Next to the number of parties in the system, their ideological spread has been perhaps the second most used measure for the characterization of party systems. It was something, that was implicitly important already for Downs, who brought the idea of party systems as single dimension ideological spaces to political science, and has been on the political science research agenda ever since. It has been related to other such central elements of political systems, like party system fragmentation, electoral turnout, volatility, voting patterns and aspects of government formation, among others. Estimates of polarization are calculated usually on the basis of party positions on the left-right dimension and their relative sizes. However, and setting aside the fact that there are also many different suggestions about how to actually calculate polarization, there are various different kinds of estimates for parties’ ideological positions. And many of these indicate rather different locations for the same parties. Each of these claims to be better than the others and therefore we should also have estimates of polarization, some of which are better than others. How would it be possible to compare these different estimates?
The objective of this paper is to provide such a comparison. It will use over a dozen different measures of political positions that have been proposed over the years, all based on the manifesto dataset, and the index of programmatic similarity, also based on the latter, to compare the corresponding estimates of party system polarization. It is assumed that a better measure is the one that is better able to capture possible associations with other closely related phenomena in the party system. The paper will model the association of polarization with other such party system characteristics like fragmentation, volatility, voter turnout and others, that have in some cases been shown or argued to have an association to the programmatic or ideological differences between parties. Models that are fitted to exactly the same set of cases, but use different estimates of party system polarization, are then compared to each other in terms of overall model fit. The model and thus the measure of polarization that fits the data better, is considered to be the better model and the better measure, as it describes reality better.
