Journal of Economic Behavior & Organization 116 (2015) 500–517

Contents lists available at ScienceDirect

Journal of Economic Behavior & Organization

j ourna l ho me pa g e: www.elsev ier .com/ locate / jebo

Well-being effects of a major natural disaster: The case ofFukushima

Katrin Rehdanza,b,∗, Heinz Welschc, Daiju Naritaa,d, Toshihiro Okuboe

a Kiel Institute for the World Economy, Kiellinie 66, 24105 Kiel, Germanyb University of Kiel, Department of Economics, Olshausenstr. 40, 24118 Kiel, Germanyc University of Oldenburg, Department of Economics, 26111 Oldenburg, Germanyd JICA Research Institute, 10-5 Ichigaya Honmuracho, Shinjuku-ku, Tokyo 162-8433, Japane Keio University, Faculty of Economics, 2-15-45 Mita, Minato-ku, Tokyo 108-8345, Japan

a r t i c l e i n f o

Article history:Received 30 October 2013Received in revised form 7 April 2015Accepted 20 May 2015Available online 29 May 2015

JEL classification:D62Q51Q54I31

Keywords:FukushimaSubjective well-beingNatural disasterNuclear accidentDifference-in-differencesWillingness to pay

a b s t r a c t

Based on a quasi-experimental difference-in-differences approach, we use panel data for5979 individuals interviewed in Japan before and after the tsunami and nuclear accident atFukushima to analyze the effects of the combined disaster on people’s subjective well-being.To conduct our analysis, we use Geographical Information Systems to merge the subjectivewell-being data with information on respondents’ distance from the Fukushima Dai-ichinuclear power plant, their proximity to nuclear power stations in general, and the spatialdistribution of radioactive fallout after the accident. Our main findings are as follows: (1)After the disaster, people living in a place affected by the tsunami or close to the FukushimaDai-ichi power plant experienced a drop in life happiness, while the effects declined withdistance to the place of the disaster. (2) No change in subjective well-being is detectablein people living close to nuclear facilities in general. (3) In contrast to happiness with lifeafter the disaster, no effect on people’s happiness with their entire life can be found amongthose affected by the disaster. (4) The drop in life happiness in municipalities affected bythe tsunami is equivalent to 72% of annual income and reaches 240% for those living in closedistance to the Fukushima Dai-ichi power plant (≤150 km).

© 2015 Elsevier B.V. All rights reserved.

1. Introduction

On 11 March 2011, following a major earthquake off the Pacific coast of Japan, a tsunami disabled the power supply andcooling systems of three reactors of the Fukushima Dai-ichi power plant, causing a major nuclear accident. The accidenttriggered substantial releases of radioactive material and resulted in one of the worst nuclear disasters ever, second onlyto the Chernobyl disaster in 1986. The earthquake and tsunami caused nearly 16,000 deaths, over 1.2 million destroyed or

damaged buildings, and temporary evacuation from their homes for over 380,000 people. The combined event (earthquake,tsunami, nuclear accident) also disrupted water supply, power distribution, and train, highway, and air transport systemsin extensive areas of eastern Japan.

∗ Corresponding author at: Kiel Institute for the World Economy, Kiellinie 66, 24105 Kiel, Germany. Tel.: +49 431 8814 407.E-mail addresses: katrin.rehdanz@ifw-kiel.de (K. Rehdanz), heinz.welsch@uni-oldenburg.de (H. Welsch), Narita.Daiju@jica.go.jp (D. Narita),

okubo@econ.keio.ac.jp (T. Okubo).

http://dx.doi.org/10.1016/j.jebo.2015.05.0140167-2681/© 2015 Elsevier B.V. All rights reserved.

ptcpowse

vlace

ctifchpdu

onimd

tiBaitSooulen

dfamt“i

ha

2

K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517 501

The disaster at Fukushima may have impacted on subjective well-being (SWB) in a variety of ways, some of which involvehysical consequences of the disaster whereas others refer to psychological mechanisms.1 The physical consequences ofhe disaster may have led to income losses, job losses or health impairments. In addition, it may have caused psychologicalosts in terms of fear, anxiety or mental distress related to fatalities, injuries or radioactive contamination. Some of thosesychological costs need not be restricted to persons directly affected by the event. Due to media coverage, they may spillver and cause distress in other, more distant places. Notably in the case of a major nuclear accident, people may becomeorried about nuclear power in general, especially if they live close to nuclear facilities themselves. Depending on how

erious consequences were, they may have implied fundamental changes in how people evaluate happiness with theirntire life up to the present.

Based on a quasi-experimental difference-in-differences (DD) approach, we use nation-wide panel data for 5979 indi-iduals interviewed in Japan before and after the disaster to analyze its impact on people’s SWB, specified as happiness withife.2 More specifically, we study the following research questions: (1) In which ways, if any, has the Fukushima disasterffected the happiness of those affected? (2) Has the nuclear accident at Fukushima reduced the happiness of people livinglose to nuclear power stations in general? (3) Has the disaster affected not just people’s happiness with their life after thevent, but happiness with their entire life? (4) What are the monetary equivalents of the disaster’s SWB effects?

To address those questions, we use several indicators that correspond to various dimensions of the disaster and of itsonsequences. In particular, we use measures of the distance from or proximity to the place of the disaster to capture the ideahat some of its effects may decay with distance, in particular effects related to the breakdown of electricity supply and othernfrastructures and to radioactivity. To capture effects of the earthquake and tsunami more specifically, we use measures ofatalities and injured persons and of destroyed or damaged buildings by geographic unit. Radioactivity-related effects areaptured by the level of radioactive substances in different areas. To explore the hypothesis that the nuclear accident mayave spurred people’s concern over nuclear power more generally we use indicators of the presence and density of nuclearlants near the places where people live. Finally, we use an alternative measure of SWB to explore the possibility that theisaster may have affected not just people’s happiness with their life after the disaster, but happiness with their whole lifep to the present.

Our identification strategy is to use those indicators as measures of (the degree of) treatment by the different dimensionsf the disaster and to study whether people’s happiness with their lives has changed after the disaster depending on theature and extent of treatment, yielding a multi-dimensional DD design. By running happiness regressions that control for

ncome, unemployment and health, and regressions that do not control for those factors, we are able to differentiate betweenechanisms that involve the physical consequences of the disaster and psychological mechanisms (fear, anxiety, mental

istress) that may affect happiness independent of those variables.Our paper is the first to find an effect of a natural disaster on SWB within a spatial quasi-experimental DD framework and

o estimate its implicit monetary value. In connection with nuclear accidents, Almond et al. (2009) have investigated thempacts of the Chernobyl disaster on health and school outcomes, but not its effect on SWB and the monetary equivalent.erger (2010) and Goebel et al. (2013) found an increase in German citizens’ concern about the environment after Chernobylnd Fukushima, respectively, but no change in SWB. Focusing on the Fukushima disaster, Yamane et al. (2013) analyze itsmpact on local property values. Ohtake and Yamada (2013) use data elicited after the disaster to explore the spatial andemporal SWB pattern post-Fukushima but are unable to analyze the difference in SWB before and after the event.3 OtherWB studies of single disasters are Kimball et al. (2006) and Metcalfe et al. (2011). The first of these analyzes the unhappinessf Americans after hurricane Katrina, while the other studies mental distress in British citizens following the 9/11 attacksn the World Trade Center. Instead of addressing particular events, Luechinger and Raschky (2009) and Carroll et al. (2009)se a correlational design to study the relationship between floods and droughts, respectively, and SWB. We extend the

iterature on natural disasters and SWB by providing evidence of an effect of the former on the latter by (i) applying a quasi-xperimental DD approach and (ii) explicitly taking account of the spatial dimension. To our knowledge, no SWB study of aatural disaster featuring (i) and (ii) has established such an effect as yet.

To conduct our analysis, we use Geographical Information Systems to merge SWB data with information on respondents’istance from the Fukushima Dai-ichi nuclear power plant and their proximity to nuclear power stations in general. Weurther use information on the spatial distribution of radioactive fallout after the accident and identify regions that wereffected by the tsunami. Our SWB data come from the Keio Household Panel Survey (KHPS), 2011 and 2012. Our maineasure of SWB is the response to a question asking individuals about their happiness with life in the previous year. Since

he interviews in the KHPS were conducted in January of the respective year, the answers from the 2011 survey refer to thepre-Fukushima” period while those from the 2012 survey refer to the “post-Fukushima” period. In addition to life happinessn the previous year, the respondents are also asked about happiness with their entire life. In our econometric analysis we

1 We use the term disaster to denote the combined event (earthquake, tsunami, and nuclear accident).2 Among various measures of subjective well-being, the primary distinction is between cognitive life evaluations, derived from questions about howappy or satisfied people are with their lives, and emotional reports (Helliwell and Wang, 2012). The measures used in this paper refer to how happy peoplere with their lives, and we refer to them throughout the paper as ‘happiness with life’ (‘happiness’ for short).3 Keio University has published two books in Japanese containing a collection of papers based on KHPS data on SWB and the disaster (Seko et al., 2012,

013). None of these papers discuss the disaster in terms of the spatial dimension.

