from one office group to another. The value of such averages is doubtful, since the overlapping of groups is so great as to preclude the possibility of making reliable individual predictions on the basis of the averages.

Link (83) reports that 935 clerks were tested in the employment office of an industrial firm, with a series of intelligence tests and tests for technique (card sorting, letter substitution, arithmetic, etc.) and follow-up reports obtained. Only 188 cases are discussed. These were clerks recommended on the basis of the tests and then followed up every month for three months. The tabulation given is:

Percentage of those called good by their superiors:
At the end of I mo...... 75 per cent.

The size of these per cents means little since they depend upon the method of rating and the kindliness of the superiors. The increases from month to month, however, are significant. Tests for typists, stenographers, and calculating machine operators are also reported by Link. More than 1,000 people were tested in the construction of these tests. Both tests of aptitude or ability and tests of proficiency were used. Results for typists and stenographers are not given, and only a few comptometer operators (in the calculating machine group) are reported upon. The tests in each instance, however, were deemed of sufficient value to be adopted for use in employment.

Among the earliest attempts to build tests for ability in special clerical occupations are those of Lahy (79), who tried out a variety of psycho-physical tests on typists and found a few of these tests that showed rather close agreement with typing ability, and those of Lough (87), who correlated scores on a substitution test with the abilities shown by commercial students in their several school studies and found fairly high correlations with typing, business correspondence, and stenography. A number of later studies have been made along more or less similar lines. Bills (12, 13), who used a series of tests on 139 students of comptometry and stenography in a technical night school, found a general intelligence test decidedly valuable in eliminating the failures and a special ability test in selecting the successful students. The tests were later found to agree well with ratings of ability of two small groups of stenographers in a business concern. The method used by Bills is especially worthy of note. Scatter diagrams are plotted between scores on two tests, and the ratings of individuals according to the criterion are shown by different markings (as circles and crosses for successes and failures respectively) on the scatter sheet. It is then possible to draw two dead-lines and study the combined effect. This study also emphasizes the fact that different tests may be useful for predicting at different levels of ability. Bregman (21) reports the use of a long series of tests on a decidedly heterogeneous group of clerical workers in a department store, using only the extremes of the group. Some relationship of tests to ability is indicated but the correlation coefficients, as already pointed out, mean nothing.1 Rogers (131) tried out 10 tests on 43 students of typing and stenography and compared these with typing tests during the training period and with instructor's estimates. Combinations of tests gave correlations of about 5 and 6. Jacques (64) reports some further work of Rogers on typists in a business firm. Several psychological tests (not proficiency tests) were found to correlate closely with the ability of 38 typists. No data are presented.

Cody (32) presents a long series of tests-all of them educational or proficiency tests for office workers. The tests are for almost all the usual operations in a business office. They have been standardized among thousands of school students and office employees, but the author includes no statistical comparisons with criteria nor any measures at all save the average scores for different groups tested. Benge (II) developed a proficiency test for extenders and verifiers in a city gas company. Henderschott and Weakley (55) devised and tried out a test for billers and found fair agreement with estimated ability (though the fallacy of the selected group is again present). Marcus (89) developed a series of tests for Hollerith card punchers, which proved to be considerably more valuable in predicting ability than the civil service examination that had been in use, though even with the new tests the correlations coefficients do not appear to be large. Carney (28), experimenting with a large number of tests for selecting time clerks, obtained only a few fairly close cases of agreement between tests and competence on the job. Flanders (40) found no significant correlation between scores on an intelligence test by express clerks and their success at work. Wells (174) developed a test for file clerks but did not proceed with its standardization. Burtt (25) used 10 tests on a group of general clerical workers but insufficient information is given to show what the obtained correlation coefficient of .56 means.

2. Only a few instances are reported in the literature of tests that were tried out for factory workers. Link's (83, 84) work in the Winchester Repeating Arms Co. is the outstanding example of thoroughgoing scientific test procedure in factory or workshop. Similar, but less extensive, research was later conducted in a Canadian rubber company by Burtt (25).

Link (83, 84) describes in considerable detail the construction and standardization of special tests for five different occupational groups in the shop. In each instance a careful study is first made of the requirements of the work; then a series of tests is tried out; comparisons are made between test scores and some measure of working efficiency-production records where available and otherwise estimates by superiors. Sixteen tests, for example, were tried for shell inspectors, three of which proved especially diagnostic, giving correlations of 6 and 7 for a group of 52 inspectors. These tests were almost equally valuable for three other groups of inspectors doing highly similar work. Follow-up reports of applicants who had been tested also showed that the tests were operating effectively. The same tests were found to be very poor for selecting shell gaugers. A series of tests for assemblers of gun parts, tried with small groups of men and women assemblers, was found fairly satisfactory. A valuable test was devised and standardized for classifying a group of machine operators according to ability. And finally, tests for apprentice machinists were found which correlated very closely with estimates of ability for three small groups of apprentices.

