- Email: [email protected]
Measuring impacts of extreme weather events using the life satisfaction approach
Ecological Economics 121 (2016) 108–116
Contents lists available at ScienceDirect
Ecological Economics journal homepage: www.elsevier.com/locate/ecolecon
Analysis
Measuring impacts of extreme weather events using the life satisfaction approach Charlotte von Möllendorff a,⁎, Jesko Hirschfeld b a b
Carl von Ossietzky University Oldenburg, Department of Economics, 26111 Oldenburg, Germany Institute for Ecological Economy Research (IÖW), Potsdamer Str. 105, 10785, Berlin, Germany
a r t i c l e
i n f o
Article history: Received 8 May 2015 Received in revised form 23 October 2015 Accepted 15 November 2015 Available online xxxx JEL classification: D61 I31 Q51 Q54
a b s t r a c t Extreme weather events cause harm among the aggrieved party that often goes beyond material damages. This paper studies the impact of extreme weather events on measures of self-reported life satisfaction. Focusing on Germany, we use representative panel data for 2000–2011 to study the effect of seven storm & hail events and five floods on subjective well-being in the affected NUTS 3 regions. Our results indicate that both weather experiences bear statistically significant negative externalities. Following an extreme weather event, life satisfaction is reduced by 0.020–0.027 on the 11-point scale. While the effect of storm & hail events is rather immediate in nature, the effect from floods persists much longer. © 2015 Elsevier B.V. All rights reserved.
Keywords: Extreme weather events Subjective well-being Life satisfaction Nonmarket valuation
1. Introduction In the past, we observed some destructive storms and floods that severely impacted on the population living in affected areas. With climate change, the magnitude and frequency of extreme weather events are expected to increase even further (Ciscar et al., 2009). The valuation of impacts imposed by extreme weather events usually focuses on economic aspects, such as damages on buildings, items and infrastructures and thereby neglects immaterial values such as mental distress, worries, health injuries or the loss of personal belongings (see e.g. Dehnhardt et al., 2008, Tapsell et al., 2002). In this paper we study the impact of extreme weather events in Germany using evidence from subjective well-being data. Life satisfaction analysis has been increasingly used to evaluate environmental attributes and non-marketed goods (see e.g. Welsch and Ferreira, 2013 or Ferreira and Moro, 2010). We analyse how life satisfaction changes in affected regions due to the occurrence of an extreme weather event using panel data from the German Socio-Economic Panel Study (SOEP, see Schupp et al., 2014). Using panel data permits to control for unobserved interpersonal ⁎ Corresponding author. E-mail addresses: [email protected] (C. von Möllendorff), [email protected] (J. Hirschfeld).
http://dx.doi.org/10.1016/j.ecolecon.2015.11.013 0921-8009/© 2015 Elsevier B.V. All rights reserved.
heterogeneity. The importance of controlling for individual fixed effects in life satisfaction analysis has been emphasized by Ferrer-i Carbonell and Frijters, (2004). Our paper ties in with a small number of studies that analyse well-being effects of climate variables (Maddison and Rehdanz, 2011) and weather events like floods (Luechinger and Raschky, 2009), droughts (Carroll et al., 2009), hurricanes (Kimball et al., 2006) and forest fires (Kountouris and Remoundou, 2011). Luechinger and Raschky (2009) study the well-being effect of floods in Europe between 1973 and 2004 on NUTS 2 level and find a sizeable negative effect on life satisfaction. In a more recent analysis, Osberghaus and Kühling (2014) study indirect and direct effects of weather experiences in Germany – namely storms, floods, heavy rain and heat waves – on subjective well-being using a specifically designed and conducted one-time survey. They find a significant negative effect of climate-change induced damage expectations on subjective wellbeing while the direct effect is only significant in the case of heat waves. In our analysis we focus on Germany using data from the SOEP for 2000–2011 which allows us to control for interpersonal heterogeneity and relevant socio-economic characteristics. We further use spatially disaggregated data (NUTS 3 level) from the German Insurance Association (GDV) on insured losses of five floods and seven storm & hail events. Hence, the main contribution of our paper is to study the well-being effects of different types of extreme weather events (floods and storm & hail) on a disaggregated level (NUTS 3) over a period of
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
12 years (2000–2011) using the rich dataset of the SOEP. Unlike earlier studies, we consider regional data on insurance density, distinguish different levels of regional impacts, analyse households in rented/owned property separately and control for movers/stayers. Our results indicate a significant negative effect of storm & hail events as well as floods on subjective well-being in affected regions, decreasing life satisfaction by about 0.02–0.027 on the 11-point scale. While the effect of storm & hail events dissipates after 6 months, floods affect life satisfaction much longer. Moreover, we find that the effect is particularly adhering to house owners and is lower in areas with high insurance rates. The following sections are structured as follows: Section 2 presents the data and Section 3 describes the econometric approach. Section 4 reports and discusses the results while Section 5 concludes.
2. Data The life satisfaction data along with the socio-economic control variables were made available by the German Socio-Economic Panel Study (SOEP) of the German Institute for Economic Research (DIW), Berlin. Since 1984, the SOEP conducts annual interviews surveying the socioeconomic situation of German households. In each annual wave approximately 20,000 persons living in 11,000 households are surveyed (see Wagner et al., 2007). As for the panel structure of the survey study, the same set of respondents is reinterviewed each year. This facilitates to make longitudinal analyses of changes on the individual level and to control for unobserved time-invariant characteristics (Andreß et al., 2013). Especially by focusing on a longer time frame, the SOEP study faces some temporary or permanent drop-outs which are offset by including new respondents, i.e. the data is unbalanced. Our analysis is conducted based on an unbalanced subset of the SOEP, which is composed of 239,209 observations from 39,679 respondents that were interviewed during 2000 and 2011. The spatial level of the analysis refers to the NUTS 3 regions of Germany which coincide with the level of Landkreise/Kreise and kreisfreie Städte.1 There are 402 NUTS 3 regions in Germany with an average size of 888 km 2. The life satisfaction data of the SOEP results from responses to the following question: “How satisfied are you with your life, all things considered?”. The question can be answered on a scale from 0 (completely dissatisfied) to 10 (completely satisfied). In addition, we considered time-varying socio-economic control variables on the individual level, that were found to cause changes in individual well-being (see e.g. Frijters et al., 2004).2 With regard to extreme weather events we created a dataset that includes floods and storm & hail events that occurred in Germany during 1999 and 2010, with the most severe being the Elbe flood in 2002 and the storm Kyrill in 2007. The events are summarized in Table 1. The selection of events is based on their intensity which is approximated by the claims expenditure they caused for insurances. In our analysis we considered five floods that caused claims expenditures higher than 55 million and seven storm & hail events that caused claims expenditures higher than 230 million. The great difference in claims expenditure between both types of events is related to the insurance density. In Germany more than 90% of all buildings are insured against storm and hail events, while only 30% of all buildings have a natural hazard insurance, which covers, inter alia, damages due to floods (GDV, 2012). So the fraction of damages rendered by the insurances is higher for storm & hail events compared to floods. Furthermore, there is regional variation in insurance densities with regard to natural hazard insurances (see e.g. Kreibich et al., 2011 or Seifert et al., 2013). For this reason, we used data on regional insurance densities, which represent for each
1
NUTS stands for Nomenclature of Units for Territorial Statistics and is a geocode standard for referencing counties and regions in the European Union. 2 See Appendix for summary statistics.
109
federal state the share of insured buildings in the highest risk zone (GDV, 2013). Though we are analysing the period from 2000 to 2011 the storm Lothar that occurred in December 1999 is included in the analysis so as to study possible effects on self-reported life satisfaction in the interviews conducted in 2000. In the same line of reasoning 2011 is included in the analysis to track potential effects of events that occurred in 2010 on stated well-being in 2011. The thunderstorm Hilal is listed twice because some of the damages were covered by the natural hazard insurance, while others were covered by the building insurance. The data we use refers to damages from floods and storm & hail events collected and provided by the German Insurance Association (Gesamtverband der deutschen Versicherungswirtschaft, GDV), Berlin. Storm & hail events are usually insured via building insurances while floods (due to extreme rainfall or overflowing) come under the natural hazard insurance. We used damage data of both insurance types to obtain information on storm & hail events as well as floods in Germany.3 For each considered event the data was made available in the form of high-resolution maps showing the classified damage frequencies per NUTS 3 region (see Fig. 1 for corresponding maps of the Elbe flood, 2002 and storm Kyrill, 2007). Damage frequency (DF) is defined as follows (GDV, 2012:43): Damage frequency ðDFÞ ¼
number of claims number of contracts
ð1Þ
Thus, the damage frequency describes the number of contracts that were deployed to claim from the insurer in relation to the total number of running contracts per NUTS 3 region. By this, the variable implicitly controls for regional variation in insurance density. As the borders of the NUTS 3 regions are displayed in the maps, the information on damage frequency per NUTS 3 region and event could be extracted using a Geographic Information System (GIS). The damage frequencies of the NUTS 3 regions are displayed in intervals which correspond to the incidence rates of the events ranging from On one day there occurred as many damages as usually occur within 1 week (storm & hail)/1 month (floods) to On one day there occurred as many damages as usually occur within 1 year (storm & hail)/12 years (floods). To illustrate, consider the NUTS 3 region in Fig. 1(a) coloured in black which was most severely affected by the Elbe flood in 2002 (Sächsische Schweiz-Osterzgebirge): It exhibits a damage frequency higher than 200 which means that over 200 out of 1000 households insured against natural hazard events claimed from the insurer, an event whose corresponding incidence rate is 12 years. Taking a look at storm Kyrill, the map shows that most of the NUTS 3 regions (129) had a damage frequency between 38.7 and 77.4 meaning that in those regions there occurred more damages on one day than usually occur within six months, or rather the incidence rate is 6 months (see Table 2). This exemplifies that the relation between damage frequency and incidence rate differs for floods and storm & hail events, respectively. So in order to combine the data on floods and storm & hail events and to get down to a common definition of extreme weather events we revert to the incidence rate. Based on this, we define an extreme event as follows: On one day there occurred more damages as usually occur within a month. This means that for each event we considered respondents as affected if they lived in a NUTS 3 region with a damage frequency higher than 0.4 in case of floods, and 6.4 in case of storm & hail events (see Table 2). This is the lowest threshold value commonly available for both types of events. If we increase the threshold level, we are likely to get down to the more severely impacted individuals which may 3 For simplicity we refer to flood events throughout the paper. Strictly speaking the data describes all other natural hazard damages as well (backwater, earthquakes, land subsidence, landslide, snowslides, snow pressure, and volcanic eruption) which are generally, however, of minor relevance in Germany (GDV, 2012).
110
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
Table 1 Extreme weather events in Germany with insured losses, 1999–2011 (following (GDV, 2012):10ff/23ff). Year
Name
Event type
Date
Average claims amount
Total claims expenditure (in million)
1999 2002 2002 2006 2007 2008 2008 2008 2008 2009 2010 2010
Lothar Elbe flood Jeanett Queeny Kyrill Emma Hilal Hilal Zsuzanna Rainer Xynthia Elbe, Neiße, Spree
Storm & hail Flood Storm & hail Storm & hail Storm & hail Storm & hail Flood Storm & hail Flood Flood Storm & hail Flood
25/12–26/12 31/07–02/09 27/10–28/10 28/06 18/01–19/01 29/02–01/03 29/05–02/06 30/05 26/07–04/08 24/06–03/07 28/02–01/03 01/08–13/08
€1359 €17,752 €783 €12,749 €1021 €881 €5805 €2151 €5639 €4998 €821 €10,846
€800 €1800 €760 €230 €2060 €390 €100 €330 €55 €85 €367 €265
Fig. 1. Damage frequencies of the Elbe Flood (2002) and Storm Kyrill (2007) (source: GDV, 2012:13/23).
have experienced higher well-being losses. However, increasing the threshold level reduces the number of respondents who have experienced a storm & hail or a flood event during the past 6 months from 67,395 (threshold: “1 month”) to 40,811 (threshold: “3 months”), 23,675 (threshold: “6 months”) or 12,782 (threshold: “1 year”). With increasing threshold levels the results become less significant. Thus, we face a trade-off between the severity on the one hand and the number of cases on the other hand. Our event variables are specified as the number of storm & hail/ flood/extreme weather events, respectively, that occurred in the respondents NUTS 3 region within a given time frame (see Section 3 for a discussion of the different model variations). Thereby, extreme weather events is simply defined as the sum of storm & hail and flood events.4 4
See Appendix for summary statistics.
