Path: Janda's Home Page > Workshop > Canonical Analysis > Return to Opening Page
Political Studies (1998), Vol. 46, pp. 61l-632
Effects of Party Organization on Performance during the 'Golden Age' of Parties
Organizational Effects on Overall Party Performance

Up to now, we have been concerned with explaining variations in three different aspects of party performance. Can we provide a more comprehensive explanation which simultaneously involves all three? If so, it would approximate scholars' efforts to 'type' parties according to similar organizational and behavioural traits. Wright, for example, distinguishes between the 'rational-efficient' and 'party democracy' models of behaviour according to their functions, structural characteristics, party processes, and evaluative criteria.[59] For Wright, rational efficient parties focus on their electoral function, engage in limited activities, are motivated by material incentives, employ organization suited to situational requirements, lack formal membership, neglect the policy role of the party, and evaluate effectiveness solely by electoral success. In contrast, those fitting his party democracy mold pursue ideological and governing functions, engage in activities beyond campaigning, stress purposive incentives, feature extensive and integrated structures, require formal party membership, emphasize policy making, and judge their effectiveness in terms of policy results. If we can somehow relate variations in all four organizational variables (complexity, centralization, involvement, and factionalism) simultaneously to all three aspects of performance (electoral success, breadth of activities, and legislative cohesion), we can give empirical content to such trait configurations, which we will call 'party syndromes'.

Canonical analysis provides a method for relating two such sets of variables on each side of an equation. It weights the variables on each side to produce two sets of composite scores and then calculates one or more canonical correlations, which are equivalent to product-moment correlations between the sets of weighted variables. The number of canonical correlations computed is equal to the number of variables in the smaller set. Rcan1 can be interpreted as the maximum correlation that can be obtained through the best linear combinations of both sets of variables. Rcan2 is the next best linear combination of the variables, under the constraint that this pair of composite scores is uncorrelated with the first pair--and so on. Whether the first or any of the subsequent correlations are significant, of course, depends on the relationships within the data.[60] In essence, the number of significant canonical regressions indicates the number of clusters of relationships in the set of elements analysed. Since we have posited two sets, more or less equivalent to the 'rational-efficient' versus 'party democracy' distinction, we expect two groupings from the canonical analysis.

Our canonical analysis of organizational characteristics and party performance is guided by the theory discussed above. To simplify the interpretation of the results, we dropped the two environmental variables, a parliamentary system and an effective legislature. The need for complete data for all variables on all cases reduced the number of parties to 44. Most of the findings discussed above reappear in the canonical results in Table 5. Of the three canonical correlations produced from the analysis, only the first two, R2can1 = 0.61 and R2can2 = 0.32, were significant at the 0.05 level.

Canonical correlations are essentially product-moment correlations between sets of weighted scores. So the squared canonicals in Table 5 express the variance in one set of variables explained by the other. Each correlation represents a different, unrelated solution to the relationship among the observations. The analyst's task is to interpret these solutions by referring mainly to two sets of values on the computer output.[61] One set is the standardized canonical variate coefficients, which are akin to the beta-coefficients in an ordinary regression equation. These coefficient can be compared for the relative effect of each variable in one set to the composite score constructed from the other set of variables.

The other and perhaps more useful indicators are the simple correlations between the canonical scores and their composite variables. These correlations are called canonical 'loadings'--like variable loadings in factor analysis. Based on the variables' standardized variate coefficients and their loadings on both canonical scores, we interpret the two canonical correlations reported in Table 5 as reflecting different syndromes of party performance. They correspond to Wright's 'party-democracy' and 'rational-efficient' party models, but we prefer to label them the 'doctrinaire' and the 'mobilizing' party syndromes. Whereas 'model' implies categorization, 'syndrome' suggests a measurable pattern of traits that are common to all parties but are exaggerated by some.

