Base Rates: Why Can’t One Striking Clue Determine the Whole Probability?

A base rate is how common a category was in the relevant population before new evidence appeared. It is not a verdict on the present case, but it is background probability that should not vanish when judgement begins.

Suppose only five of a company’s hundred employees work in cybersecurity. A new colleague is cautious, keeps to themself and often talks about passwords. That description may resemble a security engineer, but resemblance is only case-specific evidence; people in other roles may share those traits. If we ignore the original five-to-ninety-five distribution, the vivid description takes over the whole judgement. Yet refusing to move beyond the base rate is also wrong: genuinely discriminating evidence should update the starting probability.

A base rate is therefore not an average. An average describes the centre of a quantity, while a base rate gives the proportion belonging to a category. Nor is it simply a baseline, which is usually a reference state chosen for comparison. Bayesian reasoning combines a prior probability with the discriminating force of new evidence to produce an updated probability. This does not let old statistics override reality; it requires both kinds of information to count.

In hiring, medical testing, risk assessment and AI classification, an individual case is often more vivid than the population behind it. Sound judgement does not choose between “trusting the data” and “trusting what is in front of us”. It asks first: how common was this outcome, and how much should the new evidence change that probability?

https://www.itl.nist.gov/div898/handbook/apr/section2/apr1a.htm