Two learning models for responses measured on a continuous scale

Two linear operator models are presented for a class of learning situations in which the response is on a numerical scale and the subject is given the magnitude of his error on some or all of the trials. Theoretical expressions are developed for sequential dependencies, mean learning curves, variances, and covariances, which permit a number of tests of goodness of fit.This work was supported by Grant G-12986 from the National Science Foundation. The author wishes to express his appreciation to J. H. Alexander for his cogent assistance.