How Kristian Lum M.S. '08, Ph.D.' 10 makes the data matter

With the preponderance of available data has come a preponderance of concern about how the information is used and who possesses it. Kristian Lum M.S. ’08, Ph.D. ’10 counts herself among those concerned. And as lead statistician at the Human Rights Data Analysis Group, she’s in a position to help elucidate data use.

“A lot of my work touches on the ways in which data and machine learning may not work in the way you’d like or hoped,” she says.

In simple terms, she focuses on data fairness. Yes, “fairness” is a subjective term, but in Lum’s work the effort to get closer to that notion is key. Her research includes examining bias in predictive policing systems—analytical techniques used to identify likely targets for police intervention to prevent or solve crimes through statistical predictions. For instance, she’s worked with the City of New York as it seeks to design a fair bail system, by looking at the data it uses to predict the likelihood of recidivism.

Lum’s work also extends to population estimation, which includes looking at better methods to avoid the underreporting of those killed in times of conflict. It’s important not just for historical memory, she says, but also because the numbers can affect policy. “We look at who was targeted and who was responsible.”

She took an introduction to statistics class about halfway through her pursuit of a math degree at Rice University. “I thought it was fantastic,” she says, still sounding smitten. “I felt I had found my home.” Statistics, she says, had the analytical aspect of math paired with “being concrete and grounded in reality.” She added it as a major. Duke became her choice for a graduate education because, she says, she was impressed with the statistics department and the supportive environment. Still, she had no idea what she was going to do professionally.

She did know she wanted to apply statistics in a meaningful way. More clarity came after a friend told her about the Human Rights Data Analysis Group. She sent a cold e-mail to executive director Patrick Ball and asked whether she could work with him. That led to a summer job. She kept the group in her periphery as she earned her Ph.D. After working as a research professor at Virginia Tech and with a tech start-up in San Francisco, she joined the group full time.

Decision makers often use data to make policy decisions; in her role, Lum sees an opportunity to not only make those decisions explicitly consider social-justice concerns, but give those concerns the heft of typically quantifiable issues like cost. “By allowing us to demonstrate patterns of injustice in addition to individual instances of injustice, it helps to shed light on the scope of the problem,” she says.