By Arndt Leininger, PhD candidate, Hertie School of Governance, Berlin
What is the story?
Turnout in the 2014 European Parliament election in Germany has increased to 48.1% from 43.3% in 2009. Some political observers have offered as an explanation that the 4.9%-points increase in turnout is due to an unusually high number of local elections and referendums conducted in parallel to the EP election on May 25th, 2014. In total, local elections were held in 11 out of 16 German states – the capital and city state of Berlin held a state wide referendum. I put this idea to the test using official preliminary regional election returns in conjunction with demographic and socioeconomic data on regions (Wahlstrukturdaten).
I show that turnout has been higher in regions that conducted local elections in parallel to the EP elections and that a substantial portion of that difference can confidently be accorded to local elections. Secondly, I calculate counterfactual turnout, 46% at the national level, that would have obtained if no local elections would have been held. Lastly, I show that one cannot quickly dismiss the idea that local elections contributed to the increase in turnout from 2009 to 2013.
Regions that held local election saw higher turnout in EP elections than regions that did not.
Turnout in the 2014 EP election in Germany varied greatly from state to state, and from region to region (see Tab. and Fig. ). The five regions that saw the lowest turnout are all in Bavaria, a state that just recently held local elections. The five regions exhibiting the highest turnout rates are in Rhineland-Palatinate and Saarland, two states that held regional elections. It is striking that nearly all states that saw local elections also saw higher turnout in EP elections than other states. Note that Berlin, which has the highest turnout of all states that did not hold local elections, held a statewide referendum on the day of the EP elections. However, for the subsequent analysis Berlin will be coded as not having seen regional elections. The results presented hereafter therefore represent conservative estimates of the effect of local elections on turnout in EP elections.
A simple comparison of turnout rates between states with and without local elections reveals a statistically significant difference in turnout of 5.8%-points. Average turnout in states where no local elections were held was 43.8% and 49.6% in states where local elections where conducted. (For this comparison Lower Saxony was coded as not having held regional elections because regional elections were limited to a subset of regions in Lower Saxony) Turning to a comparison of regions I find an even more striking difference of 8.9%-points, with local election regions seeing an average turnout of 50.9%, compared to 41.9% for those regions that did not hold local elections.
I subsequently estimate random intercept models regressing turnout on a local elections dummy, one-by-one adding macro-level determinants of turnout readily available from the aforementioned Wahlstrukturdaten (population density, the proportion of people receiving unemployment benefits, local business tax revenues as a proxy for wealth of a region and the proportion of high school graduates in the region). The coefficient on local elections remains relatively unaffected and significant at the five percent level by the inclusion of covariates. The results suggest that, ceteris paribus, turnout in EP elections is about 3.3%-points higher when local election are held in conjunction with the EP election.
In Lower Saxony only 17 of 46 regions held direct elections for mayor or head of the district authority this provides an opportunity to corroborate the results obtained from the national sample. I now estimate garden variety OLS regressions, again one-by-one adding above mentioned covariates. The coefficient on local elections again remains substantially unchanged indicating an effect of about 2.6%-points.
A counterfactual turnout rate
Having estimated the effect of local elections on regional turnout in EP elections it is now straightforward to calculate counterfactual turnout rates that would have obtained if no local election would have been held. To do so I subtract the average effect of 3.2%-points, obtained from the prior analysis, from all regional turnout rates. Using these new turnout rates I calculate the absolute number of voters that one would expect in the absence of local elections. Aggregating absolute turnout numbers to the state and national level I am able to obtain state and federal level counterfactual turnout rates (see Fig. ). Aggregating to the national level the counterfactual national turnout rate in the absence of local elections is 46% – the actual turnout was 48.1%.
Did turnout increase because of local elections?
Still, knowing that local elections increased turnout in EP elections does not demonstrate that local elections are a source of the increase in turnout in EP elections from 2009 to 2014.One should note that only three states, Brandenburg, Hamburg and North Rhine Westphalia, that held local elections in 2014 did not hold elections on EP election day in 2009.
A proper test of the hypothesis that local elections increased turnout in 2014 vis-à-vis 2009 would require panel data, however one can still get a first sense of whether there really is something to investigate more closely using turnout in the 2009 election. Turnout in the last election is reported alongside turnout in 2014 in the official election returns. Past turnout is a strong predictor of current turnout. Introducing past turnout into the regression effectively focuses the other covariates on explaining the change from the last to the current election.
Including past turnout (2009) in regressions on 2014 turnout decreases the coefficient on local elections slightly but it remains statistically significant, and continues to be significant and largely unchanged even when controlling for further determinants of turnout. The results suggest that the additional holding of local elections in Brandenburg, North Rhine-Westphalia and Hamburg could account for up to 3/5 of the increase in turnout from 2009 to 2013. The counterfactual turnout rate reported in section 3 suggests a more moderate effect.
Concluding remarks
A quick analysis of the preliminary regional election returns suggests that regions that held local elections in parallel to the EP elections would have seen about 3%-points less in turnout had they not held local elections. The national turnout rate would have been 2.1%-points lower. Analysis of the data also shows that the popular hypothesis that local increased turnout in 2014 vis-à-vis 2009 cannot be quickly dismissed. This is politically significant as low turnout seems to indeed benefit fringe parties like the Eurosceptic Alternative für Deutschland as Steffen Zittlau shows in an analysis on this blog.
Ich sehe das wieder einmal als ein Beispiel dafür, daß „wissenschaftlich“ empirisch bewiesen wird, was man sowieso weiß.
Außerdem wird im Englischen zwischen „hot“ und „not“ unterschieden – Englischkompetenz anscheinend auch nicht so toll … (siehe letzten Absatz).
Mit besten Grüßen
Fritz Neubauer
Danke für den Hinweis auf den Schreibfehler. Diesen haben wir korrigiert.
The previous comment asked whether this is just „scientifically“ (quotation marks used by prior commenter) and empirically proving what we already know. Not surprisingly, I disagree. Here’s why:
1. I think what the comment is saying is that it is intuitive to think that simultaneously held elections raise turnout vis-à-vis single elections. Even if that is the case, intuition is nice and well but we need empirical studies to confirm or disapprove our intuitions which obviously can be wrong sometimes.
2. I don’t think it is obvious that holding one what we call a second order election in political science, an election of lesser importance to voters, in conjunction with another second order election raises turnout. Hence the need for an empirical test.
3. Obviously, turnout in states and regions with local elections was on average higher than in states and regions without such simulaneously held elections. A ’scientific‘ vs. such a ’naive‘ comparison tries to ascertain whether the observed difference can credibly be ascribed to local elections and not some other factors. The scientific approach here lies in using statistical models to answer that question, including factors in the analysis that political science theory tells us are determinants of turnout.
Lastly, a minor note: I’m not proving anything here. Proofs exists in mathematical models but empirical findings are not proofs, merely evidence. In this case I think I have quite credible evidence for a causal effect of local elections.