Furthermore, we tested the untransformed frequency metric as a continuous variable which entails information on regional variation in impact levels. Both, the SOEP data and the data on extreme weather events are merged on the basis of NUTS 3 regions. Our dataset poses several problems. First, we cannot distinguish between those individuals who are actually affected and those who are not because we revert to administrative borders. It is very difficult to circumvent this problem. Second, we cannot distinguish between individuals who are insured and those who are not. This applies especially to floods as we have an insurance density of only 30% for natural hazard events. For storm & hail events this problem is not as severe because here we have an insurance density of 90% and can assume that most people have an insurance that covers related damages. Third, we possibly have a selection bias in the sense that individuals more exposed to
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116 Table 2 Incidence rate of extreme weather events (following GDV, 2012:42). Flood events
Storm and hail events
On one day there occurred as many damages as usually occur within
Lower bound of damage frequency (in ‰)
On one day there occurred as many damages as usually occur within
Lower bound of damage frequency (in ‰)
1 month 3 months 6 months 1 year 3 years 6 years 12 years
0.4 1.2 2.4 4.8 14.5 29.0 58.0
1 week 2 weeks 1 month 3 months 6 months 1 year
1.5 3.0 6.4 19.3 38.7 77.4
risks from flooding are rather insured against natural hazards while there is also a small fraction of individuals living in high risk areas which are rejected by the insurances. There are other disparities in insurance policies that might as well reinforce different insurance densities across regions. Still, we assume that with an average insurance density of 30% most NUTS 3 regions are represented in the dataset and regional disparities are levelled out by dividing by the number of running contracts. 3. Econometric framework By conducting a longitudinal analysis, we study in a fixed effects modelling framework whether individual well-being changes due to the occurrence of an extreme weather event. An advantage of the fixed effects model is that it facilitates to capture unobserved factors like personality characteristics that were found to have an important influence on stated subjective well-being (Diener and Lucas, 1999). Thereby the risk of omitted variable bias is reduced though a bias can still arise due to the neglect of unobserved time-varying variables. On account of the fixed effects modelling framework time-invariant covariates commonly considered in life satisfaction analyses (such as gender, migration background, highest level of education) cannot explicitly be incorporated in the model, though they are implicitly controlled for via the individual fixed effect. Running a Hausman test confirms that a fixed-effects model is appropriate because unobserved personal heterogeneity cannot be considered as uncorrelated with the explanatory variable as is assumed in random effects models. Formally the estimating model can be stated as follows: LSijt ¼ aQ jt þ β ln Mijt þ D0 ijt y þ ηi þ ηt þ εijt :
ð2Þ
LSijt indicates the life satisfaction of respondent i in NUTS 3 region j in year t. Qjt is a metric variable describing the number of extreme weather events and will (i) separately capture the two considered event types (floods and storm & hail) or (ii) represent alternatively an overall term for extreme weather events, as discussed above. Mijt specifies the net monthly household income which has been equivalized according to the OECD-modified scale and is included in logarithmic form in order to take account of its decreasing marginal utility. D'ijt describes time-varying covariates at the micro-level which have been found to influence life satisfaction, namely household size, age squared, state of health, partner status, employment status and whether there is a person in the household in need of care. The variable ηi describes the individual fixed effect. Time-specific effects are captured by the year dummies ηt, which are included to account for unobserved factors pertaining to a specific year. Measurement errors are described by εijt. Age is included as a quadratic term to control for the proposed Ushaped relationship between life satisfaction and age. The linear term of age cannot be incorporated, because it correlates with the year dummies. Macro-level variables are left out of the analysis too: The regional GDP per capita was tested but its effects are insignificant; the
111
regional unemployment rate had a marginal effect on life satisfaction but was omitted because no data was available for 2010 and 2011. Moreover, population density was included in order to take account of possible migration following an event but was not found significant. We estimated Eq. (2) applying a linear fixed effects estimator. Despite the ordinal nature of the dependent variable LSijt that suggests ordinal regression techniques, previous studies confirmed that the differences in the results are negligible (see e.g. Ferrer-i Carbonell and Frijters, 2004). Furthermore, there is no consensus on how to appropriately implement individual fixed effects in ordered regression models (Baetschmann et al., 2014). Standard errors are adjusted for clustering on the individual level, which is necessary because the data contains multiple observations from each respondent (Moulton, 1990). Additionally, we tested the results with regard to clustering on the regional level. We estimated several different models. Models I to III refer to flood events, Models IV and V refer to storm & hail events and Model VI uses a common variable of extreme weather events. In Model I floods are captured as a metric variable describing the number of events, Model II uses a dummy variable in interaction with the regional insurance density, while Model III introduces a continuous variable of the regional damage frequency. Similarly, Model IV includes a variable on the number of storm & hail events, while Model V estimates a continuous variable of the regional damage frequency. Model VI summarizes the number of floods and storm & hail events using a metric scale. In order to survey the robustness of the results we included some further variations. Model VII takes a macro approach and studies changes in life satisfaction following flood events on NUTS 3 level. Model VIII excludes migrants while subsequently in Model IX standard errors are adjusted for clustering on NUTS 3 level. Finally, Model X uses interaction terms in order to analyse differences in the effects of flood events for house owners and renters, respectively. Within all models, several different specifications were tested which revert to the elapsed time between the event and the interview. As the exact dates of the interviews and the events are known, the time-lag can be calculated.5 In this analysis we test three different time-frames: First, events that date back less than 6 months; second, events that date back less than 12 months; and third, events that date back less than 18 months. Expanding the time frame instead of using disjoint intervals conduces to gradually increase the number of observations with respect to people that experienced an extreme weather event. As a result, considering different time frames caters to the question how long the effect of extreme weather events on subjective well-being prevails but has also implications for the number of cases and thus for the significance of the results as will be discussed below.
4. Results 4.1. Estimation results In the following, we present the results for the different model specifications as defined in the previous section. Table 3 displays the corresponding outputs for flood events (Model I–III) and Table 4 for storm & hail events (Model IV and V) as well as for a common term of extreme events (Model VI). When comparing the different columns, it becomes obvious that the socio-economic covariates are stable with regard to the various model specifications. In line with other findings, the results reveal a highly significant positive effect of equivalized income on life satisfaction. Concerning the effect of partner status on life satisfaction, the results indicate that being in a partnership is favoured above being single, 5 If the event ranged over a longer period of time, the last day of the event was chosen as a reference. Taking the first day or the midpoint of the event as reference instead produces largely the same results.
112
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
Table 3 Estimation results for flood events (source: SOEPv28, author's calculations). Variables Flood variables Number of events
Model I (a) 6 months (b) 12 months (c) 18 months −0.0112 (0.0165)
−0.0210⁎⁎ (0.00925)
−0.0230⁎⁎⁎ (0.00836)
Dummy Insurance density Dummy × insurance density
Model II (a) 6 months (b) 12 months (c) 18 months
−0.0561⁎⁎ (0.0266) −0.0844 (0.0920) 0.111⁎⁎ (0.0530)
−0.0226 (0.0148) −0.0821 (0.0920) 0.00702 (0.0291)
−0.0295⁎⁎ (0.0138) −0.0840 (0.0919) 0.0209 (0.0270)
0.280⁎⁎⁎
0.280⁎⁎⁎
Damage frequency Log. Equiv. Household income Household size Age squared Person needing care in HH Health status Good Satisfactory Poor Bad Employment status Not employed Unemployed Pensioner Military, community service In education Self-employed Partner status Partner outside household Partner inside household Year dummies Individual fixed effects R 2(within) Number of individuals Number of observations
0.280⁎⁎⁎ (0.0137) 0.0114⁎
0.280⁎⁎⁎
0.280⁎⁎⁎
(0.0137) 0.0115⁎
(0.0137) 0.0115⁎
0.280⁎⁎⁎ (0.0137) 0.0115⁎
(0.0137) 0.0115⁎
Model III (a) 6 months (b) 12 months (c) 18 months
0.000169 (0.000295) 0.280⁎⁎⁎ (0.0137) 0.0116⁎
0.000138 (0.000244) 0.280⁎⁎⁎ (0.0137) 0.0115⁎
0.000285 (0.000208) 0.280⁎⁎⁎ (0.0137) 0.0115⁎
(0.00632) (0.00632) (0.00632) 0.0000 0.0000 0.0000 (0.00004) (0.00004) (0.00004) −0.409⁎⁎⁎ −0.409⁎⁎⁎ −0.409⁎⁎⁎ (0.0310) (0.0310) (0.0310) Reference group: very good −0.317⁎⁎⁎ −0.317⁎⁎⁎ −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0137) 0.0115⁎ (0.00632 (0.00632) (0.00632) 0.0000 0.0000 0.0000 (0.00004) (0.00004) (0.00004) ⁎⁎⁎ ⁎⁎⁎ −0.409 −0.409 −0.409⁎⁎⁎ (0.0310) (0.0310) (0.0310)) Reference group: very good −0.317⁎⁎⁎ −0.317⁎⁎⁎ −0.317⁎⁎ (0.0117) (0.0117) (0.0117) ⁎ −0.722⁎⁎⁎ −0.723⁎⁎⁎ −0.723⁎⁎⁎
(0.00632) (0.00632) (0.00632) 0.0000 0.0000 0.0000 (0.00004) (0.00004) (0.00004) ⁎⁎⁎ ⁎⁎⁎ −0.409 −0.409 −0.409⁎⁎⁎ (0.0310) (0.0310) (0.0310) Reference group: very good −0.317⁎⁎⁎ −0.317⁎⁎⁎ −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.265⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0326) (0.0326) (0.0326) Reference group: employed −0.0576⁎⁎⁎ −0.0577⁎⁎⁎ −0.0576⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎ (0.0204) (0.0204) (0.0204) 0.0400⁎ 0.0401⁎ 0.0401⁎
(0.0326) (0.0326) (0.0326) Reference group: employed −0.0575⁎⁎⁎ −0.0576⁎⁎⁎ −0.0576⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎ (0.0204) (0.0204) (0.0204) 0.0401⁎ 0.0401⁎ 0.0400⁎
(0.0326) (0.0326) (0.0326) Reference group: employed −0.0575⁎⁎⁎ −0.0575⁎⁎⁎ −0.0575⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎ (0.0204) (0.0204) (0.0204) 0.0400⁎ 0.0400⁎ 0.0400⁎
(0.0206) (0.0206) (0.0206) 0.0538 0.0540 0.0543 (0.0505) (0.0505) (0.0505) 0.133⁎⁎⁎ 0.133⁎⁎⁎ 0.133⁎⁎⁎ (0.0205) (0.0205) (0.0205) 0.0252 0.0250 0.0251 (0.0275) (0.0275) (0.0275) Reference group: no partner 0.302⁎⁎⁎ 0.302⁎⁎⁎ 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) (0.0219) (0.0219) Yes Yes Yes Yes Yes Yes 0.0982 0.0982 0.0982 39,679 39,679 39,679 239,209 239,209 239,209
(0.0206) (0.0206) (0.0206) 0.0541 0.0540 0.0542 (0.0505) (0.0505) (0.0505) 0.133⁎⁎⁎ 0.133⁎⁎⁎ 0.133⁎⁎⁎ (0.0205) (0.0205) (0.0205) 0.0250 0.0249 0.0250 (0.0275) (0.0275) (0.0275) Reference group: no partner 0.302⁎⁎⁎ 0.302⁎⁎⁎ 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) (0.0219) (0.0219) Yes Yes Yes Yes Yes Yes 0.0982 0.0982 0.0982 39,679 39,679 39,679 239,209 239,209 239,209
(0.0206) (0.0206) (0.0206) 0.0537 0.0536 0.0534 (0.0505) (0.0505) (0.0505) 0.133⁎⁎⁎ 0.133⁎⁎⁎ 0.133⁎⁎⁎ (0.0205) (0.0205) (0.0205) 0.0252 0.0252 0.0253 (0.0275) (0.0275) (0.0275) Reference group: no partner 0.302⁎⁎⁎ 0.302⁎⁎⁎ 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) (0.0219) (0.0219) Yes Yes Yes Yes Yes Yes 0.0982 0.0982 0.0982 39,679 39,679 39,679 239,209 239,209 239,209
Standard errors in parenthesis are adjusted for clustering at the personal level. ⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.
