Estimating the Subject by Treatment Interaction in Nonreplicated Crossover Diet Studies

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Matthew Kramer*+ and Shirley C. Chen and Sarah Gebauer and David Baer
• Companies: Agricultural Research Service and Agricultural Research Service and Agricultural Research Service and Agricultural Research Service
• Keywords: nutrition ; principal components ; treatment-subject interaction ; LDL-cholesterol

Abstract

Human nutrition studies typically use crossover designs, where subjects follow the treatment diets sequentially without repeating them. If individual subjects respond differently to the diets, the subject-diet interaction can be large, but cannot be estimated since it is confounded with the residual. A proposed method for estimating this term is to use a principal components decomposition of the residuals from a two-way ANOVA with no interaction. We designed a crossover study, but with replication, to test this for nutrition studies. The study examined the effects of including pistachio nuts at 3 levels on blood compounds known to affect health (e.g. LDL-cholesterol). The first two principal components were significant and the variance they explained larger than the diet main effect, and similar to that estimated for the subject-diet interaction using a traditional mixed model.