A learning model for signal detection theory-temporal invariance of learning parameters

Biderman, Michael D.; Dorfman, Donald D.; Simpson, John C.

doi:10.3758/bf03336678

Cited by 5 publications

(4 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, as shown in Table 8, there was an interaction between the correctness of the preceding trial and a decreasing linear trend over sessions, F(l, 11) = 3.42, p = .046 (onetailed), indicating that the size of cutoff shifts after errors decreased over sessions, but not after correct responses. In contrast, Biderman et al (1975) found that the size of shifts did not diminish over trials. Otherwise the pattern of results shown by magnitudes of cutoff shifts does not correspond to the pattern observed in Table 6 for percentages of cutoff shifts.…”

Section: P(xi N < Cn)mentioning

confidence: 78%

“…Additive-operator dynamic-cutoff models. A number of additive-operator dynamiccutoff models have been proposed and studied by Biderman, Dorfman, and Simpson (1975), Dorfman (1973), Dorfman and Biderman (1971), Dorfman, Saslow, and Simpson (1975), Kac (1962Kac ( , 1969, Larkin (1971), Norman (1970Norman ( , 1972, and Thomas (1973and Thomas ( , 1975. The following formulation generalizes all the existing additive-operator dynamic-cutoff models (specifically Dorfman & Biderman, 1971;Thomas, 1973):…”

Section: Dynamic-cutoff Modelsmentioning

confidence: 99%

See 1 more Smart Citation

The decision rule in probabilistic categorization: What it is and how it is learned.

Kubovy¹,

Healy²

1977

Journal of Experimental Psychology: General

139

View full text Add to dashboard Cite

A numerical decision task, meant to be a numerical analogue of signal detection, was employed to investigate the nature of the decision rule used for probabilistic categorization. Two conditions were compared, with 12 paid volunteers participating in six 1-hr sessions in Condition 1 and 12 participating in three 1-hr sessions in Condition 2. In both conditions, one of two distributions of five-digit numbers was sampled on each trial (400 trials per session), and the research participant was to decide which distribution was sampled. Specifically, the participant was told that each stimulus represented the height of a person from an artificial population, and the participant was to decide whether that height belonged to a man or a woman. The stimulus distributions were normal and equal in variance; their means were one standard deviation apart. In Condition 1 the participants were free to adopt whatever decision procedure they preferred, including a probabilistic strategy. In contrast, in Condition 2 the participants were forced to adopt a cutoff rule: They reported a five-digit cutoff number before each trial. On each trial, participants were forced to produce a response consistent with the reported cutoff number-that is, the response had to be man if the stimulus number was larger than the reported cutoff and woman if it was smaller.For both conditions, the numbers of deviations from the best fitting staticcutoff rule were computed for each block of SO trials. Although the deviations were found to be significantly more numerous in Condition 1 than in Condition 2, in Condition 1 the numbers of deviations were much closer to those in Condition 2 than to the number of deviations predicted from the best available probabilistic model (Schoeffler, 1965). Furthermore, Conditions 1 and 2 were very similar with respect to other characteristics, such as the effects of practice and the locations of the best fitting static-cutoff point. Sequential analyses of the reported cutoff numbers in Condition 2 revealed that participants did not always shift their cutoffs in the direction predicted by additive-operator models and that, contrary to the predictions of these models, the probabilities and amounts of cutoff shift were not stationary over sessions. In contrast to the additiveoperator models, a new dynamic-cutoff model describing the behavior of an ideal learner was able to account for some of these features of the data, but additional analyses ruled out this model as well.These results led to the following conclusions: (a) In a numerical decision task a dynamic-cutoff rule is a good first approximation to the decision rule adopted by individuals, (b) The only dynamic-cutoff models previously available in the literature-the additive-operator models-are inadequate, (c) A new dynamic-cutoff model-the ideal-learner model-is also inadequate, (d) Individuals tend to shift their cutoffs in the normatively prescribed direction after errors, but tend to shift their cutoffs randomly after correct responses. 427

show abstract

Section: P(xi N < Cn)mentioning

confidence: 78%

Section: Dynamic-cutoff Modelsmentioning

confidence: 99%

The decision rule in probabilistic categorization: What it is and how it is learned.

