less of frequency, in the first and the last records. Examination of the table reveals 14 problems for which a greater number of children were reported in the last record than in the first one, but it also reveals 27 problems for which fewer children were reported the second time. In three cases the percentages on the two records are identical. For the 14 cases in which the second percentage is larger than the first, the difference between the two percentages varies from 1.3 to 9.8: the mean difference is 5.4. For the 27 cases in which the second percentage is smaller than the first, the difference between the two ranges from 1.2 to 19.7; the mean difference is 7.9. Five of these latter differences 3 are between two and three times the standard error of the difference, while none of the former reaches even this extent. The whole picture of the two records is one which indicates not only a decrease in the frequency with which a given type of problem behavior is observed in a given pupil by the teacher but also a tendency toward decrease in the number of children in which it is observed at all. Something has happened in the lives of these boys and girls which has made the group as a whole less of a problem group in the eyes of their teachers, although it has by no means eliminated altogether the manifestations of undesirable conduct.

2. What changes took place in the overt problem behavior of the NPC group from the first to the last behavior record?—The nonproblem group will now be subjected to the same technique of comparing the first and the last records and of considering the results in relation to those which have just been cited for the problem group. Eighty-five of the former were enrolled in the Berkeley schools in December, 1930. Table 8 shows the distributions of the scores and other statistical data for the initial and the final reports on these children. Two points will at once be noted: First, the increase in the range and consequent variability of the two distributions; second, the significant increase in the means.

Instead of lowering their scores, these children have raised them; instead of eliminating the highest scores of the original records, they have added still higher ones. The difference between the means of the two distributions is 22 in a positive direction. This is more than three times its standard error (7.1), and therefore stands the test of statistical reliability. These children as a group have actually developed significantly more problem behavior during the two years of the study, while the experimental problem group have shown a definite improvement.

The percentages showing these differences are italicized and underscored in Table 7.

As with the problem group, so also the change which took place in the problem scores of the nonproblem group was not a sudden one occurring in any one term, but developed from one semester to the next, as is shown by the following means, taken in successive terms: 82, 83.4, 97.6, 104.

TABLE 8.-First and last behavior problem scores of the nonproblem control group

(85 cases)

Read the table as follows: On the first record no child had a behavior problem score between 275 and 299, or between 250 and 274; 2 children had scores between 225 and 249, etc. Read similarly for scores on the

last record.

For meanings of statistical symbols, see Table 5 and text.

NOTE.-Intervals of 25 were used in this table instead of intervals of 50, as in Table 6, in order to show the differentiation more clearly. The statistical results of using either type of distribution are very similar.

The more detailed analysis, of the respective behavior records corroborates this finding. It was found for the problem group that the difficulties reported as "occurring frequently" became less numerous, while those reported as "occurring once or twice" became more numerous. Consequently one phase of improvement lay in the frequency with which a given type of conduct was observed. With the nonproblem group all degrees of frequency show some increase, as follows:

In analyzing these records still further by computing the percentage of children for whom each problem was reported without regard to frequency, the disparity between the two series may be seen even more clearly. Table 9 gives these data, and one can not help but be impressed by the fact that, whereas for the problem group 27 types of difficulties showed a decrease in the percentage reported, in

5 A comparison of the figures in this column with those of the total number of 109 nonproblem children (in Table 4) will show that the group remaining at the close of the study is in this respect representative

this group 34 types of undesirable conduct showed an increase, and in 6 of these (those italicised and underscored in the table) the difference between the percentages is from two to four times its standard error. The boys and girls who at the beginning of the study were adjudged by their teachers as evincing no serious overt behavior problems seem to have developed during the two years of study significant tendencies in that direction.

TABLE 9.-Behavior problems reported for nonproblem control group on first and last records (85 cases)

Read the table as follows: Inattention was reported on the first behavior problem record for 68.2 per cent of the nonproblem children; on the last record it was reported for 74.1 per cent. Read similarly for other behavior problems. Behavior problems are listed in descending order of occurrence on the first record.

Restlessness, talking, fidgeting, asking to leave room too frequently.

TABLE 10.-Behavior problem scores of 68 pairs of problem and nonproblem children (initial record)

Read the table as follows: Of 68 problem children, 1 child had an initial behavior problem score between 550 and 599; 1 had a score between 500 and 549; 3 had scores between 450 and 499, etc. Read similarly for 68 nonproblem children. For meanings of statistical symbols, see Table 5 and text.

3. What does the comparison between the EP and NPC groups show if they are limited to those equated pairs which continued unbroken to the end of the study?—The discussions under questions 1 and 2 have dealt with the entire group of problem and nonproblem children, respectively, for whom records were available at the beginning and again at the end of the study. One further step will now be taken in the comparative analysis by the elimination of every child in either group whose mate in the other group was lost to the study. This process left 68 of the originally equated pairs of problem and nonproblem children who remained for comparison with each other. What that comparison reveals as to behavior scores is given in Tables 10 and 11.o

• In order to insure the representative character of this group of equated pairs, the initial records of each series were compared with those of the complete group of 109 children and were found to be statistically similar, as the following data will signify:

TABLE 11.-Behavior problem scores of 68 pairs of problem and nonproblem children

(final record)

Read the table as follows: Of 68 problem children, 4 had final behavior problem scores between 400 and 2 had scores between 350 and 399, etc. Read similarly for 68 nonproblem children. For meanings of statetical symbols, see Table 5 and text.

The following points are worthy of emphasis:

(a) When the original groups have been reduced for exact comparative purposes from 109 to 68 pairs of problem and nonproblem children, the difference between the means of their initial behavior scores remains statistically of real significance, being equal to eleven times its standard error. (See Table 10.)

(b) In the final records of these 68 pairs of children, the difference between the means of behavior scores has been cut in half, and is equal only to five times its standard error. (See Table 11.) This difference is still large enough to be within the limits of statistical reliability.

(c) The marked reduction of the difference between the scores of these two groups of children has been brought about through a significant improvement in the scores of the problem children and a significant deterioration in the scores of the nonproblem children. These changes in opposite directions have brought the groups closer together in their final status.

Additional evidence is provided by the comparison of problem frequency in these same two groups at the beginning and at the end of the study, which may be briefly summarized as follows: