Thomas Lumley • 9/17/2020

A Bayesian t-test?

The article critically examines the Bayesian counterpart to the classic two-sample t-test. It argues that modern t-tests rely on Central Limit Theorem approximations, not strict Normal/location-shift assumptions. The author challenges the robustness of common Bayesian approaches that use convenience priors and proposes an alternative method using discrete (multinomial) distributions with Dirichlet priors to compare means without restrictive parametric assumptions.

0 comments

#data analysis #Bayesian Statistics #Hypothesis Testing