All Times EDT
Keywords: Statistical inference, p-value, second-generation p-value, large-scale inference, false discovery rate, FDR, positive false discovery rate
The second-generation p-value (SGPV) is a novel alternative to the p-value that accounts for scientific relevance by using a composite null hypothesis to represent null and scientifically trivial effects. A SGPV indicates when the results of a study are compatible with alternative hypotheses (SGPV = 0), null hypotheses (SPGV = 1), or are inconclusive (0 < SPGV < 1). It addresses many of the traditional p-value’s undesirable properties, and it is generally easier to interpret as a simple summary indicator. False discovery rates (FDRs) for traditional p-values are well established, and they provide an important assessment of how likely it is that the observed results are mistaken. In this talk, we derive FDRs for SGPVs focusing on the positive false discovery rate (pFDR). Ranking findings on the combination of SGPVs and their estimated pFDR emphasizes precisely estimated clinically meaningful effects, and eschews clinically uninteresting effects that are often captured as statistically significant by traditional methods. We will illustrate our methods with a large-scale example and demonstrate an R package for computing these quantities.