ingly disconfirmed by the data. The missing data problem prevented a proper testing of the goal orientation cluster, so the validation issue is unresolved for this concept. The social activities scale found among its variables does not square completely with the conceptualization and thus cannot properly be counted as a confirmation of expectations. Concerning the concept of autonomy, the five variables did not interrelate at all as expected, which serves a purpose by demonstrating that there was nothing in the data, the conceptualization, or the design that forced the variables to behave as expected. Perhaps autonomy is the exception that supports the validation.

In summary, six of the ten clusters of variables performed virtually as expected and two performed mostly as expected. One of the remaining clusters could not be properly validated, and the other simply did not bear out expectations, leading to a completely new reconceptualization. Regardless of how one counts these results at the concept level, most of the variable clusters were supported by the empirical tests.

One might also summarize the results in another way. The original conceptual framework for the comparative analysis of political parties identified 111 variables organized under eleven (revised to ten) separate concepts. Not all of these could be included in the validation due to missing information--and 21 variables were dropped for lack of funds to collect the data. Another 3 variables in the institutionalization and governmental status clusters were dropped early for conceptual reasons and were not subject to empirical testing. Subtracting these 24 from the original 111 leaves a total of 87 variables that were involved in empirical tests. Of these, 67 or 77 percent were successfully incorporated into scales composed of other variables within their original concept clusters. Thus one might say that three quarters of the variables proposed for the comparative analysis of political parties met tests for convergence among indicators conducted to validate the conceptual framework.

The strategy of "convergent validation" considered up to now sought to determine whether the indicators for each concept were intercorrelated as expected. This approach considers only the internal consistency of the proposed scale items within each concept and ignores the relationships among the concepts. But, as Campbell and Fiske point out, "Tests can be invalidated by too high correlations with other tests from which they were intended to differ" (1959, p. 81). In other words, the items should not only converge upon one another within the same scale but should be discriminating in their convergence, so that they do not also "converge" upon items supposed to measure a different concept.

The ten concepts proposed to accommodate the observable variation among political parties across nations were conceived as being analytically distinct from one another, but not necessarily empirically independent. For example, one can envision "degree of organization" as being conceptually distinct from Marxist orientation and yet being empirically related through the proposition that Marxist parties tend to be more highly organized than non-Marxist parties. Specifying the empirical relationships expected among the concepts is within the province of theory construction and is a task to be undertaken comprehensively later. Validating the conceptual framework through the process of discriminate validation requires only that conceptual independence be established, regardless of theoretical linkages. In practice, this means that the variables proposed as indictors for a given concept should correlate more highly among themselves than they correlate with indicators of any different concept.

We can investigate the discriminate validity of the items in the conceptual framework initially by examining the correlations among the scales. The lack of a strong empirical relationship between any two concepts as embodied in the scales obviously supports their conceptual independence and thus implies discriminate validity among their indicators, already demonstrated to be internally consistent. It turns out that relatively few of the scales are strongly related empirically. A total of 105 correlations are produced by intercorrelating the 15 "parent" scales, those that are not subsets of other scales (as are positive state, personal power, and structural power.) Observing that the question of discriminate validity seemed to arise only as correlations between the scales approach or exceed .50, we established .40 as our selection criterion. We found that thirteen correlations attested to a "strong" empirical relationship between two supposedly different concepts. However, nine of these involved pairs of the "social diversity" scales, which are themselves alternative measures of the concept of social support. Because this "layering" of indicators within the same concept complicates the analysis without illuminating the issue of discriminate validity among the indicators of the major concepts in the framework, the nine instances of high correlations between social diversity scales will not be considered here.

Our attention turns instead to the four correlations above .40 that involved cross-concept scales: institutionalization with urban-rural diversity, governmental status with liberalism, Marxism with involvement, and degree of organization with involvement. The question at hand is "Are these scales conceptually or analytically distinct although empirically related?" One might answer that of course they are, for they were constructed using indicators of concepts which were distinct by def-

go to page 159