No abstract

An adaptive procedure for rapid and efficient psychophysical testing is described. PEST (Parameter Estimation by Sequential Testing) was designed with maximally efficient trial-by-trial sequential decisions at each stimulus level, in a sequence which tends to converge on a selected target level. An appendix introduces an approach to measuring test efficiency as applied to psychophysical testing problems.

A series of experiments measured human ability to discriminate between durations of auditory signals presented in a noise background. Independent variables were the signal voltage, the "base" duration T, and the increment duration/xT. Separate experiments assessed the effect of each of these on discrimination.A decision-theoretical model is presented, based on a "counting mechanism," which operates on impulses generated over the relevant durations. The source of these impulses is assumed to be random. Limitations on performance come from uncertainty regarding the end points of the time interval and from limited memory. The decision processes underlying the model are presented as a general theory of duration discrimination. HESE experiments measured the ability of human observers to discriminate differences in duration between short auditory signals. An understanding of time judgment is important for theories of signal detection, since detection depends on attention during the time of the signal, and only during that time. A practical problem related to temporal discrimination arises in speech perception, where cues to linguistic meaning in some languages, and cues to stress and inflection in English, seem to depend on relative duration (cf. Peterson and Lehiste•). A quantitative theory of temporal discrimination is developed to account for discrimination in the range ordinarily covered by speech sounds and the signals used in psychoacoustic experiments.The study of time discrimination is a poor stepchild of the growth of psychophysical research. Titchener 2 characterized the area as a "microcosm, perfect to the last detail," exemplifying in miniature the course of development to that time of psychophysical methods, concepts, and empirical knowledge. Study of temporal discrimination, and the estimation of time, has remained a microcosm, somewhat isolated from the main stream of empirical research on sensory capacities.A reason for this neglect was pointed out by Nichols a in an early, scholarly, historical review of the philosophy and psychology of time. This was the problematic status of the "time sense" as an independent psychic faculty, aside from the content of sensory input. The question has not yet been resolved, and continues to haunt the researches and theoretical efforts of psychologists.Reviews of the experimental literature after Nichols' were presented by Dunlap, . et al., 7 and most recently by Wallace and Rabin. 8 Chapters on time perception were offered by Titchener, 2 by Boring, 9 and by Woodrow. •ø With all this activity, we still find that there is "... as yet no generally accepted view as to how we perceive or estimate time, "7 and fourteen years later we find ourselves still on the trail of the "hitherto elusive 'time sense'. ,,s Although an experiment by Stott n was primarily concerned with the analysis of "time errors," enough data were reported to draw some conclusions about the temporal sensitivity of his observers. The observations were made in a group setting, and the data were averaged o...

Models of discrimination based on statistical decision theory distinguish sensitivity (the ability of an observer to reflect a stimulus-response correspondence denned by the experimenter) from response bias (the tendency to favor 1 response over others). Measures of response bias have received less attention than those of sensitivity. Bias measures are classified here according to 1 characteristics. First, the distributions assumed or implied to underlie the observer's decision may be normal, logistic, or rectangular. Second, the bias index may measure criterion location, criterion location relative to sensitivity, or likelihood ratio. Both parametric and "nonparametric" indexes are classified in this manner. The various bias statistics are compared on pragmatic and theoretical grounds, and it is concluded that criterion location measures have many advantages in empirical work. We are grateful to Howard Kaplan for many useful discussions of these issues. We also thank Doris Aaronson, reviewers A. E. Dusoir and J. G. Snodgrass, and an anonymous reviewer for helpful comments on previous drafts.

For most perceptual continua, observers' ability to discriminate exceeds their ability to identify. Certain dimensions, however, particularly in speech perception, are said to be "categorically" perceived, in the sense that they can be discriminated only as well as they can be labeled. This article offers a signal detection theory analysis of categorical perception; in previous models, lowthreshold assumptions have been made. Discrimination paradigms popularly used to test the categorical perception hypothesis, such as the ABX and samedifferent designs, are analyzed, and unbiased sensitivity measures (d r ) abstracted. A Thurstonian model is used to predict discrimination from identification under the hypothesis that perception is categorical. For cases in which perception is found not to be categorical, we show how the hypothesis of dual processing of phonemic and nonphonemic information can be distinguished from alternative models.

No abstract

Can accuracy and response bias in two-stimulus, two-response recognition or detection experiments be measured nonparametrically? Pollack and Norman (1964) answered this question affirmatively for sensitivity, Hodos (1970) for bias: Both proposed measures based on triangular areas in receiver-operating characteristic space. Their papers, and especially a paper by Grier (1971)that provided computing formulas for the measures, continue to be heavily cited in a wide range of content areas. In our sample of articles, most authors described triangle-based measures as making fewer assumptions than measures associated with detection theory. However, we show that statistics based on products or ratios of right triangle areas, including a recently proposed bias index and a not-yetproposed but apparently plausible sensitivity index, are consistent with a decision process based on logistic distributions. Even the Pollack and Norman measure, which is based on non-right triangles, is approximately logistic for low values of sensitivity. Simple geometric models for sensitivity and bias are not nonparametric, even if their implications are not acknowledged in the defining publications.In many experiments in cognitive science, observers try to assign distinct labels to stimuli chosen from different classes. In the simplest case, there are two stimulus classes (signal and noise in a detection experiment, old and new items in a recognition memory study) and two corresponding responses. The experimenter wishes to abstract from the results a measure ofaccuracy, or sensitivity, that reflects the subject's ability to distinguish the stimulus classes, as well as a measure of response bias, that is, the tendency to choose one response over the other.Signal detection theory (SDT; Green & Swets, 1966;Macmillan & Creelman, 1991) is a framework for generating and evaluating both kinds ofindices. The best known SDT measures, d' (for sensitivity) and f3 (for bias), are consistent with a decision model in which stimulus classes lead to equal-variance normal distributions of a decision variable. This model can be tested by examining receiveroperating characteristic (ROC) curves-functions that relate the proportion of hits (yes responses to signal presentations) to the proportion offalse alarms (yes responses to noise presentations) as response bias is either manipulated

No abstract