Natural Disasters and Political Participation: Evidence from the 2002 and 2013 Floods in Germany

Abstract

How do natural disasters affect electoral participation? The existing social science literature offers contradictory predictions. A considerable body of research in sociology and psychology suggests that traumatic events can inspire pro-social behaviour, which might increase turnout. Yet, political science has long held that even minor changes to participation costs of low benefit activities can lead to considerable drops in civic engagement. Consequently, natural disasters should reduce electoral participation. We show how these distinct views can be jointly analysed within the Riker–Ordeshook model of voting. This paper then reports results on the impact of the 2002 and 2013 floods in Germany on turnout in federal and state elections in Saxony and Bavaria, conducted few weeks after the floods. Analysing community level turnout data, and drawing on a difference-in-differences framework, we find that flood exposure has a consistent negative effect on turnout. This indicates that the increase in the costs of voting outweighed any increase in political engagement in our case and stands in contrast to findings from developing contexts, where flood management was convincingly linked to electoral participation.

Acknowledgments

For helpful comments and feedback we want to thank the editor and an anonymous referee, Tadeusz Kugler and the panel participants at the 2015 Annual Meeting of the European Political Science Association (EPSA) as well as seminar participants and colleagues at LMU Munich and Durham University. We thank Joerg Spenkuch for the kind provision of election data. Many thanks as well to Kathrin Fischer from the Saxonian State Agency for Environment, Agriculture and Geology for her support on Saxony flood layers and to Heike Bach and Martina Hodrius from Vista Remote Sensing in Geosciences GmbH for the provision of Bavaria flood layers and help in pre-processing the GIS files.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental Data and Research Materials

The supplemental Online Appendix for this article can be accessed on the Taylor & Francis website, 10.1080/09644008.2017.1287900. The replication files for the analysis in this paper and the Online Appendix can be accessed at the Harvard Dataverse under doi:10.7910/DVN/X3VUSW.

About the Authors

Lukas Rudolph is a PhD Candidate at the Geschwister Scholl Institute, Ludwig-Maximilians-Universität München (LMU Munich), Oettingenstrasse 67, 80538 Munich, Germany.

Patrick M. Kuhn is a Lecturer in Comparative Politics at the School of Government and International Affairs (SGIA), Durham University, The Al-Qasimi Building, Elvet Hill Road, Durham, DH1 3TU, United Kingdom.

ORCID

L. Rudolph http://orcid.org/0000-0001-8977-4480

Notes

1 For 2002, Bechtel and Hainmueller (2011) report an increase of around seven percentage points in the incumbent SPD's second vote share in affected electoral districts along the Elbe, linked to effective flood management. It was generally noted that the government's flood response played an important role in the re-election of Gerhard Schröder as chancellor (see e.g. Pappi, Shikano, and Bytzek 2004).

2 For a summary see Thurner (1998).

3 For similar arguments regarding cold weather and rain see Shachar and Nalebuff (1999) and Fujiwara, Meng, and Vogl (2016).

4 Others have even argued for a non-linear theoretical set-up (Blais and Achen 2010). From this viewpoint a low level of D is a prerequisite for C to matter. With low duty levels an increase in costs will depress the individual turnout probability; with high duty levels, cost increases do not lead citizens to abstain. Whether variations in cost/duty has a linear or non-linear effect could be tested exceptionally well with variation in flood exposure and individual level panel data, as usually D cannot be varied (quasi-)experimentally. The Blais–Achen model would imply that the increase in costs leads those individuals that have low ex-ante levels of duty to abstain; the increase in duty changes the turnout likelihood of individuals disproportionately, and can overcompensate a similar increase in costs. This is a fruitful avenue for future research. However, in our case with aggregate level data, we can only look at aggregate changes. The Blais–Achen model would lead to different aggregate predictions, though. Only if would we expect a negative flood turnout effect; both D=C and D>C lead to a positive flood turnout effect.

