Journal of Experimental Psychology: Human Perception and Performance 1983, Vol. 9, No.2, 299-309 Copyright 1983 by Ibe American Psychological Association, Inc. 0096-1 523/83/0902'()299$OO. 75 Information Integration and the Identification of Stimulus Noise and Criteria! Noise in Absolute Judgment - Robert M. Nosofsky Harvard University Two main classes of theories have been proposed regarding range effects in uni- dimensional absolute-identification tasks. One class posits that as range is in- creased, .criterial noise increases but stimulus noise remains constant. Another class posits increasing stimulus noise but constant criterial noise. In this study, an effort is made to help decide this issue. Multiple observations are used in several absolute-identification tasks of varying range. A stimulus integration model is proposed in which averaging takes place over stimulus internal repre- sentations, thereby reducing stimulus variance; on the other hand, it is assumed that criterial'variance is unaffected by the number of observations. The model allows one to identify the relative amounts of stimulus noise and criterial noise inherent in observers' recognition judgments. The model yields good fits to data in several experiments, and it is concluded that both stimulus noise and criterial noise increase as range in the absolute-identification task is increased. Researchers have theorized that the re- sponse variance inherent in unidimensional absolute-judgment tasks is the result of two underlying factors: stimulus noise and crite- rial or memory noise (Durlach & Braida, 1969; Gravetter & Lockhead, 1973; Wickel- gren, 1968). Each time a stimulus is pre- sented it is assumed to give rise to some in- ternal psychological representation. These in- ternal representations, like the physical stimuli, are conceived as lying along some unidimensional psychological continuum. Because of noise in the system, a given stim- ulus will not yield the same internal repre- sentation each time it is presented; rather, the internal representations that arise are as- sumed to be normally distributed. I This source of variance in the recognition process is termed stimulus or sensory noise. To make a decision regarding the identity of a presented stimulus, the subject must ob- viously use some sort of response rule. It is assumed that the subject partitions the range This work was supported in part by grants from the Nationat Science Foundation to Harvard University. I would like to thank R. Duncan Luce for the valuable suggestions, help, and time he gave me at all stages of this project. Requests for reprints should be sent to Robert M. Nosofsky, Department of Psychology and Social Rela- tions, Harvard University, 33 Kirkland Street, Cam- bridge, Massachusetts 02138. along which the psychological representa- tions lie into as many intervals as there are responses. The locations of these partitions, or category boundaries, are presumably based on subjects' memory for past presentations of the stimuli as well as various other judg- mental and bias factors. Due to imperfect memory and other noise factors inherent in the judgmental process, the locations of the category boundaries may exhibit variability over time. Like the stimulus distributions, they are assumed to be normally distributed, although there is no very strong reason for this assumption. This source of variance in the recognition process is referred to as cri- teria/ or memory noise. As G. A. Miller (1956) made well known, there are severe limits on the ability of sub- jects to identify absolutely unidimensional stimuli. In terms of the Thurstonian frame- work just presented, errors in absolute iden- tification are due to the overlap that occurs among the various stimulus and criterial dis- 1 The assumption of normal distributions is not ab- solute, of course, it is just the assumption generally made. Another assumption sometimes made is that it is the double exponential. The rationale behind the normality assumption, both theoretical and empirical, is reviewed extensively in Green and Swets (1973); the rationale be- hind the double exponential is treated by Yellott (1977) and Wandell and Luce (1978). 299