ON THE TECHNOLOGY OF INFERENCE AND DECISION Final Technical

A persistent problem in any kind of psychological research that reachesconclusions about inaccessible processes or experiences inside a subject's head istc validate those conclusions-that is, to exhibit persuasive reasons to believethat emitted behavior is some sense faithfully reports inaccessible processes.In the mid-1950s, perceptual researchers widely adopted an approach that miqhtbe called validation by cupidity. If the experimenter is willing to define acorrect response, he can reward the subject for correct responses and not forwrong ones; suitable reward schemes combine with an assumption of rationalbehavior on tne subject's part to permit direct inference of internal processes.However, decision-theoretical maxima are flat, in the sense that seriously inappropriatebehavior produces relatively little reduction in the subject's expected payoff. Thismeans that costs and payoffs are rather feeble means of instructing subjects whatto do, or of ensuring that he does it.This argument is made specific in examples drawn from three kinds ofperceptual experiments. In some tasks, such as probability estimation, subjectsdirectly estimate subjective quantities, and receive rewards for accuracy ofestimate. An analysis of proper scoring rules for probability estimation showsthat their maxima are inevitably quite flat. An analysis of a yes-no decisiontask shows that the incorrect answer produces flat maxima; while the payoff functioncan be sharpened by increasing the magnitudes of all payoffs, a suitable relativepayoff function is intractable. In such yes-no tasks, criterion variability produces| even more flatness, so much so that it would be surprising if such variation did not