Key points are not available for this paper at this time.
ONE method for item-analyzing a scale involves the assessment of item-total correlations (see Guilford, 1954, pp. 417-443). Using this method one simply computes the correlation of each item with the sum of the items (total score) and eliminates those items that seem to be more poorly correlated with the total. The most rigorous way of pursuing this procedure would involve correlating each item with the sum of the remaining items (total minus the relevant item) in order to avoid the spurious correlation inflation caused by the relevant item’s inclusion in the total. This latter approach, how-ever, is quite time consuming and the usual practice is to assume that each item-total correlation is equally inflated. The above assumption is probably adequate for deciding between items in the measurement of one scale, but is not applicable to the problem of discriminant item analyses. A valuable extension to the convergent and discriminant approach to construct validity of Campbell and Fiske (1959) can be made in the area of test construction. In their now classic article, they stress that trait measurement must be validated by low correlations with tests from which they were intended to differ, as well as high cor-relations with tests with which they were intended to agree. Camp-bell and Fiske proposed the evaluation of this discrimination and convergence by the multitrait-multimethod matrix which involves the measurement of more than one trait by each of at least two
Howard et al. (Sat,) studied this question.