抄録
Two experiments were conducted to determine which of two hypotheses, the original Thorndikian spread hypothesis or the serial response (guessing-sequence) hypothesis, is more adequate in explaining the “spread of effect” phenomenon.
In Experiment I, inter-response intervals were varied (2.5sec. and 5sec.) in an orthodox “spread of effect” experiment. When the interval was longer (5sec.), S-R connections would be more strongly reinforced than when it was shorter (2.5sec.) and thus a spread phenomenon should show up more clearly under the former conditions if the Thorndikian view is correct. The reverse would be predicted from the hypothesis of serial response habits because the habits would be more disturbed when S-R connections were stronger, i.e., when responses were more distributed.
The results confirmed the serial response hypothesis rather than the spread hypothesis (Table 1). This was especially true when the rewarded responses were repeated (Table 2, Fig. 1), and the findings were in accordance with our previous study in which one group of S's was required to give a number (1 to 10) at a rate of 3sec. and another group at a rate of 6sec. in a non-learning situation.
Analysis of intra-trial serial response repetitions was also made. The average percentage of such repetitions was 8 or 9 per cent (Table 4). This was contrary to Smith's probability bias hypothesis according to which the serial repetition was assumed 0. Besides, two interesting facts were observed: responses immediately preceding reward were mostly frequently repeated and rewarded responses were least frequently repeated on the following positions (Table 4).
In place of inter-response intervals, inter-trial time intervals were changed (20sec. 3min., and 10min). in Experiment II which was otherwise similar to Experiment I. As the responses were massed (2.5sec.) within trials, the differences in inter-trial intervals would not significantly affect S's learnig massed “wrong” responses but isolated “right” responses would be better recalled with longer inter-trial intervals.
The Thorndikian hypothesis would predect steeper spread gradients around reward with longer inter-trial intervals but the response-oriented hypothesis expects insignificant differenses among the three conditions. Again the latter hypothesis was supported in place of the former (Table 6 and Fig. 2). Marx and Goldbeck conducting a similar experiment to ours reported strong after-gradients for those who failed to repeat the rewarded response. Analysis of the present data also shows after-gradients following the nonrepeated rewarded response in both of the two experiments (Table 3) but the gradients were always lower compared with those following the repeated rewarded response and these findings were in agreement with our previous results under a non-learning situation.
The two findings concerning intra-trial serial response repetitions in Experiment I were also observed in Expeiment II but they were less clear. Tentative explanations were given but further research is necessary.
