Journal of Statistics: Advances in Theory and ApplicationsAuthor's: Oluwagbohunmi Awosoga, Joseph Mckean and Bradley Huitema
Pages: [131] - [161]
Received Date: June 30, 2016
Submitted by:
DOI: http://dx.doi.org/10.18642/jsata_7100121682
This study develops R estimators of the fixed effects in an experiment done over correlated baseline series. Besides a simple parametric approach we investigate linear mixed models procedures also. The random errors are independent within series but are dependent between several baseline series. The JR method seems more appropriate in term of its empirical type I error and the power of the test for the case of independence within but dependence between series than other methods (i.e., CT, WW, and LME) considered in this study. We illustrated the robustness of the procedures on a real data set which contained some outliers. Our robust procedures were much less sensitive to the effect of the outliers than the traditional analysis based on LS. A simulation study over situations similar to that of the data set confirmed the validity of our new approaches. The study also showed the robustness of efficiency of our approach over that of the traditional analysis.
joint rank, meta-analysis, robustness, multiple baseline, correlated tests, weighted Wilcoxon, contaminated normal distribution.
