Graduate Student Seminar

Location

Biological Sciences : 120

Date & Time

April 2, 2014, 11:00 am12:00 pm

Description

First SpeakerRowena Bastero
Second Speaker Elande Baro
Session Chair Mingyu Xi
DiscussantDr. D. Park
PlaceBS 120

Rowena Bastero
Title
Introduction to Propensity Scoring Analysis
Abstract
In several studies, the main goal is to be able to establish the causal effect of a certain treatment relative to a control condition. In an ideal setup, the subjects are randomly assigned to a treatment or control group whereby each group's background characteristics are made as similar as possible. This ensures that significant differences in the outcome variable between groups reflect the treatment effects and not the effects of differences in covariates. However, this nature is not guaranteed in some studies such as observational studies where imbalanced covariate distributions are most likely to occur. As a consequence, direct estimation of outcome differences may not be feasible as they could easily be attributed to differences in the group's backgrounds rather than the effect of the treatment itself. To reduce biases caused by such an imbalance, propensity score analysis is proposed. The structure of the talk begins with a discussion of the basic framework of propensity scores followed by the basic matching techniques currently used in such an analysis. Stratification, one of the known matching schemes, is then illustrated using an NSAIDs data after which the current trends in this area will be briefly discussed.

Elande Baro
Title
Bayesian Latent Propensity Score Approach for Average Causal Effect Estimation Allowing Covariate Measurement Error
Abstract
In observational studies, it is often the case that covariates are measured with error. The naive approach is to ignore the error and use naive propensity score methods with observed covariates to estimate the average causal effect (ACE). It has been shown that the na=EFve approach might bias the ACE inference. Dr. Yi Huang developed a set of causal assumptions allowing covariate measurement error and extended the standard propensity scoring theory (without measurement error). She proved the consistency of ACE estimation using the proposed latent propensity scores and proposed a joint likelihood approach in finite mixture model format for ACE estimation with continuous outcomes. In Huang and etc paper, EM algorithm is used, where the numerical performance is not ideal due to the large dimensions of unknown parameters. We extend this work and use Bayesian estimation method under the latent propensity score model in finite mixture model format. The method captures the uncertainty in propensity score subclassification arising from the unobserved measurement error. Simulations studies are presented to show the performance of this newly developed Bayesian approach compared to the existing EM algorithm and naive approach ignoring the error. It shows that Bayesian method provides a more stable inference with good standard error estimates. This is a joint work with Dr Yi Huang and Dr Anindya Roy.