controlled and automated manner and hopefully somewhat reproducible. Political data, or data about voter behavior, is more like that collected in behavioral sciences or even atmospheric science or studying other large dynamic systems that are not easily controlled or reproduced. In documenting voter behavior, which might be votes cast or voter registration data, there is often a human element of data recording which always makes the records messier.
Ad Age: What barriers do you see to broader adoption of data analytics among down-ballot campaigns? Does it go beyond culture?
Ms. Preische: Many campaigns smaller than state-wide elections simply don't know yet that micro-targeting can help them put together a winning strategy. There's a lot of data out there, but they just don't know that it's relevant to them and their situation. And I would include primary elections in this as well. We did a support model for a hotly-contested congressional primary this year, and the model played a major role in determining the program the campaign put together to win.
Ad Age: There's an ongoing debate regarding whether data and tech firms serving political clients should be partisan -- or whether technology services should be agnostic. What do you think?
Ms. Preische: I have not heard that anyone in the field is actually debating this point. Data by itself may be agnostic but is not all that useful. As soon as you interpret the data and think about how to use it, and even deciding which data to look at, your work becomes political tactics and strategy. That is why as campaigns are using more complex data, it is important to have someone who understands the many facets of a campaign. It is the person with significant campaign experience who can most effectively integrate the data into a campaign as something the campaign can use day-to-day.
Ad Age: Despite the hype around political data and analytics, aren't there elections or campaigns that sophisticated approaches to data can have little to no effect on?
Ms. Preische: Open access to data can help campaigns at all levels. It may be used differently in different types of elections and situations. I know a mayor who probably knows almost every voter in his small city by name. In this situation, there is no sense in having a statistical model to 'predict' voter behavior because he can pretty effectively 'measure' each voter. Yet this mayor will still use data in his campaign, for example, vote history (elections in which each voter has previously voted) to encourage people to come out on election day.