SELECTION 



495 



experience in pilot selection is an example: 

 in a group of men screened only by a physical 

 examination, the failure rate was 77 percent, 

 but among men selected by the best combi- 

 nation of tests, the failure rate was only 24 

 percent (16, pp. 75 ff.)- Even this is far 

 short of perfect prediction; the validity co- 

 efficients around .65 leave much room for 

 improvement. 



Identification of Basic Aptitudes 



One major line of progress has been the 

 factorial technique for analyzing tests. In 

 the past, testing has been a trial-and-error 

 process. One test after another is tried to 

 measure various components of job success, 

 and those which work are retained. This 

 leads to several types of inefficiency, well 

 represented in Peterson's study of the Navy 

 Classification Battery (57, pp. 19-21). (1) 

 Tests supposed to measure different abilities 

 really measure the same thing under different 

 names. This was true for GOT and Read- 

 ing. (2) Classification personnel are di- 

 rected to consider scores on separate parts of 

 a test, when the separate parts have no real 

 meaning (for example, Mechanical Knowl- 

 edge: Mechanical, and Machanical Ivnowl- 

 edge : Electrical were nearly identical psycho- 

 logically). (3) A test supposed to measure 

 one factor may measure something else, 

 so that the intended factor is left unmeas- 

 ured. Thus, the Arithmetic test turned out 

 to be a measure of general schooling and 

 intelligence, and no measure of quantitative 

 ability was in the basic battery. 



Factor analysis is more than a technique 

 for removing inefficient tests from a battery, 

 though that outcome is significant when it 

 permits a classification interviewer to con- 

 sider only three test scores per man instead 

 of twelve. The greater contribution of fac- 

 tor analysis is that it maps out the relations 

 between abilities so that future research can 

 be more systematic. If we know what fac- 

 tors (common elements) are found in tests 

 which seem to work in job prediction, we can 

 design tests to measure those factors directly. 



Some investigators believe that tests could 

 be made for all the significant abilities of 

 men, and that in classification for any given 

 job one would merely combine such scores 

 to measure the combination of abilities 

 needed on that job (19). 



Studies such as that of the Air Force, 

 which identified nearly 30 factors in their 

 classification tests, are to be encouraged, 

 since this yields a listing of abilities which 

 ought to be considered in setting up any 

 future classification plan. It would be most 

 helpful in planning selection of men to oper- 

 ate any new gear to have an efficient list of 

 all major human abilities which might be 

 involved. But study of aviation tests is not 

 enough. Studies of all types of tests found 

 useful in classification are needed. Prob- 

 ably a good place to begin such a study 

 would be to assemble all tests now known to 

 be of use in processing, whether from Navy, 

 Air Force, industry, or schools, to administer 

 overlapping sets of them to similar groups of 

 men, and to separate out the factors. 



One of the fundamental issues remaining 

 to be settled in factor analysis is: what 

 mathematical system should be employed? 

 There are a large number of mathematically 

 sound procedures, of which mention may be 

 made of Thurstone's centroid method lead- 

 ing to correlated factors, Guilford's use of 

 the centroid method to get independent but 

 less meaningful factors, Holzinger's extrac- 

 tion of a "general" factor before isolating 

 specific aptitudes, and Kelley's principal 

 components technique. Which of these is 

 now used depends on an investigator's pref- 

 erence. Probably, for military purposes, 

 there is one method which is most efficient. 

 Since factorial studies made in the next 

 twenty years will no doubt affect all future 

 processing, it is especially desirable to have 

 some statistician analyze the military prob- 

 lem to determine which plan should be fol- 

 lowed in current studies. All methods are 

 sound, but almost certainly one of them will 

 be of greatest use in designing selection 

 batteries. 



