Changing Seasons: Kosuke Imai Returns to Harvard to Expand the Reach of Quantitative Social Science at Harvard and Abroad

April 17, 2020
Kosuke Imai

by Jonathan Colburn

When he last occupied office space at Harvard, Kosuke Imai, IQSS’s newest resident faculty affiliate, had a decidedly different experience than he does in his current location at CGIS Knafel overlooking Cambridge Street.

“IQSS didn’t exist yet, but its predecessor was housed in a little yellow house where the Ukrainian Research Institute is,” Imai says. “My office as a graduate student was upstairs in the attic. They were demolishing the old building that sat here (to build CGIS), and every time the construction equipment would start up my whole office would start shaking.”

His working conditions are just the beginning of the changes to his life since returning to Harvard after completing a Ph.D. in Political Science in 2003 and an M.A. in Statistics in 2002. After spending 15 years as a professor at Princeton, Imai rejoined Harvard as a professor appointed jointly in the Departments of Government and of Statistics, and as an affiliate of IQSS. His return has opened up a world of collaborative opportunities across Harvard and the myriad research institutions in the greater Boston area.

“My work has always been in developing statistical methods, but the methods themselves are applicable across different problems and social science fields,” says Imai. “I was looking for more interaction, not just within political science but beyond, including the professional schools like HSPH and the Law School.”

This search for interaction led to a promising collaboration with Jim Greiner, a professor at Harvard Law School, and Faculty Director of the Access to Justice Lab. Greiner had an interest in evaluating risk assessment scores used by criminal courts. These algorithms are used for a variety of purposes within the criminal justice system, such as helping to determine whether individuals charged with a crime should be released until trial, or held because of a high risk that they will commit further crimes or fail to attend hearings at all. Algorithms like these are pervasive in modern life and can affect hiring decisions at companies, or admission chances at prestigious universities. Naturally, questions of whether these algorithms can be biased against individuals on the basis of ethnicity, gender, income status or other factors have led to the creation of entire fields of research to study algorithmic fairness. But beyond the fairness of a score calculation lies a further question of how people choose to use them in their decision-making.

“What interests me is that there are human judges making these decisions, not just following the numbers,” says Imai. “Much like the way that doctors are receiving a lot of information provided by algorithms, at the end of the day both [judges and doctors] are making their own decisions. The scores themselves may be biased, and yet humans have their own biases, and these may interact in ways that we don’t understand yet.”

To help provide answers, Greiner has convinced some states to participate in a study in which some judges will have access to these scores while making decisions about bail, and some will not. Imai and his collaborators have developed new statistical methods for assessing the impacts of machine recommendations on human decision making. Using these new methods, they just started analyzing data. The road to conclusive results will necessarily rely on some assumptions. For example, it is of course impossible to know what choices individuals who were not granted bail would make if they were instead released, but the research may lead to much more equitable policy outcomes. “There is a lot of research on determining the most optimal, or fair, score, but the most fair score may not lead to the most fair decision by a judge,” Imai says.

Kosuke Imai in front of a white boardThe project is emblematic of a whole world of possible collaborations for Imai, whose research has also covered areas such as evaluation of congressional redistricting, software development for municipalities that are trying to merge voter records for more accurate information and analyzing the impact on elections in states where inactive voters are purged from rolls. His return to New England has resulted in more than just expanded research opportunities.

“I’ve been biking to my office from Newton most days,” says Imai. “I started when the weather was nice, and I wasn’t sure whether I could make it through the winter, but I did it last year. I bike in from Newton, along the Charles River; it’s beautiful in all four seasons. I’m originally from Japan, and I really like the changing seasons.”

Imai’s two teenage children have adjusted to their new community as well, helped in part by the fact that annual family ski trips in Vermont, where his children learned to ski from a young age, are now a few hours shorter.

He has also been able to pursue another multi-year project in partnership with the University of Tokyo, where he has been teaching for a month every year in an effort to train young researchers and help build a quantitative scholarly community.

“Quantitative social science is a very active area of research in the United States, but in Asia the scientific study of society is still not a major area of focus,” Imai says. “A lot of the research there tends to be more qualitative or historical, as opposed to using data to scientifically evaluate hypotheses, but there is an interest. Everyone is of course excited about data.”

In addition to his course, he also helps to lead an annual Asian Political Methodology Meeting. He admits that it’s taken some time to get used to sitting in department meetings with faculty peers who served as his instructors not so long ago. But having seen the results of his teaching firsthand, Imai hopes that some of his own students may have the same opportunity to join him in turn.

“I have some students from Japan, Korea, and elsewhere who have taken my class there and then come to Harvard as graduate students,” Imai says. “There are some very smart students out there, they just need some opportunity to get some training. My impact feels like it’s on a small-scale in some ways, but if I train one person, that person will train others and I’ll be able to see the impact grow.”