especially if the partner lives in the same household. Turning to the health status coefficients, a strong negative effect on life satisfaction is observed for a decrease in the state of health. When considering one's health as very good, a drop to bad health is associated with a loss in life satisfaction by 2.26 on an 11-point scale. Becoming unemployed affects life satisfaction negatively. This result is in line with the common findings in the happiness literature (see e.g. Di Tella et al., 2001). Turning to the extreme weather event variables of the baseline models (Model I and Model IV) we observe different patterns for the two types of events. Regarding floods, their negative effect on life satisfaction becomes most evident when we consider all events of the past 18 months (Model I). In contrast, storm & hail events display only a significant effect on life satisfaction if we consider a short period of the past 6 months for our analysis (Model IV). However, the effects are quite
similar in magnitude: While a flood that happened during the past 18 months reduces life satisfaction by 0.0230, a storm & hail event that happened during the past 6 months causes a drop in life satisfaction by 0.0268 on an 11-point scale (both significant at the 1%-level). This may indicate that the impact of a storm & hail event is rather immediate in nature and dissipates after a short period. Differently, the effect of flood events becomes most evident by considering a long period of 18 months. However, by interpreting these results we have to keep in mind that by considering a period of 18 months we are likely to observe more individuals that experienced a flood event, than in the shorter time-frames. To exemplify, the number of observations in NUTS 3 regions that experienced a flood during the past 6 months is only 7857 while the number of observations in NUTS 3 regions that experienced a flood during the past 12 months is 36,062. So in the short term it
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
113
Table 4 Estimation results for storm & hail events and extreme events (source: SOEPv28, author's calculations). Variables Storm & hail events Number of events
Model IV (a) 6 months (b) 12 months (c) 18 months −0.0268⁎⁎⁎ (0.00915)
−0.0109 (0.00798)
0.000763 (0.00799)
Damage frequency
Models V (a) 6 months (b) 12 months (c) 18 months
−0.000232⁎⁎⁎ (0.0000813)
−0.000219⁎⁎⁎ (0.0000808)
−0.000170⁎⁎ (0.0000690)
Extreme weather events Number of events Log. Equiv. Household income Household size Age squared Person needing care in HH Health status Good Satisfactory Poor Bad Employment status Not employed Unemployed Pensioner Military, community service In education Self-employed Partner status Partner outside household Partner inside household Year dummies Individual fixed effects R2 (within) Number of individuals Number of observations
0.280⁎⁎⁎ (0.0137) 0.0113⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
0.280⁎⁎⁎ (0.0137) 0.0114⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
0.280⁎⁎⁎ (0.0137) 0.0115⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
0.280⁎⁎⁎ (0.0137) 0.0114⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
0.280⁎⁎⁎ (0.0137) 0.0115⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
Models VI (a) 6 months (b) 12 months (c) 18 months
0.280⁎⁎⁎ (0.0137) 0.0115⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
−0.0145⁎⁎ (0.00589) 0.280⁎⁎⁎ (0.0137) 0.0115⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
−0.00948⁎ (0.00566) 0.280⁎⁎⁎ (0.0137) 0.0114⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
(0.0310) (0.0310) (0.0310) Reference group: very good ⁎⁎⁎ ⁎⁎⁎ −0.317 −0.317 −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0310) (0.0310) Reference group: very good ⁎⁎⁎ −0.317 −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0310) (0.0310) (0.0310) Reference group: very good ⁎⁎⁎ ⁎⁎⁎ −0.317 −0.317 −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0140) (0.0140) (0.0140) −1.264⁎⁎⁎ −1.264⁎⁎⁎ −1.264⁎⁎⁎ (0.0177) (0.0177) (0.0177) −2.260⁎⁎⁎ −2.260⁎⁎⁎ −2.260⁎⁎⁎ (0.0326) (0.0326) (0.0326) Reference group: employed −0.0574⁎⁎⁎ −0.0575⁎⁎⁎ −0.0576⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎ (0.0326)
(0.0140) (0.0140) −1.264⁎⁎⁎ −1.264⁎⁎⁎ (0.0177) (0.0177) −2.260⁎⁎⁎ −2.260⁎⁎⁎ (0.0326) (0.0326) Reference group: employed −0.0573⁎⁎⁎ −0.0574⁎⁎⁎ −0.0574⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎
(0.0140) (0.0140) (0.0140) −1.264⁎⁎⁎ −1.264⁎⁎⁎ −1.264⁎⁎⁎ (0.0177) (0.0177) (0.0177) −2.260⁎⁎⁎ −2.260⁎⁎⁎ −2.260⁎⁎⁎ (0.0326) (0.0326) (0.0326) Reference group: employed −0.0574⁎⁎⁎ −0.0575⁎⁎⁎ −0.0574⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0535 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0537 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0533 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0401⁎ (0.0206) 0.0537 (0.0505) 0.133⁎⁎⁎
(0.0205) (0.0205) (0.0205) 0.0257 0.0253 0.0251 (0.0275) (0.0275) (0.0275) Reference group: no partner ⁎⁎⁎ ⁎⁎⁎ 0.302 0.302 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎
(0.0205) 0.0251 (0.0275)
(0.0219) Yes Yes 0.0982 39,679 239,209
(0.0204) 0.0400⁎ (0.0206) 0.0536 (0.0505) 0.133⁎⁎⁎
(0.0219) Yes Yes 0.0982 39,679 239,209
(0.0219) Yes Yes 0.0982 39,679 239,209
(0.0310)
−0.0206⁎⁎⁎ (0.00746) 0.280⁎⁎⁎ (0.0137) 0.0113⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
−0.317⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0532 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0399⁎ (0.0206) 0.0537 (0.0505) 0.133⁎⁎⁎
(0.0205) (0.0205) 0.0251 0.0251 (0.0275) (0.0275) Reference group: no partner ⁎⁎⁎ ⁎⁎⁎ 0.302 0.302 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) (0.0219) (0.0219) Yes Yes Yes Yes Yes Yes 0.0982 0.0982 0.0982 39,679 39,679 39,679 239,209 239,209 239,209
(0.0204) 0.0401⁎ (0.0206) 0.0538 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0539 (0.0505) 0.133⁎⁎⁎
(0.0205) (0.0205) (0.0205) 0.0256 0.0253 0.0253 (0.0275) (0.0275) (0.0275) Reference group: no partner ⁎⁎⁎ ⁎⁎⁎ 0.302 0.302 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) Yes Yes 0.0982 39,679 239,209
(0.0219) Yes Yes 0.0982 39,679 239,209
(0.0219) Yes Yes 0.0982 39,679 239,209
Standard errors in parenthesis are adjusted for clustering at the personal level. ⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.
might just be the case that we do not have enough observations of affected individuals in order to detect a significant effect. Therefore, our results do not justify the conclusion that floods do not have an effect on subjective well-being in the short term. In a sensitivity analysis we widened the time frame of flood events to 24 months which produced a coefficient that was lower in magnitude but still slightly significant. Another difference between floods and storm & hail events refers to the insurance density. Although we observed in the past that flood damages are sometimes compensated by public authorities, the effect of floods on subjective well-being may be more persistent because the aggrieved party is rather in an insecure position with regard to compensation. Insurances may have mitigating effects in this regard which is why we introduced regional insurance density rates in interaction with a dummy of flood events (Model II). Now the flood variable becomes significant in the short time frame of 6 months where we also find a positive coefficient of the interaction term. Even if other compensation schemes might blur the picture, this result indicates that
insurances can at least partly offset negative well-being effects of flood events. However, this effect does not persist in the 12 and 18 months time frame. In Models III and V we introduced a continuous variable of damage frequency using the midpoints of the intervals shown in Fig. 1. The coefficient yields only significant results in case of storm & hail events. This result indicates that while well-being decreases when the area was flooded, the degree of regional detriment does not seem to matter as much. In Model VI, a common variable for extreme weather events is introduced which comprises both storm & hail events and floods. Here we find a negative effect of extreme weather events which decreases in magnitude and significance by widening the time frame. Tables 5 and 6 include some further estimations analysing the robustness of the results with respect to flood events. Model VII describes the results of a macro analysis where information on life satisfaction is
114
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
Table 5 Macro analysis (source: SOEPv28, author's calculations).
Variables Flood dummy Year dummies Region fixed Effects R2 Number of observations
Model VII (a) 6 months (b) 12 months 18 months −0.0311 (0.0655) Yes Yes
−0.0458⁎ (0.0268) Yes Yes
−0.0393 (0.0241) Yes Yes
0.602 4785
0.602 4785
0.602 4785
⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.
aggregated on NUTS 3 level and regressed on the occurrence of flood events. Regional and time fixed effects are captured by introducing dummies for each year and NUTS 3 region, respectively. The
estimation results are only weakly significant in the 12 months specification of flood events. This might indicate that levelling out all regional variation (e.g. on flood risks or insurance density) implies a loss in information. In Model VIII we excluded people from the analysis who moved across NUTS 3 borders because migration may be linked to the occurrence of extreme weather events or movers may be falsely counted as affected if they moved into a flooded regions within the 6/12/18 months following the event. This reduces our sample size from 239,209 to 211,313 observations. The results become slightly lower in magnitude and significance. This may be due to the smaller sample size or because the most aggrieved move away and thus drop out of our sample. Following, in Model IX we then clustered observations on NUTS 3 level instead of the personal level because observations from the same region cannot be considered as independent. This further reduces the significance of our results. Here, only the 18 month specification is significant on the 10%-level.
Table 6 Robustness checks (source: SOEPv28, author's calculations). Variables Flood variables Number of events
Model VIII Model IX Model X (a) 6 months (b) 12 months (c) 18 months (a) 6 months (b) 12 months (c) 18 months (a) 6 months (b) 12 months (c) 18 months 0.000360 (0.0173)
−0.0172⁎ (0.00976)
−0.0193⁎⁎ (0.00881)
0.000360 (0.0202)
−0.0172 (0.0125)
−0.0193⁎ (0.0111)
Dummy Owner Owner × dummy Log. Equiv. Household income 0.301⁎⁎⁎ (0.0155) Household size 0.0205⁎⁎⁎
0.301⁎⁎⁎ (0.0155) 0.0206⁎⁎⁎
0.301⁎⁎⁎ (0.0155) 0.0206⁎⁎⁎
0.301⁎⁎⁎ (0.0185) 0.0205⁎⁎
0.301⁎⁎⁎ (0.0185) 0.0206⁎⁎
0.301⁎⁎⁎ (0.0185) 0.0206⁎⁎
(0.00770) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.00770) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.00631) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.00631) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.00631) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.0323)
(0.0363) (0.0363) Reference group: very good −0.310⁎⁎⁎ −0.310⁎⁎⁎
Person needing care in HH
(0.00770) 0.0000 (0.00004) −0.433⁎⁎⁎
Health status Good
(0.0323) (0.0323) Reference group: very good −0.310⁎⁎⁎ −0.310⁎⁎⁎
Age squared
(0.0363)
0.00374 (0.0233) 0.0237 (0.0161) −0.0373 (0.0304) 0.262⁎⁎⁎
0.00760 (0.0144) 0.0286⁎ (0.0162) −0.0528⁎⁎⁎ (0.0172) 0.263⁎⁎⁎
−0.00362 (0.0133) 0.0277⁎ (0.0163) −0.0354⁎⁎ (0.0159) 0.263⁎⁎⁎
(0.0136) −0.0542⁎⁎⁎ (0.00707) 0.0000 (0.00004) −0.412⁎⁎⁎
(0.0136) −0.0543⁎⁎⁎ (0.00707) 0.0000 (0.00004) −0.412⁎⁎⁎
(0.0136) −0.0542⁎⁎⁎ (0.00707) 0.0000 (0.00004) −0.412⁎⁎⁎
(0.0310) (0.0310) Reference group: very good −0.317⁎⁎⁎ −0.317⁎⁎⁎
Poor
(0.0127) −0.712⁎⁎⁎ (0.0150) −1.248⁎⁎⁎
(0.0127) −0.712⁎⁎⁎ (0.0150) −1.248⁎⁎⁎
−0.310⁎⁎⁎ (0.0127) −0.712⁎⁎⁎ (0.0150) −1.248⁎⁎⁎
(0.0133) −0.712⁎⁎⁎ (0.0164) −1.248⁎⁎⁎
(0.0133) −0.712⁎⁎⁎ (0.0163) −1.248⁎⁎⁎
−0.310⁎⁎⁎ (0.0133) −0.712⁎⁎⁎ (0.0163) −1.248⁎⁎⁎
(0.0117) −0.723⁎⁎⁎ (0.0139) −1.265⁎⁎⁎
Bad
(0.0188) −2.242⁎⁎⁎
(0.0188) −2.242⁎⁎⁎
(0.0188) −2.242⁎⁎⁎
(0.0217) −2.242⁎⁎⁎
(0.0217) −2.242⁎⁎⁎
(0.0217) −2.242⁎⁎⁎
(0.0177) −2.260⁎⁎⁎
Satisfactory
(0.0117) −0.723⁎⁎⁎ (0.0139) −1.265⁎⁎⁎
(0.0310) −0.317⁎⁎⁎ (0.0117) −0.723⁎⁎⁎ (0.0139) −1.265⁎⁎⁎
Unemployed
(0.0340) (0.0340) Reference group: employed −0.0676⁎⁎⁎ −0.0677⁎⁎⁎ −0.0677⁎⁎⁎ (0.0213) (0.0213) (0.0213) −0.475⁎⁎⁎ −0.475⁎⁎⁎ −0.475⁎⁎⁎
(0.0399) (0.0399) Reference group: employed −0.0676⁎⁎⁎ −0.0677⁎⁎⁎ −0.0677⁎⁎⁎ (0.0221) (0.0221) (0.0221) −0.475⁎⁎⁎ −0.475⁎⁎⁎ −0.475⁎⁎⁎
(0.0177) (0.0177) −2.260⁎⁎⁎ −2.260⁎⁎⁎ (0.0326) (0.0326) (0.0326) Reference group: employed −0.0590⁎⁎⁎ −0.0592⁎⁎⁎ −0.0591⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.494⁎⁎⁎ −0.494⁎⁎⁎ −0.494⁎⁎⁎
Pensioner
(0.0217) 0.0441⁎⁎
(0.0219) (0.0219) 0.0441⁎ 0.0442⁎ (0.0241) (0.0241) 0.0594 0.0597 (0.0575) (0.0574) ⁎⁎⁎ 0.142 0.142⁎⁎⁎ (0.0256) (0.0255) 0.0148 0.0147 (0.0350) (0.0350) Reference group: no partner 0.294⁎⁎⁎ 0.294⁎⁎⁎ (0.0188) (0.0188) 0.442⁎⁎⁎ 0.443⁎⁎⁎ (0.0260) (0.0260) Yes Yes Yes Yes 0.100 0.100 36,382 36,382 211,313 211,313
(0.0204) (0.0204) 0.0403⁎ 0.0406⁎⁎ (0.0206) (0.0206) 0.0417 0.0415 (0.0505) (0.0506) 0.124⁎⁎⁎ 0.124⁎⁎⁎ (0.0205) (0.0205) 0.0256 0.0253 (0.0275) (0.0275) Reference group: no partner 0.301⁎⁎⁎ 0.301⁎⁎⁎ (0.0169) (0.0169) 0.417⁎⁎⁎ 0.417⁎⁎⁎ (0.0221) (0.0221) Yes Yes Yes Yes 0.0980 0.0980 39,679 39,679 239,205 239,205
(0.0340)
Employment status Not employed
Military, community service In education Self-employed Partner status Partner outside household Partner inside household Year dummies Individual fixed effects R 2(within) Number of individuals Number of observations
(0.0217) 0.0442⁎⁎ (0.0211) (0.0211) 0.0594 0.0597 (0.0572) (0.0572) ⁎⁎⁎ 0.142 0.142⁎⁎⁎ (0.0243) (0.0243) 0.0148 0.0147 (0.0306) (0.0306) Reference group: no partner 0.294⁎⁎⁎ 0.294⁎⁎⁎ (0.0193) (0.0193) 0.442⁎⁎⁎ 0.443⁎⁎⁎ (0.0262) (0.0262) Yes Yes Yes Yes 0.100 0.100 36,382 36,382 211,313 211,313
Clustered standard errors in parantheses. ⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.
