ABSTRACT
In this chapter, the strength of the explanatory variables with regard to the dependent variable is analysed in order to distinguish the relevant determinants of electoral system choice. Since the dependent variable is categorical, logistic regression is applied.1 The primary method for dealing with a nominal dependent variable consisting of three or more categories is multinomial logistic regression, whereas binary logistic regression is designed for dummy dependent variables. Multinomial logistic regression selects one category as the baseline and estimates sets of coefficients for contrasts with the other three variables respectively. Each set of coefficients represents the effects of a one-unit change in the explanatory variables on the logged odds of belonging to each category relative to the reference category. In order to make a complete estimation including all four categories of the dependent variable, another three comparisons must be made. In each comparison, the coefficients refer only to the subset of cases falling into the two categories used in a contrast. A categorical dependent variable that has more than two values might also lend itself to a set of separate binary logistic regressions. It is possible that four categories might involve four logistic regressions with four dummy variables: the first category versus all others, the second category versus all others, the third category versus all others, and the fourth category versus all others. However, the combination of several categories in making a comparison with one category implies that precise contrast between two categories cannot be isolated. The advantage of running several separate binary regressions instead of multinomial regression is that the number of cases is larger, which means that the results are more reliable. Therefore, the democratic sample is subject to binary regression, while multinomial regression is preferred when the total research population is analysed. Since variables of three competing perspectives are included in the study, three different regression models are presented. Party system structure and party system transformation cannot be included in the regression analyses, which means that the rational perspective is represented by two variables only: population size and ethnic/cultural diversity. Either ethnic or cultural diversity must be excluded, because they to a great extent correlate with each other.2 Cultural diversity is preferred to ethnic diversity, since the former takes several aspects
into consideration. The alternative vote is considered appropriate for ethnically and culturally diverse countries in contrast to other majority formulas – therefore, the formula is excluded from those analyses that include cultural diversity. Since only two variables of the rational perspective are analysed, I shall begin with an analysis of patterns of diffusion. Thereafter, cultural diversity and population size are added. In the last analysis of the entire data sample, variables of all three perspectives are included. Another reason to begin with the cultural and historical perspective instead of the rational perspective is to keep the number of models at a minimum. Form of government is not analysed in the large sample, because the variable concerns democracies only. To begin with, a decision must be made on which variables of the historical and cultural perspective should be included in the logistic regression. As to colonial diffusion, four variables are conceivable: British legacy, French legacy, Soviet legacy and a fourth variable that observes all former colonies. Legacy of other colonial powers than these are not considered, because they possessed only a few colonies. We noticed earlier that Soviet legacy does not correlate with electoral system choice. Accordingly, this variable is not included in the regression analysis. A control for multicollinearity shows that the variable that observes all cases of colonial diffusion correlates with both British legacy and French legacy, particularly with the first mentioned. Therefore, the variable that concerns all colonies is omitted from the analysis. This variable also correlates with regional diffusion and temporal diffusion, which further justifies the exclusion. Both regional and temporal diffusion are operationalized as dummy variables. The latter is represented by four dummies, each variable denoting a distinct period. The first period, which consists of considerably fewer cases than the latter periods, is not included in the analysis. Variables of the cultural and historical perspective are analysed in Table 7.1. Beginning with the comparison of plurality systems with other systems, we notice that British legacy stands out as the most important factor. French legacy also affects the adoption of plurality systems but not to the same extent as British legacy. The rareness of plurality choices in recent times results in a significant effect of the fourth period when plurality systems are compared to the other electoral system categories. The association is particularly strong when mixed systems are compared to plurality systems, suggesting that mixed systems have been very popular during the fourth period. Distinct trends of electoral system choice did not exist during the second and the third period. Regional diffusion plays a significant role in explaining the introduction of proportional systems in comparison with majority and mixed systems. Double-checking with four separate binary logistic regression models confirms these results. Cultural diversity and population size are added to the regression model in Table 7.2 and Table 7.3. Comparisons between plurality systems and all other systems are presented in the former table, whereas coefficients of the three remaining comparisons are given in the latter. The chi-square and the Nagelkerke values are slightly higher but the −2 log likelihood value also exceeds the one in Table 7.1, suggesting that the explanatory model is not considerably improved
by including the two variables of the rational perspective. Surprisingly enough, cultural diversity has a significant effect on the adoption of plurality systems in two comparisons – however, there is reason to believe that this relationship is spurious, since it runs contrary to the expectations. Concerning patterns of diffusion, the tendencies observed in the former model remain. When these findings are double-checked with four separate binary logistic regressions, the difference between plurality systems and other systems is once again the most prominent of all, whereas the comparison of majority systems with other systems returns little to explain. Mixed systems are linked to the fourth period, and proportional systems are associated with regional diffusion. Variables of all three perspectives are included in Table 7.4 and Table 7.5. Again, plurality systems are compared to the other systems in one table, while the remaining three comparisons are given in the other. The institutional perspective is represented by territorial organization and chamber structure. It has