Doctrinaire Parties

The first canonical solution is called the 'doctrinaire' party syndrome due to the configuration of canonical variate coefficients on the performance side in Table 5: high values for legislative cohesion and breadth of activities and a negative value for electoral strength. The simple correlations on the far right show that cohesion and breadth of activities correlated 0.81 and 0.76 with the composite score, while electoral strength barely had any correlation (0.12). The canonical correlation squared reveals that 61 % of the variance in the performance composite can be linked to the composite score of the organizational variables, for which centralization and involvement are the most important. In fact, centralization by itself correlates 0.88 with the organizational composite. Of little importance in the analysis is complexity, a condition that figured in most of the regression analyses above.

TABLE 5. Squared Canonical Correlations between Organizational Traits and Party Performance*
Correlations with composite Scoresa
Four Organizational Variables
Canonical Variate Coefficientsb

Canonical Variate Coefficients
Three Aspects of Party Performance
Correlations with Composite Scores
First Canonical Analysis: the Doctrinaire party syndrome

Electoral Strength
Breadth of Activities

Legis. Cohesion

Second Canonical Analysis: the Mobilizing party syndrome

Electoral Strength
R2can2 = 0.32
Breadth of Activities

Legis. Cohesion

Red or blue indicates high loadings that define the two uncorrelated party syndromes.
*This canonical analysis is based on 44 parties that had valid data on all seven measures.
aThe simple product-moment correlations between the variable and the composite scores computed in the canonical analysis.
bThe standardized coefficient of the variable used in computing the canonical variate that generated the composite scores.

Although involvement has a negative effect on cohesion alone (Table 5), it makes a positive contribute to the doctrinaire party syndrome (0.70), which is somewhat puzzling. The effect of involvement on party performance appears to vary considerably, depending on the control of other variables and the mix of performance indicators. It deserves closer scrutiny at a later time. A succinct verbal summary of the first canonical analysis might be that highly centralized parties with highly involved activists, moderate complexity, and little factionalism tend to be very cohesive, engage in many activities, but are not particularly successful.

This analysis is illustrated in Figure 1, which identifies and plots the composite organization and performance scores from the first canonical analysis for 44 parties. Note that the most doctrinaire parties, located in the upper right corner of the figure, according to their performance in 1957-62, were the West German SPD, the French and Indian Communists, and Peru's APRA. At the other extreme, the least doctrinaire--in the sense of pursuing electoral success at the cost of legislative cohesion marked by very low centralization and considerable factionalism--were both US parties and the Dutch CHU.

Figure 1. 'Doctrinaire' Party Syndrome: Plot of Composite Scores from the First Canonical Analysis

Mobilizing Parties

The second canonical correlation (Table 5) corresponds to the 'mobilizing' party syndrome--so named for the dominant influence of electoral strength, which by itself correlates 0.79 with the composite score, followed by breadth of activities at 0.53. The mobilizing syndrome reflects a second attempt to maximize the correlation between the two sets of variables, under the constraint that the second solution be uncorrelated with the first. The squared canonical correlation for the mobilizing syndrome explains only 32% of the trait variation--much less than that for the doctrinaire syndrome. Nevertheless, its theoretical linkage is clear. Mobilizing performance is related mainly to high complexity (0.04) and very low involvement (-0.02). Centralization and factionalism have virtually no effect. A brief summary of these results might be that very successful parties that engage in a moderate range of activities--but have little legislative cohesion--tend to be distinguished by high organizational complexity and little else in the way of party organization.

The plot for all 44 parties in Figure 2 illustrates the second analysis of party performance in 1957-62. The highest performers on the mobilizing syndrome were the Uruguayan Blancos, the German Christian Democrats, and the Swedish Social Democrats. The lowest performers were the Mayalan MIC and the Australian Country party. The Democrats placed in the upper group, while the Republicans placed in the center followed by the British Conservatives and British Labour.

Figure 2. 'Mobilizing' Party Syndrome: Plot of Composite Scores from the Second Canonical Analysis

Path: Janda's Home Page > Workshop > Canonical Analysis > Return to Opening Page