Mathematics
The Bayes Factor as a Tool for Hypothesis Testing
Quick fact
A Bayes factor of 10 means the data are 10 times more likely under one hypothesis than the other, but it does not mean the first hypothesis is true with 10/11 probability—that depends on your prior.
Why this is interesting
You've probably heard of p-values, but what if there were a more direct way to compare two hypotheses—one that tells you exactly how much the data favor one over the other?
Read the full explanation
Understanding The Bayes Factor as a Tool for Hypothesis Testing
Imagine you're a detective trying to decide whether a suspect is guilty (H₁) or innocent (H₀). You have some prior belief, perhaps based on initial clues, that the suspect might be guilty. Then you find a piece of evidence: a fingerprint. The Bayes factor is like a scale that tells you how much this new evidence shifts your belief in guilt versus innocence. It's a ratio of two likelihoods: the probability of finding this fingerprint if the suspect is guilty, divided by the probability of finding it if they are innocent. If the fingerprint is much more likely under guilt, the scale tips heavily toward guilty. In statistics, the Bayes factor does exactly this: it compares how well each hypothesis predicts the observed data. It's a number that tells you the strength of evidence provided by the data alone, without yet considering your prior. To update your belief, you multiply your prior odds (guilt vs. innocence) by the Bayes factor to get posterior odds.
A deeper explanation
The Bayes factor emerges directly from Bayes' theorem. For two hypotheses H₁ and H₀, the posterior probability of H₁ given data D is proportional to the prior probability of H₁ times the likelihood of D under H₁. The same holds for H₀. If you divide the posterior odds (P(H₁|D)/P(H₀|D)) by the prior odds (P(H₁)/P(H₀)), you get the ratio of the likelihoods: P(D|H₁)/P(D|H₀). That ratio is the Bayes factor (BF). A BF 1 favors H₁, a BF < 1 favors H₀, and a BF around 1 means the data are equally likely under both—so the evidence is inconclusive. This is why the Bayes factor is a direct measure of evidence: it isolates the contribution of the data alone. Unlike p-values, which are the probability of observing data as extreme as yours under the null, the Bayes factor compares two competing hypotheses symmetrically. It doesn't require you to pick a 'null' that you hope to reject. This makes it a powerful tool for hypothesis testing, especially in fields like psychology and medicine where replicability and evidence strength are crucial. The Bayes factor also highlights the importance of specifying both hypotheses fully—including priors on parameters—which forces researchers to think more carefully about what their hypotheses actually predict.