Recently, wehave been able to do the factor analysis of the correlation matrix with great many variables quite easily with the aid of the electronic computer. However, the uniqueness of the meanings of factors extracted by the different reseachers using the same variables, which has been for long argued, stillr emains unsolved. This problem is attributable to the fac tthat the informations of the subjects such as sex or developmental stage are neglected. In view of this point, this study aims to apPly the L. J. Cronbach'sgeneralizability theory to the multivariate variables, so that the correlation matrix can be divided into the sum of serveral component matrices, which is cgnsidered to depend largely on the variance analysis technique.
Through this technique, the correlation matrix R is diveded as the fullowing formula shows.
(1) R=IRA+RAe (in case of one classificatior category)
(2) R=RA+RB+RAB+RABe (in case of two classification categories)
(3) R=RA+RB+Rc+RAB+RBc+RcA+RABC+RABCe (in case of three categories)
(A, B, C denotes the classification categories such as sex. AB, BC, CA denotes the interaction effect. RAe, RABe, RABce denotes the residual ma trix closely related to the individual differences.)
In ordinary analysis, the correlation matrix R is directly analyzed, but in this. analysis RA, RB,... RABce are separately analyzed by the principal axis method.
Factors extracted from RA, RB,..., RABC are closely related to the meaning of the classification category and the factors extracted froM the residual matrik RABCe are free of the meaning of the cla-ssification category. This technique was applied to the data of the General Vocational Aptitude Test (GATB) consisting of II scales, subjects of which are 950 Japanese students who are classified with respect to sex, developmental (high school and middle school) and regional (rural and urban) differences.
Comparing the factors extracted from nine component matrices shown in formula (3) with the factors extracted from the correlation matrix R, theprincipal factor loadings of the residual matrix RAce decreased and no small factor loadings of the developmental difference matrix were found. This fact means that the factors related to the developmental dif ference are excluded from the principal factor loadings of the correlation matrix R.
Applying this techniqu widely, it is expected that the unique factor solutions independent of the difference of the subjects and the experimental conditions will be acquired in the near future.

View full abstract

[in Japanese]

View full abstract

This experiment was designed to examine the effects of verbal reinforcement combinations on three-alternative learning in normal and mentally retarded children.
The Ss saw in front of them a series of cards, each consisting of three figures arranged horizonally. The shapes of figures were circle, triangle and square at random, each of which was painted randomly red, green or yellow. The Ss were asked to choose a figure in each card which they thought correct and to push the button in front of this figure. One of three colors was designed as correctvalue randomly for each S. We set 9 trials as one block and defined the criterion of learning as the state of all correct responses in a block. The Ss were reinforced until reaching the criterion in one of five combinations of verbal reinforcement, that is, RW, RN and NW, and RNw and NrW in which S was instructed about meaning of N before the beginning of learning. After the acquisition, 10 blocks were continued without any reinforcement.
The results were as follows:
1) There was the tendency that both mentally retarded children and normal children learned faster under RW condition than under RNw and NrW, and faster under RN and NW than under NrW (see Table 3 and Table 4).
2) Generally, for mentally retarded children the mean rate of error responses decreased largely toward the end of learning, and for normal children it decreased linearly. And under RW, RN and RNw they decreased toward the end of learning, and under NW and NrW they decreased suddenly at the end of learning (see Fig. 1, Table 5 and Table 6).
3. In the extinction, under RW, NrW and NW in both subject groups the mean rates of error responses were almost zeros. Under RNw and RN, they are larger in normal children than in mentally retarded children (see Fig. 2, Table 7, Table 8 and Table 9).
These findings in the extinction support all suggestions about properties of N in our previous two-alternative learning, that is, 1) N is a positive reinforcer at the start of experiment, 2) the positive reinforcement values increased under NW, while it is weakened to turn into a negative reinforcer under RN, 3) the acquisition of negative reinforcement value of N is more difficult for mentally retarded children than for normal children. But the findings in acquisition are never accounted for by these properties of N.
Now we must note that as at the learning with three and above alternations of responses W cannot point to what is the correct response, we must consider probabilities of an wrong response and an right response on each trial and quantities of informations by R and W. In the learning of this experiment, quantity of information by R is much larger than that by W, though probability of an wrong response is larger than that of an right one at least at the early stage of learning. Thus it is not unreasonable to expect that performances under RW is the best and those under RN and RNw are rather better than under NW and NrW. But at present we have no direct evidence to account for why N has no effect on performance in three and above-alternative learning like on that in two-alter native learning.

