CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: Estimating the prevalence of peer effects and other spillovers Authors:  David Choi - Carnegie Mellon University (United States) [presenting]
Abstract: In randomized experiments with arbitrary and unknown interference, we show that hypothesis tests for the sharp null of no effect can be inverted with no assumptions on interference, producing one-sided interval estimates (or lower bounds) -- not for the treatment effect, but rather for the number of units who were affected by treatment. Similarly, tests for the null of no interference can be inverted with no assumptions beyond randomization to estimate the number of units that were affected by the treatment of others. This does not fully identify the treatment effect, but may be used to show that a peer effect exists, and to estimate whether it is widely prevalent.