338 



MEMOIRS NATIONAL ACADEMY OF SCIENCES. 



[Vol. XV, 



Section 9. — -Correlations with total score. 



Correlations of each test with, total score had been made with inferior groups, unselected 

 groups, amd superior groups. These are summarized on page 32Sff. From the data there 

 given it will be seen that the scale as a whole is in general best represented by tests 6, 9, 5, and 

 4, while test 2 shows distinctly the lowest correlation with total score. Tests 2 and 10 correlate 

 relatively better with total score in the lower than in the higher ranges, while tests 3, 8, and 9 

 give higher correlations in the upper ranges. 



The amount of such correlation is not due to a single cause and is subject to a variety of 

 interpretations. Generally speaking, it may be said that the higher a test correlates with 

 the total score of a battery of good tests, the better that test is when taken alone as a single 

 measure. On the other hand, the objection to making up an intelligence scale out of tests which 

 are all extremely highly intercorrelated is obvious, for in such a case the multiplication of tests 

 beyond a very limited number becomes sheer waste. At the same time, low correlation of a 

 given test with total score or with the other tests is not itself a recommendation for such test; 

 it is the reverse unless it can be shown to correlate well with other measures of intelligence. 

 The latter condition is one which must not be overlooked. A test which will not correlate fairly 

 well with the total score of a good battery of tests is ipso facto under grave suspicion; there is 

 little likelihood that it will consistently correlate well with any other proved measure of intel- 

 ligence. 



The average correlation of each test with the remaining tests is shown elsewhere for exam- 

 ination a and for alpha (p. 540ff). Tests 2, 8, and 10 are most unlike the other tests. In the 

 higher ranges tests 3, 8, and 9 become more like, and tests 1 and 7 more unlike, the other tests. 

 It will be seen that the tests of alpha give slightly higher intercorrelations than do those of 

 examination a. It is, of course, possible that the committee on revision was too much influ- 

 enced by the point of view expressed in the preceding paragraph, though data presented in 

 chapter 1 1 show that the elimination of the two tests which yielded low correlations with the other 

 tests did not, as far as could be determined, lower the correlations of the total score with out- 

 side measures. 



Section 10. — Time limits of the tests. 



Table 10 shows the distributions of zero and perfect scores, and table 12 certain data 

 regarding attempts for each test in the case of 920 unselected men (Dix). Tests 3, 7, 8, and 9 

 are too difficult at the beginning. The most nearly normal distributions are given by tests 2, 

 4, 5, and 10. 



Such facts, however, only indirectly throw light on the proper time limits of the tests. 

 Certain data gained by the experimental lengthening of time limits have been set forth in 

 chapter 2, and still more important results are set forth in chapter 9. The Lee experiment, 

 in which the time of tests 3 to 10 was lengthened to allow 50 per cent of the unselected group to 

 finish, showed that only in tests 3, 9, and 10 was the score appreciably increased. The correla- 

 tions of the separate tests with total score were also little affected, as shown by the figures 

 below. Here it should be recalled, time was not extended for tests 1 and 2. The differences 

 found for these tests call in question the significance of the differences apparent in the other 

 cases. 



The figures suggest that only test 9 would benefit considerably by extended time. 

 Terman gave double time on tests 3 and 10 in the case of one group of 107 school children 

 who had been given the Stanford-Binet examination. The correlation of each test with mental 