(0.0217) 0.0442⁎⁎ (0.0211) 0.0600 (0.0572) 0.142⁎⁎⁎ (0.0243) 0.0148 (0.0306) 0.294⁎⁎⁎ (0.0193) 0.443⁎⁎⁎ (0.0262) Yes Yes 0.100 36,382 211,313
(0.0399)
(0.0219) 0.0442⁎ (0.0241) 0.0600 (0.0574) 0.142⁎⁎⁎ (0.0255) 0.0148 (0.0350) 0.294⁎⁎⁎ (0.0188) 0.443⁎⁎⁎ (0.0260) Yes Yes 0.100 36,382 211,313
(0.0204) 0.0404⁎⁎ (0.0206) 0.0419 (0.0505) 0.124⁎⁎⁎ (0.0205) 0.0255 (0.0275) 0.301⁎⁎⁎ (0.0169) 0.417⁎⁎⁎ (0.0221) Yes Yes 0.0980 39,679 239,205
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
In Model X we estimated an interaction effect for house owners so as to test whether they are differently affected by flood events than renters. In this model, the dummy becomes insignificant, while the interaction effect is negative and (strongly) significant in the 12 and 18 month specification. These results indicate that negative effects from floods are particularly adhering to house owners. 4.2. Discussion The results of the baseline regressions (Model I, IV and VI) indicate negative well-being externalities of extreme weather events suggesting that life satisfaction of residents living in affected NUTS 3 regions drops by about 0.0206–0.0268 on the 11-point scale following an extreme event. Though statistically significant, this value is rather low as compared to other studies (see e.g. Luechinger and Raschky, 2009). Also one might have expected a stronger effect for floods as compared to storm & hail events. There are several reasons that might explain these results. It becomes apparent from Fig. 1 that storms struck rather area-wide while impacts of floods are usually limited to particular areas. On the whole this means that we have less observations of respondents who lived in a NUTS 3 region that has been flooded as compared to respondents that lived in an area hit by a storm. This can be seen from the estimation results in Model I where the flood coefficient becomes only significant if we take into account all events of the past 12 or 18 months. In other respects it could also be the case that people living in flood prone areas adapt to flood events in the course of time so that flood impacts do not cause as much grief among the affected population any more. Last but not least, we observe a large variation of flood damages within NUTS 3 regions. This means that we are likely to have many people in the sample that were not flooded themselves but are considered as affected because they lived in a NUTS 3 region that has faced damages due to a flood. This might cause an underestimation of the true wellbeing effect of flood events. Thus, it is the limitation of our analysis that we used administrative borders in order to determine whether an individual was affected by an event, or not. A more precise distinction between the really affected and the non-affected as well as between the compensated and non-compensated respondents might reveal stronger effects than we could identify in our current study. To study mitigating effects of insurances, we incorporated federal data on natural hazard insurance densities in Model II. Insurance rates vary widely across federal states; while Baden-Württemberg has 91% of all buildings in the highest risk zone insured against natural hazard events, the corresponding value for Lower Saxony is only 7%. The results indicate that well-being losses due to extreme events are significantly reduced with higher insurance rates. As NUTS 3 regions can be affected with varying severity, we introduced damage frequency as a continuous variable in Models III and V for floods and storm & hail events, respectively. While the coefficients for storm & hail events are (strongly) significant in all variations, there is no significant effect for floods. This result indicates that the degree of regional detriment does not seem to play the major role in comparison to the question if a respondent was affected or not. Furthermore, we tested whether the intensity of an event is important in explaining wellbeing losses by including the average claims amount of each event (see Table 1) as an interaction term but could not determine any significant results either. Most likely, our database is too coarse in order to refine the analysis in this regard. In our analysis we combined data on individual and NUTS 3 level. As observations from the same individual over time or from individuals of the same NUTS 3 region can be correlated, it is necessary to calculate robust standard errors. In the baseline regressions we report standard errors that are clustered according to the individual level. As can be seen from the estimation results in Model IX, clustering on regional level reduces the significance of the results by one level. Adjusting standard errors for clustering on the regional level in the other model
115
specifications shows the same tendency. Similarly to the macro analysis, this result indicates that (indirectly) accounting for regional attributes, such as flood risk or insurance density, reduces the significance of the effect. Finally, the estimation results of Model X show that well-being losses are particularly adhering to house owners. This suggests that well-being losses rather originate from damages on (immobile) property that needs to be renovated by the house owners, a fact that means considerable financial and psychological stress for them and thus cause well-being losses. In contrast, appliances and personal belongings of renters are rather mobile and may potentially be saved from destruction to a larger share.
5. Conclusion Despite compensation through insurances and public post-hoc disaster relief funds, life satisfaction is significantly negatively affected by extreme weather events. In our analysis, we found that floods and storm & hail events decrease life satisfaction by about 0.02 on the 11-point scale. Due to the study design it may well be that we underestimate the effect of extreme events on well-being because we cannot distinguish respondents which have been personally affected from those who lived in a affected region but did not incur any damages themselves. Counting the latter as affected might cause a downwards bias in the estimated effect. For our analysis we used a rich dataset based on the German Socio-Economic Panel (SOEP) and regionally disaggregated data on extreme weather events and insurance density. By enhancing the database with regard to the aforementioned distinction between affected and unaffected as well as compensated and non-compensated individuals, the informative value of such an analysis could be further increased. Nevertheless, our results underline that there remain negative wellbeing effects that are not compensated for by existing monetary compensation schemes—be it arranged on private markets (insurances) or publicly provided (disaster funds). These effects are usually not accounted for in conventional cost–benefit analyses. Decisions on the design and quantity of investments into flood and storm protection measures might therefore neglect a share of the benefits that could be generated by stronger efforts to improve protection, adaptation and overall risk management. With climate change and the possible aggravation of extreme weather events and the ongoing increase of damage potentials within risk prone areas, immaterial values should be carefully taken into consideration in risk management as well as in the design of climate change mitigation and adaptation policies. To that effect, our analysis showed that the loss of life satisfaction due to extreme events is significantly reduced with higher insurance rates. This poses a strong argument to intensifying efforts to increase insurance rates covering extreme events or even introduce mandatory insurance schemes. The life satisfaction approach presents an interesting extension to common methods of damage assessment or cost-benefit analysis as it allows for the valuation of intangible values. With its methodology being still further elaborated, it can already serve to supplement insights from economic valuations and to point out the relevance of other value dimensions (see Fujiwara and Campbell, 2011).
Acknowledgments This article gives an account of results from the project Economics of Climate Change Adaptation (econCCadapt) funded by the German Federal Ministry of Education and Research (FKZ 01LA1137A). Furthermore, we would like to thank Heinz Welsch, Eugen Pissarskoi and two anonymous referees for their helpful comments.
116
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
Appendix A
References
Table A.I: Summary statistics of SOEP data. Variable
Description
Mean
Life satisfaction Equivalized income (in €) HH size Age Person needing care in HH Very good health Good health Satisfactory health Poor health Bad health Employed Not employed Unemployed Pensioner Military/community service In education Self employed No partner Partner living in HH Partner living outside HH
11-point scale Log. of equivalized net monthly HH income Number of persons in HH Years Dummy variable
S.D.
Min Max
6.99 7.21
1.78 0 0.51 0
2.72 47.93 0.04
1.28 1 17.65 16 0.20 0
10 11.12 14 101 1
Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable
0.10 0.40 0.33 0.13 0.04 0.48 0.07 0.06 0.24 0.003
0.29 0.49 0.47 0.34 0.19 0.50 0.26 0.23 0.43 0.06
0 0 0 0 0 0 0 0 0 0
1 1 1 1 1 1 1 1 1 1
Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable
0.08 0.06 0.22 0.7 0.08
0.27 0.24 0.41 0.46 .27
0 0 0 0 0
1 1 1 1 1
Abbreviation: HH = household.
Table A.II: Summary statistics of event variables. Variable
Description
Mean
S.D.
Min
Max
Flood events (6 months) Flood events (12 months) Flood events (18 months) Storm & hail events (6 months) Storm & hail events (12 months) Storm & hail events (18 months) Extreme events (6 months) Extreme events (12 months) Extreme events (18 months) Flood events (6 months) Flood events (12 months) Flood events (18 months) Storm & hail events (6 months) Storm & hail events (12 months) Storm & hail events (18 months) Extreme events (6 months) Extreme events (12 months) Extreme events (18 months)
Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Number of events Number of events Number of events Number of events Number of events Number of events Number of events Number of events Number of events
0.03 0.11 0.14 0.25 0.33 0.55 0.26 0.38 0.58 0.03 0.11 0.15 0.25 0.34 0.61 0.28 0.46 0.76
0.17 0.31 0.34 0.43 0.47 0.50 0.44 0.48 0.49 0.17 0.35 0.39 0.44 0.49 0.59 0.50 0.65 0.77
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Andreß, H.-J., Golsch, K., Schmidt, A.W., 2013. Applied Panel Data Analysis for Economic and Social Surveys. Springer, Berlin; Heidelberg. Baetschmann, G., Staub, K.E., Winkelmann, R., 2014. Consistent estimation of the fixed effects ordered logit model. J. R. Stat. Soc. A. Stat. Soc. (n/a–n/a). Carroll, N., Frijters, P., Shields, M., 2009. Quantifying the costs of drought: new evidence from life satisfaction data. J. Popul. Econ. 22 (2), 445–461. Ciscar, J.-C., Soria, A., Goodess, C., Christensen, O., Iglesias, A., et al., 2009. Climate change impacts in Europe. Final report of the PESETA research project JRC Scientific and Technical Reports. Dehnhardt, A., Hirschfeld, J., Drünkler, D., Petschow, U., Engel, H., Hammer, M., 2008. Kosten-Nutzen-Analyse von Hochwasserschutzmaβnahmen. Forschungsbericht 204 21 212. UBA Texte 31/2008, Umweltbundesamt, Dessau-Roßlau. Di Tella, R., MacCulloch, R.J., Oswald, A.J., 2001. Preferences over inflation and unemployment: evidence from surveys of happiness. Am. Econ. Rev. 91 (1), 335–341. Diener, E., Lucas, R.E., 1999. Personality and subjective well-being. In: Diener, E., Kahneman, D., Schwarz, N. (Eds.), Well-Being: The Foundations of Hedonic Psychology. Russell Sage Foundation. Ferreira, S., Moro, M., 2010. On the use of subjective well-being data for environmental valuation. Environ. Resour. Econ. 46 (3), 249–273. Ferrer-i Carbonell, A., Frijters, P., 2004. How important is methodology for the estimates of the determinants of happiness? Econ. J. 114 (497), 641–659. Frijters, P., Haisken-DeNew, J.P., Shields, M.A., 2004. Money does matter! Evidence from increasing real income and life satisfaction in east Germany following reunification. Am. Econ. Rev. 94 (3), 730–740. Fujiwara, D., Campbell, R., 2011. Valuation Techniques for Social Cost-Benefit Analysis: Stated Preference, Revealed Preference and Subjective Well-Being Approaches: A Discussion of the Current Issues. HM Treasury. GDV, 2012. Naturgefahrenreport 2012. Gesamtverband der Deutschen Versicherungswirtschaft e.V, GDV, Berlin. GDV, 2013. Versicherungsdichte “Elementargefahren”. Gesamtverband der Deutschen Versicherungswirtschaft e.V, GDV, Berlin. Kimball, M., Levy, H., Ohtake, F., Tsutsui, Y., 2006. Unhappiness after Hurricane Katrina. Working paper 12062. National Bureau of Economic Research. Kountouris, Y., Remoundou, K., 2011. Valuing the welfare cost of forest fires: a life satisfaction approach. Kyklos 64 (4), 556–578. Kreibich, H., Christenberger, S., Schwarze, R., 2011. Economic motivation of households to undertake private precautionary measures against floods. Nat. Hazards Earth Syst. Sci. 11 (2), 309–321. Luechinger, S., Raschky, P.A., 2009. Valuing flood disasters using the life satisfaction approach. J. Public Econ. 93 (3–4), 620–633. Maddison, D., Rehdanz, K., 2011. The impact of climate on life satisfaction. Ecol. Econ. 70 (12), 2437–2445. Moulton, B.R., 1990. An illustration of a pitfall in estimating the effects of aggregate variables on micro units. Rev. Econ. Stat. 72 (2), 334–338. Osberghaus, D., Kühling, J., 2014. Direct and indirect effects of weather experiences on life satisfaction: which role for climate change expectations? ZEW Discussion Paper. ZEW—Zentrum für Europäische Wirtschaftsforschung/Center for European Economic Research. Schupp, J., Kroh, M., Goebel, J., Bartsch, S., Giesselmann, M., et al., 2014. Sozio-oekonomisches Panel (SOEP), Daten der Jahre 2000–2011. Version: 28. Dataset, SOEP—Soziooekonomisches Panel. http://doi.org/10.5684/soep.v28doi:10.5684/soep.v28 Seifert, I., Botzen, W.J.W., Kreibich, H., Aerts, J.C.J.H., 2013. Influence of flood risk characteristics on flood insurance demand: a comparison between Germany and The Netherlands. Nat. Hazards Earth Syst. Sci. 13 (7), 1691–1705. Tapsell, S., Penning-Rowsell, E., Tunstall, S., Wilson, T., 2002. Vulnerability to flooding: health and social dimensions. Proc. R. Soc. London, Ser. A 360 (1796), 1511–1525. Wagner, G.G., Frick, J.R., Schupp, J., 2007. The German socio-economic panel study (soep)—scope, evolution and enhancements. Schmollers Jahr. 127 (1), 139–169. Welsch, H., Ferreira, S., 2013. Environment, well-being, and experienced preference. Int. Rev. Environ. Resour. Econ. 7, 205–239.
