Location
Mathematics/Psychology : 103
Date & Time
April 4, 2014, 11:00 am – 12:00 pm
Description
Speaker
Dr. Yi Huang
Department of Mathematics and Statistics
University of Maryland Baltimore County
Title
Latent Propensity Score Method for Average Causal Effect Estimation Allowing Covariate Measurement Error in large observational studies
Abstract
Many covariates in biomedical and policy studies are measured with error. The naive approach is to ignore the error and use the observed covariates in current propensity score framework for average causal effect (ACE) estimation. First, I developed a set of causal assumptions to extend the standard one allowing unobserved nondifferential covariate measurement error. Then, under this extended causal inference framework, we showed that the naive approach typically produces biased ACE inference. In this talk, we proved the consistency of ACE estimation using the proposed latent propensity scores under this extended causal framework, and developed a joint likelihood based approach using finite mixture model for ACE estimation, which will take the uncertainty from measurement error into consideration in propensity score subclassification. Its performance will be evaluated by large simulations studies. One real application on the post-marketing evaluation of a low-incidence medical device will be shown at the end - the impact of long-term breast pump usage on mother and infant’s health during 1st year infancy.
This is a joint work with Xiaoyu Dong, Andrew Raim, Karen Bandeen-Roche, and Cunlin Wang.
Abstract
Many covariates in biomedical and policy studies are measured with error. The naive approach is to ignore the error and use the observed covariates in current propensity score framework for average causal effect (ACE) estimation. First, I developed a set of causal assumptions to extend the standard one allowing unobserved nondifferential covariate measurement error. Then, under this extended causal inference framework, we showed that the naive approach typically produces biased ACE inference. In this talk, we proved the consistency of ACE estimation using the proposed latent propensity scores under this extended causal framework, and developed a joint likelihood based approach using finite mixture model for ACE estimation, which will take the uncertainty from measurement error into consideration in propensity score subclassification. Its performance will be evaluated by large simulations studies. One real application on the post-marketing evaluation of a low-incidence medical device will be shown at the end - the impact of long-term breast pump usage on mother and infant’s health during 1st year infancy.
This is a joint work with Xiaoyu Dong, Andrew Raim, Karen Bandeen-Roche, and Cunlin Wang.
Tags: