Abstract
A central question in psychophysical research is how physical differences along a dimension of interest translate into perceptual differences. Knowing the true psychophysical scale would reveal how an external stimulus is converted to an internal event (Krueger, 1989). It would be particularly instructive to know whether and how the conversion changes under varying viewing conditions. Various methods have been advocated to derive perceptual scales, ranging from Fechner’s integration of just noticeable differences to Stevens’ direct scaling techniques. Yet, practically, the scale estimation is often bypassed by simply measuring appearance matches, which are an expression of the underlying scaling function but do not allow to estimate the scale directly. More recently two methods have been developed, MLDS (Maximum Likelihood Difference Scaling) and MLCM (Maximum Likelihood Conjoint Measurement), that promise to reliably estimate perceptual scales. Here we explore the requirements and limitations of these methods. We adopt three different observer models in the domain of lightness perception and, using simulations, we predict the response behavior for an MLDS, MLCM and matching experiment. We also tested observers with all three methods in two tasks favoring two different theoretical observer models. Both simulation and empirical data favor the scales estimated by MLCM. Scales estimated in MLCM do not suffer from arbitrary anchoring as those in MLDS, but both methods make strong assumptions about the noise, and we will discuss their justification.
Acknowledgement: Supported by the German Research Foundation (DFG MA5127/3-1 and /4-1 to Marianne Maertens)