5 The larger distance between flood and election in 2013 especially allowed for disaster relief to be more effectively distributed and reconstruction to begin to a larger extent in the Bavarian case. Given our theoretical arguments, this could potentially mitigate the effect of the floods. We do not expect large temporal effects, however. First, empirically even a three month period is very short compared to existing research in this area. For example, Cole, Healy, and Werker (2012) estimated for Indian voters a cut-off of a one-year time period when disasters do no longer impact aggregate electoral outcomes significantly. Similarly, Lazarev et al. (2014) noted increased government support one year after forest fires in Russia. Bechtel and Hainmueller (2011) and Eriksson (2016) even argued for a persistent influence of natural disasters on vote choice over several electoral cycles. But second, even if we empirically expect an especially high impact of the disaster just before election day (as, e.g., argued by Chen 2013), this should make it harder for us to find any effects (especially in the Bavaria case), as we would expect time to off-set both any negative costs and positive duty effects on turnout. We therefore do not consider the potentially differential implementation of flood relieve policy in the following, other than noting that effects in Bavaria might already be dampened to an unknown extent relative to effects in Saxony.

6 In Saxony 2002, €500 per affected person, maximum €2,000 per household, were handed out to all households that applied (restriction: household income < €40,000); in Bavaria 2013, this ‘instant flood support’ amounted to €1,500 per household without any income restriction (BMI 2013b).

7 We looked at the responses of Politbarometer survey respondents to the question ‘What is according to your opinion currently the most important problem in Germany?’ right before the election. The top five topics mentioned (reported in Appendix Figure 1) were mainly economic and/or social security related, with unemployment the most frequently raised topic in both Saxony (65 per cent) and Bavaria (13 per cent). While community-level economic factors explicitly enter our regressions as control variable, we do not expect that other frequently mentioned topics such as unification or the Euro crisis/Grexit debate have a differential impact within Bavaria/Saxony, as they are not geographically focused topics. Similarly, important last minute campaign topics mentioned in the literature, such as the Iraq war (discussed, e.g., in Bechtel and Hainmueller 2011) or a 2013 public scandal in Bavaria (discussed, e.g., in Rudolph and Däubler 2016) are not correlated with flood exposure.

8 For the floods in Bavaria 2002 and Saxony 2013 there currently are no similarly high-quality flood layers.

9 Actually, with our research design we can only estimate compound treatment effects, that is, the effects of flood exposure and subsequent disaster relief/reconstruction policy. This is a problem barely addressed in the literature on disaster effects. To estimate both effects independently, we would need regions exogenously exposed to a disaster and at the same time exogenous variation in relief efforts. This is an empirical challenge, as relief is nearly always a function of exposure. We therefore have to define our treatment as combined disaster exposure and relief effort. Theoretically, however, as the government response was publicly judged to have been very effective, we deem it plausible in our case that both the shock to the C term and to the D term in our model are attenuated towards zero on aggregate, attenuating any estimation of flood treatment effects. For details on compound treatments, see, for example, Hernán and VanderWeele (2011); Keele and Titiunik (2015).

10 By identifying treatment effects within Saxony and within south-eastern Bavaria we as well sidestep confounding that might arise from the differential party systems in Bavaria and Eastern Germany – the latter, for example being referred to as ‘“different world” in electoral terms’ (Saalfeld 2004, 197).

11 State elections generally experience lower average turnout rates, although we can expect that voter decision making for these elections, especially if so close together temporally, is not independent from one another (Bechtel 2012).

12 For Bavaria, as our placebo tests show that the parallel trend assumption likely holds, we also regressed the continuous indicator and its square on turnout. Although there is, as expected, a negative relationship between the flooded area and turnout, there does not seem to be a strong non-linear relationship. The squared flood indicator showed no significant effect on turnout.

13 Note that alternative explanations, especially the spatial distribution of the ethnic composition of communities as a confounder, have been suggested for the u-shape in the results of Sinclair, Hall, and Alvarez (2011) (Vanderleeuw, Liu, and Williams 2008).

14 This has been similarly highlighted in other post-disaster contexts, see, for example, Stein (2015).

Additional information

Funding

Lukas Rudolph acknowledges financial support by a research scholarship of the German National Academic Foundation [Studienstiftung des deutschen Volkes].