Making reliable decisions in the study of wildlife diseases: using hypothesis tests, statistical power, and observed effects.

The increasing importance of wildlife diseases in conservation efforts places an additional importance on research study design, data analysis, and interpretation. In this paper, we explore the design and analysis of wildlife disease data with regard to hypothesis testing, statistical power, sample sizes, the relative costs of type I versus type II errors, and effect size. To illustrate these ideas, we conducted a literature review of the Journal of Wildlife Diseases (JWD), ran computer simulations that estimate type II error rates for statistical techniques commonly used in JWD, and reanalyzed previously published data on disease prevalence. Many studies published in JWD used chi-squared analysis on prevalence data, but only 19% reported estimates of the observed effect size. Furthermore, 10% of studies had pooled sample sizes < or =40, and many had potentially high costs of type II relative to type I errors. Our computer simulations suggest that many articles published in JWD lack sufficient statistical power, and this, coupled with our findings that many studies often ignore high costs of type II errors, argues for increased attention to statistical power. Finally, our data reanalysis shows how the presentation of observed effect sizes could allow a better assessment of the biologic significance of findings reported in JWD. We conclude with some general guidelines to assist wildlife disease researchers in the design of future studies and the statistical analysis of their data.