Page 378 - Discrimination at Work The Psychological and Organizational Bases

14. LAW AND PSYCHOLOGY
for example, may have some predictive validity beyond that provided by
an interview, but unless the predictive validity of the exercise is substan­
tial enough to demonstrate business necessity, it could produce disparate
impact problems that the employer might rather avoid. 345
Yet another problematic aspect of disparate impact analysis arises when
the employer uses an employment practice to screen a population that
is heterogeneous along a key dimension. Factors other than the employ­
ment practice may be associated both with the pass rate and with the
protected group status. For example, a recruiting technique that has a
disparate impact on Black persons may have differential pass rates de­
pending on the socioeconomic status or education level of the persons it
is applied to. Suppose that education level is associated with race in the
sense that persons of a given race with differing educational levels tend
to have different outcomes from the recruiting technique. Because of ag­
gregation paradoxes, it is possible that within each educational level, there
is no disparate impact of the recruiting technique on Black persons—or
perhaps there is even an advantage to Black persons. Traditional disparate
impact analysis would ignore these latter possibilities—i.e., the Black plain­
tiffs could establish a disparate impact for the one selection mechanism—
yet the stratification by educational level would seem to shed statistical
doubt on that finding. Statistical inferences based on the initial aggrega­
tion across educational level is misleading in either of these two cases,
because educational level confounds the relationship between race and
recruiting pass rate. Because the recruiting technique is the only selection
mechanism used by the employer (i.e., educational level is not a selec­
tion criterion), it is unlikely that the employer could defend by showing
that the confounding factor—educational level—eliminated the impact
(see Paetzold & Willborn, 1996, who discuss the "blindness" of the dis­
parate impact model). The relationship between recruiting method pass
rate and educational level should be examined by the employer as part
of the determination as to whether the recruiting method should be em­
ployed.
Obviously, not all possible dimensions of the population can be identi­
fied as being relevant and important to consider. Nonetheless, some clearly
have the potential to be related to selection mechanisms and are easy
enough to consider. Because employers do not want to create a situation
in which they could be held liable for a disparate impact when a practice
that is useful but perhaps difficult to justify as meeting business necessity
is used, they should take these heterogeneous factors into account in their
decision making.
The stratification problem here could go the other way as well. Just be­
cause a practice does not appear to have a disparate impact when evaluated