chapter eleven

11 Bayesian hypothesis testing: Asking the right questions when hypothesizing

 

This chapter covers

  • The Bayesian approach to hypothesis testing
  • The Bayes factor as an alternative to 𝑝-values
  • Moving beyond hypothesis testing

In classical (frequentist) statistics, hypothesis testing is built around the famous 𝑝-value, a statistical measure widely used to determine whether the results of an experiment are meaningful or just occurring by chance. We start with a null hypothesis, compute how surprising the data would be if that hypothesis were true, and reject the null if the result crosses a predetermined threshold.

This framework is widely used, but it often leaves important questions unanswered. For example, a 𝑝-value tells us how unusual the data would be under the null hypothesis (for example, what’s the chance that you see this many clicks on a new website that’s no better than the current website), but it doesn’t tell us how likely that hypothesis, or any other, is. A 𝑝-value below the predetermined threshold also doesn’t directly inform our decision making; that is, we can’t directly answer questions about which action to take given our knowledge about a problem, such as whether the new website is worth deploying.

The frequentist approach and 𝑝-values

Is the new design better?

The traditional hypothesis testing procedure

Questions that remain unanswered

Hypotheses as competing models and the Bayes factor

Advantages of Bayesian hypothesis testing

Directly measuring what we are interested in

Accounting for priors

Accumulation of evidence

Working with posterior models

Summary