Academic journal article
By Duffy, Sean; Huttenlocher, Janellen; Hedges, Larry V.; Crawford, L. Elizabeth
Psychonomic Bulletin & Review , Vol. 17, No. 2
The category adjustment model (CAM) proposes that estimates of inexactly remembered stimuli are adjusted toward the central value of the category of which the stimuli are members. Adjusting estimates toward the average value of all category instances, properly weighted for memory uncertainty, maximizes the average accuracy of estimates. Thus far, the CAM has been tested only with symmetrical category distributions in which the central stimulus value is also the mean. We report two experiments using asymmetric (skewed) distributions in which there is more than one possible central value: one where the frequency distribution shifts over the course of time, and the other where the frequency distribution is skewed. In both cases, we find that people adjust estimates toward the category's running mean, which is consistent with the CAM but not with alternative explanations for the adjustment of stimuli toward a category's central value.
This article explores a well-known finding in the memory literature: that estimates of categorized stimuli are often remembered as being more typical members of their categories than they actually are. Known variously as the central tendency bias or schema effect, this phenomenon has been described by several psychologists as perceptual or memory distortions (Bartlett, 1932; Estes, 1997; Hollingworth, 1910; Poulton, 1979). Alternatively, Huttenlocher and colleagues (e.g., Crawford, Huttenlocher, & Engebretson, 2000; Huttenlocher, Hedges, & Vevea, 2000) have proposed a rational basis for these effects. They argued that this bias arises from an adaptive Bayesian process that improves accuracy in estimation. The category adjustment model (CAM) proposes that stimuli are encoded at two levels of detail: as members of a category, and as fine-grain values. In reconstructing stimuli, people combine information from both category and fine-grain levels of detail. This combination results in estimates adjusted toward the central region of their categories. This adjustment reduces the mean square error of estimates at any given stimulus value enough to more than compensate for the bias introduced into individual estimates (Huttenlocher et al., 2000, p. 240).
In their model, a category is a bounded range of stimulus values that vary along a stimulus dimension, such as size, weight, or intelligence. Memory for a stimulus is a fine-grain value along a dimension, such as a specific person's height. If the remembered value is inexact, the model proposes that the estimate (R) of the stimulus is a weighted combination of a category's central value ( ρ) and the inexact fine-grain memory (M) for a particular stimulus. The weight λ given to the fine-grain and category levels varies as a function of the dispersion of the category (σ^sup 2^^sub ρ^) and the degree of inexactness surrounding the fine-grain memory (σ^sup 2^^sub M^ ) and is derived from Bayes's theorem. To illustrate the Bayesian principle underlying the model, consider the example in Figure 1. The figure depicts a category with a normal frequency distribution of instances that varies along a continuous dimension. If a fine-grain memory for a stimulus falls at value M, and there is uncertainty surrounding the stimulus's true value, it is more likely that the stimulus's true value is in Direction A (toward the direction where the majority of instances fall) than in Direction B (where there are fewer instances). The combination of information about the prior distribution with the present distribution of inexactness surrounding the true value for M results in biased estimates that are more likely to fall in Region A than in Region B.
Huttenlocher et al. (2000) tested the CAM with an experimental task in which participants learned an inductive category by observing and reproducing a series of stimuli, such as lines that varied in length. On each trial, a target line briefly appeared; participants estimated its length, after a delay, by adjusting a response line to be the same length. …