How to Measure Ideological Polarization in Party Systems

In addition to fragmentation, polarization is one of the most established and discussed indicators of party systems (Dalton 2008, Sartori 1976, Curini/Hino 2012). Although a lot of research has been done on this subject, political studies vary concerning the specific operationalization (e.g. Taylor/Herman 1971, Gross/Sigelman 1984, Dalton 2008; in summary Rehm/Reilly 2010: 40). The common ground is that polarization conceptually represents some kind of ideological dispersion of parties’ positions within the electoral competition. However, there is a significant disagreement about the specific measurement and, apart from that, most studies lack a substantial reasoning for their used operationalization. At least, I clarify three distinguishing features: (1) the underlying measure of dispersion, (2) the usage of a weighting function, and (3) the considered number of dimensions. Hence, the study aims to compare the potential measures of polarization and to show the differences between them. The findings improve our understanding about the indicators of party system polarization. Referring to recent empirical studies, the most frequently used operationalization consists of a variance based measure combining one left-right dimension with a weighting by parties’ electoral success (e.g. Taylor/Herman 1971, Ezrow 2007, Dalton 2008, Steiner/Martin 2012). Alternatively, the summed distances between parties (e.g. Gross/Sigelman 1984) or the range between the maximum and minimum ideological position (e.g. Mair 2001) are also taken as a basis. Furthermore, ideological polarization can also be interpreted as electoral success (or presence) of ideological extreme parties. Finally, most of the mentioned measurements still stick to a one-dimensional approach even though this limitation is frequently discussed (e.g. Franzmann 2009, Laver/Sergenti 2012). Hence, I compare multi-dimensional approaches with their one-dimensional counterpart. The comparison and evaluation of the different measurements take place in two steps. First, I analyze a formal, theoretical model of spatial party competition to illustrate the differences referring to counterfactual scenarios. As a result, I illustrate that the pictured disparities systematically vary due to the characteristics of the party competition. Thus, the ranking of polarization between cases partially depends on the used measure. Second, I analyze the empirical observable differences in a time-series cross-section analysis based on one the Manifesto Project Data (Volkens et al. 2015, in addition Franzmann/Kaiser 2006). In particular, the position of the major parties, the fragmentation and, as a consequence, the electoral system influence the disparity between these measures. Hence, the differences between measurements of polarization are also empirically context-sensitive. Lastly, the consequences of choosing a specific measurement of polarization are discussed. Referring to study’s findings, the peculiarities of the measures have to be considered to understand empirical analyses and their conclusions regarding the causes and consequences of party system polarization.