Thus, opposite conditions lead to maximization
of Equations 7.1 and 7.2. Equation
7.1 shows that internal consistency is maximized
when items are highly correlated, and
Equation 7.2 shows that criterion-related
validity is maximized when items are uncorrelated.
In practical terms, this means it is
mathematically impossible to simultaneously
maximize reliability and validity of a scale.
Therefore, test developers must choose
between making very homogeneous scales
that reliably predict only a narrow set of
behaviors versus creating more heterogeneous
scales that measure more comprehensive
scale content. Because the predictive
validity of a scale is the ultimate measure of
its worth, internal consistency reliability
should not be the main criterion used in scale
development.