PER Distinguished Lecture Series: Susan Athey

March 25, 2016 - 1:00pm - 4:00pm
404 International Affairs Building
420 W 118th Street
New York, NY 10027
United States

PER Distinguished Lecture Series: Susan Athey


Friday, March 25, 2016


404 International Affairs Building


Talk 1: Machine Learning and Econometrics: Estimating Treatment Effect Heterogeneity in Experiments and Observational Studies

This talk will begin with an introduction to supervised machine learning and some observations on how ideas from that literature can be used to improve the practice of empirical analysis in economics, while maintaining the economist’s focus on estimating causal effects and constructing confidence intervals for estimates.  The main part of the talk will focus on two recent papers that modify popular machine learning methods to address the problem of estimating treatment effect heterogeneity in randomized experiments and observational studies.  Applications include personalized medicine and large-scale A/B tests in technology firms.  The first paper provides a data-driven way to identify subgroups with differing levels of treatment effects, while preserving nominal coverage rates for confidence intervals even when there are many covariates relative to the number of observations.  The second paper adapts the widely used random forest method.  We first establish that for a modified version of the random forest prediction algorithm, the predictions are asymptotically centered and normally distributed.  The modified random forest substantially outperforms K-nearest neighbor matching in both mean-squared error and coverage of confidence intervals, so that the modified random forest is a better-performing alternative to KNN and kernel regression in settings with more than a very small number of covariates.  We second show how the algorithm can be further modified to provide personalized predictions about treatment effects, together with confidence intervals.  


Talk 2: The Internet and the News Media

This talk will review a series of recent papers on the impact of the internet on the news media.  The internet has changed how people consumer news, particularly how they select among outlets and how many different outlets they frequent.  This has implications for advertising markets, both by increasing competition and also by changing the structure of advertiser demand in a two-sided market setting.  As consumers increase their switching behavior, advertisers self-select into groups that multi-home across outlets and those that single-home.  From an empirical perspective, we explore how aggregators and intermediaries affect consumer demand and the type of news that is consumed.  Using two natural experiments, we show that Google News is a complement for news consumption at small-to-midsized outlets and a substitute for the largest outlets.  Finally, we show how consuming news through social media changes the types of articles people read on the same topic.