Using Propensity Scores for Nonresponse Adjustment with Covariate Measurement Error

Research question/goal: 

The proposed project will advance knowledge about the use of propensity scores for nonresponse adjustment when measurement error is present in the covariates used for adjustment. In particular, this project will (1) demonstrate, via simulations, the consequences of covariate measurement error for nonresponse adjustments as they are currently performed, (2) investigate the amount and structure of measurement error present in readily available auxiliary variables and paradata collected through interviewers, (3) examine the effect of known differential measurement error on nonresponse adjustment, and (4) develop new methods to perform propensity score nonresponse adjustments in the presence of covariate measurement error. Addressing the issue of measurement errors in nonresponse adjustment variables will affect population estimates of key statistics spanning a wide range of topics, such as welfare recipiency, reproductive behaviour, and health.  Our goal is to understand the amount and consequences of these errors and to propose practical steps for addressing them.  This work will also push propensity score methods more generally in important new directions, in particular by assessing the effects of measurement error on the performance of propensity score approaches, and by developing methods to handle differentially measured covariates.

Current stage: 

Our article in the Annual Review of Statistics and Its Application 2019 discusses the usability of samples with unknown selection probabilities for various research questions. In doing so, we review assumptions necessary for descriptive and causal inference and discuss research strategies developed to overcome sampling limitations. Based on this work, we are currently exploring practical steps and user guides to explore effect heterogeneity and inform researchers about possible inferential limits of their work.

Fact sheet

2015 to 2020