Activity Number:
|
196
|
Type:
|
Contributed
|
Date/Time:
|
Monday, August 1, 2016 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Social Statistics Section
|
Abstract #321328
|
|
Title:
|
Adjustment by Minimum Discriminant Information in the Presence of Measurement Error
|
Author(s):
|
Daniel F. McCaffrey* and J.R. Lockwood and Shelby Haberman and Lili Yao
|
Companies:
|
Educational Testing Service and Educational Testing Service and Educational Testing Service and Educational Testing Service
|
Keywords:
|
causal inference ;
post stratification weighting ;
exponential tilting ;
NPMLE
|
Abstract:
|
Weighting is a common approach to removing differences in observed covariates among groups in observational studies. Examples include estimation of causal effects and test score equating. Minimum discriminant information adjustment (MDIA) yields weights with the desirable properties of exactly matching weighted sample moments in a study population to the corresponding values for a target population, and being closest to constant among weights that yield such balance. However, existing results and methods for estimating MDIA weights do not apply to covariates measured with error. We derive results for weights of error-prone data that exactly match moments of the unobserved error-free variables in the target population in expectation and which are closest to constant among all weights that meet this criterion. Methods for estimating the weights that rely on the measurement error distribution being known or accurately estimated are developed and tested via a simulation study to demonstrate their feasibility.
|
Authors who are presenting talks have a * after their name.