(De)Centralization and voter turnout: theory and evidence from German municipalities

LAU stands for ‘Local Authority Unit’ and is a classification issued by Eurostat. While LAU 2 is the municipal level, LAU 1 captures different forms of inter-municipal cooperation.

For simplicity, we assume throughout the paper that there are two types of public goods: (i) one where the costs/benefits are localized to a small subgroup within the municipality (e.g., child care centers, local parks) and (ii) one where the costs/benefits cover the entire municipality (e.g., municipal roads).

In the German setting analyzed below, these communities often reflect previously independent municipalities that merged into one new (larger) municipality.

Although we introduce only two parties (Y and Z) in our model formulation, the Downsian model—and our use of it here—can easily be extended to a setting with multiple parties and coalition formation (see McKelvey and Ordeshook 1972; Geys and Heyndels 2006). As this does not add additional theoretical insights, but complicates the notation, we refrain from doing so here.

For more details on Germany’s federal system, municipal tasks and comparisons of municipal types, we refer the interested reader to Biehl (1994), Zimmermann (1999) and Rosenfeld et al. (2007), respectively.

This issue is strengthened by the fact that the 2005 national election was originally scheduled for 2006. As the rescheduling was announced by Chancellor Schroeder in May 2005, the campaign for the national election of 2005 did not affect the local elections contained in our sample that were held in 2004. This restriction excludes observations from Bavaria (which held local-level elections in 2002 and 2008) and Lower Saxony (with local-level elections in 2001 and 2006).

Although restricting ourselves to the relative size of the first two parties in our measure of closeness may be overly restrictive in our multi-party setting, measures explicitly aimed at incorporating the vote (or seat) shares of multiple parties such as the ‘entropy’-measure proposed in Kirchgässner and Schimmelpfennig (1992) and Kirchgässner and Zu Himmern (1997), i.e., \(E = - \sum_{i = 1}^{n} p_{i} \ln(p_{i})\) with p _i the vote (or seat) share of party i, are problematic as well since they generally depend on the number of parties included in its calculation (which clearly contaminates their measurement of size inequalities between the parties). As there is no commonly accepted solution to this problem, to the best of our knowledge, we gave preference to the relative simplicity of the two-party vote difference.

Since some federal states aggregate information on the vote share of non-partisan candidates in their official statistics, the share of such non-partisans can become quite large in our sample. To avoid this biasing our estimate of the election closeness effect, we enter a dummy variable equal to 1 when the share of non-partisan votes exceeds 33 % of all votes.

In most previous work, population homogeneity is approximated by either racial/ethnic or income diversity. Unfortunately, we lack detailed information for both these indicators. Hence, we instead rely on age dispersion as measured by the Herfindahl-Hirschman index: \(\sum_{i} x_{i}^{2}\), where x _i is the share of citizens in a specific age class (age: <3, 3–6, 6–10, 10–15, 15–18, 18–20, 20–25, 25–30, … , 65–75, >75).

We carried out both estimation procedures, but focus on the quasi-maximum likelihood (QMLE) results below. The reason is that some evidence suggests that even if the beta assumption is valid, the maximum likelihood approach outperforms the QMLE estimator only in certain circumstances (Ramalho et al. 2011). Moreover, to verify that our results are not driven by this particular choice of econometric method, we also performed a further robustness check building on a traditional OLS estimation with a logit-transformed dependent variable (i.e., log(turnout/(1−turnout))) and Huber-White corrected standard errors. The results remained almost identical (details available upon request).