How does Bayesian updating work in everyday reasoning?

Bayesian updating is a way of changing your mind in proportion to the evidence. You begin with a prior, which is what you believed before the new information, and you move it by how strongly the evidence points one way or the other. The result is called the posterior. As one guide puts it, you always move from the prior and never from scratch, and strong evidence moves you a lot while weak evidence moves you a little. [1] [2]

The prior should reflect the base rate, meaning how common the thing is in general. Useful priors often come from past experience, comparable cases, historical patterns, or known incentives. If most similar projects take three months, that should shape your expectation of the next one. [1] [3]

The likelihood asks how probable the evidence would be if the claim were true. The posterior is the updated judgement. One everyday guide summarizes the relationship as posterior proportional to prior times likelihood, and suggests keeping the order of the steps even if you skip the formula. [4]

Spencer Greenberg of ClearerThinking.org describes a single question for judging evidence: how likely would I be to see this evidence if my hypothesis is true, compared with if it is false? The ratio of those two likelihoods is called the Bayes factor. A ratio of three to one is moderate evidence, thirty to one is strong, and one to one is no evidence at all. A ratio below one pushes in the opposite direction. [5]

Greenberg notes that people often drop the comparison. They reason that evidence is likely if their hypothesis is true, and so conclude it is strong, without asking how likely it would be if the hypothesis were false. He also says you usually will not get a hard number, but you can often form a rough gut sense of whether the evidence is much more likely one way than the other. The Bayes factor then combines with your prior to give a new probability. [5]

Suppose a colleague has not replied for half a day. One explanation is that he is busy, and another is that he is deliberately ignoring you. Being busy is far more common, so it starts much higher. If you then learn he replied instantly to someone else, that clue fits the second explanation better, so it rises and the first falls. But if the second explanation started very low, it may still be low afterward, which is why skipping the prior is an error. [4]

The same source describes the typical mistake as asking only how much the evidence looks like the dramatic explanation, and never how rare that explanation is in the first place. [4]

The guides list several recurring errors. One is confusing resemblance with probability: an explanation that sounds right speaks to likelihood, not to the posterior. Another is concluding from a single clue, since weak evidence should leave the posterior close to the prior. A third is ignoring reliability, because a rumor passed through five people counts for much less than a credible, disinterested eyewitness or a result reproduced three times. [4] [6]

Comparing several alternative explanations also helps. Low signups after a product launch, for example, fit a weak product, an unclear page, the wrong audience, or low trust, so more evidence is needed to tell them apart. Updating is meant to be gradual and ongoing, with beliefs staying revisable. [3]