Validation of selection tests or other selection variables is most frequent
ly accomplished through correlation and regression. It is quite common to n
ote that both sexes or two ethnic groups were included in the sample. Somet
imes in the combined sex or ethnic groups the correlation of the predictor
and criterion may be moderate or large, but within each group the correlati
on of predictor and criterion is low or teres. This seeming paradox is not
a psychological phenomenon, but a consequence of the mathematics of regress
ion and correlation. Examples of this two-group phenomenon are provided and
discussed. The problem would seem to require a Within and Between Analysis
(WABA), but having only two groups prohibits the use of WABA methodology.
A general hierarchical linear models analysis is proposed and demonstrated.