We tried to make a model of error factors so as to be able to analyze error responses on the n-dimensional discrimination learning (see Table 1). On the basis of the model, we worked out a way to calculate the rate of appearance of each error factor. We could not determine ene error factor for each error response, But we could determine possible error factors for each error response and also possible error factors for each correct response. In a block of trials the ratio of number of each error factor which possibly corresponds to error responses to that which possibly corresponds to all responses was used as measure of the rate of appearance. Then, we applied the model to an analysis of the process of color discrimination learning and color discrimination shift learning in normal and mentally retarded children ((1) Matsuda & Matsuda, 1967) and examined if the model was useful to clarify differences in response patterns between both groups of subjects and among five reinforcement combinations, RW, RNw, NrW, RN and NW. Actual error factors in these learnings are shown in Table 2, and a sample of analysis and calculation of rate of appearance of error factors is shown in Table 3. The results were as follows: 1) In (1) performance in the discrimination learning had been generally worse in mentally retarded children than in normal children. From analysis of error factors it was found that stimulus perseveration (error factor I), position alternation (error factor II), and win-stay-lose-shift and lose-stay-win-shift with regard to position (error factors III and III′) were stronger for mentally retarded children than normal children (see Table 4 and 5, and Fig. 2 and Fig. 3). 2) In 5 combinations of verbal reinforcement for mentally retarded children performance of NrW was found to be the best, and RN and RNw the worst ((1)). From analysis of error factors, it was found that error factor III was negligible only under NrW. Under RN and RNw, it seemed that subjects showed rather strong stimulus preference and were apt to respond in a stereotyped way (see Table 7). 3) In normal children, there was little difference in performance among 5 combinations of verbal reinforcement ((1)), but there was large difference in response patterns especially in regard to error factors II and II′. Error factor III was negligible under all 5 conditions. 4) It seemed that error factors which had been strong in discrimination learning were depressed in the discrimination shift learning and vice versa (see Tables 4 and 5, Fig. 4, Fig. 5, Fig. 6 and Fig. 7).
The legibility of a letter may be influenced by a lot of factors, in which both structural characteristics of a letter and psychological factors would be involved. Many relevant researches have been reported so far, but they could not afford evidence as to how those factors determine the legibility and also what quantitative combinations of those factors have a share in the legibility. The purpose of this paper is to clarify these points by the employment of a multiple regression method. The quantized 46 Kata-Kana letters contaminated by visual noise (Fig. 1), for which 4 noise levels, labelled as P6, P9, P135, and P203, were prepared, were used to give an index of the legibility of a letter (Table 3). The obtained legibility was intended to be explained by 9 factors, which were named as the internal redundancy (Factors 1 to 4), the component of direction (Factors 5 to 8), and the frequency (Factor 9). The multiple regression analysis done at 4 noise levels provided us with standard regression coefficients and regression correlation coefficients of 9 predictor variables (Table 8), which revealed relative importance of factors determining the legibility, and also multiple correlation coefficients R (Table 8), showing an over-all aspect of discrepancies between the actual proportion of correct identification and the one estimated by the equation (4). The following results were obtained. 1. The highest multiple correlation coefficient was shown at the P135 noise level, i.e., when noise cells are contained in a mesh with the probability 0.135. 2. The longitudinally-scanned redundancy (factor 2), the left-directionallyscanned redundancy (factor 4), the component of horizontal direction (factor 5), and the component of longitudinal direction (factor 6) were identified as positive factors participating in the legibility. 3. The right-directionally-scanned redundancy (factor 3) and the component of right direction (factor 8) were regarded as trivial factors. 4. The factor of frequency (factor 9), which was a sole psychological factor, among the analyzed variables, negatively affected the legibility through all of the noise levels.
The purpose of this experiment was to investigate whether the superiority of post-rest performance of the distributed practice group (DP) over the mass practice group (MP) was caused by SIR or dependent on the difference of the level of achieved skill. A work limit and a time limit method were employed. It was expected that if SIR existed, the difference at post-rest performance might appear in both practice methods, otherwise if the different levels of skill were true, it would take place in the time limit method alone. Ss of four groups received a task of printing throughout three periods that consisted of practice, pre-rest and post-rest periods. The number of trials were 5, 15 and 5 in abovementioned periods respectively for the two groups under the work limit method. On the other hand, for the two groups of the time limit method, 11, 23 and 6 trials were adopted respctively in order to make equal the number of printed letters between the two methods on the basis of the data of the work limit method. In both methods, the four groups practiced under the mass practice condition in each of the three periods, except that 2 DP were given a distributed practice with a 30 sec rest between each trial in the pre-rest period only. The direct comparison among the four groups was carried out by the transformed measure-the time required to write one letter. In all groups, the number of responses at each trial was accumulated from the first trial of the practice period (Table 1), and the corresponding trials between the 2 methods were obtained. The performance curves for all the groups are shown in Fig. 1. A significant difference between the two performances was not detected at the practice period. Therefore, it seemed that all groups were homogeneous. An analysis of variance was given among a few corresponding trials as regards the number of printed letters in the latter half of the pre-rest period (Table 2). From Table 2, a remarkable effect of distribution in the pre-rest period was noticed and it was found that the work limit method was somewhat advantageous. Two MP showed a significant reminiscence over a 10 min interpolated rest interval. DP of the time limit method, unexpectedly, gained significantly in performance at the first trial of the post-rest period. In the post-rest period, the two groups under the work limit method showed different levels of performance over the last period. From our results, it may be not appropriate to suppose that difference between the performances at the post-rest period between DP and MP is explained as the index of SIR. The relation of the level of skill to the number of responses was not clear. This result was considered to be due to motivational factors including the knowledge of result and from the adaptive responsiveness to the practice condition etc.