Abstract
We present an approach to smoothing balanced, single-error term analysis of variance (ANOVA), descended from Smith, that also allows spatial, temporal, or spatiotemporal smoothing. The approach addresses unreplicated designs, masked contrasts in effects with many degrees of freedom, and subgroup analysis, demonstrated using a study of denture-lining materials. Our approach is Bayesian but can be viewed as a way to generate frequentist procedures. A simulation experiment compares four priors, unsmoothed ANOVA, and dropping nonsignificant interactions. Three priors have advantages when some interactions are absent; dropping nonsignificant interactions has serious flaws. We contrast our approach with the approaches of Nobile-Green and Gelman.
Original language | English (US) |
---|---|
Pages (from-to) | 12-25 |
Number of pages | 14 |
Journal | Technometrics |
Volume | 49 |
Issue number | 1 |
DOIs | |
State | Published - Feb 2007 |
Bibliographical note
Funding Information:This work was supported in part by National Institute of Allergy and Infectious Diseases (NIAID) contract NO1-AI05073 and NIAID Graduate Training in Biostatistics grant 1-T32- A107432-01A1. The authors thank Tom Louis, the RAND Statistics group, and two referees for helpful comments and Dr. Igor Pesun of the School of Dentistry, University of Minnesota, for permission to use the polishability data.
Keywords
- Bayesian analysis
- Degrees of freedom
- Masking
- Prior distribution
- Shrinkage
- Sub-group analysis