Why does bayesian updating fail?

The most common reason bayesian updating fails: Two symmetric failures bracket the Bayesian ideal. Conservatism: you anchor to your prior belief and treat new evidence as noise, updating far less than the evidence warrants. This is the more common failure — Edwards (1968) found that people update at roughly half the rate that Bayes' theorem prescribes. The opposite failure is base rate neglect: you overweight vivid new evidence and forget everything you already knew, swinging wildly from one position to another. Both failures share the same root cause — an inability to hold prior knowledge and new evidence in proper proportion.

The fix: Pick a belief you currently hold with moderate confidence — a prediction about your career, a judgment about a colleague's competence, an assumption about how a project will unfold. Write it down with a probability: 'I am X% confident that Y.' Now identify the single most important piece of evidence that could change this belief. What would you expect to see if your belief were correct? What would you expect to see if it were wrong? Assign your best estimate of how much that evidence should move your number, and in which direction. Check back in one week. Did the evidence arrive? Did you actually update? By how much? Compare what you planned to do with what you actually did. The gap between planned and actual updating is your conservatism signature.

The underlying principle is straightforward: Update the strength of your beliefs proportionally to the strength of new evidence.