Special Events:
Pre-Conference Workshop: Deploying Machine Learning Tools for Public Policy Impact
Wednesday, November 6, 2019: 11:00 AM-5:00 PM
I.M Pei Tower: Majestic Level, Majestic Ballroom (Sheraton Denver Downtown)
Speakers: Jason Anastasopoulos1, Peter Bergman2, George Chen3, Alexandra Chouldechova4, Jennifer Doleac5, James Evans6, Jens Ludwig6, Sendhil Mullainathan6, Aaron Roth7, Chenhao Tan8 and Stefan Wager9, (1)University of Georgia(2)Columbia University(3)Carnegie Mellon University(4)Carnegie Melon University(5)Texas A&M University(6)University of Chicago(7)University of Pennsylvania(8)University of Colorado, Boulder(9)Stanford University
Join us the day before the Fall Research Conference kicks off for this year's pre-conference workshop: Deploying Machine Learning Tools for Public Policy Imapct. Attendees can register for the workshop during the full conference registration process, opening on July 15th. Registration for the workshop will be $55 for APPAM members and $65 for non-members.
Workshop Description:
Organized by: Alexandra Chouldechova, Carnegie Melon University; Jens Ludwig, University of Chicago; and Sendhil Mullainathan, University of Chicago
While public policy analysis as a field has had considerable impact by helping to answer key causal questions, many policy decisions hinge not on a causal inference but instead on a prediction: Which defendants are too high risk for a judge to release from jail as they await adjudication of their case? Which calls of potential child abuse are most likely to reflect actual abuse? Which students are at elevated risk for dropping out of school, and so should be prioritized for academic supports? Which households are most likely to be eligible for social services but unlikely to be enrolled in them? Which chronically ill low-income patients are most likely to miss doctor’s appointments or forget to refill or take their prescription medications, and so might particularly benefit from home visits or reminders?
The growing availability of government administrative records (‘big data’) combined with new tools from the computer science field of machine learning create important new possibilities for substantial impact across a broad range of policy problems. These new machine learning tools share some similarities with the usual policy analysis tool-kit (regression, matching, etc.), but do have some important differences in their goals and methods. They also make it possible to draw on new sources of data that policy analysts historically have not even recognized as data, such as written text, audio clips, or video images.
At the same time, re-deploying these machine learning tools to policy problems raises new challenges that are quite different from those associated with canonical computer science applications, which create important opportunities for the field of policy analysis to add value. These include: challenges in evaluating the social impact of algorithms in the presence of outcome data that is non-randomly missing, as the result of the decisions of humans in the existing policy system; potential mis-alignment between what is being predicted and the objectives that policymakers are trying to optimize; and the possibility of algorithmic bias, and how that compares to human bias.
The workshop seeks to provide an accessible introductory overview to machine learning tools, illustrate the range of policy problems to which they can be applied, develop understanding of what makes for a good policy application for these tools, what can go right (and wrong), and where and how policy analysts can add value to making progress on these problems. We will also include some discussion about how machine learning tools can be useful for solving the sort of causal inference problems that have traditionally been the focus of policy analysis work.
