A full logarithmic plot of magnitude estimation of loudness of tones against the physical intensity of these tones gives rise to a curve that is linear over much of its extent (S. S. Stevens, 1956). The slope of the linear portion of this curve is characteristic of the frequency of the tone studied and varies from above 0.4 for low and high frequencies to approximately 0.3 for tones at 1000 Hz (e.g., Marks, 1974;West, Ward, & Khosla, 2000). We shall show here that the slope of this straight line (the power function exponent) can be obtained from the properties of a stimulus-response matrix constructed from an experiment on absolute identification of sound intensity.Our stimulus-response matrix is a slight generalization of the one inaugurated by Garner and Hake (1951). It is designed to encode the results of an experiment on stimulus categorization, which includes absolute identification, and is described in detail by Norwich, Wong, and Sagi (1998). In brief summary, our experiments dealt with judgments of tone intensity and were conducted in the following manner: Tones of 1.5-sec duration at 1000 Hz were delivered to subjects binaurally at intervals of 20 sec. The sound intensity was distributed uniformly from 1 dB HL to an upper bound of R dB HL. The subjects were trained to identify tones to the nearest decibel. Feedback was not required (Mori & Ward, 1995). We confirmed the results of Mori and Ward repeatedly: Feedback to the subjects regarding the accuracy of their estimates did not improve overall performance. No more than 160 identifications were made by any subject in a single day. Five hundred identifications were made by each subject for a given range, R. Each of 5 subjects was tested over several ranges (1-10 dB, 1-30 dB, 1-50 dB, 1-70 dB, and 1-90 dB).Experiments over a given range could be analyzed by dividing the range into any reasonable number of categories of equal width. For example, an experiment carried out over the range 1-30 dB could be divided into 6 categories each of 5 dB in width, 10 categories 3 dB in width, 30 categories 1 dB in width, and so on. If there were m stimulus categories and m response categories, a stimulus-response matrix could be assembled in the usual way, with stimulus categories represented as rows and response categories as columns, as shown in Figure 1. Then the entry in the jth row and k th column represents the number of times the subject identified a stimulus in the jth category as belonging to the k th category. Each of the m stimulus categories has equal a priori probability.Analysis of these experiments revealed that the distribution of responses to a tone of m dB was essentially normal with mean m and variance s 2 . The mean, m, is known as the row mean, and the variance, s 2 , as the row variance. Row mean was measured in matrix units. Thus, if the range, R, were divided into m categories, the mean responses for row 1, 2, . . . m would be approximately equal to 1, 2, . . . m, respectively. Similarly, row variances are expressed in square matrix units. The ro...