analyzing. the patterns of intercorrelations among the remaining six items. If they all measured the same concept, as proposed, then they should all be relatively highly intercorrelated, and the directions of their intercorrelations should be in accord with the posited relationships. Because all the variables except "leadership competition" are posited to be inversely related to institutionalization, the interrelationships among the variables should all be positive except for any involving leadership competition, which should be negative.

In general, the directions of the intercorrelations were as expected, but the magnitude averaged only .26, with the coefficients involving "name changes" and "organizational discontinuity" being especially low. A factor analysis of the correlation matrix showed these two variables to reflect a different factor from the other four, whose average intercorrelation was .48. A reanalysis of the other four variables extracted only a single factor explaining 62 percent of their total variance.[9]When the scores on these items are transformed into standard form (with means of 0 and standard deviations of 1), adjusted for direction of sign, and summed into a composite scale score, the reliability of the scale is a relatively high .79 as computed according to the formula for Cronbach's alpha, a measure of scale reliability (Bohrnstedt 1970).

The institutionalization scale demonstrates an impressive degree of reliability in the "internal consistency" sense, for example, the scale items are strongly and consistently intercorrelated among one another. But what does this mean for validating the conceptual framework? The measurement literature points out that demonstrating reliability (that items appear to be alternative indicators of the same thing) does not demonstrate validity (that the items are measuring institutionalization as claimed). At this point, one might appeal to face validity, for whatever it is worth, to bolster the measurement claim. Recall that variable 1.07 was eliminated soon after its inclusion in the conceptual framework because it was judged to lack face validity; it did not directly tap a party property but rather a consequence of that property. Ignoring for the moment the face validity of the two items that were dropped from the scale on empirical grounds, one is left to decide the face validity of the scale according to the remaining four indicators. Given the definition of institutionalization as a combination of endurance and stability, does it seem reasonable to assume that (1) older parties that (2) have undergone frequent and orderly change of leaders and that show little fluctuation in (3) legislative and (4) electoral strength over time are high in institutionalization?

Although this interpretation of the combined indicators might seem reasonable enough, one longs for a more objective interpretation to support the subjective judgment. Construct validity offers a more objective approach by demanding that relationships among the variables are observed as predicted. The prediction was that six multiple indicators of institutionalization would converge on one another through overlapping variance when applied to the same parties. But is this consistency in measurement nothing more than reliability all over again? Whether intercorrelations among indicators are viewed as evidence of reliability or construct validity seems to be partly in the mind of the viewer, but a distinction can be made according to the timing of the empirical research. Reliability is commonly an a posteriori finding. Given any set of items purporting to be a scale, one can determine internal consistency and hence the reliability of the designated scale. Construct validity, on the other hand, involves an a priori element by specifying in advance the set of items to be evaluated, which is the same as predicting them to be intercorrelated. To be sure, construct validity includes more than confirmation of expected convergence among indicators, but a convincing demonstration of intercorrelations among variables as expected is one manifestation of construct validity.

Because the conceptual framework specifies in advance the indicators that are supposed to be interrelated, the convergence among indicators as assessed by correlational analysis, factor analysis, and especially scale reliabilities will be interpreted hereinafter as assessments of construct validity, except as otherwise noted. In the case of the institutionalization indicators, the evidence for construct validity is strong but not unassailable. "Name changes" and "organizational discontinuity," which seem acceptable on their face as measures of institutionalization, did not intercorrelate strongly enough with the other four variables to be included in the scale. Nevertheless, the scale reliability of .79 seems adequate for supporting the conceptualization, and the empirical findings largely support this section of the framework.

Governmental Status. Parties vary in the nature and extent of their participation in national politics. The concept of governmental status, defined as the amount of access to the governmental structure that a party enjoys, is offered to subsume various expressions of party participation in politics. Originally, eight variables were proposed as indicators of governmental status:

2.01 Government Discrimination. The greater the positive discrimination in favor of the party, the

---

9. All factor analyses reported herein were performed with pairwise deletion of cases due to missing values and with unities in the diagonals. Very rarely were negative eigenvalues obtained, and they were very small when encountered. Comparisons of factor structures for listwise deletions found little difference between them.

advance to page 145