Bayesian Analysis for Big Data, Lecture

I introduce Bayes factors as a way of indicating how strong the evidence is for the null hypothesis versus the alternative. Many problems have resulted from researchers, almost ubiquitously, using non-significance as a basis for saying there is no effect: e.g.“The p-value map shows nothing going on here..”, Alternatively, researchers, or reviewers, say because a key result was non-significant, the study or test of an effect was not informative (and hence can’t be published). None of these conclusions actually follow from the fact of non-significance itself. I show how Bayes factors can be used to show whether the data provide evidence for the alternative and against H0, for H0 and against H1, or are insensitive and do not discriminate H0 from the alternative. I will give practical guidance on how to do this for real data, what pitfalls to avoid, and on what basis to choose the different Bayes factor calculators. Examples will be used throughout which can be checked with free online software as we go along. At the end of the workshop participants will be able to determine for themselves for their own data whether a non-significant result was evidence for H0, or just insensitive; and also how much evidence a significant result actually provided for H1 over H0.