Path: Janda: Political Parties, Home Page > Part 1: Table of Contents > Chapter 14
Chapter 14: Validating the Conceptual Framework (pp. 135-161), p. 141
(you can navigate to other pages by clicking on page numbers below)
p. 135
p. 136
p. 137
p. 138
p. 139
p. 140
p. 142
p. 143
p. 144
p. 145
p. 146
p. 147
p. 148
p. 149
p. 150
p. 151
p. 152
p. 153
p. 154
p. 155
p. 156
p. 157
p. 158
p. 159
p. 160
p. 161

parties coded, and the third gives the mean AC codes assigned over all the parties that were coded on the variables. It can be seen that we were able to code 100 percent of the parties for only 9 of the 111 variables. Our rates of coding success varied widely over the other variables.

One can see that we were least successful in scoring parties on a number of the Goal Orientation variables in cluster 6. The entries in Table 14.2 tagged with footnote 5 identify variables that were coded for less than 30 percent of the parties. This low rate of success was not due to the lack of applicability of the framework--in the sense that these variables did not translate readily into observed phenomena--but rather may be attributed to the dwindling of research funds. Coding on these variables was stopped about one-third of the way through the parties, which explains the low percentages scored.[7]

There are only two other places in the conceptual framework where we were unable to code more than half of the parties in the study. One place is in another part of the goal orientation cluster, variables 6.51 through 6.55. These variables pertain to a party's efforts in providing for the welfare of its members, activities that tend not to be discussed at length in the literature--as witnessed by their relatively low AC codes, averaging only 5.2. The other place is in the social support cluster where we found it difficult to obtain information for coding the composition of party support for religious, ethnic, and educational groupings. One might wish to attribute these low success rates to the empirical inapplicability of the conceptual framework. On the other hand, one can argue--perhaps in all these cases--that the weakness is evidential rather than conceptual, that is, in the literature, not the framework.

Excluding the variables already discussed, we found it possible to score most of the parties in the study on each of the variables in the framework. In fact, more than two-thirds of the parties were scored on all but 6 of the remaining 82 variables. Cluster by cluster, it can be seen that we were most successful in scoring parties on variables pertaining to their institutionalization and governmental status, while least successful for degree of organization--along with the goal orientation and the social support clusters already discussed. Note that the percentage of parties coded is not an accurate guide to the quality of the data coded. The most striking case is the involvement cluster. Although an average of 81 percent of the parties were scored on these variables, the average of the AC codes attached to those scores was the lowest for any of the clusters. This was because of the problem of determining the motivational bases of party militants. There were few hard data relating to these variables, and the AC codes reflected the heavy reliance on coder judgment and inference in the scoring.

Recognizing that the literature is unclear about the parties' activities and characteristics on many points, one should ask how reliably the variables in the conceptual framework were scored. Coding reliabilities were assessed at the beginning stages of the project by having two coders independently score the same parties on common variables. A total of 557 such "blind pairs" were generated over many variables in the framework. The variables with roughly comparable means and standard deviations were grouped together to calculate several associated correlation coefficients as measures of reliability. When the resulting coefficients are summed and weighted according to the number of cases involved in each, the weighted mean correlation was .79, indicating a substantial degree of reliability in the scoring. One might also compute a single, unweighted correlation for the pooled cases, but only after eliminating variables with large variances and high coding reliabilities. Otherwise, these variables dominated the measure of reliability, yielding an inflated value of .98 over all 557 cases. Dropping the extremes produces a correlation of .87 for 536 cases. While merging the double-coded data produces higher intercoder correlations, computing the weighted mean correlation for cluster of variables with comparable dispersions provides a more realistic estimate of coding reliability. Moreover, an overall coding reliability of .79 seems readily acceptable.

Is the proposed conceptual framework applicable to empirical research on political parties throughout the world? Clearly, it is not perfectly workable with information that is currently collectable, but its data demands are far more often met from existing information than denied. One might suspect that there is a pervasive bias throughout the framework that makes it more empirically applicable to "Western" parties--which generate most of the methodologically and analytically impressive research literature--than non-Western parties. But, for most of the variables, straightforward historical studies and descriptive reports so typical of the literature on non-Western parties proved more useful for scoring purposes than did more theoretically oriented research. Moreover, the sheer quantity of the available literature on many Western parties is often less than that on party politics elsewhere. Our information files of Iceland, Ireland, and Luxembourg--for example--were much smaller than our files on Ghana, Ecuador, and North Korea. Nevertheless, it is true that the Western parties were coded at a somewhat higher rate than the non-Western, but the differences are not dramatic.

7. Reference to the AC codes can test this explanation, If these 21 variables whose coding were curtailed were equally applicable to party phenomena as the 12 which were coded, the mean AC codes for each set would expect to be comparable. Indeed, the AC codes for the omitted variables averaged 6.8 compared with 6.2 for the other variables in the goal orientation cluster.

go to page 142