502 K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517

test different specifications, using one or the other as dependent variables testing whether people’s evaluation of the qualityof their lives in the previous year or of their entire lives has changed after the disaster.

Referring to the research questions stated above our results suggest the following outcome: (1) After the disaster, peopleliving in a place affected by the earthquake/tsunami or close to the Fukushima Dai-ichi power plant experienced a dropin SWB, while the SWB effects declined with distance to the place of the disaster. Effects related to radioactivity levelsin the air cannot be established, suggesting an absence of short-term radiation-related impairments. An effect through(uncompensated) income losses or increased unemployment can also not be detected. (2) No change in SWB is detectablein people living close to nuclear facilities in general. (3) While those affected by the disaster experienced a decline inhappiness with their life after the disaster, no effect on happiness with their entire life can be found. (4) The drop in lifehappiness in municipalities affected by the tsunami is equivalent to 72% of annual income and reaches 240% for those livingin close distance to the Fukushima Dai-ichi power plant (≤150 km). The SWB effect for Japan as a whole amounts to %of GDP.

The paper is organized as follows: Section 2 provides a review of the literature on economics and SWB and some infor-mation on the Fukushima nuclear accident. Section 3 presents the empirical approach and the data. Section 4 reports theresults. Section 5 discusses and concludes.

2. Background

2.1. Economics and subjective well-being

In economics, the interest in SWB has increased rapidly over the last decade (for overviews, see, e.g., Frey and Stutzer,2002; Dolan et al., 2008; Van Praag and Ferrer-i-Carbonell, 2008; MacKerron, 2012). The rationale for using data on SWB ineconomic analysis is that they are considered to be an empirical approximation of what Kahneman et al. (1997) have labeled“experienced utility.”

Research on SWB has identified a number of personal, demographic, and socio-economic covariates that explain observedSWB (see e.g. Clark et al., 2008; Dolan et al., 2008; Frey and Stutzer, 2002). Important factors at the societal level aremacroeconomic conditions (unemployment rate, inflation rate), institutional conditions (political freedom, democracy, ruleof law), public bads (terrorism, civil war, corruption), and environmental (dis)amenities. The unemployment rate and theinflation rate affect SWB negatively (Di Tella et al., 2001), whereas good institutional quality yields greater SWB (Frey andStutzer, 2000). Public bads such as terrorism, civil war, and corruption have sizeable negative effects on happiness (Freyet al., 2009; Welsch, 2008a, 2008b, respectively). With regard to terrorism, Metcalfe et al. (2011) have shown that the 9/11attacks on the World Trade Center even had sizeable and statistically significant effects on SBW in the UK, a place remotefrom the event.

With regard to environmental (dis)amenities, the issues addressed so far represent a considerable range of environmentalproblems and natural disasters. They comprise air pollution (Welsch, 2002, 2006; Luechinger, 2009; MacKerron and Mourato,2009; Ferreira and Moro, 2010; Levinson, 2012), airport noise (Van Praag and Baarsma, 2005), climate parameters (Rehdanzand Maddison, 2005; Maddison and Rehdanz, 2011; Murray et al., 2013), flood events (Luechinger and Raschky, 2009), anddrought events (Carroll et al., 2009). Kimball et al. (2006) investigated changes in happiness for US adults after hurricaneKatrina. The results of their descriptive analysis indicate a nation-wide effect of the event. All of these studies establishedthat SWB is positively related to environmental quality and negatively related to several kinds of natural disaster. However,no effect of the Fukushima disaster on SWB has been established as yet (Goebel et al., 2013; Ohtake and Yamada, 2013).

In addition to measuring effects of (dis)amenities on SWB, well-being regressions have been used as a non-market val-uation tool. Since, in addition to the non-market goods or extreme events under study, they include income among theexplanatory variables, well-being regressions permit calculating the monetary equivalents of those goods or events. Thatis, the change in income that would have the same effect on SWB as a change in the variable considered. For marginalchanges in the respective variable, the monetary measure is the marginal rate of substitution of income for that vari-able while for non-marginal changes it is the compensating variation (see Welsch and Ferreira, 2014 for a discussionand review).

2.2. The Fukushima disaster

The earthquake we are concerned with here had a magnitude of 9.0 and was the fourth largest event of its kind anywherein the world since 1990. It occurred on 11 March 2011 off the coast of Miyagi prefecture in the north-eastern region of Japan.Soon after the earthquake, the Fukushima Dai-ichi nuclear power plant, which is sited close to the Pacific Ocean and 180 kmaway from the epicenter, experienced failures in the cooling systems of the reactors, and four of its six reactors releasedsubstantial quantities of radioactive material in the aftermath of meltdowns and gas explosions. These releases blew morethan 500,000 TBq (terabecquerel) of highly toxic radioactive material (iodine-131, cesium-134, and cesium-137) into the

K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517 503

FtS

ai

1h(on

coCtouss

tTumc

p

ig. 1. Distribution and location of nuclear power plants in Japan. Note: Information on the location of existing nuclear power plants was obtained fromhe National Land Numerical Information Download Service (http://nlftp.mlit.go.jp/ksj-e/gml/gml datalist.html, last accessed, 29-04-2013).ource: Own presentation.

tmosphere and the ocean.4 The accident was one of the worst nuclear disasters ever, second only to the Chernobyl disastern 1986.

According to the Reconstruction Agency,5 the combined disaster (earthquake, tsunami, nuclear accident) caused nearly6,000 deaths, over 1.2 million destroyed or damaged buildings, temporary evacuation of over 380,000 people from theiromes, most of them residents of the Iwate, Miyagi, and Fukushima prefectures on the north-east coast of the Pacific Oceansee Fig. 1). It also disrupted water supply, power distribution, and train, highway, and air transport systems in muchf eastern Japan. The reconstruction of infrastructure has been partly hindered by radioactive contamination around theuclear power plant.

Since the nuclear accident, no deaths from radiation exposure have been reported. But radioactivity has added a spe-ial dimension to the problem.6 To keep radiation exposure to a minimum, all residents within an approx. 20-km radiusf the Fukushima Dai-ichi power plant were forced to leave their homes (approximately 113,000 as estimated by theabinet Office in February 2012). Due to contamination of radioactive iodine, many local health authorities, includinghat of Tokyo, 220 km away from the Fukushima site, issued recommendations not to give tap water to infants. Mostf those warnings were however lifted within a month of the disaster. Also, radioactive contamination of farming prod-cts was widely detected in Fukushima and the neighboring prefectures. Those products did not get onto the market, butuch cases of contamination were a source of economic harm for farmers and also provoked public anxieties about foodafety.

Before March 2011, nuclear power had provided about 30% of electricity in Japan. After the Fukushima disas-er, all the nuclear power plants in the country were taken off stream one by one for comprehensive safety tests.he electricity supply was particularly throttled in summer 2011, when demand was particularly high due to these of air conditioning. Electricity rationing was imposed on major industrial customers (>500 kW), and the govern-ent also mounted a large-scale public campaign aimed at persuading individual customers to cut down on electricity

onsumption.

Fig. 1 shows the distribution and location of the 21 nuclear power stations in Japan, highlighting Fukushima Dai-ichi. The

ower plants are distributed across the whole country. All of them are located near the coast, and there is some clustering.

4 http://www.tepco.co.jp/cc/press/betu12 j/images/120524j0105.pdf.5 http://www.reconstruction.go.jp/topics/000046.html.6 The facts presented in the following come from the Nuclear Emergency Response Headquarters (2011).

504 K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517

3. Methodology and data

3.1. General approach and empirical strategy

As described above, the disaster at Fukushima had a variety of effects that may have impacted on subjective well-being.The earthquake and tsunami caused injuries and fatalities, destroyed or damaged buildings and infrastructures, and mayhave led to income losses or job losses due to the destroyed structures. The nuclear accident may have affected peoplethrough radioactivity-related health impairments or the fear of such impairments to arise in the future. In addition, thenuclear accident may have caused income losses through the radioactive contamination of produce and fish and throughthe breakdown of the electricity supply system.

