Masters candidate in Biostatistics, Zheng Wang, will present:
“Nonparametric Methods for the Propensity Scores”
Plan B Adviser: Chap Le
Abstract: Effect size is a standardized measure of the effect, and it is widely used in comparing the intervention effects of different groups, such as the treatment group versus the control group in clinical trials. A common parametric method of estimating the effect size is taking the difference of group means and divided by a common or average standard deviation. The main problem of this parametric method is that it is susceptible to extreme values and outlying observations, which may be common in clinical trial studies. In this study, we proposed a nonparametric estimator of effect size based on the nonparametric estimate of the area under the ROC curves and applied this nonparametric estimator to form a nonparametric method for use with propensity scores in the analysis of observational studies. The simulation showed that the performance of this the nonparametric estimator of effect size was comparable to that of a parametric estimator. Despite the 3.3% loss in relative efficiency, the nonparametric estimator is less influenced by extreme values and outlying observations. Real data from an observational study on throat cancer was used as a numerical example to illustrate and compare the performance of the nonparametric method and the traditional parametric method in propensity score stratification. The proposed nonparametric method for propensity scores provides an alternative way to estimate the overall effect size and the confidence interval in propensity score stratification and it is more robust when the distribution is skewed or there is extreme observation.