Contents lists available at ScienceDirect
Ecological Economics journal homepage: www.elsevier.com/locate/ecolecon
Analysis
Measuring impacts of extreme weather events using the life satisfaction approach Charlotte von Möllendorff a,⁎, Jesko Hirschfeld b a b
Carl von Ossietzky University Oldenburg, Department of Economics, 26111 Oldenburg, Germany Institute for Ecological Economy Research (IÖW), Potsdamer Str. 105, 10785, Berlin, Germany
a r t i c l e
i n f o
Article history: Received 8 May 2015 Received in revised form 23 October 2015 Accepted 15 November 2015 Available online xxxx JEL classification: D61 I31 Q51 Q54
a b s t r a c t Extreme weather events cause harm among the aggrieved party that often goes beyond material damages. This paper studies the impact of extreme weather events on measures of self-reported life satisfaction. Focusing on Germany, we use representative panel data for 2000–2011 to study the effect of seven storm & hail events and five floods on subjective well-being in the affected NUTS 3 regions. Our results indicate that both weather experiences bear statistically significant negative externalities. Following an extreme weather event, life satisfaction is reduced by 0.020–0.027 on the 11-point scale. While the effect of storm & hail events is rather immediate in nature, the effect from floods persists much longer. © 2015 Elsevier B.V. All rights reserved.
Keywords: Extreme weather events Subjective well-being Life satisfaction Nonmarket valuation
1. Introduction In the past, we observed some destructive storms and floods that severely impacted on the population living in affected areas. With climate change, the magnitude and frequency of extreme weather events are expected to increase even further (Ciscar et al., 2009). The valuation of impacts imposed by extreme weather events usually focuses on economic aspects, such as damages on buildings, items and infrastructures and thereby neglects immaterial values such as mental distress, worries, health injuries or the loss of personal belongings (see e.g. Dehnhardt et al., 2008, Tapsell et al., 2002). In this paper we study the impact of extreme weather events in Germany using evidence from subjective well-being data. Life satisfaction analysis has been increasingly used to evaluate environmental attributes and non-marketed goods (see e.g. Welsch and Ferreira, 2013 or Ferreira and Moro, 2010). We analyse how life satisfaction changes in affected regions due to the occurrence of an extreme weather event using panel data from the German Socio-Economic Panel Study (SOEP, see Schupp et al., 2014). Using panel data permits to control for unobserved interpersonal ⁎ Corresponding author. E-mail addresses: [email protected] (C. von Möllendorff), [email protected] (J. Hirschfeld).
http://dx.doi.org/10.1016/j.ecolecon.2015.11.013 0921-8009/© 2015 Elsevier B.V. All rights reserved.
heterogeneity. The importance of controlling for individual fixed effects in life satisfaction analysis has been emphasized by Ferrer-i Carbonell and Frijters, (2004). Our paper ties in with a small number of studies that analyse well-being effects of climate variables (Maddison and Rehdanz, 2011) and weather events like floods (Luechinger and Raschky, 2009), droughts (Carroll et al., 2009), hurricanes (Kimball et al., 2006) and forest fires (Kountouris and Remoundou, 2011). Luechinger and Raschky (2009) study the well-being effect of floods in Europe between 1973 and 2004 on NUTS 2 level and find a sizeable negative effect on life satisfaction. In a more recent analysis, Osberghaus and Kühling (2014) study indirect and direct effects of weather experiences in Germany – namely storms, floods, heavy rain and heat waves – on subjective well-being using a specifically designed and conducted one-time survey. They find a significant negative effect of climate-change induced damage expectations on subjective wellbeing while the direct effect is only significant in the case of heat waves. In our analysis we focus on Germany using data from the SOEP for 2000–2011 which allows us to control for interpersonal heterogeneity and relevant socio-economic characteristics. We further use spatially disaggregated data (NUTS 3 level) from the German Insurance Association (GDV) on insured losses of five floods and seven storm & hail events. Hence, the main contribution of our paper is to study the well-being effects of different types of extreme weather events (floods and storm & hail) on a disaggregated level (NUTS 3) over a period of
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
12 years (2000–2011) using the rich dataset of the SOEP. Unlike earlier studies, we consider regional data on insurance density, distinguish different levels of regional impacts, analyse households in rented/owned property separately and control for movers/stayers. Our results indicate a significant negative effect of storm & hail events as well as floods on subjective well-being in affected regions, decreasing life satisfaction by about 0.02–0.027 on the 11-point scale. While the effect of storm & hail events dissipates after 6 months, floods affect life satisfaction much longer. Moreover, we find that the effect is particularly adhering to house owners and is lower in areas with high insurance rates. The following sections are structured as follows: Section 2 presents the data and Section 3 describes the econometric approach. Section 4 reports and discusses the results while Section 5 concludes.
2. Data The life satisfaction data along with the socio-economic control variables were made available by the German Socio-Economic Panel Study (SOEP) of the German Institute for Economic Research (DIW), Berlin. Since 1984, the SOEP conducts annual interviews surveying the socioeconomic situation of German households. In each annual wave approximately 20,000 persons living in 11,000 households are surveyed (see Wagner et al., 2007). As for the panel structure of the survey study, the same set of respondents is reinterviewed each year. This facilitates to make longitudinal analyses of changes on the individual level and to control for unobserved time-invariant characteristics (Andreß et al., 2013). Especially by focusing on a longer time frame, the SOEP study faces some temporary or permanent drop-outs which are offset by including new respondents, i.e. the data is unbalanced. Our analysis is conducted based on an unbalanced subset of the SOEP, which is composed of 239,209 observations from 39,679 respondents that were interviewed during 2000 and 2011. The spatial level of the analysis refers to the NUTS 3 regions of Germany which coincide with the level of Landkreise/Kreise and kreisfreie Städte.1 There are 402 NUTS 3 regions in Germany with an average size of 888 km 2. The life satisfaction data of the SOEP results from responses to the following question: “How satisfied are you with your life, all things considered?”. The question can be answered on a scale from 0 (completely dissatisfied) to 10 (completely satisfied). In addition, we considered time-varying socio-economic control variables on the individual level, that were found to cause changes in individual well-being (see e.g. Frijters et al., 2004).2 With regard to extreme weather events we created a dataset that includes floods and storm & hail events that occurred in Germany during 1999 and 2010, with the most severe being the Elbe flood in 2002 and the storm Kyrill in 2007. The events are summarized in Table 1. The selection of events is based on their intensity which is approximated by the claims expenditure they caused for insurances. In our analysis we considered five floods that caused claims expenditures higher than 55 million and seven storm & hail events that caused claims expenditures higher than 230 million. The great difference in claims expenditure between both types of events is related to the insurance density. In Germany more than 90% of all buildings are insured against storm and hail events, while only 30% of all buildings have a natural hazard insurance, which covers, inter alia, damages due to floods (GDV, 2012). So the fraction of damages rendered by the insurances is higher for storm & hail events compared to floods. Furthermore, there is regional variation in insurance densities with regard to natural hazard insurances (see e.g. Kreibich et al., 2011 or Seifert et al., 2013). For this reason, we used data on regional insurance densities, which represent for each
1
NUTS stands for Nomenclature of Units for Territorial Statistics and is a geocode standard for referencing counties and regions in the European Union. 2 See Appendix for summary statistics.
109
federal state the share of insured buildings in the highest risk zone (GDV, 2013). Though we are analysing the period from 2000 to 2011 the storm Lothar that occurred in December 1999 is included in the analysis so as to study possible effects on self-reported life satisfaction in the interviews conducted in 2000. In the same line of reasoning 2011 is included in the analysis to track potential effects of events that occurred in 2010 on stated well-being in 2011. The thunderstorm Hilal is listed twice because some of the damages were covered by the natural hazard insurance, while others were covered by the building insurance. The data we use refers to damages from floods and storm & hail events collected and provided by the German Insurance Association (Gesamtverband der deutschen Versicherungswirtschaft, GDV), Berlin. Storm & hail events are usually insured via building insurances while floods (due to extreme rainfall or overflowing) come under the natural hazard insurance. We used damage data of both insurance types to obtain information on storm & hail events as well as floods in Germany.3 For each considered event the data was made available in the form of high-resolution maps showing the classified damage frequencies per NUTS 3 region (see Fig. 1 for corresponding maps of the Elbe flood, 2002 and storm Kyrill, 2007). Damage frequency (DF) is defined as follows (GDV, 2012:43): Damage frequency ðDFÞ ¼
number of claims number of contracts
ð1Þ
Thus, the damage frequency describes the number of contracts that were deployed to claim from the insurer in relation to the total number of running contracts per NUTS 3 region. By this, the variable implicitly controls for regional variation in insurance density. As the borders of the NUTS 3 regions are displayed in the maps, the information on damage frequency per NUTS 3 region and event could be extracted using a Geographic Information System (GIS). The damage frequencies of the NUTS 3 regions are displayed in intervals which correspond to the incidence rates of the events ranging from On one day there occurred as many damages as usually occur within 1 week (storm & hail)/1 month (floods) to On one day there occurred as many damages as usually occur within 1 year (storm & hail)/12 years (floods). To illustrate, consider the NUTS 3 region in Fig. 1(a) coloured in black which was most severely affected by the Elbe flood in 2002 (Sächsische Schweiz-Osterzgebirge): It exhibits a damage frequency higher than 200 which means that over 200 out of 1000 households insured against natural hazard events claimed from the insurer, an event whose corresponding incidence rate is 12 years. Taking a look at storm Kyrill, the map shows that most of the NUTS 3 regions (129) had a damage frequency between 38.7 and 77.4 meaning that in those regions there occurred more damages on one day than usually occur within six months, or rather the incidence rate is 6 months (see Table 2). This exemplifies that the relation between damage frequency and incidence rate differs for floods and storm & hail events, respectively. So in order to combine the data on floods and storm & hail events and to get down to a common definition of extreme weather events we revert to the incidence rate. Based on this, we define an extreme event as follows: On one day there occurred more damages as usually occur within a month. This means that for each event we considered respondents as affected if they lived in a NUTS 3 region with a damage frequency higher than 0.4 in case of floods, and 6.4 in case of storm & hail events (see Table 2). This is the lowest threshold value commonly available for both types of events. If we increase the threshold level, we are likely to get down to the more severely impacted individuals which may 3 For simplicity we refer to flood events throughout the paper. Strictly speaking the data describes all other natural hazard damages as well (backwater, earthquakes, land subsidence, landslide, snowslides, snow pressure, and volcanic eruption) which are generally, however, of minor relevance in Germany (GDV, 2012).
110
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
Table 1 Extreme weather events in Germany with insured losses, 1999–2011 (following (GDV, 2012):10ff/23ff). Year
Name
Event type
Date
Average claims amount
Total claims expenditure (in million)
1999 2002 2002 2006 2007 2008 2008 2008 2008 2009 2010 2010
Lothar Elbe flood Jeanett Queeny Kyrill Emma Hilal Hilal Zsuzanna Rainer Xynthia Elbe, Neiße, Spree
Storm & hail Flood Storm & hail Storm & hail Storm & hail Storm & hail Flood Storm & hail Flood Flood Storm & hail Flood
25/12–26/12 31/07–02/09 27/10–28/10 28/06 18/01–19/01 29/02–01/03 29/05–02/06 30/05 26/07–04/08 24/06–03/07 28/02–01/03 01/08–13/08
€1359 €17,752 €783 €12,749 €1021 €881 €5805 €2151 €5639 €4998 €821 €10,846
€800 €1800 €760 €230 €2060 €390 €100 €330 €55 €85 €367 €265
Fig. 1. Damage frequencies of the Elbe Flood (2002) and Storm Kyrill (2007) (source: GDV, 2012:13/23).
have experienced higher well-being losses. However, increasing the threshold level reduces the number of respondents who have experienced a storm & hail or a flood event during the past 6 months from 67,395 (threshold: “1 month”) to 40,811 (threshold: “3 months”), 23,675 (threshold: “6 months”) or 12,782 (threshold: “1 year”). With increasing threshold levels the results become less significant. Thus, we face a trade-off between the severity on the one hand and the number of cases on the other hand. Our event variables are specified as the number of storm & hail/ flood/extreme weather events, respectively, that occurred in the respondents NUTS 3 region within a given time frame (see Section 3 for a discussion of the different model variations). Thereby, extreme weather events is simply defined as the sum of storm & hail and flood events.4 4
See Appendix for summary statistics.
