|
Abstract:
|
The Bayes Factor has stronger theoretical justification than p-values for quantifying statistical evidence, but when the goal is hypothesis testing, the Bayes Factor yields no insight about false positive vs. false negative results. I introduce the False Positive Bayes Factor (FPBF) and the False Negative Bayes Factor (FNBF), show that they are approximately the two components of the Bayes Factor, and provide additional insight not obvious from the Bayes Factor. FPBF & FNBF require only the p-value and the power under an alternative hypothesis, forging a new link of p-values to Bayes Factors. This link can be exploited to understand different inferences drawn by Bayes Factors versus p-values. In a genome-wide association study of prostate cancer, FPBF & FNBF help reveal the two SNPs declared positive by p-values and Bayes Factors that with future data turned out to be false positives.
|
