Biometrics & Biostatistics

Biography

Biography: Rowena F Bastero

Abstract

In observational studies, systematic differences in the covariates of the treatment and control groups may exist which pose a problem in estimating the average treatment effect. Although propensity score analysis provides a remedy to this issue, assessment made on the matched pairs or groups formed through these scores continue to reflect imbalance in the covariates between the two groups. Hence, a modified method is proposed that guarantees a more balanced group with respect to some, if not all, possible covariates and consequently provide more stable estimates. This matching procedure estimates the average treatment effect using techniques that infuse “swapping” of models based on classical regression and meta-analyses procedures. The “swapping” procedure allows for the imputation of the missing potential outcome Y(1) and Y(0) for units in the control and treatment groups, respectively while meta-analysis provides a means of combining the effect sizes calculated from each matched group. Simulated and real data sets are analyzed to evaluate comparability of estimates derived from this method and those formulated based on propensity score analysis. Results indicate superiority of the estimates calculated from the proposed model given its smaller standard errors and high power of the test. The proposed procedure ensures perfect balance within matched groups with respect to categorical variables and addresses issues of homogeneous effect sizes. It also identifies and incorporates relevant covariate information into the estimation procedure which consequently allows derivation of less biased estimates for the average treatment effect.