To avoid discrimination in the use of tests, an employer should make sure that tests are not a subterfuge for discrimination. It might be important for an individual to be able to read, and an employer might have the requirement of a high school diploma. Such a requirement could have a substantial adverse effect on minorities and would, therefore, be discriminatory because although literacy was required, a very high level of literacy might not be proportionally valuable to employees.

Diamond (1976) has located the sources of gender bias in testing in society, test content, and biased use of test results. She suggests that changes should be made at the level of item-writing. In one study (Scheuneman, 1987) it was found that test items asking "for the one false response" (which of the following is not true) had a disparate impact on minorities. Therefore, one could remove such items from a knowledge test. In an article in Personnel Psychology, Arvey (1986) outlines issues pertaining to possible gender bias in job evaluation procedures and reviews relevant research in this area.

The format of a test can also affect discrimination. Schmitt and Mills (2001) found adverse impact was significantly lower where simulation tests were given (.30) as opposed to paper-and-pencil tests (.61).

Test Impact

Employers are required by Equal Employment Opportunity Commission and other governmental guidelines to keep records of the impact of selection procedures. This means that the information on test performance must be kept for whites, blacks, Hispanics, Asian-Americans, American Indians, and for males and females. An example of a computerized report one might use for employment tests of reading, arithmetic, and job knowledge is shown below.

Yearly Test Analysis

Generally, the employer would compare these various groups to determine whether they met the so-called four-fifths or 80% rule. So long as one group or another does not fall below 80% of parity with the comparison group, discrimination is not said to take place. Obviously, however, if the numbers are very, very large and the 80% rule is barely met, the Equal Employment Opportunity Commission would probably file charges.

Another procedure is to assess the significance of the difference between two percentages. One can use the chi square statistic or other statistics for the significance of the difference between two percentages (Marascuilo and Serlin, 1988, pp. 323-325).