An Assessment of Denominator Degrees of Freedom Approximations in the Analysis of Binary Endpoints in Veterinary Clinical Efficacy Trials
Qing Kang, The Statistical Intelligence Group, LLC  *Christopher I. Vahl, Kansas State University 

Keywords: veterinary medicine, binary endpoint, generalized linear mixed models

Many veterinary clinical efficacy trials utilize a binary endpoint as the primary outcome variable. When a generalized mixed model analysis is applied, the test statistic for evaluating treatment effect is regularly assumed to follow an F distribution with a certain choice for its denominator degrees of freedom. In contrast, practitioners usually employ simple closed-form calculations derived from the chi-square distribution to estimate sample sizes for their trials. To assess the impact of this practice, we conduct a simulation study under several practical designs with realistic parameter settings. The implications of this investigation on future study design and analysis will be discussed.