View full abstract

[in Japanese]

View full abstract

The purpose of the present experiment was to test the hypothesis that ease of reversal shift learning in kindergarten children is positively related to the S' s abstraction ability. This hypothesis was deduced from the verbal mediating response theory (Kendler & Kendler, 1962).
160 kindergarten Ss were given an abstract word test consisting of 20 items of two objects in each. The Ss were required to guess an abstract word common to the two objects, e. g., What

View full abstract

The purpose of the present experiments was to examine the relations between personality traits and sensory-motor learning abilities in adults, under the two conditions of neutral (Exp. 1) and stressful instructions (Exp. 2).
Twenty one college boys and girls served as Ss in Exp. 1, and 18 students in Exp. 2, and the apparatus used was a stylus maze with a buzzer which would sound each time Ss made an error. Measurements taken were a) number of repetitions req-u ired to reach the criterion of learning (R), b) total number of errors made in reaching the criterion (E), and c) total time spent (T).
Under the neutral instruction, Ss were required to learn three tasks with free speed, and after 24 hours they were again required to learn them. A month later, they were instructed to learn the same mazes with the instruction that would make them stressful. In Exp. 2, they were told to learn as correctly and quickly as possible, and were informed time spent every five seconds.
The summary of the present experiments was as follows.
1. In Exp. 1, Ss who obtained a high score in Rhathymia tended to take a small Learning Index (LI=_??_), while in Exp. 2, Ss who were lacking in objectivity and had inferiority feelingshad a tendency to take a small LI. It might be stated that Ss with such characteristics would be apt to be much motivated by the psychological stress.

View full abstract

We applied our model of error factors to analyze of process of three-alternative discrimination learning in normal and mentally retarded children and to clarify differences in response patterns among five reinforcement combinations, RW, RNw, NrW, RN and NW Fig. 1 and Matsuda & Matsuda, 1969 Actual error factors in the learni ng are shown in Table 1, and a sample of analysis and calculation of rates of appearance of error factors is shown in Table 2.
The results were as follows:
1. There were few subjects who showed error factor I 1 or I 3. The number of subjects who showed error factor I 2 were a little more than the expectation, but the rate of its appearance went down to the level of the rate of error responses within 2 or 3 blocks (see Table 4, Fig. 2 and Fig. 3). Therefore it seems that error factor I has little effect on performance of this learning.
2. In all groups, error factor II 1 was stronger than II'1 and error factor II 3 was stronger than II'3. The rates of appearance of II 1 or II 3decreased relatively smoothly in all groups, but in III 1 or 3 the rates of appearance under NW and NrW did not decrease smoothly and were significantly higher than those under the others (see Fig. 4, Table 5, and Table 9). Error factor II and II', which do not relate to right or wrong of responses, may be strong when subjects can't sufficiently discern right or wrong of their responses or cannot fully use information which oright or wrong of their responses have. Therefore these findings suggest that the information by W which is thought to be smaller than by R is scarcely used because. using this small information may be too difficult for the subjects. In this respect, we can understand that the fact that the acquirement of negative reinforcement value of N is difficult especially in mentally retarded children (see results of extinction in Matsuda & Matsuda, 1969) has little effect on the performance of this learning.
3. Especially under RW, NrW and NW in mentally retarded children and under RW, RNw and NrW in normal children, in which subjects might be able to discern right and wrong of their responses (see results of extinction in Matsuda & Matsuda, 1969) error factor III 3 was stronger than III'3 (see Fig. 5, Table 6, and Table 10). Therefore in respect of these error factors which never appear if subjects do not know right and wrong of their responses, it is probable that being able to easily acquire the reinforcement value of N affects rather negatively the performance of this learning
4. It seems that error factor II 2, II'2, III2 and III'3 which relate to shape of the stimulus figures were not so important as the others (see Fig. 4, Fig. 5, Table 5, Table 6, Table 9, and Table 10).
5. Stereotypical response patterns were observed more often in mentally retarded children than in normal children (see Table 7 and Table 8). Presumably this is one of the reasons why learning curves with rates of error responses in mentally retarded children were positively acclelated see Matsuda & Matsuda, 1969.

View full abstract

[in Japanese]

View full abstract