Furthermore, we tested the untransformed frequency metric as a continuous variable which entails information on regional variation in impact levels. Both, the SOEP data and the data on extreme weather events are merged on the basis of NUTS 3 regions. Our dataset poses several problems. First, we cannot distinguish between those individuals who are actually affected and those who are not because we revert to administrative borders. It is very difficult to circumvent this problem. Second, we cannot distinguish between individuals who are insured and those who are not. This applies especially to floods as we have an insurance density of only 30% for natural hazard events. For storm & hail events this problem is not as severe because here we have an insurance density of 90% and can assume that most people have an insurance that covers related damages. Third, we possibly have a selection bias in the sense that individuals more exposed to
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116 Table 2 Incidence rate of extreme weather events (following GDV, 2012:42). Flood events
Storm and hail events
On one day there occurred as many damages as usually occur within
Lower bound of damage frequency (in ‰)
On one day there occurred as many damages as usually occur within
Lower bound of damage frequency (in ‰)
1 month 3 months 6 months 1 year 3 years 6 years 12 years
0.4 1.2 2.4 4.8 14.5 29.0 58.0
1 week 2 weeks 1 month 3 months 6 months 1 year
1.5 3.0 6.4 19.3 38.7 77.4
risks from flooding are rather insured against natural hazards while there is also a small fraction of individuals living in high risk areas which are rejected by the insurances. There are other disparities in insurance policies that might as well reinforce different insurance densities across regions. Still, we assume that with an average insurance density of 30% most NUTS 3 regions are represented in the dataset and regional disparities are levelled out by dividing by the number of running contracts. 3. Econometric framework By conducting a longitudinal analysis, we study in a fixed effects modelling framework whether individual well-being changes due to the occurrence of an extreme weather event. An advantage of the fixed effects model is that it facilitates to capture unobserved factors like personality characteristics that were found to have an important influence on stated subjective well-being (Diener and Lucas, 1999). Thereby the risk of omitted variable bias is reduced though a bias can still arise due to the neglect of unobserved time-varying variables. On account of the fixed effects modelling framework time-invariant covariates commonly considered in life satisfaction analyses (such as gender, migration background, highest level of education) cannot explicitly be incorporated in the model, though they are implicitly controlled for via the individual fixed effect. Running a Hausman test confirms that a fixed-effects model is appropriate because unobserved personal heterogeneity cannot be considered as uncorrelated with the explanatory variable as is assumed in random effects models. Formally the estimating model can be stated as follows: LSijt ¼ aQ jt þ β ln Mijt þ D0 ijt y þ ηi þ ηt þ εijt :
ð2Þ
LSijt indicates the life satisfaction of respondent i in NUTS 3 region j in year t. Qjt is a metric variable describing the number of extreme weather events and will (i) separately capture the two considered event types (floods and storm & hail) or (ii) represent alternatively an overall term for extreme weather events, as discussed above. Mijt specifies the net monthly household income which has been equivalized according to the OECD-modified scale and is included in logarithmic form in order to take account of its decreasing marginal utility. D'ijt describes time-varying covariates at the micro-level which have been found to influence life satisfaction, namely household size, age squared, state of health, partner status, employment status and whether there is a person in the household in need of care. The variable ηi describes the individual fixed effect. Time-specific effects are captured by the year dummies ηt, which are included to account for unobserved factors pertaining to a specific year. Measurement errors are described by εijt. Age is included as a quadratic term to control for the proposed Ushaped relationship between life satisfaction and age. The linear term of age cannot be incorporated, because it correlates with the year dummies. Macro-level variables are left out of the analysis too: The regional GDP per capita was tested but its effects are insignificant; the
111
regional unemployment rate had a marginal effect on life satisfaction but was omitted because no data was available for 2010 and 2011. Moreover, population density was included in order to take account of possible migration following an event but was not found significant. We estimated Eq. (2) applying a linear fixed effects estimator. Despite the ordinal nature of the dependent variable LSijt that suggests ordinal regression techniques, previous studies confirmed that the differences in the results are negligible (see e.g. Ferrer-i Carbonell and Frijters, 2004). Furthermore, there is no consensus on how to appropriately implement individual fixed effects in ordered regression models (Baetschmann et al., 2014). Standard errors are adjusted for clustering on the individual level, which is necessary because the data contains multiple observations from each respondent (Moulton, 1990). Additionally, we tested the results with regard to clustering on the regional level. We estimated several different models. Models I to III refer to flood events, Models IV and V refer to storm & hail events and Model VI uses a common variable of extreme weather events. In Model I floods are captured as a metric variable describing the number of events, Model II uses a dummy variable in interaction with the regional insurance density, while Model III introduces a continuous variable of the regional damage frequency. Similarly, Model IV includes a variable on the number of storm & hail events, while Model V estimates a continuous variable of the regional damage frequency. Model VI summarizes the number of floods and storm & hail events using a metric scale. In order to survey the robustness of the results we included some further variations. Model VII takes a macro approach and studies changes in life satisfaction following flood events on NUTS 3 level. Model VIII excludes migrants while subsequently in Model IX standard errors are adjusted for clustering on NUTS 3 level. Finally, Model X uses interaction terms in order to analyse differences in the effects of flood events for house owners and renters, respectively. Within all models, several different specifications were tested which revert to the elapsed time between the event and the interview. As the exact dates of the interviews and the events are known, the time-lag can be calculated.5 In this analysis we test three different time-frames: First, events that date back less than 6 months; second, events that date back less than 12 months; and third, events that date back less than 18 months. Expanding the time frame instead of using disjoint intervals conduces to gradually increase the number of observations with respect to people that experienced an extreme weather event. As a result, considering different time frames caters to the question how long the effect of extreme weather events on subjective well-being prevails but has also implications for the number of cases and thus for the significance of the results as will be discussed below.
4. Results 4.1. Estimation results In the following, we present the results for the different model specifications as defined in the previous section. Table 3 displays the corresponding outputs for flood events (Model I–III) and Table 4 for storm & hail events (Model IV and V) as well as for a common term of extreme events (Model VI). When comparing the different columns, it becomes obvious that the socio-economic covariates are stable with regard to the various model specifications. In line with other findings, the results reveal a highly significant positive effect of equivalized income on life satisfaction. Concerning the effect of partner status on life satisfaction, the results indicate that being in a partnership is favoured above being single, 5 If the event ranged over a longer period of time, the last day of the event was chosen as a reference. Taking the first day or the midpoint of the event as reference instead produces largely the same results.
112
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
Table 3 Estimation results for flood events (source: SOEPv28, author's calculations). Variables Flood variables Number of events
Model I (a) 6 months (b) 12 months (c) 18 months −0.0112 (0.0165)
−0.0210⁎⁎ (0.00925)
−0.0230⁎⁎⁎ (0.00836)
Dummy Insurance density Dummy × insurance density
Model II (a) 6 months (b) 12 months (c) 18 months
−0.0561⁎⁎ (0.0266) −0.0844 (0.0920) 0.111⁎⁎ (0.0530)
−0.0226 (0.0148) −0.0821 (0.0920) 0.00702 (0.0291)
−0.0295⁎⁎ (0.0138) −0.0840 (0.0919) 0.0209 (0.0270)
0.280⁎⁎⁎
0.280⁎⁎⁎
Damage frequency Log. Equiv. Household income Household size Age squared Person needing care in HH Health status Good Satisfactory Poor Bad Employment status Not employed Unemployed Pensioner Military, community service In education Self-employed Partner status Partner outside household Partner inside household Year dummies Individual fixed effects R 2(within) Number of individuals Number of observations
0.280⁎⁎⁎ (0.0137) 0.0114⁎
0.280⁎⁎⁎
0.280⁎⁎⁎
(0.0137) 0.0115⁎
(0.0137) 0.0115⁎
0.280⁎⁎⁎ (0.0137) 0.0115⁎
(0.0137) 0.0115⁎
Model III (a) 6 months (b) 12 months (c) 18 months
0.000169 (0.000295) 0.280⁎⁎⁎ (0.0137) 0.0116⁎
0.000138 (0.000244) 0.280⁎⁎⁎ (0.0137) 0.0115⁎
0.000285 (0.000208) 0.280⁎⁎⁎ (0.0137) 0.0115⁎
(0.00632) (0.00632) (0.00632) 0.0000 0.0000 0.0000 (0.00004) (0.00004) (0.00004) −0.409⁎⁎⁎ −0.409⁎⁎⁎ −0.409⁎⁎⁎ (0.0310) (0.0310) (0.0310) Reference group: very good −0.317⁎⁎⁎ −0.317⁎⁎⁎ −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0137) 0.0115⁎ (0.00632 (0.00632) (0.00632) 0.0000 0.0000 0.0000 (0.00004) (0.00004) (0.00004) ⁎⁎⁎ ⁎⁎⁎ −0.409 −0.409 −0.409⁎⁎⁎ (0.0310) (0.0310) (0.0310)) Reference group: very good −0.317⁎⁎⁎ −0.317⁎⁎⁎ −0.317⁎⁎ (0.0117) (0.0117) (0.0117) ⁎ −0.722⁎⁎⁎ −0.723⁎⁎⁎ −0.723⁎⁎⁎
(0.00632) (0.00632) (0.00632) 0.0000 0.0000 0.0000 (0.00004) (0.00004) (0.00004) ⁎⁎⁎ ⁎⁎⁎ −0.409 −0.409 −0.409⁎⁎⁎ (0.0310) (0.0310) (0.0310) Reference group: very good −0.317⁎⁎⁎ −0.317⁎⁎⁎ −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.265⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎
(0.0326) (0.0326) (0.0326) Reference group: employed −0.0576⁎⁎⁎ −0.0577⁎⁎⁎ −0.0576⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎ (0.0204) (0.0204) (0.0204) 0.0400⁎ 0.0401⁎ 0.0401⁎
(0.0326) (0.0326) (0.0326) Reference group: employed −0.0575⁎⁎⁎ −0.0576⁎⁎⁎ −0.0576⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎ (0.0204) (0.0204) (0.0204) 0.0401⁎ 0.0401⁎ 0.0400⁎
(0.0326) (0.0326) (0.0326) Reference group: employed −0.0575⁎⁎⁎ −0.0575⁎⁎⁎ −0.0575⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎ (0.0204) (0.0204) (0.0204) 0.0400⁎ 0.0400⁎ 0.0400⁎
(0.0206) (0.0206) (0.0206) 0.0538 0.0540 0.0543 (0.0505) (0.0505) (0.0505) 0.133⁎⁎⁎ 0.133⁎⁎⁎ 0.133⁎⁎⁎ (0.0205) (0.0205) (0.0205) 0.0252 0.0250 0.0251 (0.0275) (0.0275) (0.0275) Reference group: no partner 0.302⁎⁎⁎ 0.302⁎⁎⁎ 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) (0.0219) (0.0219) Yes Yes Yes Yes Yes Yes 0.0982 0.0982 0.0982 39,679 39,679 39,679 239,209 239,209 239,209
(0.0206) (0.0206) (0.0206) 0.0541 0.0540 0.0542 (0.0505) (0.0505) (0.0505) 0.133⁎⁎⁎ 0.133⁎⁎⁎ 0.133⁎⁎⁎ (0.0205) (0.0205) (0.0205) 0.0250 0.0249 0.0250 (0.0275) (0.0275) (0.0275) Reference group: no partner 0.302⁎⁎⁎ 0.302⁎⁎⁎ 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) (0.0219) (0.0219) Yes Yes Yes Yes Yes Yes 0.0982 0.0982 0.0982 39,679 39,679 39,679 239,209 239,209 239,209
(0.0206) (0.0206) (0.0206) 0.0537 0.0536 0.0534 (0.0505) (0.0505) (0.0505) 0.133⁎⁎⁎ 0.133⁎⁎⁎ 0.133⁎⁎⁎ (0.0205) (0.0205) (0.0205) 0.0252 0.0252 0.0253 (0.0275) (0.0275) (0.0275) Reference group: no partner 0.302⁎⁎⁎ 0.302⁎⁎⁎ 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) (0.0219) (0.0219) Yes Yes Yes Yes Yes Yes 0.0982 0.0982 0.0982 39,679 39,679 39,679 239,209 239,209 239,209
Standard errors in parenthesis are adjusted for clustering at the personal level. ⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.
