An Evaluation of an Incomplete Block Design for the Analysis of Treatment Efficacy for Clinical Mastitis

The purpose of this paper is to suggest a method of analyzing data from clinical mastitis trials when the response of interest is binary (success/failure) in nature, and the type of experimental design adheres to an incomplete block form. This approach is referred to as an "Empirical Bayes" strategy in statistical literature and incorporates over-dispersion commonly associated with binomial trials by developing a likelihood expression based on the beta-binomial distribution. As a result, this procedure recovers interblock information that logistic regression, conditional logistic regression, and chi-square methods are unable to utilize, thereby obtaining a more efficient estimate of the parameter of interest, the log-odds ratio between two competitive mastitis therapies. This particular application of the methodology focuses on evaluating the usefulness of different analytical techniques for comparing treatment efficacy for clinical mastitis in a typical Holstein dairy herd.