Psy/Orf 322: PROBABILISTIC THINKING • We think about probabilities, because actions depend on what is likely to happen. • Theorists disagree about how we make these inferences. • Probability calculus: self-evident rules, e.g. extensional notion that prob(event) = sum of probabilities of different ways in which it can occur. • Do people reason probability calculus? according to “Someone with only the most modest knowledge of probability mathematics could have won himself the whole of Gaul in a week.” -- Ian Hacking (1975) SOME SIMPLE JUDGMENTS 1. In the U.S. which is most probable: death in automobile accident, by stroke, or by stomach cancer? 2. What’s the probability of a civil war in Iraq? 3. In a box, there is a red marble or a green marble, or both. What’s the probability that there is both the red and the green marble in the box? THE MEANING OF PROBABILITY • What do such assertions mean: probability (civil war in Iraq) = 0.6? In what circumstances would it be true? or false? • Philosophers argue seriously about interpretation of probabilities: subjective belief (assertion above is sensible) limit on a relative frequency (assertion above is meaningless) partial logical entailment (?) ‘Naive’ performance • How do you infer probability of: death in auto accident, by stroke, or by stomach cancer? Correct answer: stroke > stomach cancer > auto accident Method: use available evidence, e.g. frequency in media (use of heuristics studied by Tversky and Kahneman). • How do you infer probability of: red & green marble in box? Method: rules or models. MENTAL MODELS & PROBABILITIES Three assumptions: 1. Truth: people use what’s true, not false [last lecture]. 2. Equiprobability: if no information to the contrary, each model represents an equiprobable alternative. Cf. Laplace’s ‘indifference’ over events by which he proved that the odds that the sun will rise to-morrow are 1,826,214 to 1. 3. Proportionality: p(event) = proportion of models in which it occurs. A problem If one of the following assertions is true then so is the other: A green if and only if a blue. There is a green. Which is more likely to be in the box, green or blue? 90% say: equiprobable. It’s an illusion! Both assertions true: G Both false: not-G B B Blue more probable than green. • Moral: people use models, not (valid) formal rules from probability calculus. A problem • Phil has two children. One is a girl. What’s the probability that the other is a girl? Most people say: approx 1/2 Conditional probability: prob(A/B) i.e. probability of A, given that B is the case. Correct answer: 1st born girl girl boy boy 2nd born girl boy girl boy prob(other is girl/one girl) = 1/3 • Why do people go wrong? FIRST ERROR IN REASONING ABOUT CONDITIONAL PROBABILITIES • Failure to detect that question is about conditional probability, as opposed to simple probability. Hence, inappropriate models for problem: girl boy • prob(A) = 1/2, and prob(B) = 1/2. What is probability of A and B? Answer depends on p(A/B): prob(A & B) = p(A)p(B/A) or equivalently = p(B)p(A/B) Because p(A)p(B/A) = p(B)p(A/B), we have: p(B/A) = p(B)p(A/B) [Bayes’s theorem] p(A) A PROBLEM The suspect’s DNA matches the crime sample. The probability of a DNA match is 1 in a million if the suspect is not guilty. Is the suspect likely to be guilty? Why do people go wrong? p(DNA matches/not guilty) = 1 in a million They build models with frequencies: ¬ Guilty Frequencies 1 999,999 DNA matches . . . and flesh them out: ¬ Guilty Guilty DNA matches DNA matches Suppose the PARTITION is: ¬ Guilty DNA matches ¬ Guilty ¬ DNA matches Guilty DNA matches Guilty ¬ DNA matches p(DNA matches/not guilty) BUT: p(not guilty/DNA matches) = = Frequencies 1 999,999 Frequencies 1 999,999 9 0 1 in million 1 in 10 • SECOND ERROR: hard to hold all models in mind. BAYESIAN INFERENCE • One bag contains 70 red and 30 blue chips; another bag contains 30 red and 70 blue chips. One bag chosen at random. From it, 12 chips are selected at random with replacement. Result: 8 red and 4 blue chips. What’s prob that ‘70 red’ bag was chosen? People’s estimates average around: 0.8 Bayes’s theorem: 0.967 Moral: people don’t use Bayes’s theorem to infer posterior probabilities. • But, people can infer probabilities. How do they do it? posterior CONCLUSIONS • Naïve individuals have some ability to reason about probabilities. • It appears to be based, not on the probability calculus, but on mental models. • Frequencies make the arithmetic easier; but no evidence for an innate module for reasoning about frequencies.