MSc Stats Project Presentation "Evaluation of Confidence Intervals for Stratified Standardized Event Ratios"

CANDIDATE:   ANASTASIA GAYOWSKY

ABSTRACT:

The effectiveness of different types of confidence intervals for stratum-specific standardized event ratios (SERs) are evaluated and compared in this project. The five methods for constructing confidence intervals included are Byar's method, the Rothman and Greenland method, the percentile and BCa bootstrap methods, and the asymptotic normal method. These confidence intervals are constructed for SERs calculated from both simulated time to event data and real data. Overall, the Byar, Rothman and Greenland, and asymptotic normal methods perform quite well, while the bootstrap confidence intervals have much lower coverage probabilities for the simulated SERs than expected. The bootstrap confidence intervals are much more similar to the other intervals constructed for the SERs calculated from the real data, where the proportion of censoring is higher than in the simulated data. The bootstrap methods may perform better overall in situations with higher rates of censoring.