The Asian Political Methodology Meeting Best Poster Award

Soichiro proposes a general identification assumption and estimation procedure for the average treatment on the treated that utilizes multiple pre-treatment periods. Specifically, he assumes the bias for the post-treatment period as a function of past biases. By imposing smoothness on the mean function for potential outcomes, he proposes a feasible estimator that combines difference-in-means with bias-correction under this identification assumption. He demonstrates usefulness of the proposed method by simulations and an application.

Existing methods for missing data have their shortcomings: information loss, over-fitting due to cyclic usage of data and excessive distributional assumption. To address these concerns, Naijia proposes an honest inference method which assumes a probability distribution of the data and a missing at random scheme as well as conducts double machine learning procedure on the observed and missing groups, to obtain the bias-correction term for mean imputation. Multiple sample-splitting is conducted throughout the method for the consideration of variance estimation. Her simulation result illustrates superiority of her method.