Kubovy¹,

Healy²

1977

Journal of Experimental Psychology: General

139

View full text Add to dashboard Cite

show abstract

“…A partial reconciliation of these views came in the form of deterministic dynamic-criterion models (Biderman, Dorfman, & Simpson, 1975;Dorfman, 1973;Dorfman & Biderman, 1971;Kac, 1962Kac, , 1969, in which the criterion varied systematically from trial to trial on the basis of the stimulus, response, and outcome. These models outperformed models with static criteria (Dorfman & Biderman, 1971;Larkin, 1971) but did not account for a relatively large amount of apparently nonsystematic variability (Dorfman, Saslow, & Simpson, 1975).…”

Section: Deterministic Versus Probabilistic Response Criteriamentioning

confidence: 99%

Signal detection with criterion noise: Applications to recognition memory.

Benjamin¹,

Diaz²,

Wee³

2009

Psychological Review

138

226

View full text Add to dashboard Cite

A tacit but fundamental assumption of the Theory of Signal Detection (TSD) is that criterion placement is a noise-free process. This paper challenges that assumption on theoretical and empirical grounds and presents the Noisy Decision Theory of Signal Detection (ND-TSD). Generalized equations for the isosensitivity function and for measures of discrimination that incorporate criterion variability are derived, and the model's relationship with extant models of decision-making in discrimination tasks is examined. An experiment that evaluates recognition memory for ensembles of word stimuli reveals that criterion noise is not trivial in magnitude and contributes substantially to variance in the slope of the isosensitivity function. We discuss how ND-TSD can help explain a number of current and historical puzzles in recognition memory, including the inconsistent relationship between manipulations of learning and the slope of the isosensitivity function, the lack of invariance of the slope with manipulations of bias or payoffs, the effects of aging on the decision-making process in recognition, and the nature of responding in Remember/Know decision tasks. ND-TSD poses novel and theoretically meaningful constraints on theories of recognition and decision-making more generally, and provides a mechanism for rapprochement between theories of decision-making that employ deterministic response rules and those that postulate probabilistic response rules.

show abstract

“…It might be argued that our subjects have not reached asymptote in Experiment 1. However, Biderman, Dorfman, and Simpson (1975) have shown that the estimates of parameters from signal detection experiments during the first 200 trials are almost identical to those from the last 200. The main difference is in the variance of the cut point (larger in the early part).…”

Section: Discussionmentioning

confidence: 93%

Estimates of utility function parameters from signal detection experiments.

Kornbrot¹,

Donnelly²,

Galanter³

1981

Journal of Experimental Psychology: Human Perception and Perfor

View full text Add to dashboard Cite

A theory is developed to make it possible to estimate the slope parameter of a presumed power law utility function from an analysis of data from signal detection experiments. The theory can be applied to detection data analyzed either according to constant likelihood ratio models or according to linear learning models of bias. Experiments to test the consequences of the theory are described. A linear learning model is found to give the best account of the data. The utility function obtained from this model is a power function with exponent equal to .48.Currently, there is a dualism between the behaviorist tradition in human experimental psychology, concerned exclusively with performance, and other more "cognitive" approaches that include the observers' introspective reports. The scope of theory should therefore be widened to produce models that account for performance and verbal report within a common framework.It is with such a goal in mind that we propose to develop a single coherent theory of the perception of "utility" that will describe not only the results of scaling experiments of both utility and number judgments but also the way people actually use number in such diverse tasks as adding and comparing numbers, gambling, decision making, and signal detection. Some of the earliest systematic attempts to investigate the perception of number and utility are in the domain of scaling. Attneave (1962) has suggested that magnitude esti-This research was supported in part by the Engineering Psychology Programs, Office of Naval Research. This publication replaces Psychophysics Laboratory Report PLR-32. The authors are indebted to the referees for their thoughtful and helpful comments.

show abstract

A learning model for signal detection theory-temporal invariance of learning parameters

Cited by 5 publications

References 6 publications

The decision rule in probabilistic categorization: What it is and how it is learned.

The decision rule in probabilistic categorization: What it is and how it is learned.

Signal detection with criterion noise: Applications to recognition memory.

Estimates of utility function parameters from signal detection experiments.

Contact Info

Product

Resources

About