496 



SELECTION AND TRAINING 



Factor analysis is not a path to be followed 

 blindly. It can identify abilities found in 

 present tests, but will overlook abilities not 

 now tested. An unending search for further 

 abilities must be made; for example, a psy- 

 chologist could profitably employ his time 

 inquiring "what types of individual differ- 

 ences in hearing capacities might conceivably 

 be tested?" Work prior to World War II 

 had stemmed from Seashore's interest in 

 music. He had identified numerous hearing 

 aptitudes found in music, and had made 

 tests. Two of them, the rhythm test and 

 the pitch test, turned out to be useful in 

 selecting code operators and sonar operators, 

 respectively. These tests were used because 

 someone had designed them; no one knows 

 how many other significant hearing apti- 

 tudes were overlooked because group tests 

 had not been considered worth developing. 

 Searches for types of differences, without 

 regard to their present practical significance, 

 may well be made in motor ability, reason- 

 ing, vision, and so on. 



Preliminary research is needed to demon- 

 strate the effectiveness of factorial test bat- 

 teries. While many workers have hopes 

 that combining "pure" tests of different fac- 

 tors will give effective prediction, this has 

 not been tried. It may be that such a 

 battery will be inefficient in classification, 

 if it requires some men to take dozens of 

 tests to get the same information that one 

 test designed to predict a single job could do. 

 It is imperative that workers educated in 

 factorial methods set up a small-scale classi- 

 fication program to determine what theoreti- 

 cal and practical problems will arise. Such 

 a program must be carefully validated. Fac- 

 tor studies so far have shown that factorial 

 tests measure different things, but no study 

 has yet shown that the factor scores give 

 superior predictions of anything. 



A departure to be encouraged is factorial 

 studies based on criteria of job performance. 

 The present technique is to make a map, or 

 coordinate system, which most efficiently de- 

 scribes the tests used in selection; then a 



given job is located relative to the test fac- 

 tors. It might be more efficient, although 

 more difficult, to design a factor pattern to 

 describe Navy jobs most efficiently, and to 

 locate tests in terms of job factors. This 

 is so radical a departure that no methodology 

 exists. It might be profitable enough to 

 justify even the most extreme labor. So 

 far, no study has considered more than one 

 criterion at a time. If all Navy jobs were 

 considered at once, and a listing were made 

 of the factors most prominent in all Navy 

 jobs, an exceedingly useful classification bat- 

 tery could be designed. The effect of this 

 plan on test construction is discussed below. 



Factorial Design vs. Job Replicas 



One of the gravest doubts about the fac- 

 torial approach is that it tends to disregard 

 "unique" abilities, found in only one test. 

 The Air Force list of factors reports elements 

 found in several tests, but it discards, for 

 example, the special factor found in the com- 

 plex coordination test. This test, in which 

 the pilot operates a stick and rudder in re- 

 sponse to light signals, had nearly the highest 

 validity of any test in the program (18). 

 But 27 percent of the saturation of this test 

 was a factor found only in this test, a factor 

 which does not appear in the list resulting 

 from analysis. Other tests having superior 

 validity, but heavily influenced by factors 

 not in the master list, are Map Distance (for 

 navigators), Biographical Data (pilot), and 

 Discrimination Reaction Time (valid for 

 pilot, bombardier, and navigator). Tests 

 known as job replicas, which are complex 

 tests closely resembling the job to be pre- 

 dicted, have frequently been excellent pre- 

 dictors. Factor analysts assume that tests 

 of elements, rather than of complex combina- 

 tions of elements, are the most efficient tests. 

 But when one integrates many abilities into 

 a complex pattern, he does more than add 

 them. Concentrated research is necessary 

 to determine whether factorial tests can re- 

 produce the predictive effectiveness of job 

 replicas. If they cannot, research should 



