Page 345 - Discrimination at Work The Psychological and Organizational Bases

ARTHUR AND DOVERSPIKE
312
including altering test weights, random selection after a specified cut score,
race-based selection, and banding. The literature on altering test weights
suggests that this may reduce the degree of adverse impact in selection sys­
tems caused by cognitive ability tests. However, the differences in weights
must be fairly substantial to result in a practical difference in adverse im­
pact (Doverspike, Winter, Healy, & Barrett, 1996). Random selection after
a specified cutoff and banding are relatively effective methods of reducing
adverse impact, but the greatest reduction in adverse impact occurs when
race-based selection is used after banding or after setting cutoffs (Sackett
& Roth, 1991). Unfortunately, in addition to questions concerning the lost
utility of the test, there are legal, political, and ethical ramifications in­
volved in the use of race-based selection (Barrett, Doverspike, & Arthur,
1995; Campion et al., 2001).
Use of Nonability Constructs The use of nonability constructs has been
considered as an approach to reducing subgroup differences. Common
nonability constructs that have been investigated in attempts to reduce
adverse impact include personality variables (Hogan, Hogan, & Roberts,
1996) such as conscientiousness (Schmitt et al., 1996) and integrity (Sack­
ett & Wanek, 1996). Although subgroup differences have been shown to
be lower on some of these constructs (e.g., personality variables; Hogan
et al., 1996; Hough et al., 2001), the use of nonability predictor constructs
in reducing adverse impact has not been very successful (Ryan, Ployhart,
& Friedal, 1998), and the lower validity resulting from the use of these con­
structs may result in a considerable reduction in utility (Schmitt, Rogers,
Chan, Sheppard, & Jennings, 1997).
Combine Cognitive Ability with Other Predictors A variation of the con­
struct change approach has been to combine cognitive ability with other
predictors—either constructs (e.g., personality variables, integrity) or
methods (e.g., structured interviews, performance tests). Although the ad­
ditional predictors add incremental validity beyond cognitive ability, the
empirical evidence suggests that combining cognitive ability with these
alternative predictors does not necessarily eliminate subgroup differences
for a wide range of selection ratios (Sackett & Ellingson, 1997; Schmitt et al.,
1997).
Changing the Test Method Another approach to reducing subgroup dif­
ferences and adverse impact is to change the test method with the intention
of altering test perceptions and attitudes, and also of reducing nonjob­
related reading demands. This approach to reducing adverse impact rec­
ognizes that cognitively loaded paper-and-pencil tests of knowledge, skill,
ability, and aptitude are the most valid predictors of job performance
but posits that subgroup differences on cognitive ability may partially
arise from the mode or method of testing—specifically, paper-and-pencil