The effect of number of training problems and total training trials on amount of transfer to a new problem

Abstract

The present experiment examined the effects of the number of training problems and the total training trials on amount of transfer to a new problem. Reviewed research showed the effects of varying these two variables to depend upon a number of other factors. Three such factors were incorporated into the present task: (1) the existence of a rule common to all problems, (2) variability within each problem, and (3) inconsistency of S-R relationships between problems. Four hypotheses were generated which were consistent both with the relevant literature and with the above three factors. The hypotheses, somewhat abridged, are the following: 1. The first hypothesis consisted of three contradictory parts, each part consistent with a different set of experimental findings. 1a. The fewer the training problems, the better should be the transfer to a new problem. 1b. The number of training problems should have no effect upon subsequent transfer. le. The greater the number of training problems, the better should be the transfer to a new problem. 2. The greater the total training trials beyond a certain point, the greater should be the transfer to a new problem. 3. Performance on a second problem should first decrease (show negative transfer) and then increase (show positive transfer) as a function of increasing training trials on the preceding problem. 4. Practice on a succession of interfering problems should eventually, perhaps after initial negative transfer, produce positive transfer to later problems. In the investigation of Hypotheses 1 and 2, a two way factorial design was used in which Ss received training on 1, 2, or 3 problems at 24 or 44 total trials before they were transferred to a new problem. Hypotheses 3 and 4 were evaluated by examining the performance of certain groups within the factorial design. Below are summarized the findings pertaining to the hypotheses: 1. Hypothesis 1b was supported. 2. Hypothesis 2 was supported. 3. Hypothesis 3 was partly supported. Performance on a second problem showed increasing positive transfer as a function of increasing training trials on the preceding problem. The initial negative transfer predicted in Hypothesis 3 was not found. 4. Hypothesis 4 was partly supported. Practice on a succession of interfering problems showed only positive transfer between problems. Contrary to Hypothesis 4, no negative transfer was found. Taken together, the results suggested the following interpretations: subjects show positive transfer between problems because they learn an abstract rule and also learn to ignore variability within each problem. Such learning is increased as a function of total training trials and not number of training problems. Subjects also show a lack of negative transfer in part because they learn the abstract rule and learn to ignore variability within problems, but also because they learn to ignore inconsistencies between problems.