especially if the partner lives in the same household. Turning to the health status coefficients, a strong negative effect on life satisfaction is observed for a decrease in the state of health. When considering one's health as very good, a drop to bad health is associated with a loss in life satisfaction by 2.26 on an 11-point scale. Becoming unemployed affects life satisfaction negatively. This result is in line with the common findings in the happiness literature (see e.g. Di Tella et al., 2001). Turning to the extreme weather event variables of the baseline models (Model I and Model IV) we observe different patterns for the two types of events. Regarding floods, their negative effect on life satisfaction becomes most evident when we consider all events of the past 18 months (Model I). In contrast, storm & hail events display only a significant effect on life satisfaction if we consider a short period of the past 6 months for our analysis (Model IV). However, the effects are quite
similar in magnitude: While a flood that happened during the past 18 months reduces life satisfaction by 0.0230, a storm & hail event that happened during the past 6 months causes a drop in life satisfaction by 0.0268 on an 11-point scale (both significant at the 1%-level). This may indicate that the impact of a storm & hail event is rather immediate in nature and dissipates after a short period. Differently, the effect of flood events becomes most evident by considering a long period of 18 months. However, by interpreting these results we have to keep in mind that by considering a period of 18 months we are likely to observe more individuals that experienced a flood event, than in the shorter time-frames. To exemplify, the number of observations in NUTS 3 regions that experienced a flood during the past 6 months is only 7857 while the number of observations in NUTS 3 regions that experienced a flood during the past 12 months is 36,062. So in the short term it
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
113
Table 4 Estimation results for storm & hail events and extreme events (source: SOEPv28, author's calculations). Variables Storm & hail events Number of events
Model IV (a) 6 months (b) 12 months (c) 18 months −0.0268⁎⁎⁎ (0.00915)
−0.0109 (0.00798)
0.000763 (0.00799)
Damage frequency
Models V (a) 6 months (b) 12 months (c) 18 months
−0.000232⁎⁎⁎ (0.0000813)
−0.000219⁎⁎⁎ (0.0000808)
−0.000170⁎⁎ (0.0000690)
Extreme weather events Number of events Log. Equiv. Household income Household size Age squared Person needing care in HH Health status Good Satisfactory Poor Bad Employment status Not employed Unemployed Pensioner Military, community service In education Self-employed Partner status Partner outside household Partner inside household Year dummies Individual fixed effects R2 (within) Number of individuals Number of observations
0.280⁎⁎⁎ (0.0137) 0.0113⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
0.280⁎⁎⁎ (0.0137) 0.0114⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
0.280⁎⁎⁎ (0.0137) 0.0115⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
0.280⁎⁎⁎ (0.0137) 0.0114⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
0.280⁎⁎⁎ (0.0137) 0.0115⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
Models VI (a) 6 months (b) 12 months (c) 18 months
0.280⁎⁎⁎ (0.0137) 0.0115⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
−0.0145⁎⁎ (0.00589) 0.280⁎⁎⁎ (0.0137) 0.0115⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
−0.00948⁎ (0.00566) 0.280⁎⁎⁎ (0.0137) 0.0114⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
(0.0310) (0.0310) (0.0310) Reference group: very good ⁎⁎⁎ ⁎⁎⁎ −0.317 −0.317 −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0310) (0.0310) Reference group: very good ⁎⁎⁎ −0.317 −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0310) (0.0310) (0.0310) Reference group: very good ⁎⁎⁎ ⁎⁎⁎ −0.317 −0.317 −0.317⁎⁎⁎ (0.0117) (0.0117) (0.0117) −0.722⁎⁎⁎ −0.722⁎⁎⁎ −0.722⁎⁎⁎
(0.0140) (0.0140) (0.0140) −1.264⁎⁎⁎ −1.264⁎⁎⁎ −1.264⁎⁎⁎ (0.0177) (0.0177) (0.0177) −2.260⁎⁎⁎ −2.260⁎⁎⁎ −2.260⁎⁎⁎ (0.0326) (0.0326) (0.0326) Reference group: employed −0.0574⁎⁎⁎ −0.0575⁎⁎⁎ −0.0576⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎
(0.0140) −1.264⁎⁎⁎ (0.0177) −2.260⁎⁎⁎ (0.0326)
(0.0140) (0.0140) −1.264⁎⁎⁎ −1.264⁎⁎⁎ (0.0177) (0.0177) −2.260⁎⁎⁎ −2.260⁎⁎⁎ (0.0326) (0.0326) Reference group: employed −0.0573⁎⁎⁎ −0.0574⁎⁎⁎ −0.0574⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎
(0.0140) (0.0140) (0.0140) −1.264⁎⁎⁎ −1.264⁎⁎⁎ −1.264⁎⁎⁎ (0.0177) (0.0177) (0.0177) −2.260⁎⁎⁎ −2.260⁎⁎⁎ −2.260⁎⁎⁎ (0.0326) (0.0326) (0.0326) Reference group: employed −0.0574⁎⁎⁎ −0.0575⁎⁎⁎ −0.0574⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.490⁎⁎⁎ −0.490⁎⁎⁎ −0.490⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0535 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0537 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0533 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0401⁎ (0.0206) 0.0537 (0.0505) 0.133⁎⁎⁎
(0.0205) (0.0205) (0.0205) 0.0257 0.0253 0.0251 (0.0275) (0.0275) (0.0275) Reference group: no partner ⁎⁎⁎ ⁎⁎⁎ 0.302 0.302 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎
(0.0205) 0.0251 (0.0275)
(0.0219) Yes Yes 0.0982 39,679 239,209
(0.0204) 0.0400⁎ (0.0206) 0.0536 (0.0505) 0.133⁎⁎⁎
(0.0219) Yes Yes 0.0982 39,679 239,209
(0.0219) Yes Yes 0.0982 39,679 239,209
(0.0310)
−0.0206⁎⁎⁎ (0.00746) 0.280⁎⁎⁎ (0.0137) 0.0113⁎ (0.00632) 0.0000 (0.00004) −0.409⁎⁎⁎
−0.317⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0532 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0399⁎ (0.0206) 0.0537 (0.0505) 0.133⁎⁎⁎
(0.0205) (0.0205) 0.0251 0.0251 (0.0275) (0.0275) Reference group: no partner ⁎⁎⁎ ⁎⁎⁎ 0.302 0.302 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) (0.0219) (0.0219) Yes Yes Yes Yes Yes Yes 0.0982 0.0982 0.0982 39,679 39,679 39,679 239,209 239,209 239,209
(0.0204) 0.0401⁎ (0.0206) 0.0538 (0.0505) 0.133⁎⁎⁎
(0.0204) 0.0400⁎ (0.0206) 0.0539 (0.0505) 0.133⁎⁎⁎
(0.0205) (0.0205) (0.0205) 0.0256 0.0253 0.0253 (0.0275) (0.0275) (0.0275) Reference group: no partner ⁎⁎⁎ ⁎⁎⁎ 0.302 0.302 0.302⁎⁎⁎ (0.0169) (0.0169) (0.0169) 0.437⁎⁎⁎ 0.437⁎⁎⁎ 0.437⁎⁎⁎ (0.0219) Yes Yes 0.0982 39,679 239,209
(0.0219) Yes Yes 0.0982 39,679 239,209
(0.0219) Yes Yes 0.0982 39,679 239,209
Standard errors in parenthesis are adjusted for clustering at the personal level. ⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.
might just be the case that we do not have enough observations of affected individuals in order to detect a significant effect. Therefore, our results do not justify the conclusion that floods do not have an effect on subjective well-being in the short term. In a sensitivity analysis we widened the time frame of flood events to 24 months which produced a coefficient that was lower in magnitude but still slightly significant. Another difference between floods and storm & hail events refers to the insurance density. Although we observed in the past that flood damages are sometimes compensated by public authorities, the effect of floods on subjective well-being may be more persistent because the aggrieved party is rather in an insecure position with regard to compensation. Insurances may have mitigating effects in this regard which is why we introduced regional insurance density rates in interaction with a dummy of flood events (Model II). Now the flood variable becomes significant in the short time frame of 6 months where we also find a positive coefficient of the interaction term. Even if other compensation schemes might blur the picture, this result indicates that
insurances can at least partly offset negative well-being effects of flood events. However, this effect does not persist in the 12 and 18 months time frame. In Models III and V we introduced a continuous variable of damage frequency using the midpoints of the intervals shown in Fig. 1. The coefficient yields only significant results in case of storm & hail events. This result indicates that while well-being decreases when the area was flooded, the degree of regional detriment does not seem to matter as much. In Model VI, a common variable for extreme weather events is introduced which comprises both storm & hail events and floods. Here we find a negative effect of extreme weather events which decreases in magnitude and significance by widening the time frame. Tables 5 and 6 include some further estimations analysing the robustness of the results with respect to flood events. Model VII describes the results of a macro analysis where information on life satisfaction is
114
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
Table 5 Macro analysis (source: SOEPv28, author's calculations).
Variables Flood dummy Year dummies Region fixed Effects R2 Number of observations
Model VII (a) 6 months (b) 12 months 18 months −0.0311 (0.0655) Yes Yes
−0.0458⁎ (0.0268) Yes Yes
−0.0393 (0.0241) Yes Yes
0.602 4785
0.602 4785
0.602 4785
⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.
aggregated on NUTS 3 level and regressed on the occurrence of flood events. Regional and time fixed effects are captured by introducing dummies for each year and NUTS 3 region, respectively. The
estimation results are only weakly significant in the 12 months specification of flood events. This might indicate that levelling out all regional variation (e.g. on flood risks or insurance density) implies a loss in information. In Model VIII we excluded people from the analysis who moved across NUTS 3 borders because migration may be linked to the occurrence of extreme weather events or movers may be falsely counted as affected if they moved into a flooded regions within the 6/12/18 months following the event. This reduces our sample size from 239,209 to 211,313 observations. The results become slightly lower in magnitude and significance. This may be due to the smaller sample size or because the most aggrieved move away and thus drop out of our sample. Following, in Model IX we then clustered observations on NUTS 3 level instead of the personal level because observations from the same region cannot be considered as independent. This further reduces the significance of our results. Here, only the 18 month specification is significant on the 10%-level.
Table 6 Robustness checks (source: SOEPv28, author's calculations). Variables Flood variables Number of events
Model VIII Model IX Model X (a) 6 months (b) 12 months (c) 18 months (a) 6 months (b) 12 months (c) 18 months (a) 6 months (b) 12 months (c) 18 months 0.000360 (0.0173)
−0.0172⁎ (0.00976)
−0.0193⁎⁎ (0.00881)
0.000360 (0.0202)
−0.0172 (0.0125)
−0.0193⁎ (0.0111)
Dummy Owner Owner × dummy Log. Equiv. Household income 0.301⁎⁎⁎ (0.0155) Household size 0.0205⁎⁎⁎
0.301⁎⁎⁎ (0.0155) 0.0206⁎⁎⁎
0.301⁎⁎⁎ (0.0155) 0.0206⁎⁎⁎
0.301⁎⁎⁎ (0.0185) 0.0205⁎⁎
0.301⁎⁎⁎ (0.0185) 0.0206⁎⁎
0.301⁎⁎⁎ (0.0185) 0.0206⁎⁎
(0.00770) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.00770) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.00631) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.00631) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.00631) 0.0000 (0.00004) −0.433⁎⁎⁎
(0.0323)
(0.0363) (0.0363) Reference group: very good −0.310⁎⁎⁎ −0.310⁎⁎⁎
Person needing care in HH
(0.00770) 0.0000 (0.00004) −0.433⁎⁎⁎
Health status Good
(0.0323) (0.0323) Reference group: very good −0.310⁎⁎⁎ −0.310⁎⁎⁎
Age squared
(0.0363)
0.00374 (0.0233) 0.0237 (0.0161) −0.0373 (0.0304) 0.262⁎⁎⁎
0.00760 (0.0144) 0.0286⁎ (0.0162) −0.0528⁎⁎⁎ (0.0172) 0.263⁎⁎⁎
−0.00362 (0.0133) 0.0277⁎ (0.0163) −0.0354⁎⁎ (0.0159) 0.263⁎⁎⁎
(0.0136) −0.0542⁎⁎⁎ (0.00707) 0.0000 (0.00004) −0.412⁎⁎⁎
(0.0136) −0.0543⁎⁎⁎ (0.00707) 0.0000 (0.00004) −0.412⁎⁎⁎
(0.0136) −0.0542⁎⁎⁎ (0.00707) 0.0000 (0.00004) −0.412⁎⁎⁎
(0.0310) (0.0310) Reference group: very good −0.317⁎⁎⁎ −0.317⁎⁎⁎
Poor
(0.0127) −0.712⁎⁎⁎ (0.0150) −1.248⁎⁎⁎
(0.0127) −0.712⁎⁎⁎ (0.0150) −1.248⁎⁎⁎
−0.310⁎⁎⁎ (0.0127) −0.712⁎⁎⁎ (0.0150) −1.248⁎⁎⁎
(0.0133) −0.712⁎⁎⁎ (0.0164) −1.248⁎⁎⁎
(0.0133) −0.712⁎⁎⁎ (0.0163) −1.248⁎⁎⁎
−0.310⁎⁎⁎ (0.0133) −0.712⁎⁎⁎ (0.0163) −1.248⁎⁎⁎
(0.0117) −0.723⁎⁎⁎ (0.0139) −1.265⁎⁎⁎
Bad
(0.0188) −2.242⁎⁎⁎
(0.0188) −2.242⁎⁎⁎
(0.0188) −2.242⁎⁎⁎
(0.0217) −2.242⁎⁎⁎
(0.0217) −2.242⁎⁎⁎
(0.0217) −2.242⁎⁎⁎
(0.0177) −2.260⁎⁎⁎
Satisfactory
(0.0117) −0.723⁎⁎⁎ (0.0139) −1.265⁎⁎⁎
(0.0310) −0.317⁎⁎⁎ (0.0117) −0.723⁎⁎⁎ (0.0139) −1.265⁎⁎⁎
Unemployed
(0.0340) (0.0340) Reference group: employed −0.0676⁎⁎⁎ −0.0677⁎⁎⁎ −0.0677⁎⁎⁎ (0.0213) (0.0213) (0.0213) −0.475⁎⁎⁎ −0.475⁎⁎⁎ −0.475⁎⁎⁎
(0.0399) (0.0399) Reference group: employed −0.0676⁎⁎⁎ −0.0677⁎⁎⁎ −0.0677⁎⁎⁎ (0.0221) (0.0221) (0.0221) −0.475⁎⁎⁎ −0.475⁎⁎⁎ −0.475⁎⁎⁎
(0.0177) (0.0177) −2.260⁎⁎⁎ −2.260⁎⁎⁎ (0.0326) (0.0326) (0.0326) Reference group: employed −0.0590⁎⁎⁎ −0.0592⁎⁎⁎ −0.0591⁎⁎⁎ (0.0197) (0.0197) (0.0197) −0.494⁎⁎⁎ −0.494⁎⁎⁎ −0.494⁎⁎⁎
Pensioner
(0.0217) 0.0441⁎⁎
(0.0219) (0.0219) 0.0441⁎ 0.0442⁎ (0.0241) (0.0241) 0.0594 0.0597 (0.0575) (0.0574) ⁎⁎⁎ 0.142 0.142⁎⁎⁎ (0.0256) (0.0255) 0.0148 0.0147 (0.0350) (0.0350) Reference group: no partner 0.294⁎⁎⁎ 0.294⁎⁎⁎ (0.0188) (0.0188) 0.442⁎⁎⁎ 0.443⁎⁎⁎ (0.0260) (0.0260) Yes Yes Yes Yes 0.100 0.100 36,382 36,382 211,313 211,313
(0.0204) (0.0204) 0.0403⁎ 0.0406⁎⁎ (0.0206) (0.0206) 0.0417 0.0415 (0.0505) (0.0506) 0.124⁎⁎⁎ 0.124⁎⁎⁎ (0.0205) (0.0205) 0.0256 0.0253 (0.0275) (0.0275) Reference group: no partner 0.301⁎⁎⁎ 0.301⁎⁎⁎ (0.0169) (0.0169) 0.417⁎⁎⁎ 0.417⁎⁎⁎ (0.0221) (0.0221) Yes Yes Yes Yes 0.0980 0.0980 39,679 39,679 239,205 239,205
(0.0340)
Employment status Not employed
Military, community service In education Self-employed Partner status Partner outside household Partner inside household Year dummies Individual fixed effects R 2(within) Number of individuals Number of observations
(0.0217) 0.0442⁎⁎ (0.0211) (0.0211) 0.0594 0.0597 (0.0572) (0.0572) ⁎⁎⁎ 0.142 0.142⁎⁎⁎ (0.0243) (0.0243) 0.0148 0.0147 (0.0306) (0.0306) Reference group: no partner 0.294⁎⁎⁎ 0.294⁎⁎⁎ (0.0193) (0.0193) 0.442⁎⁎⁎ 0.443⁎⁎⁎ (0.0262) (0.0262) Yes Yes Yes Yes 0.100 0.100 36,382 36,382 211,313 211,313
Clustered standard errors in parantheses. ⁎ p b 0.1. ⁎⁎ p b 0.05. ⁎⁎⁎ p b 0.01.
