Signature requirements and citizen initiatives: Quasi-experimental evidence from Germany

See Lupia and Matsusaka (2004) and Matsusaka (2005) for general reviews of the functioning of direct democracy.

For example, the German Pirate Party uses an online voting tool called “Liquid Democracy” to arrive at all party positions. Also, major infrastructural projects in Germany are often planned in cooperation with citizen associations to avoid public protest at a later stage.

Recently, Boehmke et al. (2013) have shown that in US states, the length of the gridlock interval (the area in the policy space where no reforms are possible) is a better (positive) predictor of initiative use than signature requirements, which turn out to be insignificant.

Population thresholds have also been used to identify the effects of wage changes for politicians (Ferraz and Finan 2009; Gagliarducci and Nannicini 2009), the allocation of transfers (Litschig and Morrison 2010; Brollo et al. 2009), representative versus direct democracy (Hinnerich and Pettersson-Lidbom 2010), fiscal rules and policy (Grembi et al. 2012), female participation in politics (Campa 2012), and voter turnout (Eggers 2013; Tukiainen and Lyytikäinen 2013).

Below the federal level there are 16 states, followed by about 12,500 municipalities which are organized into about 450 counties.

Note that German municipalities are often also in charge of administering funds allocated to them by higher tiers of government, e.g., for public schools or social services. In these areas, municipalities may have only limited discretion.

The introduction of direct legislation in German states is often accompanied by other institutional changes, as noted by Blume et al. (2011). In this regard, Bavaria is an ideal setting for evaluating the functioning of direct democracy. Bavaria offers a setting in which other features like direct election of mayors, vote aggregation and splitting or the treatment of small parties (explicit electoral hurdles) have been stable for many years before the introduction of the tool of citizen initiatives. In contrast, other states often changed those features and those reforms sometimes coincide with the direct democracy reforms. The introduction of the initiative is thus “unique” in Bavaria compared to the other German states, which is why we refrain from pooling states and concentrate on Bavaria.

See www.​gesetze-bayern.​de for more information.

Each state has a list of topics that are eligible for initiatives. Note that one reason that we observe a relatively large number of initiatives in Bavaria is that the state law allows for many different aspects of local government policy to be decided by citizen initiatives.

This so-called “quorum” was introduced on April 1, 1999 and hence 3.5 years after it became possible to propose initiatives. Twenty percent (15 %) of the electorate has to vote “yes” in municipalities below 50,000 (100,000) inhabitants for the proposal to become law. Above 100,000 inhabitants, 10 % is sufficient.

Around 500,000 inhabitants, there is a 2 % jump, but this affects only the cities of Munich and Nuremberg.

An anonymous referee correctly pointed out that fixed costs in setting up the signature collection campaign might play a crucial role and lead to economies of scale. Our argument applies only to the variable costs of collecting signatures. On the margin, proposers in municipalities above and below the cutoff incur the same fixed costs.

See http://​www.​mehr-demokratie.​de/​bb-datenbank.​html.

We use the official population figure for the current year as the crucial measure. On the basis of this information we then determine what kind of signature requirement was in place and whether a municipality belongs to the treatment or control group.

Distance to the threshold is measured in percentage points. For the threshold of 10,000 inhabitants, the ±5 % sample will include all municipalities that have a population between 9,500 and 10,500 inhabitants.

Here, we follow Imbens and Kalyanaraman (2012).

When we write initiative in this section, we always mean initiative petition – i.e., the first stage of the process (see Sect. 2).

We also estimated a model in which we give individual dummies to each of the five population thresholds and estimate separate effects (results not reported). Owing to the smaller sample size at the individual thresholds, the results are mostly insignificant. However, the point estimates are very similar to our main effects and range from 6.6 % points at the 10,000 to 10.0 % points at the 30,000 thresholds. The point estimates seem to increase slightly with the size of the population, but they are statistically not significantly different from one another.

Note that we implement the RDD using OLS for all outcomes. Alternatively, we also tested logit and ordered logit models when appropriate (results not reported). The results came out to be very similar, which is why we chose to keep the model parsimonious.

The interested reader can find all tables and figures in the working paper version of this article (Arnold and Freier 2013). Also, all results are available upon request.

Also, based on a formal test (McCrary 2008), we cannot reject the null hypothesis of no sorting. Note that our results are markedly different from the evidence on sorting in German municipalities presented by Ade and Freier (2011), who use data that include earlier periods. In fact, the sorting in Ade and Freier (2011) is observed mostly in the 1980s (which is before our data period). We can think of three explanations for the non-existence of sorting in our (later) data. First, the large influx of migrants from former East Germany after 1990 made it harder to exert control over population statistics (Freier et al. 2013). Also, because of that influx, municipalities that had negative population growth before 1990 (and hence strong incentives to manipulate the population data) are now showing solidly positive growth. Second, the introduction of electronic administration with automatic voter registration and de-registration makes it more difficult to cheat. Third, any existing sorting incentive is likely to be most prevalent just before elections. Initiatives are also proposed, however, at non-election times when there is no benefit in sorting.

In particular, we test the number of valid votes, voter turnout, as well as council seats held by the Christian Social Union (CSU) and the Social Democratic Party (SPD), all in the 1990 elections.

The idea of this test is the following. Council size changes both at the thresholds where the signature requirement changes and at the other thresholds of 1,000, 2,000, 3,000, and 5,000. If council size would indeed drive our effects, we would expect to observe this influence at the lower thresholds as well, where the signature requirement does not change.

Egger and Koethenbuerger (2010) and Ade and Freier (2011) are concerned with the effect of council size on public spending in the same Bavarian data—although in earlier years. The former paper argues that large spending effects occur as a result of council size changes at population thresholds. To ensure that our results on the probability of observing an initiative are not driven by council size changes or resulting changes in spending, we test whether directly controlling for these variables makes a difference in our analysis. Note that directly controlling for spending may be problematic in our setting. Asatryan et al. (2013) estimate a RDD using spending as an outcome variable, showing that initiative use positively impacts municipal expenditures. Thus, spending may be considered endogenous. For this reason, we include spending only as a robustness check.

We control for the quorum in two different approaches: First, we extend our model to include indicator variables for higher/lower quorums and before/after introduction of the quorum. Second, we estimate our model including the quorum as a continuous variable. In both cases, the treatment effect stays similar in size and retains its significance.

To the extent that other confounding discontinuous rules at the population thresholds affect processes of direct democratic decision making, they should also have influenced the outcome of council-initiated referendums. The fact that this is not the case supports our analysis.