# Probability **The model:** A numerical measure of how likely an event is, ranging from 0 (impossible) to 1 (certain). Modern probability theory formalizes intuitions about likelihood, randomness, and uncertainty into a mathematical structure that supports rigorous reasoning. The framework is foundational to statistics, decision theory, finance, insurance, epidemiology, climate science, and modern AI. The mental-model importance is partly about the mathematics and partly about probability *intuition*. Humans are systematically bad at thinking probabilistically: we overestimate small probabilities of vivid events (plane crashes, terrorist attacks), underestimate small probabilities of dispersed events (climate-change consequences, accumulated risks), and confuse different forms of probability (frequentist vs. Bayesian, conditional vs. unconditional). ## Three Interpretations **Frequentist:** Probability is the long-run frequency of an event in repeated trials. The probability of heads on a fair coin is 0.5 because if you flip it many times, the proportion of heads approaches 0.5 per the Law of Large Numbers. Works well for repeated experiments; works poorly for one-time events ("probability of Trump being re-elected" has no well-defined frequentist interpretation because the election happens once). **Bayesian:** Probability is a degree of belief, formalized as a number between 0 and 1. New evidence updates the belief per [[Bayes Theorem]]. Works well for one-time events and for incorporating prior knowledge; criticized for the subjectivity of the prior. **Logical:** Probability is the rational degree of confidence given the evidence, considered as a property of the evidence rather than the believer. Less common in contemporary statistics; more common in philosophy of probability. The interpretations differ on philosophical grounds but produce similar mathematical machinery for most everyday applications. The political and rhetorical relevance is mostly about which interpretation underlies a specific claim — and whether the claim is using probability appropriately. ## KB Applications ### Probability Talk in Politics Election forecasts ("Trump has a 65% chance of winning") are usually Bayesian probabilities — degrees of belief based on polling, fundamentals, and historical patterns. They're sometimes misinterpreted as frequentist (as if elections would be run many times). The 2016 forecasts (28% Trump win probability in 538's model) were widely reported as "Clinton will win" — a confusion of low-probability with zero-probability. 28% events happen 28% of the time, by definition. ### Conditional Probability and Political Predictions Many political claims are conditional probability claims that don't mention the conditioning. "Probability of authoritarian transition under second Trump term" is conditional on many other things (court composition, military behavior, foreign-policy events, economic conditions). Reasoning honestly requires being explicit about what the probability is conditional on, and how those conditioning factors might themselves be uncertain. ### Why Tail Probabilities Are Hard Estimating small probabilities is much harder than estimating moderate ones. "1 in 100" is much easier to estimate than "1 in 10,000." The mathematical machinery (CLT, standard sampling) works less well at the tails. Most tail-event analysis is inherently model-dependent and contested. This is one reason existential-risk debates are so difficult — the probabilities at stake are in the tails where standard estimation fails. ## See Also - [[Bayes Theorem]] — how to update probabilities with evidence - Law of Large Numbers — what frequentist probability requires - Central Limit Theorem — the inferential bridge from samples to probabilities - [[Black Swan]] — why tail probabilities are systematically underestimated - [[Risk]] — probability + magnitude