Essentials of Personnel Assessment and Selection by Scott Highhouse & Dennis Doverspike & Robert M. Guion

Author:Scott Highhouse & Dennis Doverspike & Robert M. Guion [Scott Highhouse]
Language: eng
Format: epub
Publisher: Routledge

The multiple correlation, R, will equal the sum of the individual correlations only when the predictors are uncorrelated (highly unlikely). If the predictors are correlated, then R < r1 + r2.

Suppressor and Moderator Variables

Suppressors. By those principles, each test in a well-developed battery is a valid predictor of the chosen criterion and has low correlations with other variables. A valid predictor may contain an invalid, contaminating variance component. A variable that does not predict the criterion but is correlated with the contamination may actually improve prediction. To see how this works, look again at Equation 3. If ryx2 = 0, but if both of the other two correlations are not zero, then the numerator of that equation becomes simply ryx1 (the other two terms being zero). The denominator is less than 1.0 (because rx1x2 is not zero); therefore, Ry·x1x2 is greater than the validity of the one valid predictor alone. The reason is that variable X2 removes from the composite (suppresses) the unwanted variance in X1 not associated with the criterion. In a regression equation, it has a negative weight.

Consider, for example, a case in which a paper-and-pencil test of law enforcement knowledge is used to hire security guards. This test is valid, but it requires a relatively high level of reading ability to complete—a level of ability not necessary for a security job. A reading ability test would correlate with the law enforcement knowledge test, but not with ability to perform the security job. The reading ability test would, therefore, receive a negative weight in the regression equation. Although it may slightly improve prediction to add a reading test to the security guard selection system, it would be hard to explain to the company (and the test taker) why a candidate is rejected for scoring too high on it!

Moderators. Moderator variables influence the relation between other variables; they are correlated with correlation. Frederiksen and Melville (1954) found prediction of academic performance from interests better for noncompulsive students than for those classed as compulsive. Although it is easier to think about validities in subgroups, validity should change systematically and continuously as the level of the moderating variable changes. A regression equation for one predictor and one moderator has the following form:

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.