Statistics Colloquium

Location

Mathematics/Psychology : 103

Date & Time

May 9, 2014, 11:00 am12:00 pm

Description

Speaker
Dr. Claude Messan Setodji
Senior Statistician & Professor
Pardee RAND Graduate School
The RAND Corporation

Short bio:
Dr. Setodji is a senior statistician and the co-director of the RAND Center for Causal Inference and he has interest in applications of statistics to public policy, causal inferences, sampling techniques, data reduction and visualization. Dr. Setodji has done extensive work in the analysis of complex longitudinal data, especially in the value-added methodology commonly used to isolate the contribution of clusters on outcomes of interest over time. He was part of the RAND research team that developed the value-added model framework and statistical tools to capture for example the persistent of teacher effects in light of changing classroom contexts that result when students move on to other teachers. Recently, Dr. Setodji has been working on the issue of causal inference in the potential outcome framework in the presence of measurement error in covariates.

Title
Propensity Score Weighting in the Presence of Error Prone Covariates

Abstract
Inverse probability weighted estimates are widely used in applications where data are missing due to nonresponse or censoring and in the estimation of causal effects from observational studies. The current estimators rely on ignorability assumptions for response indicators or treatment assignment, and outcomes, conditional on observed covariates which are assumed to be measured without error. However, measurement error is common in variables collected for many applications. For example, in studies of educational interventions, student achievement as measured by standardized tests is almost always used as the key covariate for removing hidden biases but standardized test scores often have substantial measurement errors for many students. This presentation provides an approach that can be used to correct propensity score weighting estimators for covariate measurement error and yield a consistent estimator for population means.