Probability matching is contraindicated by the data of individual subjects (Dusoir, 1974) and by group data (Creelman & Donaldson, 1968); in general, response frequencies are not as extreme (i.e., distant from one-half) as the corresponding stimulus frequencies. This deviation from probability matching can be accounted for by at least two versions of an error-correcting model in which the subject’s decision criterion may shift only after an error, but not by a model which permits criterion shifts after correct responses. One version of the error-correcting model allows us to specify the dependence of response frequency on payoff magnitude and on signal strength. The deviation from matching is consistent also with a model that assumes that, on each trial, the criterion shifts in the direction of the sensory information, regardless of response accuracy. This model is thus applicable to situations in which trial-by-trial feedback is omitted.
[1]
A. E. Dusoir,et al.
Thomas and Legge’s matching hypothesis for
,
1974
.
[2]
Ewart A. C. Thomas.
On a class of additive learning models: Error-correcting and probability matching
,
1973
.
[3]
Donald D. Dorfman,et al.
A learning model for a continuum of sensory states
,
1971
.
[4]
C D Creelman,et al.
ROC curves for discrimination of linear extent.
,
1968,
Journal of experimental psychology.
[5]
T. Parks.
Signal-detectability theory of recognition-memory performance.
,
1966,
Psychological review.
[6]
Mark Kac,et al.
A note on learning signal detection
,
1962,
IRE Trans. Inf. Theory.
[7]
Carl F Saslow,et al.
Learning models for a continuum of sensory states reexamined
,
1975
.
[8]
Ewart A. C. Thomas,et al.
Probability matching as a basis for detection and recognition decisions.
,
1970
.