Abstract
Statistical models of optimal visuo-motor performance presuppose that visual processing of probability information conforms to standard probability theory (Maloney & Zhang, VisRes, 2010). We test whether observers combine the visually-estimated probabilities of two independent events normatively.
Task: During each trial, a subject chose between a single roulette wheel with probability r of success and a pair of independent roulette wheels with probabilities p and q of success. The observer’s estimates of p, q, and r were based on visual judgments of the fraction of each roulette wheel colored gold. The wheels were spun and the observer won a small monetary prize (i) if he chose the single wheel and it stopped in the gold, or (ii) chose the pair and both stopped in the gold. We used a staircase procedure to estimate the r for which the subject chose the pair as often as the single wheel: r~(p,q).
Conditions: Twelve experiment conditions corresponded to twelve choices of (p,q), with p being greater than or equal to q. They included four homogeneous groups of conditions of three pairs each, where p1q1 = p2q2 = p3q3. Ten naïve subjects participated.
Results: We tested whether observers’ indifference points satisfied r = pq, implying the observer was combining probabilities normatively. All observers had r > pq (Bonferroni-corrected t-tests): they overestimated the conjunction probability. We next tested whether observers assigned the same (erroneous) probability to each of the three pairs in a homogeneous group: did they converge on the same r for each of p1q1 = p2q2 = p3q3? 9/10 observers passed this test of homogeneity.
Conclusions: Observers’ combination of probabilities in a simple visual task showed systematic deviations from normative theory.
Meeting abstract presented at VSS 2012