A practical exploration of using Bayesian methods (MCMC, multinomial likelihood with Dirichlet prior) to analyze small datasets of 1-5 ratings, contrasting the posterior distribution against a naive normal approximation. Using PyMC3 and Seaborn, the author demonstrates why normal approximations break down for small, integer-constrained rating data, drawing on the historical context of Gossett's t-test and small-sample statistics.
Sort: