Bayesian Updating

Bayesian updating revises probabilities by combining priors with likelihoods for observed evidence and normalizing across hypotheses. Its conclusions are conditional on the model and information assumptions.

Bayesian updating combines prior probabilities with the likelihood of evidence under each hypothesis, then normalizes the result. The evidence matters through how well it distinguishes alternatives, not simply how vivid or recent it is. Sequential updating also needs a model of dependence so the same information is not counted repeatedly.

Updating is conditional on the hypothesis space and likelihood model. A hypothesis given exactly zero prior probability cannot gain probability through ordinary conditioning, and a misspecified model can become confidently wrong. Useful practice therefore includes checking alternatives, information quality, and predictive performance rather than assuming honest arithmetic guarantees eventual truth.

When to use it

When new information arrives that could change your assessment. When you need to integrate multiple uncertain data sources. When arguing about beliefs—check whether each side is updating on evidence.

How it can help

Compare evidence across alternatives, account for dependent sources, and check predictions. Reconsider the hypothesis space and likelihood model when observations do not fit.