Like many Harvard students in their senior spring, Nathán Goldberg ’18 and William Long ’18 had big ideas for their future as they prepared for graduation. Both had a background in Big Data. Both were brought deeper into national politics as the 2016 election unfolded while they were at Harvard. And both happened to discover that, despite the pronounced divisions between the Democratic and Republican parties that presidential race is remembered for, the two parties did have one unfortunate problem in common: unexpectedly poor data infrastructure.
Long, who now runs his own data platform for political campaigns, was knocking on doors for Marco Rubio in New Hampshire during that election. Volunteers used a canvasing app for their mission, although Long noticed it was “very, very bare bones.”
“It was really surprising because it’s a pretty major presidential campaign,” he said, “but it looked like something an undergrad could have put together in a weekend.”
When Goldberg was working on Beto O’Rourke’s senate campaign in Texas two years later, he was struck by a similar problem. “Even the most expensive U.S. Senate race in history did not have the proper data support eight months out from election day,” he said, referencing the unprecedented donations O’Rourke’s campaign received at the time. Both students seemed to have stumbled into a gap in the political infrastructure—and a major one, too, given what they knew about how effectively harnessing data could change the fate of a campaign.
Now, Goldberg and Long are running major organizations attempting to close that gap.
For Goldberg, a Democrat, that means training ranks of college undergraduates already versed in data analysis to apply their knowledge in a political setting. His nonprofit—which he named Bluebonnet, after the Texas state flowers which literally “turn Texas blue” upon blooming each spring—trains more than 300 fellows each year. Once they’re trained and ready to enter the field, Bluebonnet, co-founded with Danielle Strasburger ’18 and Paul Dingus M.P.P. ’23, matches them with progressive Democratic campaigns to assist, and ideally help win.
For Long, a Republican, closing the gap has meant creating a platform for conservative campaigns to process data and transform it into phone calls, texts, and surveys. The platform has a canvasing app, too, and can model voter behavior and preferences. Since starting Numinar, Long’s company has worked with more than 900 campaigns. He cites its work with Glenn Youngkin’s 2021 campaign for Virginia governor, where its model for mapping the state’s undecided voters and their major issues guided the campaign’s strategy for who to target with what issues—such as abortion, or election integrity. Come election day, “the modeling ended up being almost completely correct,” Long said. The margin of error was just two percent.
It’s that relatively novel ability—to take immense amounts of publicly available information about the public, break it down, and process it into predictions of the behavior of large demographic groups—that has been driving new and more efficient strategizing in modern political campaigns. As funding and manpower help them aggregate more data and analyze it better, campaigns are more closely able to map which voters to target, with what messaging, and by which means of communications, all to maximize impact.
More recently, as the science has improved, data analysis has enabled campaigns to tell not only what voters will do (such as predicting whether a voter will vote for your candidate) but what they might do: how different interventions will affect that possibility and make them more likely to vote one way or another.
Even beyond understanding voters, the ability to analyze data is a valuable tool. In one 2019 race for the Virginia House of Delegates, Goldberg said Bluebonnet wanted to help a Harvard alumnus advance a key argument to voters: that Republican policies had led to hospital closures in the state. To make it stick, they were able to map current hospitals, those that had closed, and the addresses of voters within the race’s district, and then present voters with statistics about how much the drive time to the nearest hospital had increased following hospitals’ closures. This is one example of how campaigns have gotten creative in harnessing the mountain of personal information available commercially online and from the government to create more effective campaigns. That includes looking at people’s age, zip code, class, age, gender, and more to produce an “expected vote value” for them, Goldberg explained. (An early application of these techniques closer to home was Harvard Forward’s campaign to qualify candidates for the board of overseers by petition, and to elect several of them—a campaign Goldberg helped lead.)
Having found success in the once-emerging field of political data, Long and Goldberg credit the resources and connections they found at Harvard for part of their success. While preparing to launch careers in political data during their senior spring, both relied on meetings with fellows from the Institute of Politics to better understand the data gap, and how they could close it. Two of Bluebonnet’s other co-founders were also Harvard graduates, and Long first began building Numinar in a school of engineering and sciences class specifically for creating start-ups.
For Goldberg, his Harvard background remains a potential advantage for Bluebonnet’s future. He’s aiming to expand Bluebonnet further, and said he was hopeful the enterprise could continue benefiting from the school’s “connections and support.”
And as Harvard—already renowned for its computer science and statistics programs—continues expanding its offerings in computer and data science, the cases of Numinar and Bluebonnet suggest the ecosystem will be there to keep its graduates at the forefront of political technology, whatever their partisan preference, in elections to come.
