Category ratings and magnitude judgments are affected by four range biases, the centering bias, the stimulus and response equalizing biases, and the contraction bias; by three nonlinear biases, the local contraction bias, the stimulus spacing bias, and the logarithmic bias; and by bias from transfer. Models of the biases are described. The biases are most marked in sensory dimensions that students are not taught to handle, such as loudness and brightness. Avoiding all the biases requires exceedingly rigorous investigations. Why Sensory MagnitudesSensory magnitudes are selected for this review of biases in judgment because the stimuli can be measured on a physical scale. Judgments of the quality of life or of the likableness of people lack a precise measure of the stimulus. Thus the biases are more difficult to specify exactly. Anderson (1974), Kelson (1964), Parducci (1963, and the late S. S. Stevens (1975) all use sensory magnitudes when presenting their theories of judgment. Kinds of StimuliAveraging two weights and multiplying two lengths to calculate the area of a rectangle are a part of common knowledge. Students can handle these problems with relatively little bias (Anderson, 1974, Figures 1 and 8) because they are taught models of the way the sensory dimensions work.Common knowledge does not include averaging two sound intensities or two light in-The author is grateful to R. F. Fagot and R. Teghtsoonian for permission to make use of their unpublished information, to R. D. Patterson for commenting on a draft of the article, and to the British Medical Research Council for financial support.Requests for reprints should be sent to E.

6 pictorial models describe the effect upon magnitude estimation of the choices of the values of the independent variables: (a) the range of stimuli, (b) whether the range includes the threshold region, (c) the position of the standard (1st stimulus) within the range, (d) the distance of the 1st variable (2nd stimulus) from the standard, (e) whether the set of numbers used is infinite or finite, and (f) the size of the modulus (the number given to the standard), (a) alone accounts for about J of the variance in S. S. Stevens' table of exponents. Effects are classified under headings of response bias, level of adaptation, and a mathematical artifact. They are more compatible with a learned-calibration theory than with a simple transducer theory, but neurophysiological data are too varied to decide between the 2 types of theory. Transfer effects within and between experiments are described. The approximately logarithmic relationship usually found between partition or category scales and magnitude scales can be explained in terms of (b) and (e). The exact form also depends on experimental design and history of the observers, and these points need more attention both in executing and reporting experiments.Attempts have been made to explain the discrepant results which are found when different methods are used to estimate sensory magnitudes:

When each man receives a number of conditions in a balanced or random order, an unwanted range effect can sometimes reverse the rank order of the experimental results. With a range effect, responses are influenced by the range of stimuli, by the range of responses used by the man, or by both the range of stimuli and the range of responses. Range effects generally involve a central tendency but not always. There is no way of discovering whether a within-subject design has introduced an unwanted range effect, except by repeating parts of the experiment using a separate-groups design. Textbooks and courses on experimental design in psychology should emphasize the dangers of within-subject designs. Conflicts between experimental results can sometimes be resolved by discarding the results of within-subject designs. Revisions of theory may then be necessary.In a within-subject design, each man receives more than one experimental condition. The order of the conditions is balanced or randomized over the group of men. Edwards (1968, chap. 10) described Latin-square designs, with each man represented by a row or column. Lindquist (1953, chap. 6) called them Treatments X Subject designs. Winer (1971, chap. 4) called them repeated-measures designs on the same men.An experiment with a within-subject design should be regarded as a learning experiment. The men are trained, or partly trained, on all the conditions included in the experiment. The results apply only to men trained like this. The results do not apply to men trained on a different set of experimental conditions, nor to untrained men. In order to produce results which apply to men who have not been trained previously, it is necessary to use separate groups of men for each experimental condition.It is not possible to ensure unbiased results even by restricting each man to the two experimental conditions to be compared. This is because transfer between the two conditions may not be equal in the two directions. The

Within-subjects designs have obvious advantages, but they allow asymmetric transfer from influential companion conditions to bias the results. Examples are given both where asymmetric transfer occurs and where asymmetric transfer probably occurs but is not mentioned. The object of experimenting in the laboratory is to get away from variables that bias the results in unknown ways. Using a within-subjects design puts back the variables, all neatly balanced for subjects and order, but it still biases the results in unknown ways as a result of asymmetric transfer. The asymmetric transfer is attributed to a strategy that is learned in one condition and used in a subsequent condition where it is not . appropriate. Asymmetric transfer probably has its greatest influence when two or more conditions normally require different strategies, when the strategies are unobtrusive, and when the conditions are interleaved randomly within a block of trials.Experimental results have two kinds of explanation. Experimenters like to interpret their results in terms of a theory that is too general to consider all the experimental details. Yet the results may be specific to experimental details that are not incorporated in the theory. The experimental details may either be part of a particular condition or be other conditions included in a within-subjects design when one is used. If so, it may not be valid to generalize the results to performance with different details.An example of a result produced by an experimental detail within a particular condition comes from Broadbent's (1954) early work on noise. An effect of noise that is attributed to distraction can be accounted for simply by the masking of the acoustic cues from the response equipment (Poulton, 1980).The present review is concerned with the second kind of experimental details, the I am grateful to B. B. Murdock for kindly comparing between-and within-subjects designs, to U. Neisser, J. B. Barclay, and D. A. Allport for answering questions about their experiments, and to the British Medical Research Council for financial support.Requests for reprints should be sent to E.

During World War II, the late S. S. Stevens (1972) concluded that continuous intense noise does not degrade human performance, except by masking auditory cues. In the 1950s (1953, 1954, 1955, 1957, 1958), D. E. Broadbent claimed that continuous intense noise does affect people directly, by a mechanism other than masking. But recent experimental checks indicate that masking of the auditory feedback cues occurred in Broadbent's early experiments and in experiments reported subsequently by others. The auditory feedback tells the man that his response has been recorded. This is a help when there is a confusing directional relationship between control and display, when the control buttons are difficult to locate, and when considerable control pressure is required. Sometimes the auditory feedback helps to augment inadequate visual feedback. The remaining experiments in which continuous intense noise reliably degrades performance involve verbal working memory. Here, the noise can be said to interfere with or mask inner speech. Yet current explanations of the detrimental effects of continuous intense noise usually follow Broadbent and ignore masking in favor of nonspecific concepts like distraction, the funneling of attention, or overarousal. (70 ref)