Date:
Location:
Jens Ludwig, McCormick Foundation Professor of Social Service Administration, Law, and Public Policy, University of Chicago.
Social scientists increasingly use randomized controlled trials (RCTs) to inform policy decisions relevant to inequality and other social problems. We usually compare outcomes for those assigned to a treatment condition (or one of multiple treatment conditions) versus control, then do something like recommend the intervention with the largest average treatment effect—a “one-size-fits-all” policy.
But with respect to the determinants of people’s life outcomes there is no reason to believe one size always fits all. A step in the direction of personalization is the growing use of machine learning to understand treatment heterogeneity, which lets us move away from average treatment effects to optimal assignment rules.
One limitation of this approach is the modest sample sizes of our RCTs prevent us from fully exploiting machine learning tools that leverage “big data.” Perhaps an even more important limitation is that so many critical policy decisions do not depend on a causal question—they hinge instead on a prediction. Indeed a growing share – perhaps even now a majority—of the problems that local government agencies ask for help with from the research centers I help run at the University of Chicago (the Crime Lab and the Education Lab) hinge on predictions. In most cases these predictions are currently made by a human, even though we know from behavioral science that this type of probabilistic thinking is difficult for most people.
The potential gains from using machine learning predictions instead can be substantial, as I illustrate for pre-trial bail decisions (Kleinberg et al., QJE, forthcoming) and for hiring decisions for teachers and police (Chalfin et al., AER P&P, 2016). These tools also seem capable of improving fairness as well as efficiency, given the implicit (or explicit) biases by human decision makers.
But realizing the potential of these tools requires solving a number of challenges at the intersection of social science and computer science. For example our work helping the Chicago PD build an early intervention system to prioritize officers for supports raises challenges in evaluating the quality of the machine’s versus human’s predictions, while our work with the Mayor’s Office in New York City to help build and implement a new pre-trial risk tool raises challenges around how to help judges learn where and when they have comparative advantage relative to the new statistical decision aid, and vice versa.
About the speaker
Jens Ludwig is the McCormick Foundation Professor of Social Service Administration, Law, and Public Policy at the University of Chicago, director of the University of Chicago Crime Lab, and co-director of the University of Chicago Urban Education Lab.
He also serves as a non-resident senior fellow in economic studies at the Brookings Institution, research associate of the National Bureau of Economic Research (NBER), and co-director of the NBER's working group on the economics of crime. His research focuses on social policy, particularly in the areas of urban poverty, crime, and education.
In the area of urban poverty, Ludwig has participated since 1995 on the evaluation of a HUD-funded randomized residential-mobility experiment known as Moving to Opportunity (MTO), which provides low-income public housing families the opportunity to relocate to private-market housing in less disadvantaged neighborhoods. In the area of crime, He has also carried out research on early childhood interventions like Head Start, the effects of social conditions on children’s schooling outcomes and risk of violence involvement, and on ways of preventing gun violence.
In 2008, Ludwig helped found the University of Chicago Crime Lab, which partners with local, state, and federal government agencies to carry out randomized policy experiments to learn more about how to prevent youth violence and closely related social problems such as high school dropout. The Crime Lab recently received a $1 million award from the MacArthur Foundation as recognition for effectiveness and creativity. Crime Lab research has helped shape policy in Chicago with projects that have helped identify cost-effective ways to prevent youth from dropping out or becoming involved in violence.
His research has been published in leading scientific journals across a range of disciplines including Science, New England Journal of Medicine, Journal of the American Medical Association, American Economic Review, Quarterly Journal of Economics, the Economic Journal, and the American Journal of Sociology. His co-authored article on race, peer norms, and education with Philip Cook was awarded the Vernon Prize for best article in the Journal of Policy Analysis and Management. He is also co-author with Cook of Gun Violence: The Real Costs (Oxford University Press, 2000), co-editor with Cook of Evaluating Gun Policy (Brookings Institution Press, 2003), and co-editor with Cook and Justin McCrary of Controlling Crime: Strategies and Tradeoffs (University of Chicago Press, 2012).
In 2006 he was awarded APPAM's David N. Kershaw Prize for Contributions to Public Policy by Age 40. In 2012 he was elected to the Institute of Medicine of the National Academies of Science.
Learn more about Jens Ludwig's work
Jens Ludwig homepage