We capture those effects through several indicators that correspond to various dimensions of the disaster. Some of theeffects of the earthquake/tsunami and the nuclear accident can be expected to decay with distance to the place of the events,in particular effects related to the breakdown of electricity supply and other infrastructures and to radioactivity. Therefore,we use proximity (or distance) to Fukushima Dai-ichi as one of our indicators for the effects of the combined event. Tocapture effects of the tsunami more specifically, we use measures of fatalities and injured persons and of destroyed ordamaged buildings by geographic unit. Radioactivity-related effects will be captured by the level of radioactive substancesin different municipalities.

In addition to the direct effects of the nuclear accident, we explore the possibility that it may have spurred concern overnuclear power in those not directly affected, and we test this hypothesis using indicators of the presence and density ofnuclear plants near the places where people live. Finally, we explore the possibility that the (combined) disaster may haveaffected not just people’s happiness with their life after the disaster, but their happiness with their entire life.

The possible effects of the disaster may be of a tangible or an intangible (psychological) nature. Well-being regressions cancapture both categories to the extent that the tangible effects (related to losses of assets and income) are not compensated(by insurance or government payments). We, therefore, experiment with regressions that omit income in order to testwhether or to what extent uncompensated tangible effects exist. To learn more about the channels through which SWB isaffected, we also experiment with omitting people’s employment and health status.

We use a quasi-experimental DD approach to measure the effect of the disaster on life happiness whilst controlling for arange of other factors. The DD design enables us to isolate the effect attributable to the disaster from other contemporaneousvariables (e.g. macroeconomic changes), since the control group experiences some or all of the contemporaneous influencesthat affect life happiness in the treatment group without being affected by the event. In some of our specifications, we use acontinuous treatment design, i.e., we assume that SWB effects decrease in a continuous fashion with the distance from theplace of the event; for further discussion on individual specifications see below.

The general model employed is

SWBijt = ̨ + ˇ1 log Iit + ˇ2Hit + ˇ3Gjt + ˇ4Ei + ˇ5Yt + ˇ6(Ei × Yt) + εijt (1)

where SWBijt is the happiness with life of respondent i in location j at time t, Iit is respondent i’s annual household income,Hit represents other socioeconomic and demographic characteristics, and Gjt is region-specific information. Ei denotes eventvariables indicating whether or to what extent an individual has been affected (”treated”) by the disaster, Yt is a dummyvariable representing the year of the interview. For reasons of data availability (see below), our analysis is restricted to twoyears. We set Yt = 1 if the interview took place in the year after the disaster, Yt = 0 otherwise. The symbol ε represents theerror term.

The interaction term between the event (treatment) variable and the year of the interview tells us how the disaster mayhave changed the life happiness of persons affected by the disaster. Specifically, a negative coefficient on the interactionterm indicates that life happiness has declined in those affected by the disaster but not in others or, in the case of continuoustreatment, that life happiness has declined in proportion to the intensity of treatment.

Within this general framework we estimate a set of different specifications that differ by the way the various dimensionsof treatment are captured. We follow a general-to-specific approach, starting with specifications that include distance orproximity measures which, as explained above, may reflect impacts of the earthquake/tsunami and the nuclear accidentcombined. This will be followed by specifications that focus on the earthquake/tsunami and, respectively, nuclear accidentmore specifically. In addition to the main specifications, we test whether our results are robust to alternative specificationsconsidering e.g. selection and relocation effects.

Regarding distance, our first step was to map the location of existing nuclear power stations using GIS (see Fig. 1 above).Next, we measured the distance of each existing nuclear power station from each municipality.7 Of particular interest was,of course, the distance from the Fukushima Dai-ichi power station. For our sample, this distance varies between 57 and1771 km, with 487 km as the average.

To estimate the impact of the accident on life happiness in terms of distance measured, we use the following equation:

SWBijt = ̨ + ˇ1 log Iit + ˇ2Hit + ˇ3Gjt + ˇ4Dj + ˇ5Yt + ˇ6(Dj × Yt) + εijt (2)

7 We used the centroid of each municipality.

wtiaTpte

i

ppet

i

(fir

amts

wc

warhdi

a

icfnh

2aet

K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517 505

here the key variable of interest, Dj, represents the distance from the Fukushima Dai-ichi power plant (Ei = Dj). The parame-er ˇ4 measures whether people’s life happiness differs according to this distance. Since this specification presumes that thentensity of being “treated” by the disaster is distance-dependent, insignificance of this parameter indicates that there is no

priori difference between the life happiness of people treated more intensively and that of people treated less intensively.he parameter ˇ5 indicates the idiosyncratic difference in life happiness in the year after and before the accident. The crucialarameter is ˇ6. It measures the change (after/before) in life happiness in the less affected (more distant) regions relativeo strongly affected (less distant) regions. It is expected to be positive: SWB decreased less in people more distant from thevent than in people closer to the event.

An alternative functional form for capturing location relative to the Fukushima site includes a measure of proximity ands given by

SWBijt = ̨ + ˇ1 log Iit + ˇ2Hit + ˇ3Gjt + ˇ4

(1Dj

)+ ˇ5Yt + ˇ6

(1Dj

× Yt

)+ εijt (3)

This functional form presumes that the effect declines with distance in a hyperbolic fashion (Ei = 1/Dj), whereas therevious formula involves linear decay with distance. The parameters are to be interpreted in an analogous way as in therevious specification. Specifically, ˇ6 is expected to be negative: life happiness decreased more in people closer to thevent than in people more distant from the event. We use a Box–Cox transformation to determine the most appropriateransformation of distance measured as continuous variable (see below).

As an alternative to specifying distance or proximity as continuous variables, we measure distance by dummies thatndicate mutually exclusive radius rings around the place of the event:

SWBijkt = ̨ + ˇ1 log Iit + ˇ2Hit + ˇ3Gjt +∑

k

ˇ4kRik + ˇ5Yt +∑

k

ˇ6k(Rik × Yt) + εijt (4)

In this specification Rik is a dummy variable that takes the value 1 if individual i lives in radius ring k and 0 otherwiseEi = Rik). In contrast to the continuous specification of treatment (in terms of distance or proximity), this approach requiresxing one distance category K as the (untreated) control group and omitting the respective interaction term RiK × Yt. Theemaining parameters ˇ6k measure the treatment effect in radius ring k. They are expected to be negative.

It should be noted that the continuous specifications (2) and (3) and the discrete specification (4) both have their strengthsnd weaknesses. While specification (4) requires assuming that one distance category is unaffected by the disaster, it avoidsaking assumptions on the functional form of the happiness-distance relationship. Specifications (2) and (3) involve assump-

ions on the functional form but avoid imposing an unaffected control category. In this sense, the two approaches should beeen as complementary to each other.

In order to focus on the earthquake/tsunami more specifically and to identify how severely a region was affected by it,e collected information on the number of missing, injured, or dead people in that region as well as the number of buildings

ompletely or partly destroyed. This information is available at municipality level.8

To estimate the impact of the tsunami on life happiness we use the following equation:

SWBijt = ̨ + ˇ1 log Iit + ˇ2Hit + ˇ3Gjt + ˇ4Tj + ˇ5Yt + ˇ6(Tj × Yt) + εijt (5)

here the key variable of interest, Tj, represents a dummy variable denoting 1 if the individual was living in a municipalityffected by the tsunami and 0 otherwise (Ei = Tj). Specifically, the value is 1 if one or more of the following impacts wereeported: dead or injured persons, destroyed or damaged buildings. The parameter ˇ4 represents the a priori difference in lifeappiness in the municipalities affected by the tsunami over and against unaffected municipalities, whilst ˇ5 indicates theifference in life happiness in the year after and before the accident. The parameter ˇ6 measures the change (after/before)

n life happiness in the tsunami regions relative to unaffected regions and is expected to be negative.As an alternative to this broad measure of a municipality being affected by the tsunami, we will present results of

lternative specifications of Tj that differentiate between fatalities and damaged buildings.In addition to affecting life happiness through its physical consequences, the Fukushima accident may have led to

ncreased concern about nuclear power in general, which may particularly have affected the life happiness of people livinglose to nuclear power plants. The specifications testing this hypothesis directly follow Eqs. (2)–(4), now measuring distancerom the closest existing nuclear power plant instead of Fukushima. In our sample, the minimum distance from the closestuclear power station varies between 18 and 675 km with an average of 124 km. Within a radius of 100 km, 68% of theouseholds have at least one nuclear power station in the vicinity, some have up to four.

