Tutorial

A tutorial will take place on Wednesday 23rd of June 2021. The number of participants to the tutorials is limited and restricted only to those who attend the conference.

Title: Robust methods for sample selection models and treatment effects
Venue: TBA
Prof. Elvezio Ronchetti, University of Geneva, Switzerland, and Dr. Mikhail Zhelonkin, Erasmus University Rotterdam, Netherlands.
Email: Contact

Description:

A sample selectivity problem occurs when an investigator observes a non-random sample of a population, that is, when the observations are present according to some selection rule. Consider, for instance, the analysis of consumer expenditures, where typically the spending amount is related to the decision to spend. More specifically, the selection bias arises if controlling for explanatory variables, the spending amount is not independent from the decision to spend, i.e. they are dependent through unobservables. This type of problems arise in many research fields, including economics, sociology, political science, finance, and many others. The classical estimators introduced by Heckman (1979) are the backbone of the standard statistical analysis of these models. However, these estimators are very sensitive to small deviations from the distributional assumptions (typically the normality assumption on the error terms) which are often not satisfied in practice.

We first give a general introduction of the basic concepts and tools of robust statistics and we then develop a general framework to study the robustness properties of estimators and tests in sample selection models. We propose a procedure for robustifying the estimators and we construct a simple robust alternative to the sample selection bias test. We further extend the methods for robust estimation of endogenous treatment model and other treatment effects, including the case when the conditional independence assumption does not hold. This is implemented in the R package ssmrob, which can be used both to produce a complete robust statistical analysis of these models which complements the classical one and as a set of useful tools for exploratory data analysis. The package therefore provides additional useful information to practitioners in different fields of applications by enhancing their statistical analysis of these models. We illustrate the performance of our new methodology in the analysis of several empirical examples.

Programme - Wednesday, 23th June 2021
14:00 - 16:00Session I
16:00 - 16:30Coffee break
16:30 - 18:30Session II