been emphasized that these two variables are connected to each other. More specifically, bicameralism is often considered necessary in federal systems. A test of covariance in this data sample shows that they correlate to some extent but the coefficient is smaller than 0.3 – hence, both variables can be included in the regression model. The model chi-square and the Nagelkerke values are somewhat higher, whereas the −2 log likelihood value is modestly lower than in the previous model, indicating that the model is slightly improved by adding variables of the institutional perspective. Still, territorial organization and chamber structure have no influence on electoral system choice in any comparison. Again, the adoption of plurality systems is associated with British and, to a smaller extent, French colonial legacy. The adoption of mixed systems and the unpopularity of plurality systems is strongly related to temporal diffusion during the fourth
period. The model also emphasizes the importance of regional diffusion in explaining the adoption of proportional systems. As in previous models, no variable explains anything when majority and mixed systems are compared with each other. Double-checking with four binary logistic regressions confirms these findings. Normally, a final model that constitutes a parsimonious compromise is presented. In this case, the most parsimonious model would consist of British legacy, French legacy, regional diffusion and the fourth period, that is, a model similar to the first model. When the second and the third period are excluded
from the first model, the model chi-square value (130.7) and the Nagelkerke value (0.408) are only modestly affected, whereas the −2 log likelihood value is considerably improved from 136 to 80.1. In comparison with Table 7.1, the most parsimonious model strengthens the significance of temporal diffusion in recent times. British and French legacy are related to the introduction of plurality systems, whereas regional diffusion is associated with the adoption of proportional systems. As a whole, these four variables explain electoral system choice in the total research population. In general, the effect of French legacy is weaker than the
effect of British legacy, regional diffusion and temporal diffusion. The importance of temporal diffusion is confined to the frequent adoption of mixed electoral systems and the decline of plurality system choices between 1990 and 2007. In the following, the democratic sample is analysed. Majority systems are not compared to the other electoral systems, because there are only five democratic cases of majority systems. There is, in other words, no variance to explain concerning majority systems in relation to other electoral systems. Again, patterns of diffusion are initially analysed. French legacy is not included in the model, because there are only two (Gabon and Vanuatu) former French colonies in the democratic sample, none of which inherited electoral provisions from France. Coefficients for British legacy, regional diffusion and temporal diffusion in three separate regression models are presented in Table 7.6. British legacy stands out as the main determinant of choosing plurality systems in comparison with other electoral systems. Contrary to the findings in the total research population, temporal diffusion does not explain the adoption of mixed systems in this model. No variable has a significant effect when mixed systems are compared to other
systems. Regional diffusion explains the adoption of proportional systems relative to other electoral systems. The highest chi-square value is observed in the comparison between plurality systems and other systems. Regarding the democratic sample, population size and cultural diversity cannot be included in the same regression analysis with British legacy, because all three variables are interrelated. First, the relationship between population size and electoral system choice is contaminated by the peculiarities of microstates. In contrast to the statements by Rokkan (1970), Katzenstein (1985) and Rogowski (1987), D. Anckar (2002) asserts that plurality systems are related to smallness. The pattern is particularly evident among diminutive microstates. After a discussion on other potential explanations, D. Anckar (2002: 16) concludes that the thesis of diffusion cannot be evaded. British influence, rather than population size or other rational explanations, is the critical factor in this respect. The covariance between these two variables is considerably smaller in the total data sample. This matter is more evident among democracies, because a large part of the former British colonies have had a high level of democracy since independence. Second, in the large sample, we found a significant, positive association between cultural diversity and the adoption of plurality systems. However, there is good reason to assume that this relationship is spurious. Certainly, the possibility of affirmative gerrymandering having taken place in some countries should not be rejected. By drawing constituency borders according to ethnic lines and concentrating minority groups to a few electoral districts, it is possible to provide representation of several minorities in plurality elections. In addition, many highly fragmented countries are agrarian societies, which, according to Joel D. Barkan (1998), are better served by plurality systems than PR. He argues that the basic needs of the local community take precedence over ethnic cleavages when it comes to elections in agrarian societies. Still, the main theoretical arguments contradict the observed pattern. If the ethnic, linguistic and religious composition of society matters to electoral system choice, fragmented countries should, contrary to the empirical findings, choose proportional systems. By looking around, further light is shed on the complex of problems. Among those 25 countries that are characterized by a high level of fragmentation, 14 have adopted plurality systems, whereas only five have adopted proportional systems. Most of the plurality systems are inherited from colonial powers. The Philippines, Palau and the Federated States of Micronesia inherited their systems from the United States, whereas the United Kingdom exported SMP to Malaysia, Jamaica, Trinidad and Tobago, Fiji, Dominica, and the Solomon Islands. In addition, 13 of a total of 17 cases with low fragmentation are related to mixed or proportional systems. Thence, low-fragmented countries with mixed or proportional systems cause a great deal of the correlation between cultural diversity and electoral system choice. Accordingly, cultural diversity and population size are not included in the second model. In Table 7.7, variables of the institutional perspective are added to the first model. The explanatory model is somewhat improved, the model chi-square and
the Nagelkerke values now being larger, and the −2 log likelihood values being smaller than in the previous table. Yet, no variable of the institutional perspective has any significant influence on the dependent variable. In accordance with the previous regression models, the adoption of plurality systems is associated with British colonial legacy, proportional systems are related to regional diffusion, and no variable has any significant effect when mixed systems are compared to other systems. Again, the most parsimonious model is rather similar to the model in Table
7.6. Electoral system choice in democracies is mainly explained by British colonial legacy and regional diffusion. Temporal diffusion is of no importance in the democratic sample. Neither is French colonial legacy of any relevance in explaining electoral system choice among democracies. The findings among democracies resemble those in the total research population.