In estimating these models, happiness will be measured by “happiness with one’s life in the previous year”, which in012 refers to the time after the disaster. However, it may be the case that the consequences of the tsunami and the nuclearccident have influenced people’s assessment not only of their happiness in the year following the event but also of their

ntire lives. To test this hypothesis, we re-estimated some of the above models with “happiness with one’s whole life” ashe dependent variable.

8 Data was available from the National Research Institute for Earth Science and Disaster Prevention (http://www.j-risq.bosai.go.jp/ndis).

506 K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517

3.2. Data and empirical background

Most of the data used to investigate the well-being effect of the disaster is taken from the KHPS. The KHPS is a represen-tative Japanese household panel survey conducted by Keio University based on a set of pre-tested questionnaires for bothhouseholds and individuals. The same individuals are interviewed with some refreshment over the period in question. Thefirst wave of KHPS was assembled in 2004 and covered 4005 households. The usual sample size ranges between 3000 and3500 households. Interviews are always carried out in January.

KHPS provides information on various aspects of the individuals participating and the respective households. The ques-tionnaires assemble comprehensive information on household composition, income, occupation, employment history,school attendance, lifestyle, and location. In addition to a stable set of core questions, the survey focuses on special top-ics each year. The 2011 and 2012 questionnaires were the first to contain questions on different aspects of SWB. To takeadvantage of this information and because we are interested in the SBW effects of the accident in March 2011, our analysisrelies exclusively on the surveys of 2011 and 2012 (fielded in January of the respective years). Taken together, our datasetcontains a total number of 5979 observations. Table A1 in the Appendix provides information on the variables included andthe summary statistics.

SWB, measured as happiness with life, is the answer to the following question: Please choose a number on a scale of 0to 10, where “0” means having no feeling of happiness at all and “10” means having a feeling of complete happiness overthe last one year. The average for our sample is 6.24. Life happiness understood as happiness with one’s whole life up tothe present is the answer to the question: Please choose a number on a scale of 0–10, where “0” means having no feeling ofhappiness at all and “10” means having a feeling of complete happiness for your whole life up to the present. The averageof this variable is 6.46. Comparing the two measures over time, average life happiness decreased from 2011 to 2012 (from6.25 to 6.23) for the first measure but increased for the second (from 6.45 to 6.47).9 The explanatory variables generallyfound significant in models explaining differences in SWB are individual characteristics such as age, gender, employmentstatus, education level plus income, number of children, physical condition, or marital status. Income is converted usingequivalence scales to account for the fact that the needs of a household grow with additional household members, albeitnot proportionally (due to economies of scale in consumption).10 The equivalent income enters the regression equation inits natural logarithm to account for declining marginal utility of income.

Controls at the municipality level include population density, and the elevation of the municipality.11 At the prefecturelevel, we include information on the rate of unemployment.12 Note that in a macroeconomic sense, the effect of the March2011 disasters on employment is generally minor. The number of job seekers in the three prefectures most seriously affected(Fukushima, Miyagi, Iwate with a total population of about 5.7 million) peaked at about 164,000 in June 2011 but declinedafterwards.13 To capture further regional differences, dummy variables indicate which of the 47 prefectures the individuallives in. A further dummy variable indicates whether observations are drawn from the 2011 or 2012 survey.

Table A2 in the Appendix reports summary statistics broken down by sub-sample. There are no significant differencesbetween the sub-samples regarding income, age, sex, household size, number of children, education level, marital or healthstatus. In particular, important outcome variables (income, job, health, divorce) have not changed after the disaster.

4. Results

As is common in the SBW literature our empirical analysis mainly uses OLS. To avoid the cardinality assumption which thismethod requires, we ran all our regressions using an ordered probit as an alternative. We found the qualitative estimationresults (signs and significance) to be robust to this modification. In addition, the ratio of coefficients of our main variables

of interest and the coefficient of income was very similar in the OLS and ordered probit estimations. This suggests therobustness of ordinal preferences (WTP) to the method used. For ease of interpretation, the discussion of our results largelyrelies on the OLS estimates. However, Section 4.5 explicitly addresses the implied ordinal preferences.14

9 Note, again, it is assumed that happiness is cardinal and interpersonally comparable.10 Following the modified OECD scale, we assign a value of 1 to the first person in the household, 0.5 to every other person aged 14 and older, and 0.3 to all

children below the age of 14. Net household income is divided by the sum of this value, resulting in needs-adjusted values (see Atkinson and Bourguignon,2000). The advantage of using equivalence scales is that the effects of marginal changes in these variables can be interpreted on a hypothetical per-personbasis.

11 Information on population density was taken from the Population Census. Information on elevation was provided by the National Land Numeri-cal Information Download Service (http://nlftp.mlit.go.jp/ksj-e/gml/gml datalist.html). Elevation refers to the centroid of a municipality. There are 1719municipalities altogether, as of January 2013. For 2011 and 2012, KHPS covers 455 municipalities. In our sample, the average municipality has a size ofabout 200 km2.

12 Data was taken from the Labor Force Survey (Ministry of Internal Affairs and Communications of Japan).13 http://www.mhlw.go.jp/stf/houdou/2r9852000001z9f4.html.14 An alternative method would be to include individual fixed effects. We tested this approach and found the main coefficients of interest (interactions

of “2012” with the various “treatmentv̈ariables) to retain their signs and significance and to be of similar magnitude. However, with only two consecutiveobservations, fixed-effects estimation prevents an accurate identification of the effects of those variables that exhibit little variation between the two years,in particular income. Since we are interested in the income coefficients, we disregard fixed effects in our main analysis.

wa

4

dIi

Coomlow

4

to

tiadLcnsaifaM

lpad

dowacDb

q

e

(doh

K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517 507

To take account of a potential correlation of residuals when observations are taken from the same individual over time,e report robust standard errors. The effect is to increase the standard errors of the parameter coefficients. This procedure

lso produces robust variance estimates in the face of heteroskedasticity.

.1. Socio-demographic correlates of subjective well-being

Table 1 presents the estimation results for model specifications using life happiness referring to the previous year as theependent variable. Model 1 is a truncated version of Equation (1), which omits the event variables, variables Ei, and Ei × Yi.

t serves to check whether our SWB data are “well-behaved”, i.e., whether SWB is related to the usual explanatory variablesn ways known from the literature.

The results from Model 1 confirm those of earlier studies using data from other countries (see, e.g., Dolan et al., 2008).onsistent with earlier analyses, we find a U-shaped relationship with age and positive and diminishing marginal utilityf income. Unemployed, sick, and single individuals are significantly less happy than their counterparts. A greater degreef happiness is found among individuals with higher levels of education, those currently in the education process, anden/women who do the housekeeping or look after their children. While the presence of children has a positive effect on

ife happiness, the size of the household has a negative effect. Being male has a negative effect on life happiness. None of thether variables are statistically significant either at the 1% or 5% level. A comparison of observations from the 2011 surveyith the 2012 survey reveals no significant difference.15

.2. Direct effects of the disaster

We first focus on distance-dependent combined effects of the earthquake/tsunami and nuclear accident. As in Model 1,he dependent variable is happiness with one’s life in the preceding year. Results are reported in Table 1. Note that we focusn statistical significance either at the 1% or 5% level of confidence.

According to Model 2 (Eq. (2)), the coefficient on distance from the Fukushima site is insignificant: There was no statis-ically significant relationship between the distance and life happiness before the disaster. The interaction effect, however,s positive, as expected, and statistically significant. Post-Fukushima, distance from the power plant mattered: individualsssessed their lives to be happier the further away they lived from the place of the disaster. A 1-unit (1000 km) increase inistance from the Fukushima site was associated with happiness being one half point (0.5007) greater (on the 0–10 scale).ooking at the results for Model 3 – the inverse distance (proximity) specification (Eq. (3)) – we find a statistically signifi-ant positive relationship between proximity and life happiness before the disaster.16 Importantly, the interaction term isegative and significant, indicating that after the disaster people were less happy the closer they lived to the Fukushimaite. Quantitatively, a 1-unit increase in proximity was associated with happiness being 0.0816 points lower. To illustrate,t the minimum of (observed) proximity, which corresponds to a distance from the Fukushima site of 1771 km, a 1-unitncrease in proximity corresponds to 1132 km. At the maximum of (observed) proximity, which corresponds to a distancerom the Fukushima site of 57 km, a 1-unit increase in proximity corresponds to 3 km. Both of these increases in proximityre associated with happiness being 0.0816 points lower. The other results from Models 2 and 3 are comparable to those ofodel 1.Interpreting the estimation results in a different way, the common message from Models 2 and 3 is that decreases in

ife happiness were greater the more proximate the individual lived to the place of the disaster, independent of whetherroximity is measured in a linear or non-linear fashion. However, a Box–Cox transformation of the distance variable measureds a continuous variable reveals that the linear specification is rejected but not the inverse specification. For this reason weo not consider the linear specification further.

