ABSTRACT
Appendix C PEST Procedure A psychometric Junction is the probability o f choosing x over y, P r(x , y ) , as a function of x when the reference stimulus y is held fixed Forany designated probability level 7r, the idea of a PEST (parameter estimation by sequential testing) procedure is to locate the stimulus x(y, it) for which Pr[x(y, 7r),y] = 7T. This method first arose in psychophysics where the concern was to estimate the slope and mean of the psychometric function by establishing points close to 7r = 0.25 and 7r = 0.75 and interpolating linearly between them Initially a simple updown method was used, but to increase experimental efficiency ’E y lor and Creelman (1967) introduced the PEST procedure, and Pollack (1968) investigated some of its properties. In the context of estimating certainty equivalents, one wishes to establish which stimulus cor responds to the median response value, i.e., ir — 0.50. Bostic, Hermstein, and Luce (1990) suggested an adaptation that was based on the following reasoning. In the general neighbor hood of the 0.50 point, empirical psychometric functions appear to be approximately linear and so can be treated as if they are symmetric. Thus, if C E is the median of a lotteiy g and e is a pertuibation on it, then approximately ~Pr(CE+e, g) = l —P r ( C E —e,g). This suggested that, unlike the original PEST algorithms, we should use a perfectly symmetric algorithm to conveige on the CE.