Burtt (25) used 32 different tests on several groups of workers. He does not mention the number of people in any of his groups. Correlations of tests with ratings of ability (using multiple correlation procedure and corrections for attenuation) ranged from .7 for "workers who hand out stock" to almost zero for "tire builders". With 400 workers in a silk mill, Otis (III) found no correlation between general intelligence test scores and ability at work. Thompson's very early work, reported by Taylor (156), consisted in the selection of inspectors of ball bearings by means of a reaction time test. Remarkably valuable practical results were obtained, 35 of the selected individuals doing the work that 120 unselected ones had been performing. 1

3. On the selection of salesmen even less test research has been reported than in the case of factory workers. Scott (139) published a little material on the selection of salesmen by tests as far back as 1915. He reports the use of an intelligence test including arithmetic, opposites, proverbs, directions, etc., with 10 travelling salesmen of a large tobacco company. The results are not unequivocal but show some agreement between test scores and sales records. A later series of tests for salesmen worked out under Scott's supervision is included in the Aids in Selecting Salesmen, of the Bureau of Salesmanship Research at Carnegie Institute of Technology (27). No results have been reported from the use of these tests. Further advances in tests for selecting salesmen, especially life insurance salesmen, have been made at Carnegie Institute of Technology but are not yet published.

In the field of retail salesmanship, also, certain tests have been developed at Carnegie Institute of Technology and tried out in department stores in Pittsburgh. The only published results concerning retail sales tests, however, are those of Oschrin-Bregman (21, 110). In the main study reported, very high negative correlations were gotten between scores on a variety of mental tests and ratings of ability of some 50 salespersons by their superiors. Only the direction and not the size of the coefficients is significant due to the way in which the groups were selected and the correlations calculated. A number of other statistical processes are used in this study which appear to the reviewer unwarranted and misleading. The same writer, moreover, had arrived at opposite conclusions in an earlier study (110) where she found high positive coefficients of correlation between the estimates of ability of 18 saleswomen and the scores on a series of tests. The earlier results are not reconsidered in the later study. Burtt (26), reporting some results from data collected by Münsterberg, presents evidence that at least one of a number of tests tried on Harvard students and on several industrial groups is diagnostic of salesmanship ability. Hollingworth and Poffenberger (60) present some average scores in comparison with salaries for a group of 55 salesmen "engaged in selling all manner of commodities in all manner of ways. ". They found some tendency for higher scores and higher salaries to go together.

..

4. Far more effort has gone into the study of aviation from the point of view of vocational selection than into any other single occupation, military or civil. The principal results of American research are described in a number of articles (4, 34, 36, 56, 153, 154, 163, 181), and certain prominent features of European investigations are summarized by Dockeray and Isaacs (34) and touched upon by Stratton (153). Elaborate physical examinations and painstaking interviewing methods had been used in selecting aviation candidates from the beginning, but the predictions of future success had proved sufficiently unreliable to warrant intensive research on the problem by psychologists as well as by physiologists and medical men. Despite the tremendous difficulties involved in analyzing the abilities that enter into aviation and the corresponding difficulties of determining these capacities, considerable success was attained. Mental alertness tests and evaluations of a variety of personal information concerning the applicant were shown to be useful, on the one hand, and a number of psycho-physical tests were evolved to supplement these and get at the candidate's probable ability to fly and to endure such special conditions as partial asphyxiation. Correlation statistics presented by Henmon (56) and by Stratton, McComas, Coover, and Bagby (154) show rather low coefficients between scores on several tests and ratings of flying ability, but not so low as to be without promise. The latest report of research (154) in which a number of tests were used, both ones that had been found useful before and a number of new ones, concludes that: "The tests, as a whole, and some of them singly, are to some extent diagnostic. Their precise value, however, can be known only after trial under more favorable conditions."

Careful experimental studies were made of other military and naval tasks, one or two examples of which are briefly to be noted. Dodge, in a report edited by Yerkes (181), describes the methods he developed which proved extraordinarily valuable in the United States Navy for the selection and training of gun-pointers by means of specially constructed apparatus. He tells also of the tests devised for selecting recruits of relative fitness for plotting room service and of the comprehensive system of examinations of candidates for the Listeners' School in the Navy. In all these re