We next consider Model 4, which includes radius ring dummies instead of continuous variables (Eq. (4)). For reasons to beiscussed below, we compare three radius rings: (1) up to 150 km (5% of our sample), (2) over 150 km and up to 300 km (35%f our sample) and (3) above 300 km. Using the third category (above 300 km) as the reference category, we find that thereas a drastic and significant drop in life happiness in the range up to 150 km whereas it was much smaller but still significant

t distances between 150 and 300 km. Interpreting these changes as effects of different degrees of treatment, these resultsonfirm that the distance-dependent combined effects of the disaster declined in a monotonous though non-linear fashion.ue to the disaster, the happiness difference between the range 0–150 km (150–300 km) and the range >300 km dropped

y 0.9069 (0.2630) points. The other results are comparable to those of Model 1.

As argued in Section 3.1, distance-dependent changes in life happiness may reflect effects of both the earth-uake/tsunami and the nuclear accident. We now turn to the effects of the earthquake and the tsunami more

15 Estimating Model 1 separately for 2011 and 2012, the results are comparable regarding size and significance of variables. A notable difference is theffect of unemployment on SWB. For 2012 the variable indicating if a respondent is unemployment is significant.16 In Japan, areas in the vicinity of nuclear power plants enjoy various economic benefits, such as subsidies based on the Power Source Siting Lawsdengen san pou) and large property-tax revenues from the sitting power company. Also, the power company and related businesses tend to constitute aominant proportion of the local economy around a nuclear power plant. For example, before the Fukushima accident, more than 60% of the economicutput in Futaba county (where the Fukushima Dai-ichi power plant is located) was electricity-related (source: the Reconstruction Agency website,ttp://www.reconstruction.go.jp/topics/20120904 sangyoukoyouplan.pdf).

508

K.

Rehdanz

et al.

/ Journal

of Econom

ic Behavior

& O

rganization 116

(2015) 500–517

Table 1Main regression results.

Model 1 – base Model 2 – distance Model 3 – proximity Model 4 – radius Model 5 – tsunami Model 6 – deaths

Coeff. Standard errora Coeff. Standard errora Coeff. Standarderrora

Coeff. Standarderrora

Coeff. Standarderrora

Coeff. Standarderrora

Event2012 −0.0467 (0.0754) −0.2658 (0.1156)** 0.1848 (0.0996)* 0.0535 (0.0868) 0.0210 (0.0822) −0.0085 (0.0801)DistFukushima −0.2007 (0.9652)2012*DistFukushima 0.5007 (0.1747)***

ProxFukushima 0.1645 (0.0578)***

2012*ProxFukushima −0.0816 (0.0232)***

Ring 1 (≤150 km) 0.8407 (0.4402)*

Ring 2 (>150, ≤300) 0.4901 (0.3275)Ring 3 (>300) (reference category)2012*Ring 1 −0.9069 (0.2436)***

2012*Ring 2 −0.2630 (0.1155)**

2012*Ring 3 (reference category)Tsunami 0.1137 (0.1294)2012*Tsunami −0.2742 (0.1266)**

Deaths 0.1030 (0.1768)2012*Deaths −0.4070 (0.2085)*

Individual informationIncome 0.3782 (0.0478)*** 0.3799 (0.0477)*** 0.3799 (0.0477)*** 0.3784 (0.0476)*** 0.3790 (0.0476)*** 0.3790 (0.0477)***

Age −0.0689 (0.0176)*** −0.0690 (0.0176)*** −0.0680 (0.0176)*** −0.0685 (0.0176)*** −0.0688 (0.0176)*** −0.0689 (0.0176)***

Age (squared) 0.0008 (0.0002)*** 0.0008 (0.0002)*** 0.0008 (0.0002)*** 0.0008 (0.0002)*** 0.0008 (0.0002)*** 0.0008 (0.0002)***

Male −0.3061 (0.0621)*** −0.3044 (0.0620)*** −0.3003 (0.0620)*** −0.3056 (0.0622)*** −0.3015 (0.0620)*** −0.3007 (0.0620)***

Size −0.0754 (0.0273)*** −0.0746 (0.0273)*** −0.0733 (0.0273)*** −0.0743 (0.0273)*** −0.0751 (0.0273)*** −0.0746 (0.0273)***

Single (reference category) (reference category) (reference category) (reference category) (reference category) (reference category)Married 0.8353 (0.0882)*** 0.8354 (0.0882)*** 0.8365 (0.0881)*** 0.8367 (0.0882)*** 0.8362 (0.0882)*** 0.8344 (0.0882)***

Other −0.0155 (0.2028) −0.0114 (0.2013) −0.0062 (0.2013) −0.0014 (0.2011) −0.0095 (0.2019) −0.0107 (0.2013)Children 0.0864 (0.0381)** 0.0857 (0.0381)** 0.0851 (0.0381)** 0.0846 (0.0381)** 0.0857 (0.0381)** 0.0861 (0.0381)**

Edulevel 1 (reference category) (reference category) (reference category) (reference category) (reference category) (reference category)Edulevel 2 0.2015 (0.1208)* 0.1921 (0.1210) 0.1903 (0.1203) 0.1922 (0.1205) 0.1922 (0.1208) 0.1932 (0.1209)Edulevel 3 0.2782 (0.1439)* 0.2639 (0.1439)* 0.2684 (0.1433)* 0.2683 (0.1435)* 0.2668 (0.1438)* 0.2696 (0.1438)*

Edulevel 4 0.4039 (0.1329)*** 0.3904 (0.1330)*** 0.3929 (0.1324)*** 0.3932 (0.1326)*** 0.3912 (0.1330)*** 0.3952 (0.1330)***

Edulevel 5 −0.0358 (0.2752) −0.0488 (0.2755) −0.0451 (0.2752) −0.0407 (0.2751) −0.0516 (0.2760) −0.0493 (0.2756)Edulevel 6 0.4722 (0.1358)*** 0.4609 (0.1359)*** 0.4657 (0.1354)*** 0.4644 (0.1356)*** 0.4611 (0.1359)*** 0.4614 (0.1359)***

Edulevel 7 −0.2183 (0.2330) −0.2410 (0.2327) −0.2367 (0.2333) −0.2389 (0.2329) −0.2384 (0.2328) −0.2365 (0.2335)Employed (reference category) (reference category) (reference category) (reference category) (reference category) (reference category)Student 0.8686 (0.3245)*** 0.8550 (0.3247)** 0.8603 (0.3251)*** 0.8617 (0.3250)*** 0.8575 (0.3244)*** 0.8662 (0.3271)***

Retired 0.2094 (0.1167)* 0.2002 (0.1164)* 0.1919 (0.1161)* 0.1967 (0.1159)* 0.1995 (0.1162)* 0.1987 (0.1165)*

Unemployed −0.6255 (0.2259)*** −0.6148 (0.2268)*** −0.6143 (0.2267)*** −0.6226 (0.2272)*** −0.6222 (0.2263)*** −0.6236 (0.2264)***

K.

Rehdanz

et al.

