Online Program

Experimental Design in Pre-Clinical Animal Studies: Time for Bayes? (305352)

Keywords: experimental design, reproducible research, Bayesian analysis

Reproducibility of preclinical animal research has become a source of great concern in recent years. Null hypothesis statistical testing (NHST) and p-values receive a quite a bit of criticism as contributors to the reproducibility crisis. However, statisticians have not been able to move the biomedical research community away from NHST- in fact statisticians are far from unified on the need for alternatives to NHST.

Meanwhile, NIH guidelines for new grant applications call for rigorous experimental design. Biomedical investigators tend to view such guidelines as requiring power analysis to justify sample size decisions, but they rarely have solid information about effect sizes. In some cases, very small scale pilot data exists; in others, researchers rely on the literature. Some researchers resort to power-hacking, choosing effect sizes that coincide with experience-based sample size choices.

The availability of prior data, together with the uncertainty in effect sizes, suggests a role for Bayesian methods in sample sizing and experiment design. In this poster, we describe our efforts in working with biomedical researchers to develop some Bayesian sample sizing tools.