Associate Professor of Biostatistics and Biomedical Informatics
Vice-Chair for Education, Biostatistics
Director of Graduate Education, Data Science Institute
Vanderbilt University
Despite decades of controversy, p-values remain a popular tool for assessing when the data are
incompatible with the null hypothesis. While it is widely recognized that p-values are imperfect,
the consequences of ignoring their flaws remain elusive and p-values continue to flourish in the
scientific literature. In this talk, I will introduce the second-generation p-value, a novel and
intuitive extension that better serves the intended purpose. Second-generation p-values are the
proportion of data-supported hypotheses that are contained in an interval null hypothesis that
consists of all null, practically null, and scientifically uninteresting effects. This emphasis on
scientific relevance obviates the need for multiple comparisons adjustments and reduces false
discovery rates. I will discuss the statistical properties of second-generation p-values and
illustrate their use in a high-dimensional analysis of genetic markers for Leukemia.