413 – Consulting, Collaboration, Communication, and Impact
A New Look at Evidence Testing and Measures of Error
Robert Riffenburgh
Naval Medical Center
Recently, considerable attention has been paid to the misuse of statistical testing outcomes, particularly p-values. I see three stages to rectify this problem: conceptualization (a clear, unequivocal exposition of proper procedures), personalization (making it accessible to the general non-statistician statistics-using public in a readily understandable form), and implementation (public adoption of the second stage). The conceptualization (1st) stage was admirably addressed in 2016 by a policy statement from the American Statistical Association clarifying the process in traditional statistical terms. My paper attempts to start debate on the personalization (2nd) stage. Five components are proposed and discussed: Use terms to make the statistical discovery process meaningful to non-statistician statistics-users, concentrate on the effect of an experiment rather than a test, view a test result as a measure of belief in the effect, provide joint difference/equivalence tests, and provide a scaled ranking of beliefs relating p-values to practical real-life interpretations. The components are shown in a practical example. A template to help users follow proper procedure is given.