340 MEMOIRS NATIONAL ACADEMY OF SCIENCES. [vol. xv. 



The question was raised whether weighting should not be abandoned. For an unselected 

 group of 900 men at Camp Dix a correlation of 0.994 was found between total raw and total 

 weighted scores. In the case of a more homogeneous group of 300 men at Camp Lee the correla- 

 tion between total raw and total weighted scores was only 0.93. Correlations at Lee between 

 officers' estimates and the two methods of scoring were as follows for three different groups : 



With the same group the plan was tried of weighting only by halving the score of test 5 and 

 doubling the scores of tests 6 and 9. The result was a slightly higher correlation with officers' 

 ratings for each of the three batteries — 53, 58, and 59, respectively. 



It was generally agreed that the weighting system which had been used was at fault in allow- 

 ing too much weight for tests 5 and 10, especially in the lower ranges, and too little weight for 

 tests 6 and 9. Two methods were suggested whereby an individual's score would be defined in 

 terms of its relation to the average score of an unselected group : (1) The percentile method and 

 (2) multiples of the standard deviation. The former was rejected as unsound, since it would 

 involve the averaging of percentiles to find a total score; the latter was rejected because of the 

 difficulty incident to the use of the necessary conversion tables for translating raw scores into 

 their weighted equivalents. It was finally decided to let the revision of weighting be determined 

 after the revised scale had been given to a group of unselected men. 



Section 12. — Summary of modifications of examination "a" made in the construction of exami- 

 nation alpha. 



In the light of the data presented in the foregoing sections revision of examination a was 

 undertaken along the f o flowing lines : 



Test 1. — Changes: (1) The first item eliminated and three harder items added at the end. 

 (2) Order of items altered to conform more strictly with the order of difficulty. 



The test did not differentiate well between normal and feeble-minded subjects, as 20 per 

 cent of the feeble-minded succeeded in four items or more. The test was more valuable in the 

 middle ranges of intelligence. Too many officers made perfect scores. An additional reason 

 for increase of items was to reduce the importance of chance errors due to distraction. The 

 three additional items were prepared chiefly by Whipple and Otis. Remained test 1 of exami- 

 nation alpha. 



Test 2. — Changes : It was decided to eliminate this test. It was found to correlate poorly 

 with all other measures of intelligence, particularly in the data furnished by Terman. Its corre- 

 lations with officers' estimates were also poor, except in the lower ranges, and it was thought 

 that with a literacy test which would segregate all below fourth grade literacy for a beta examina- 

 tion, little value would be left for test 2. The difficulty of administration was also an objection 

 to the test. It gave more opportunity for cheating than any other test of the scale and was time 

 consuming both in giving and scoring. 



Test S. — Changes: (1) Instructions slightly altered in the direction of simplification. 

 Easier samples were inserted. (2) Two much easier problems were substituted at the beginning 

 of the test and enough harder problems of an abstract type were added at the end to make up a 

 total of 24 items instead of 20. (3) Several sentences were eliminated or altered because of 

 ambiguity. (4) All items were arranged in order of difficulty according to ratings made by the 

 office force, and the items were equally distributed among the five forms according to difficulty. 



The test is good in the upper ranges. The large number of zero scores, even with officers, 

 indicates that the instructions were unsatisfactory. The addition of harder items and the sim- 

 plification of instructions were intended to increase the value of the test both in the higher and 

 in the lower ranges. Became test 5 of examination alpha. 



