Jeudi 17 février 2022 | Jason Poulos (Harvard University) : Counterfactual Imputation via Matrix Completion with Staggered Treatment Implementation.
Heure : 15:30 ; Par zoom : https://uqam.zoom.us/j/87540246412
Résumé : An important problem in the social sciences is estimating the causal effect of a binary treatment on a continuous outcome over time. A recently proposed matrix completion method for counterfactual imputation decomposes observed outcomes into matrices of latent factors and factor loadings and imputes missing potential outcomes based on the estimated factors and loadings. The estimator uses matrix norm regularization to produce a low-dimensional representation of the observed outcomes and thereby improve generalizability when imputing the missing (counterfactual) values. I focus on a novel “retrospective” framework that uses units exposed to treatment throughout the panel (always-treated) to form a control group when never-treated units are unavailable. The target population consists of switch-treated units that enter treatment after an initial time, which varies across units. Two extensions to the estimator are proposed: (i.) weighting the loss function by the propensity score to correct for imbalances in the covariate distributions between the observed and missing values; and (ii.) imputing endogenous covariate values when estimating potential outcomes. An evaluation of the effect of European integration on cross-border employment illustrates the method and framework. This talk is based on joint work with Andrea Albanese (LISER), Andrea Mercatanti (University of Verona), and Fan Li (Duke).