/ Journal

of Econom

ic Behavior

& O

rganization 116

(2015) 500–517

509

Home 0.1776 (0.0811)** 0.1768 (0.0810)** 0.1780 (0.0809)** 0.1766 (0.0810)** 0.1820 (0.0809)** 0.1818 (0.0809)**

Health 1 (reference category) (reference category) (reference category) (reference category) (reference category) (reference category)Health 2 −0.6704 (0.0919)*** −0.6703 (0.0919)*** −0.6694 (0.0919)*** −0.6731 (0.0919)*** −0.6689 (0.0920)*** −0.6724 (0.0918)***

Health 3 −1.4494 (0.0898)*** −1.4509 (0.0899)*** −1.4491 (0.0897)*** −1.4550 (0.0897)*** −1.4476 (0.0899)*** −1.4540 (0.0896)***

Health 4 −2.2936 (0.1157)*** −2.2956 (0.1156)*** −2.2951 (0.1154)*** −2.3013 (0.1154)*** −2.2912 (0.1158)*** −2.2963 (0.1155)***

Health 5 −3.4163 (0.3509)*** −3.4003 (0.3502)*** −3.3800 (0.3507)*** −3.3776 (0.3509)*** −3.3982 (0.3496)*** −3.3909 (0.3517)***

Geographical and other informationUnemployment −0.0969 (0.1943) −0.0026 (0.2189) −0.1879 (0.2243) −0.2550 (0.2275) −0.1108 (0.2237) −0.0688 (0.2215)Popdens 0.0000 (0.0000) 0.0000 (0.0000) −0.0000 (0.0000) 0.0000 (0.0000) 0.0000 (0.0000) −0.0000 (0.0000)Elevation −0.0000 (0.0002) −0.0001 (0.0002) −0.0000 (0.0002) −0.0000 (0.0002) −0.0001 (0.0002) −0.0001 (0.0002)

Constant 3.6662 (1.3019)*** 4.0832 (2.4137)* 5.1670 (1.7379)*** 5.8226 (1.7584)*** 4.8309 (1.7402)*** 4.5637 (1.7264)***

Region YES YES YES YES YES YESAdjusted R2 0.1708 0.1721 0.1732 0.1730 0.1716 0.1715N 5979 5979 5979 5979 5979 5979

Note: Dependent variable is happiness with life in previous year (0–10). Method: Least Squares. DistFukushima is the distance to the Fukushima Dai-ichi nuclear plant and is measured in 1000 km. ProxFukushimais 1/DistFukushima. Tsunami is a dummy variable that takes the value 1 if damaged/destroyed buildings or dead or injured persons were reported in the respondent’s municipality and 0 otherwise. Deaths is adummy variable that takes the value 1 if fatalities were reported in the respondent’s municipality and 0 otherwise.

a Reported are robust standard errors.* Significance at the 10% level.

** Significance at the 5%.*** Significance at the 1% level.

510 K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517

Fig. 2. Distribution and location of affected municipalities.Source: Own presentation.

specifically, as specified in Model 5 (Eq. (5)). The results, presented in Table 1, indicate that before the accident lifehappiness in the municipalities affected by the earthquake/tsunami was not significantly different from life happinessin other municipalities. As anticipated, the interaction effect is however negative and statistically significant. In thepost-Fukushima period location did matter: individuals located in municipalities affected by the earthquake/tsunamiexperienced lower life happiness. Put differently, decreases (increases) in life happiness were greater (smaller) in munic-ipalities affected by the earthquake/tsunami than in municipalities not affected. Other results are comparable to those ofModel 1.

The earthquake/tsunami dummy in Model 5 aggregates several effects (dead or injured persons or destroyed or damagedbuildings). When we replace it with a dummy with value one if dead persons were reported in the respective municipality(Model 6), the results are comparable to those of Model 5, but the ˇ6 coefficient is now −0.4070, instead of −0.2742, but thevariable is statistically significant only at the 10% level of confidence.

The specifications considered so far focused on either distance or the tsunami. Since the two may be related,omission of one arguably affects the coefficient of the other. To examine these interrelationships, Table 2 presentsresults in which distance and the earthquake/tsunami dummy as well as distance and the fatality dummy are includedjointly.

Regardless of how distance is measured, the earthquake/tsunami dummy (Models 7a–9a) and the fatality dummy(Models 7b–9b) are never significant. The results for the different distance measures are comparable in size andsignificance to those presented in Table 1 (Models 2–4). The results of the other control variables are compa-rable as well, but in the interest of clarity results are not shown. Jointly with Table 1, the results in Table 2suggest that distance is an encompassing indicator for the severity of the combined disaster in different places,which includes but is not restricted to buildings and fatalities. Figure 2 provides further evidence on how the inci-dence of damaged buildings and fatalities, respectively, relates to distance to Fukushima. It can be seen that mostmunicipalities within a range of 300 km were affected by building damages and/or fatalities while damages and fatal-ities were much less prevalent outside that range. In addition, the proportion of municipalities with fatalities ismuch greater in the range 0–150 km than in the range 150–300. These differences support our choice of distancecategories.

Focusing more specifically on the nuclear accident, we analyze the effect of the radioactivity level on SWB. Replacingthe event variable by a variable measuring the distribution of radioactive substances across municipalities, we accountfor the fact that due to the prevailing wind direction after the accident radioactivity affected regions close to Fukushima

K.

Rehdanz

et al.

/ Journal

of Econom

ic Behavior

& O

rganization 116

(2015) 500–517

511

Table 2Results for distance and event variables combined.

Model 7a –distance & tsunami

Model 8a – proximity &tsunami

Model 9a – radius& tsunami

Model 7b –distance & deaths

Model 8b –proximity & deaths

Model 9b – radius& deaths

Coeff. Standard errora Coeff. Standard errora Coeff. Standard errora Coeff. Standarderrora

Coeff. Standarderrora

Coeff. Standarderrora

Event2012 −0.2104 (0.1401) 0.1822 (0.1019)* 0.0534 (0.0868) −0.2283 (0.1202)* 0.1765 (0.1030)* 0.0531 (0.0869)DistFukushima −0.1943 (0.9792) −0.2796 (0.9771)2012*DistFukushima 0.4206 (0.2097)** 0.4404 (0.1826)**

ProxFukushima 0.1627 (0.0589)*** 0.1794 (0.0593)***

2012*ProxFukushima −0.0794 (0.0294)*** −0.0776 (0.0269)***

Ring 1 (≤150 km) 0.8322 (0.4484)* 0.8696 (0.4479)*

Ring 2 (>150, ≤300) 0.4866 (0.3308) 0.4654 (0.3282)Ring 3 (>300) (reference category) (reference category)2012*Ring 1 −0.9125 (0.2768)*** −0.9233 (0.2914)***

2012*Ring 2 −0.2653 (0.1424)* −0.2646 (0.1171)**

2012*Ring 3 (reference category) (reference category)Tsunami 0.0176 (0.1392) −0.0288 (0.1417) −0.0475 (0.1426)2012*Tsunami −0.1056 (0.1520) −0.0207 (0.1600) 0.0040 (0.1636)Deaths 0.0191 (0.1813) −0.1469 (0.1894) −0.1237 (0.1920)2012*Deaths −0.2564 (0.2179) −0.0772 (0.2401) 0.0190 (0.2479)

Region YES YES YES YES YES YESAll other controls YES YES YES YES YES YESAdjusted R2 0.1719 0.1729 0.1728 0.1721 0.1732 0.1728N 5979 5979 5979 5979 5979 5979

Note: Dependent variable is happiness with life in previous year (0–10). Method: Least Squares. DistFukushima is the distance to the Fukushima Dai-ichi nuclear power plant and is measured in 1000 km.ProxFukushima is 1/DistFukushima. Tsunami is a dummy variable that takes the value 1 if damaged/destroyed buildings or dead or injured persons were reported in the respondent’s municipality and 0otherwise. Deaths is a dummy variable that takes the value 1 if fatalities were reported in the respondent’s municipality and 0 otherwise.

a Reported are robust standard errors.* Significance at the 10% level.

** Significance at the 5%.*** Significance at the 1% level.

512 K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517

Table 3Results for happiness and nearest nuclear power plant.

Model 10 – distance Model 11 – proximity Model 12 – radius

Coefficient Standard errora Coefficient Standard errora Coefficient Standard errora

Event2012 −0.0528 (0.1325) 0.0001 (0.1087) −0.0203 (0.0922)DistNuclear −1.6088 (1.2869)2012*DistNuclear 0.2249 (0.7632)ProxNuclear 0.0086 (0.0079)2012*ProxNuclear −0.0021 (0.0075)Ring 1a (≤25 km) 0.2074 (0.3383)Ring 2a (>25, ≤50) −0.2140 (0.2387)Ring 3a (>50, ≤75) 0.2070 (0.1386)Ring 4a (>75, ≤100) −0.0511 (0.1268)Ring 5a (>100) (reference category)2012*Ring 1a −0.1415 (0.3862)2012*Ring 2a 0.0221 (0.2619)2012*Ring 3a −0.1836 (0.1414)2012*Ring 4a 0.1271 (0.1480)2012*Ring 5a (reference category)

Region YES YES YESAll other controls YES YES YESAdjusted R2 0.1711 0.1711 0.1711N 5979 5979 5979

Note: Dependent variable is happiness with life in previous year (0–10). Method: Least Squares. DistNuclear is the distance to the nearest nuclear powerplant and is measured in 1000 km. ProxNuclear is 1/DistNuclear.

a Reported are robust standard errors.

