6 ideas
People expect small samples to mirror the properties of the population as closely as large samples do, treating randomness as self-correcting at every scale. This leads to over-trusting results from tiny samples and underestimating how much more variable small samples genuinely are.
When facing a hard question (like a probability or future value), the mind quietly swaps in an easier question (like similarity or retrievability) and answers that instead. The judgment feels confident and complete because the substitution is unconscious, masking that a different question was actually answered.
Subjects watched a wheel of fortune land on 10 or 65. They then estimated the percentage of African countries in the UN, and the median estimates were 25 and 45 respectively. An obviously irrelevant starting number pulled the final judgment toward itself, because people adjust away from an anchor and stop too early.
People estimate how frequent or likely an event is by how easily instances come to mind. Retrievability is shaped by vividness, recency, and media coverage as well as by true frequency. Dramatic causes of death are therefore overestimated and quiet, common ones underestimated.
People judge the probability that something belongs to a category by how closely it resembles the category's stereotype, not by how common the category is. In the engineer-lawyer experiments, a personality sketch drove judgments almost identically whether engineers made up 30% or 70% of the pool. The prior probability is ignored as soon as individuating information of any kind, even worthless information, is available.
The shortcuts people use to judge probability are usually efficient, but they deviate from normative rules in consistent, directional ways. Because the errors are systematic, they can be predicted and anticipated, and experts are prone to them too when they reason intuitively. Deviations from rationality are therefore a map of the underlying cognitive process, not random mistakes.