Page 25 - Contributed Paper Session (CPS) - Volume 1
P. 25

CPS653 Chang-Yun L.
               1.0


               0.8


               0.6


               0.4

               0.2


               0.0


                                                    Effect

              Figure 1: Marginal posterior probabilities of the effects being active for the SPDS design.

            large  enough.  Different  from  the  existing  Bayesian  methods  for  multi-
            stratum  designs,  the  proposed  method  selects  more  reasonable  models
            that follow  the  effect  heredity principle.  To  make  the  proposed  method
            easy to use, we conduct the MCMC and Gibbs sampling with the WinBUGS
            software and provide a general code that can be applied to analyze data
            for any SPDS and BDS designs. Applying the SSVS method, we reanalyze
            two datasets from the 17 x 8 SPDS design in Lin and Yang (2015) and the
            15  x  4  BDS  design  in  Jones  and  Nachtsheim  (2016).  To  compare  the
            performance of the SSVS method and the regression methods MSR and
            FSR,  we  conduct  simulation  studies  for  SPDS  designs  with  strong  effect
            heredity models and BDS designs with weak effect heredity models. Results
            show that the proposed SSVS method well controls the false discovery rate
            and  has  higher  power  than  the  two  regression  methods  on  identifying
            active effects. Although in this paper the SSVS method and the WinBUGS
            code are developed under the framework of the SPDS and BDS designs,
            they can be easily extended and applied to analyze data for any other multi-
            stratum deigns.

















                                                                14 | I S I   W S C   2 0 1 9
   20   21   22   23   24   25   26   27   28   29   30