492 



SELECTION AND TRAINING 



selection. But a reconsideration of the 

 problem by Taylor and Russell (45) before 

 World War II suggests a more realistic 

 standard. As they point out, it is obvious 

 that a test with any validity at all is more 

 useful for classification than a blind guess, or 

 taking every tenth man out of line to fill a 

 job. They derived tables to show just how 

 much the quality of a working force would be 

 raised if a test of given validity is used. 

 This led to the conclusion that tests of very 

 low validity can be helpful and profitable if 

 it is possible to skim off the cream of the re- 

 cruits. If only one recruit out of 100 is 

 needed for a draft for School A, a test with 

 validity .10 will raise the quality of men sent 

 to the school; if half the recruits must be 

 sent to School A, a test with validity of .10 

 will not help enough to be worth using. 



In selecting, one may choose 10 men out of 

 100 for a job and discard the rest. In place- 

 ment, including military training, one must 

 place 99 men out of 100 in some job or other; 

 very few men can be discarded. Placement 

 is a different problem from selection. It no 

 doubt has its mathematical logic which, if 

 formulated, would provide a guide for choos- 

 ing or rejecting tests. This logic must be 

 worked out. 



In broad terms, the problem is this: One 

 has X rates (or specialties, or stations) to fill ; 

 some of them are more important tactically 

 than others, and error in filling them must be 

 avoided. For each rate, selection methods 

 are available which predict the success of 

 each man in each job, with a certain margin 

 of error. What processing formula can con- 

 sider both the prediction and the importance 

 of the job, to place each man so that the total 

 force will have maximum effectiveness? If 

 all men not better fitted for something else 

 must be assigned to the job lowest in im- 

 portance, it does not appear worthwhile to 

 test aptitude for that job even if a test of high 

 validity is available. For the most impor- 

 tant job, a selection method should be used 

 even if its validity is so low that it would de- 

 crease failures by only one percent. What 



is the relation between test validity, job 

 importance, proportion of recruits required 

 for each job, and other variables? This is a 

 mathematical question, open to research, 

 which would suggest how to design process- 

 ing routines in wartime (16, pp. 58-61). It 

 might lead to definite rules, e.g., "One should 

 fill the quota for the most important job with 

 the best men before making any other assign- 

 ment," or "The classifying office should place 

 every man in the specialty for which he is 

 best fitted until half the billets in each spe- 

 cialty are filled; after that, quotas should be 

 filled in order of importance." 



All present statistical methods predict a 

 man's success in one field at a time. Prob- 

 lems of differential prediction likewise re- 

 quire statistical solution. The classifying 

 officer must consider the man's relative suc- 

 cess in many fields. Suppose a man has an 

 aptitude score of 60 for torpedoman, and an 

 aptitude score of 60 for radioman. He is not 

 likely to do equally well in both jobs; errors 

 of measurement and errors of prediction 

 make him a better risk for one job or the 

 other. No present statistical routine is ade- 

 quate for giving a score showing "likelihood 

 that this man will be a better torpedoman 

 than radioman." Yet that is predictable, 

 and it is a judgment crucial in manpower 

 utilization. Probably it will use different 

 tests from the present single-job prediction; 

 intelligence is needed for both jobs, but an 

 intelligence test would not help decide to 

 which job to assign the man. The only 

 direct helps now available on the differential 

 prediction problem are a set of formulas on 

 the reliability of test profiles, recently simpli- 

 fied by Bennett and Doppelt into practical 

 forati (4), and a preliminary proposal by 

 Brogden (8) for the determination of differ- 

 ential cutting scores. A thorough mathe- 

 matical study of prediction is called for. 



Billet Analysis 



Processing fills billets, but the psychologi- 

 cal problems begin earlier, when billets are 

 defined (60). What constitutes a job? 



