When judging probabilities or likelihoods, people rely on heuristics producing systematic biases. The representativeness heuristic causes overestimation of small-sample probabilities; the availability heuristic causes frequency estimates biased by memory accessibility; anchoring bias shows initial values disproportionately influence final judgments. These biases persist despite awareness and remain difficult to overcome.
From your study of reasoning biases, you know that human judgment systematically deviates from normative probability theory. Kahneman and Tversky's heuristics and biases program (1970s–2000s) catalogued these deviations — not as random errors but as predictable, replicable patterns that arise from specific cognitive shortcuts. Three heuristics account for the most practically significant biases: representativeness, availability, and anchoring.
The representativeness heuristic means we judge probability by how well something matches a prototype or stereotype, ignoring base rates. The classic case: "Linda is 31, single, outspoken, concerned with social justice. Which is more probable — Linda is a bank teller, or Linda is a bank teller and a feminist?" Most people choose the conjunction (teller AND feminist), even though elementary probability says P(A) ≥ P(A and B) always. The narrative fit of "feminist teller" feels more probable than "teller" because it matches the description better — representativeness overrides logic. The same heuristic causes base rate neglect: if a disease affects 1 in 1,000 people and a test is 99% accurate, most people say a positive test result means you almost certainly have the disease — forgetting that with such a low base rate, false positives vastly outnumber true positives. Representativeness also produces the gambler's fallacy: after five heads in a row, tails feels "due" because HHHHHT is more representative of a fair coin than HHHHHH, even though the coin has no memory.
The availability heuristic means we estimate the frequency of events by how easily examples come to mind. Deaths by shark attack are massively overestimated relative to deaths by falling vending machines — because shark attacks are vivid, media-covered, and memorable. Deaths by vending machine are neither dramatic nor covered. The heuristic is useful (frequent events are usually easier to recall) but fails when memorability is driven by factors other than frequency: novelty, emotional salience, recency, and personal relevance all inflate availability without reflecting actual rates. This creates predictable policy distortions — societies allocate vastly disproportionate resources to dramatic, visible risks while underinvesting in chronic, statistical ones.
Anchoring is perhaps the most surprising bias because it operates even when the anchor is obviously arbitrary. When asked to estimate the percentage of African nations in the UN, subjects first spin a wheel rigged to land on 10 or 65; those who saw 10 guessed ~25%, those who saw 65 guessed ~45%. The wheel should be irrelevant — but it isn't. The anchor establishes a starting point, and adjustment from anchor is typically insufficient, leaving final estimates clustered near the starting value. Anchoring affects salary negotiations (whoever names first captures the anchor), legal sentencing (prosecutors' numerical recommendations influence judges' sentences), and medical diagnosis (the first diagnosis mentioned biases subsequent evaluation). The disturbing implication is that these biases persist even when people are aware of them, even with financial incentives for accuracy, and even in experts in their domains. Awareness debiases marginally; structural changes (checklists, explicit base rate information, consider-the-opposite exercises) help more.