*Significance at the 10% level.**Significance at the 5%.***Significance at the 1% level.

very differently.17 We did not, however, find the interaction of this new variable with the post-Fukushima dummy to besignificant, not even at the 10% level of confidence.

The models considered so far control for (bad) health, income, and the regional unemployment level. Since it is possiblethat the disaster affected life happiness through these channels, we estimated a version of Model 3 in which these controlsare omitted (results not shown). If the disaster affected life happiness through uncompensated income losses, health impair-ments or higher unemployment, the coefficient of the interaction term 2012*ProxFukushima should increase. We found it tobe somewhat decreased (−0.0747 compared to −0.0814), but the difference is not statistically significant. Results for Model4 are comparable. Effects through induced deteriorations in the health, income or employment situation, if any, thus seemnot to have contributed to the well-being effects of the disaster.

In sum, we found that people living in a place affected by the earthquake/tsunami or close to the Fukushima Dai-ichipower plant experienced a drop in life happiness, while other citizens of Japan experienced no such drop or a significantlysmaller one. In a spatial perspective, the effect of the disaster was particularly strong within a radius of 150 km. Effectsrelated to radioactivity levels could not be established. The latter is consistent with a lack of evidence for health impairmentsplaying a role for the effect of the disaster on life happiness. An effect through uncompensated income losses or increasedunemployment could also not be detected. Overall, we found significant SWB effects of the combined disaster, but they seemto be mainly unrelated to the nuclear accident. It should be noted, however, that we are only able to measure short-runeffects.

4.3. Testing for additional effects of the disaster

A first additional hypothesis to be explored is that, in addition to being directly affected by the disaster, life happiness mayhave dropped due to increased concern about nuclear power among those living close to nuclear power stations. Accordingto Models 10–12 (Table 3), none of the interactions between Post-Fukushima and distance from nuclear power plants orpresence of nuclear power stations in the vicinity is significant. This is explicable in the light of our finding that most of theSWB effects of the combined disaster seem to be unrelated to the nuclear accident per se (see above).

A second additional hypothesis relates to our dependent variable, which up to this point was people’s happiness in theyear before the respective interview. In the following, we analyze whether the accident has an effect on life happinessunderstood as happiness with one’s whole life up to the present. Table 4 presents the results of specifications equivalent

17 Information is taken from the Database on the Research of Radioactive Substances Distribution provided by the Ministry of Education, Culture, Sports,Science and Technology (http://radb.jaea.go.jp/mapdb/en/). The average of total cesium-134 and -137 depositions per municipality on 29 April 2011 wasused. This is the earliest day after the accident for which data of large-scale airborne measurements are available.

K.

Rehdanz

et al.

/ Journal

of Econom

ic Behavior

& O

rganization 116

(2015) 500–517

513

Table 4Results for happiness with entire life.

Model 1a – base Model 2a – distance Model 3a – proximity Model 4a – radius Model 5a – tsunami Model 6a – deaths

Coefficient Standard errora Coefficient Standarderrora

Coefficient Standarderrora

Coefficient Standarderrora

Coefficient Standarderrora

Coefficient Standarderrora

Event2012 −0.0209 (0.0681) −0.0565 (0.1055) 0.0159 (0.0893) −0.0051 (0.0782) −0.0024 (0.0740) −0.0115 (0.0722)DistFukushima 0.3306 (0.8996)2012*DistFukushima 0.0834 (0.1578)ProxFukushima 0.1318 (0.0504)***

2012*ProxFukushima −0.0132 (0.0210)Ring 1 (≤150 km) 0.5740 (0.4055)Ring 2 (>150, ≤300) 0.3679 (0.3087)Ring 3 (>300) (reference category)2012*Ring 1 −0.2065 (0.2175)2012*Ring 2 −0.0372 (0.1051)2012*Ring 3 (reference category)Tsunami 0.0030 (0.1202)2012*Tsunami −0.0860 (0.1152)Deaths 0.2241 (0.1515)2012*Deaths −0.1248 (0.1816)

Region YES YES YES YES YES YESAll other controls YES YES YES YES YES YESAdjusted R2 0.1545 0.1551 0.1559 0.1551 0.1551 0.1553N 5979 5979 5979 5979 5979 5979

Note: Dependent variable is happiness with entire life (0–10). Method: Least Squares. DistFukushima is measured in 1000 km. ProxFukushima is 1/DistFukushima. Tsunami is a dummy variable that takes thevalue 1 if damaged/destroyed buildings or dead or injured persons were reported in the respondent’s municipality and 0 otherwise. Deaths is a dummy variable that takes the value 1 if fatalities were reportedin the respondent’s municipality and 0 otherwise.

a Reported are robust standard errors.*Significance at the 10% level.**Significance at the 5%.

*** Significance at the 1% level.

514 K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517

Table 5Monetary equivalent of the treatment effect.

¥ % income

Model 3 – proximity 1 km at av. distance (487 km) 2160 <11 km at max. distance (1771 km) 160 <11 km at min. distance (57 km) 163,120 7

Model 4 – radius ≤150 km 5,708,780 240

150–300 km 1,655,540 70

Model 5 – tsunami 1,723,310 72

Note: Evaluated at average annual net household equivalence income (¥2,382,000 in our sample).

to Models 1–6, where the replacement of the dependent variable is the only difference. Comparing the results to thosepresented in Table 1, we find no change in the sign and significance of most variables previously included (results not shown).Important exceptions are the interaction terms of post-Fukushima and the Tsunami and distance (proximity) indicators. Inall five models – Models 2a–6a – we find no statistically significant effect for the respective interaction variables.18 Thoughthe accident has affected people’s evaluation of the quality of their lives after the event, it does not seem to have influencedtheir assessment of their entire lives.

4.4. Robustness checks

So far, we have not considered the possibility of selection bias, i.e. whether people most seriously affected by the eventmay have refused to participate in the 2012 round of the survey. In addition, given that all residents within a certain radius ofthe Fukushima Dai-ichi power plant were evacuated to more distant places after the event, it is likely that the whereaboutsof some of the interviewees in 2012 was more distant than it was at the time of the disaster. This relocation effect mayintroduce a downward bias in the relationship between distance and change in life happiness. Both the selection and therelocation effects, if present, will lead to a downward bias in our estimates of the SWB effect of the Fukushima disaster.Further, people living in the affected municipalities or closer to Fukushima before the disaster are perhaps more likely tolose their jobs or have health-related problems.

Focusing on Model 3, similar results as in Table 1 are obtained when we restrict the sample to those individuals thatparticipated in the KHPS panel in both years. Restricting the sample further by excluding those who have moved doesnot change the results.19 Including a term interacting unemployment with inverse distance from the Fukushima site to testwhether unemployed persons closer to Fukushima show lower levels of life happiness, we find that the term has the expectednegative sign and is significant at the 5% level. Finally, by including a term interacting health status with inverse distance fromthe Fukushima site, we test whether individuals reporting lower levels of health status are unhappier if they live closer to theFukushima site. The term is positive, as expected, and significant at the 10% level, thus confirming the hypothesis.20 Otherresults are unchanged. In summary, our findings as presented in Section 4.2 are robust; in every alternative specification theinteraction term of the post-Fukushima dummy and inverse distance is negative and highly significant. Results for Model 4are comparable.

4.5. Valuing the treatment effect

As stated earlier, the results of SWB analyses can be used to calculate the monetary equivalents of non-market goods/badsor extreme events, that is, marginal rates of substitution or the compensating variation. We now use our results to computethe amount a person would be willing to pay to avoid being affected by the Fukushima disaster. For this purpose, we convertthe treatment effect (parameter ˇ6 in Eqs. (1)–(5)), which is measured in units of life happiness, into monetary units bydividing it by ∂SWB/∂I.

Table 5 summarizes the treatment effect for a one kilometer increase in distance from the Fukushima site (Model 3) aswell as the monetary equivalent for not being treated (Models 4–6).21

Applying the results from Model 3, the effects of a 1-km increase in distance correspond to about ¥2200 for mean distance(487 km), about ¥160,000 for minimum distance (57 km), and about ¥160 for maximum distance (1771 km), evaluated at

average annual net household equivalence income. The figure of ¥160,000 for a move from 57 km (minimum distance) to58 km from the Fukushima site corresponds to 6.6% of average annual income. For a non-marginal change in distance, e.g.from minimum distance to a distance of 100 km, the compensating surplus is about ¥6,771,000 or 280% of income (Model 3).

