CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Robust causal inferences in small area estimation Authors:  Setareh Ranjbar - HEC Lausanne (Switzerland) [presenting]
Nicola Salvati - University of Pisa (Italy)
Barbara Pacini - University of Pisa (Italy)
Abstract: When doing impact evaluation and making causal inferences in many cases, it is important to acknowledge the heterogeneity of the treatment effects for different domains. Where certain geographic, socio-demographic, or socio-economic unplanned domains may benefit from a program/policy intervention, others may be worse off. If the domain for which we are interested in the impact, is small with regards to its sample size (or even zero in some cases), then the evaluator has entered the small area estimation (SAE) dilemma. In addition small area estimators are intrinsically very sensitive to the presence of outliers due to the small sample sizes. Therefore it is important to develop or make use of robust methods in SAE. Based on the modification of inverse propensity weighting and the robust small area estimators, we propose new methods that allows one to robustly estimate the area specific average treatment effects for the unplanned domains. The Mean Squared Error (MSE) of the proposed predictors are analytically approximated for the situations that propensity scores are taken as known and a bias calibration method is also provided. By means of these methods we can provide a map of policy impacts that can help to better target the treatment group(s).