inition. But the indicators proposed for the concepts are not the concepts themselves, and one cannot establish the distinctiveness of the concepts simply by appealing to the original definitions. One must turn instead to the indicators involved in both scales to determine if they discriminate between the concepts in their patterns of interrelationships.
Adapting the reasoning and terminology of Campbell and Fiske (1959) to our context, we can require that multiple indicators of the same concept (trait) be more highly intercorrelated among themselves than they are with indicators of a different concept. In the ideal situation, the lowest correlation among indicators of the same concept would be greater than the highest correlation among any of the indicators between concepts. But in practice this turns out to be a very stringent requirement that will be relaxed to the lesser requirement that the mean correlation among indicators of the same concept be greater than the mean correlation among indicators of different concepts.
The four pairs of highly related scales that give rise to this analysis of discriminate validation are presented in Table 14.8 along with the mean correlations among the items within each scale and between each scale. In each of these problematic cases, the within-scale correlations tend to be greater than the correlations between items in one scale and those in the other, although the margin of difference is very small for the involvement scale in the last case. Nevertheless, these summary statistics support the argument that the scales are measuring different concepts which are nevertheless strongly related empirically.
Evidence of Discriminate Validation among Indicators in Scales That Were Highly Correlated
What the summary statistics in Table 14.8 do not reveal, however, is whether any of the particular indicators in one scale tend to correlate more highly with those in the other scale than with its own companions. In fact, this is not generally a problem. All the indicators in all of the scales tend to relate better with their "own kind" than with those in the other scale--except when the degree of organization scale is matched against involvement. Recall from Table 14.8 that the degree of organization indicators tended to intercorrelate among themselves only marginally stronger (difference of .08) than with the involvement indicators, which in turn showed only slightly more clustering within the scale than between scales (difference of .07). Table 14.9 portrays the situation in greater detail, and we see now that two of the "organization" indicators and two of the "involvement" indicators correlate as well or better with items in the other scales. "Intensiveness of organization" (cell, branch, city, or larger), "frequency of local meetings," severity of "membership requirements," and "doctrinism" (appealing to party doctrine) appear to measure aspects of both degree of organization and involvement.
Clearly, this finding poses a problem concerning the conceptual independence of these two concepts, or at least their proposed indicators. Degree of organization was defined as "the complexity of regularized procedures for mobilizing and coordinating the efforts of party supporters in executing the party's strategy and tactics." Involvement was defined as "the intensity of psychological identification with the party and the commitment to furthering its objectives by participating in party activities." With the former pertaining to party structure and the latter to individual psychology and behavior, these concepts seem to be distinct enough. The fault then must be in the choice of the indicators. Upon reflection, one can see "organizational" features in the severity of membership requirements (e.g., does the party require the payment of dues?) and in appeals to party doctrine (e.g., is someone authorized to interpret party doctrine?). However, the "involvement" features of the