MCMC Bootstrap Based Approach to Power and Sample Size Evaluation
Oleksandr Mykolayovich Ocheredko
National Pirogov Memorial Medical University, 21008 Vinnytsya, Ukraine
https://www.vnmu.edu.ua/en/department/department/10#
Ocheredko@vnmu.edu.ua
Abstract. Power calculation is an important and evergreen applied statistical avenue. This study delivers suggestions on enrichment of the statistical tools by a combination of bootstrap and MCMC modeling. Novelty suggests application of possible data generation mechanism using MCMC and power estimation in the bootstrap procedure. We delineated further generalizations not incorporated in statistical software yet and demonstrated basic applications using SAS/STAT POWER Procedure examples (SAS Institute Inc., 2004). One concerns ANOVA, and the other deals with the survival process. The third example deals with the loglinear modeling of thromboembolism data (Congdon, 2005). An illustrious advantage of using MCMC is the possibility to exploit distributions of parameters of interest instead of ubiquitously used point estimates. The other methodological advancement though not demonstrated in the paper is the possibility to combine preliminary or historically observed data with experts' views. The foremost appealing advantage to application environment is the flexibility that is not confined to several basic situations rendered by statistical software. We have chosen the BUGS language to demonstrate the program code that can be run on WinBUGS, OpenBUGS and JAGS engines (Lunn, Jackson, Best, Spiegelhalter, & Thomas, 2012). We have used the R2WinBUGS package (Gelman., 2015) to run the script with n.chains=1, n.iter=5000, bugs.seed=1966 specification.
Keywords: Power Analysis • Data Generation Mechanism • MCMC • Bootstrap.
DOI: https://doi.org/10.35566/isdsa2019c5
To buy this chapter or the book, use the form below