Eugene Galanter Thomas E. Wiegand Columbia University
Since the early development of direct estimation scaling methods ( Stevens & Galanter, 1957), the constraints imposed on the behavior of experimental subjects have been limited to various forms of instruction designed to help subjects express their numerical estimates of perceived stimuli. Bush, Galanter , and Luce ( 1963) went so far as to suggest that because these methods lacked explicit payoff functions they deserved a special status--experiments without an identification function. Galanter ( 1989) has elsewhere suggested, in a technical agreement with Graham and Ratoosh ( 1962), that these experiments constitute a form of "numerical introspection."
Part of the problem has disappeared as experience has shown us that the most plausible functions for prothetic stimuli are power functions. But we also know that a representative subject may exhibit response biases that distort this function. Some of these biases were described by Stevens and Galanter ( 1957) and attributed by them to range effects and other stimulus presentation characteristics. Another class of biases are thought to be attempts by subjects at response consistency (i.e., faulty absolute stimulus identification leads to a reduction in response variability). These "good subject" attempts were part of the motivation for the application of the "shifty modulus method" ( Galanter, 1988). Similar phenomena were a central source of concern of Luce and Green ( 1978) and led them to reject the power function for individual subjects. Models of "round number" biases by Baird and Noma ( 1975) with some simulated confirmation by the Teghtsoonian ( 1988) may help explicate these response effects.