(0.0217) 0.0442⁎⁎ (0.0211) 0.0600 (0.0572) 0.142⁎⁎⁎ (0.0243) 0.0148 (0.0306) 0.294⁎⁎⁎ (0.0193) 0.443⁎⁎⁎ (0.0262) Yes Yes 0.100 36,382 211,313
(0.0399)
(0.0219) 0.0442⁎ (0.0241) 0.0600 (0.0574) 0.142⁎⁎⁎ (0.0255) 0.0148 (0.0350) 0.294⁎⁎⁎ (0.0188) 0.443⁎⁎⁎ (0.0260) Yes Yes 0.100 36,382 211,313
(0.0204) 0.0404⁎⁎ (0.0206) 0.0419 (0.0505) 0.124⁎⁎⁎ (0.0205) 0.0255 (0.0275) 0.301⁎⁎⁎ (0.0169) 0.417⁎⁎⁎ (0.0221) Yes Yes 0.0980 39,679 239,205
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
In Model X we estimated an interaction effect for house owners so as to test whether they are differently affected by flood events than renters. In this model, the dummy becomes insignificant, while the interaction effect is negative and (strongly) significant in the 12 and 18 month specification. These results indicate that negative effects from floods are particularly adhering to house owners. 4.2. Discussion The results of the baseline regressions (Model I, IV and VI) indicate negative well-being externalities of extreme weather events suggesting that life satisfaction of residents living in affected NUTS 3 regions drops by about 0.0206–0.0268 on the 11-point scale following an extreme event. Though statistically significant, this value is rather low as compared to other studies (see e.g. Luechinger and Raschky, 2009). Also one might have expected a stronger effect for floods as compared to storm & hail events. There are several reasons that might explain these results. It becomes apparent from Fig. 1 that storms struck rather area-wide while impacts of floods are usually limited to particular areas. On the whole this means that we have less observations of respondents who lived in a NUTS 3 region that has been flooded as compared to respondents that lived in an area hit by a storm. This can be seen from the estimation results in Model I where the flood coefficient becomes only significant if we take into account all events of the past 12 or 18 months. In other respects it could also be the case that people living in flood prone areas adapt to flood events in the course of time so that flood impacts do not cause as much grief among the affected population any more. Last but not least, we observe a large variation of flood damages within NUTS 3 regions. This means that we are likely to have many people in the sample that were not flooded themselves but are considered as affected because they lived in a NUTS 3 region that has faced damages due to a flood. This might cause an underestimation of the true wellbeing effect of flood events. Thus, it is the limitation of our analysis that we used administrative borders in order to determine whether an individual was affected by an event, or not. A more precise distinction between the really affected and the non-affected as well as between the compensated and non-compensated respondents might reveal stronger effects than we could identify in our current study. To study mitigating effects of insurances, we incorporated federal data on natural hazard insurance densities in Model II. Insurance rates vary widely across federal states; while Baden-Württemberg has 91% of all buildings in the highest risk zone insured against natural hazard events, the corresponding value for Lower Saxony is only 7%. The results indicate that well-being losses due to extreme events are significantly reduced with higher insurance rates. As NUTS 3 regions can be affected with varying severity, we introduced damage frequency as a continuous variable in Models III and V for floods and storm & hail events, respectively. While the coefficients for storm & hail events are (strongly) significant in all variations, there is no significant effect for floods. This result indicates that the degree of regional detriment does not seem to play the major role in comparison to the question if a respondent was affected or not. Furthermore, we tested whether the intensity of an event is important in explaining wellbeing losses by including the average claims amount of each event (see Table 1) as an interaction term but could not determine any significant results either. Most likely, our database is too coarse in order to refine the analysis in this regard. In our analysis we combined data on individual and NUTS 3 level. As observations from the same individual over time or from individuals of the same NUTS 3 region can be correlated, it is necessary to calculate robust standard errors. In the baseline regressions we report standard errors that are clustered according to the individual level. As can be seen from the estimation results in Model IX, clustering on regional level reduces the significance of the results by one level. Adjusting standard errors for clustering on the regional level in the other model
115
specifications shows the same tendency. Similarly to the macro analysis, this result indicates that (indirectly) accounting for regional attributes, such as flood risk or insurance density, reduces the significance of the effect. Finally, the estimation results of Model X show that well-being losses are particularly adhering to house owners. This suggests that well-being losses rather originate from damages on (immobile) property that needs to be renovated by the house owners, a fact that means considerable financial and psychological stress for them and thus cause well-being losses. In contrast, appliances and personal belongings of renters are rather mobile and may potentially be saved from destruction to a larger share.
5. Conclusion Despite compensation through insurances and public post-hoc disaster relief funds, life satisfaction is significantly negatively affected by extreme weather events. In our analysis, we found that floods and storm & hail events decrease life satisfaction by about 0.02 on the 11-point scale. Due to the study design it may well be that we underestimate the effect of extreme events on well-being because we cannot distinguish respondents which have been personally affected from those who lived in a affected region but did not incur any damages themselves. Counting the latter as affected might cause a downwards bias in the estimated effect. For our analysis we used a rich dataset based on the German Socio-Economic Panel (SOEP) and regionally disaggregated data on extreme weather events and insurance density. By enhancing the database with regard to the aforementioned distinction between affected and unaffected as well as compensated and non-compensated individuals, the informative value of such an analysis could be further increased. Nevertheless, our results underline that there remain negative wellbeing effects that are not compensated for by existing monetary compensation schemes—be it arranged on private markets (insurances) or publicly provided (disaster funds). These effects are usually not accounted for in conventional cost–benefit analyses. Decisions on the design and quantity of investments into flood and storm protection measures might therefore neglect a share of the benefits that could be generated by stronger efforts to improve protection, adaptation and overall risk management. With climate change and the possible aggravation of extreme weather events and the ongoing increase of damage potentials within risk prone areas, immaterial values should be carefully taken into consideration in risk management as well as in the design of climate change mitigation and adaptation policies. To that effect, our analysis showed that the loss of life satisfaction due to extreme events is significantly reduced with higher insurance rates. This poses a strong argument to intensifying efforts to increase insurance rates covering extreme events or even introduce mandatory insurance schemes. The life satisfaction approach presents an interesting extension to common methods of damage assessment or cost-benefit analysis as it allows for the valuation of intangible values. With its methodology being still further elaborated, it can already serve to supplement insights from economic valuations and to point out the relevance of other value dimensions (see Fujiwara and Campbell, 2011).
Acknowledgments This article gives an account of results from the project Economics of Climate Change Adaptation (econCCadapt) funded by the German Federal Ministry of Education and Research (FKZ 01LA1137A). Furthermore, we would like to thank Heinz Welsch, Eugen Pissarskoi and two anonymous referees for their helpful comments.
116
Möllendorff, C. von, Hirschfeld, J. / Ecological Economics 121 (2016) 108–116
Appendix A
References
Table A.I: Summary statistics of SOEP data. Variable
Description
Mean
Life satisfaction Equivalized income (in €) HH size Age Person needing care in HH Very good health Good health Satisfactory health Poor health Bad health Employed Not employed Unemployed Pensioner Military/community service In education Self employed No partner Partner living in HH Partner living outside HH
11-point scale Log. of equivalized net monthly HH income Number of persons in HH Years Dummy variable
S.D.
Min Max
6.99 7.21
1.78 0 0.51 0
2.72 47.93 0.04
1.28 1 17.65 16 0.20 0
10 11.12 14 101 1
Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable
0.10 0.40 0.33 0.13 0.04 0.48 0.07 0.06 0.24 0.003
0.29 0.49 0.47 0.34 0.19 0.50 0.26 0.23 0.43 0.06
0 0 0 0 0 0 0 0 0 0
1 1 1 1 1 1 1 1 1 1
Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable
0.08 0.06 0.22 0.7 0.08
0.27 0.24 0.41 0.46 .27
0 0 0 0 0
1 1 1 1 1
Abbreviation: HH = household.
Table A.II: Summary statistics of event variables. Variable
Description
Mean
S.D.
Min
Max
Flood events (6 months) Flood events (12 months) Flood events (18 months) Storm & hail events (6 months) Storm & hail events (12 months) Storm & hail events (18 months) Extreme events (6 months) Extreme events (12 months) Extreme events (18 months) Flood events (6 months) Flood events (12 months) Flood events (18 months) Storm & hail events (6 months) Storm & hail events (12 months) Storm & hail events (18 months) Extreme events (6 months) Extreme events (12 months) Extreme events (18 months)
Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Dummy variable Number of events Number of events Number of events Number of events Number of events Number of events Number of events Number of events Number of events
0.03 0.11 0.14 0.25 0.33 0.55 0.26 0.38 0.58 0.03 0.11 0.15 0.25 0.34 0.61 0.28 0.46 0.76
0.17 0.31 0.34 0.43 0.47 0.50 0.44 0.48 0.49 0.17 0.35 0.39 0.44 0.49 0.59 0.50 0.65 0.77
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Andreß, H.-J., Golsch, K., Schmidt, A.W., 2013. Applied Panel Data Analysis for Economic and Social Surveys. Springer, Berlin; Heidelberg. Baetschmann, G., Staub, K.E., Winkelmann, R., 2014. Consistent estimation of the fixed effects ordered logit model. J. R. Stat. Soc. A. Stat. Soc. (n/a–n/a). Carroll, N., Frijters, P., Shields, M., 2009. Quantifying the costs of drought: new evidence from life satisfaction data. J. Popul. Econ. 22 (2), 445–461. Ciscar, J.-C., Soria, A., Goodess, C., Christensen, O., Iglesias, A., et al., 2009. Climate change impacts in Europe. Final report of the PESETA research project JRC Scientific and Technical Reports. Dehnhardt, A., Hirschfeld, J., Drünkler, D., Petschow, U., Engel, H., Hammer, M., 2008. Kosten-Nutzen-Analyse von Hochwasserschutzmaβnahmen. Forschungsbericht 204 21 212. UBA Texte 31/2008, Umweltbundesamt, Dessau-Roßlau. Di Tella, R., MacCulloch, R.J., Oswald, A.J., 2001. Preferences over inflation and unemployment: evidence from surveys of happiness. Am. Econ. Rev. 91 (1), 335–341. Diener, E., Lucas, R.E., 1999. Personality and subjective well-being. In: Diener, E., Kahneman, D., Schwarz, N. (Eds.), Well-Being: The Foundations of Hedonic Psychology. Russell Sage Foundation. Ferreira, S., Moro, M., 2010. On the use of subjective well-being data for environmental valuation. Environ. Resour. Econ. 46 (3), 249–273. Ferrer-i Carbonell, A., Frijters, P., 2004. How important is methodology for the estimates of the determinants of happiness? Econ. J. 114 (497), 641–659. Frijters, P., Haisken-DeNew, J.P., Shields, M.A., 2004. Money does matter! Evidence from increasing real income and life satisfaction in east Germany following reunification. Am. Econ. Rev. 94 (3), 730–740. Fujiwara, D., Campbell, R., 2011. Valuation Techniques for Social Cost-Benefit Analysis: Stated Preference, Revealed Preference and Subjective Well-Being Approaches: A Discussion of the Current Issues. HM Treasury. GDV, 2012. Naturgefahrenreport 2012. Gesamtverband der Deutschen Versicherungswirtschaft e.V, GDV, Berlin. GDV, 2013. Versicherungsdichte “Elementargefahren”. Gesamtverband der Deutschen Versicherungswirtschaft e.V, GDV, Berlin. Kimball, M., Levy, H., Ohtake, F., Tsutsui, Y., 2006. Unhappiness after Hurricane Katrina. Working paper 12062. National Bureau of Economic Research. Kountouris, Y., Remoundou, K., 2011. Valuing the welfare cost of forest fires: a life satisfaction approach. Kyklos 64 (4), 556–578. Kreibich, H., Christenberger, S., Schwarze, R., 2011. Economic motivation of households to undertake private precautionary measures against floods. Nat. Hazards Earth Syst. Sci. 11 (2), 309–321. Luechinger, S., Raschky, P.A., 2009. Valuing flood disasters using the life satisfaction approach. J. Public Econ. 93 (3–4), 620–633. Maddison, D., Rehdanz, K., 2011. The impact of climate on life satisfaction. Ecol. Econ. 70 (12), 2437–2445. Moulton, B.R., 1990. An illustration of a pitfall in estimating the effects of aggregate variables on micro units. Rev. Econ. Stat. 72 (2), 334–338. Osberghaus, D., Kühling, J., 2014. Direct and indirect effects of weather experiences on life satisfaction: which role for climate change expectations? ZEW Discussion Paper. ZEW—Zentrum für Europäische Wirtschaftsforschung/Center for European Economic Research. Schupp, J., Kroh, M., Goebel, J., Bartsch, S., Giesselmann, M., et al., 2014. Sozio-oekonomisches Panel (SOEP), Daten der Jahre 2000–2011. Version: 28. Dataset, SOEP—Soziooekonomisches Panel. http://doi.org/10.5684/soep.v28doi:10.5684/soep.v28 Seifert, I., Botzen, W.J.W., Kreibich, H., Aerts, J.C.J.H., 2013. Influence of flood risk characteristics on flood insurance demand: a comparison between Germany and The Netherlands. Nat. Hazards Earth Syst. Sci. 13 (7), 1691–1705. Tapsell, S., Penning-Rowsell, E., Tunstall, S., Wilson, T., 2002. Vulnerability to flooding: health and social dimensions. Proc. R. Soc. London, Ser. A 360 (1796), 1511–1525. Wagner, G.G., Frick, J.R., Schupp, J., 2007. The German socio-economic panel study (soep)—scope, evolution and enhancements. Schmollers Jahr. 127 (1), 139–169. Welsch, H., Ferreira, S., 2013. Environment, well-being, and experienced preference. Int. Rev. Environ. Resour. Econ. 7, 205–239.