18 This is true of the other model specifications as well, which are not presented here.19 In general, few individuals moved between the two survey years (only 30) and no regional pattern is discernible. The drop-out rate between the 2011

and 2012 samples shows no bias toward locations that were more severely affected. It reduces our sample to 5008 observations. Also, the 2011 KHPSsample includes no households located in the evacuation zone.

20 We tried both health measured as a set of categorical dummies, as in Model 5, and as a continuous variable.21 We use average net household equivalence income for all calculations; ¥2,382,000 in our sample (the average exchange rate in January 2012 was

¥100 = US$1.30).

Uoi

1emr

li

sd(

dWime

2tc

wtMob

5

t

sosoo

traaa

tc

r

p

oa

K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517 515

sing results from Model 4, the monetary equivalent for someone living more than 300 km away from Fukushima insteadf living in the range 150–300 km is about ¥ 1,650,000 (70% of income) while the value of living more than 300 km awaynstead of living in the range up to 150 km is ¥5,700,000 (240% of income).

In order to compare Model 4 with Model 3, the latter findings can be restated as saying that a person living in the range50–300 km experienced a drop in happiness worth 70% of income whereas a person living in the range up to 150 kmxperienced a drop in happiness worth 240% of income. Computing the monetary happiness effects from Model 3 at theidpoints of the respective distance categories (that is, at 75 km and 225 km) they amount to 64% and 572% of income,

espectively.The results from Model 5 suggest that for a person with average equivalence income the monetary equivalent of not

iving in a municipality affected by the tsunami is about ¥1,700,000. This is equivalent to 72% of a person’s annual averagencome.

Numbers are comparable when using results of ordered probit regressions, lending support to the adequacy of the OLSpecifications. More explicitly the changes in net equivalence income for a 1-km change are as follows: 6.9% for minimumistance (Model 3), 243% for Ring 1 and 72% for Ring 2 (Model 4), 73% for not living in a municipality affected by the tsunamiModel 5).

To put our valuation results in perspective, we compare them to those of other researchers who have investigated naturalisasters or extreme events using SWB regressions. Lueinger and Raschky (2009), for example, calculate that the averageTP for the avoidance of one certain flood event in the region of residence corresponds to 24% of average annual household

ncome. Carroll et al. (2009) calculate that a drought in Australia is equivalent to an income loss of 38% for a household withean household income. Given the nature and extent of the Fukushima disaster, it is not surprising that its monetized SBW

ffect tends to be greater than the effect of those events.Following a different approach, Yamane et al. (2013) use hedonic regressions of property values within a range of

9–80 km from the Fukushima site. They find that property values dropped from July 2010 to July 2011 in proportiono the level of radiation after the disaster, the effect being about 4% for the highest radiation category relative to the lowestategory. In addition, property values dropped by 3% in areas that had been inundated due to the tsunami.

The equivalent loss in life happiness for Japan as a whole goes up to about ¥217,535 bln (46% of GDP or US$ 2828 bln)hen using inverse distance (Model 3). The equivalent loss corresponds to ¥113,565 bln (24% of GDP or US$ 1476 bln) for

he model using radius rings (Model 4), where the difference between Models 3 and 4 mainly reflects the circumstance thatodel 4 rules out any effect in the most distant radius ring (>300 km). The total monetized effect for Model 5, which focuses

n municipalities affected by the tsunami through damaged buildings or fatalities, is ¥85,585 bln (18% of GDP or US$ 1113ln).22 For comparison, nominal GDP (seasonally adjusted) in 2012 was about ¥475,529 bln or US$6182 bln.

. Discussion and conclusion

Using data from KHSP, we have conducted a quasi-experimental SWB study of the combined disaster (earthquake,sunami, nuclear accident) near Fukushima on 11 March 2011 taking into account the spatial dimension.

With regard to general SWB factors, we found evidence that is highly consistent with findings from western countries inpite of the differences in cultural backgrounds.23 The plausibility of these results enhances our confidence in the suitabilityf the KHSP database for a study of the main issue addressed in this paper, the effects of the Fukushima events on SWBpecified as happiness with life. However, one important limitation of our study is that we are unable to investigate the rolef adaptation. This inability is due to the fact that the interviews in the Keio survey are conducted only within a short periodf time, i.e. the January of each year, and we have only two years of observations.

Possible SWB effects of the combined disaster relate, in the first place, to its direct physical consequences (in particularhe death of relatives and friends, destruction of buildings and infrastructural resources, release of radiation, evacuation ofesidents, shortage of basic necessities such as electricity and water). In the second place, it is conceivable that the nuclearccident may have reduced SWB through increased concern about nuclear power even if individuals were not directlyffected by the accident. Finally, those most seriously affected by the (combined) disaster may have reduced not only thessessment of their happiness after the event but also the assessment of their entire lives.

The main finding from our empirical analysis is that life happiness dropped significantly after the disaster in places closeo the event and, in particular, in municipalities affected by the tsunami. This result has withstood a number of robustness

hecks. In particular, it is not affected by selection bias or relocation bias.

Effects related to radioactivity levels could not be established. The latter is consistent with a lack of evidence for health-elated effects of the disaster on SWB. It should be noted, however, that we were only able to measure short-term effects.

22 Calculations are carried out at the municipality level using average net equivalence income per person (¥2,382,000 in our sample). Information onopulation size at the municipality level was taken from http://www.econstats.com/japcab/japcab a9.htm.23 Despite the ongoing shift toward individualism caused by globalization, unique culture-specific characteristics are still present in Japan (and East Asia),riginating either from prevailing philosophical traditions (e.g., Buddhism) or traits of collectivism. For a review, see Tov and Diener (2007) and Uchidand Ogihara (2012).

516 K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517

This may explain why we found no effects from the level of radiation, which are likely to be more long-term in nature. Aneffect through uncompensated income losses or increased unemployment could also not be detected.

Unlike the physically based effects of the disaster, we found no evidence of a psychological effect on SWB in terms ofincreased concern about nuclear power in those living close to nuclear power plants. This absence of effect is consistent withour finding that the nuclear accident per se did not seem to affect the life happiness of those living close to the place of theevent. Another explanation could be that all the nuclear power plants in the country were taken off stream after the nucleardisaster. Finally, in spite of the significant and sizeable drop in life happiness in affected municipalities after the disaster,people’s assessments of the quality of their entire lives do not seem to have been negatively affected.

An explanation for some of the missing effects may be the philosophical traditions influential in East Asia (e.g., Buddhismand Daoism) emphasizing the dialectical nature of things. East Asians do in fact display a relatively high degree of equanimityin the face of negative emotions and events. This also has a bearing on the way in which East Asians regard happiness. Forexample, Uchida and Kitayama (2009) find that the Japanese participants in their experiment tended to associate happinesseven with some negative factors, such as jealousy on the part of others and lack of concern for one’s surroundings. Thisfinding contrasted with the responses of the American participants in the experiment, whose view of happiness tends to beentirely positive.

Based on the model in which the effect size decreases with distance in a hyperbolic fashion, we find the monetaryequivalent of a 1-km increase in distance from the Fukushima site to be about ¥2200 for someone located at the meandistance. At the minimum distance from the Fukushima nuclear plant (57 km), the monetary equivalent of a 1-km increasein distance is about ¥160,000 or 6.6% of average annual income. The well-being effect of the disaster for Japan as a wholecorresponds to up to 46% of GDP. The monetary equivalent of living in a municipality affected by the tsunami in terms ofdamaged buildings, injured persons or fatalities is about 70% of annual average income and reaches 240% for those livingin close distance to the Fukushima Dai-ichi power plant (≤150 km). Similar as the figures established by other researchersanalyzing single events like floods and droughts, these figures should be taken as tentative only due to possible biases in theunderlying estimate of the SWB effects of income.

Acknowledgements

We would like to thank Susana Ferreira, Arik Levinson, David Maddison, Uwe Jensen and two anonymous referees forhelpful comments on an earlier version of the paper. The data analysis in this paper utilizes Keio Household Panel Survey(KHPS) data provided by the Panel Data Research Center at Keio University. For Heinz Welsch financial support by the SwissFederal Office of Energy is acknowledged.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jebo.2015.05.014.

References

Almond, D., Edlund, L., Palme, M., 2009. Chernobyl’s subclinical legacy: prenatal exposure to radioactive fallout and school outcomes in Sweden. Quart. J.Econ., 1729–1772.

Atkinson, A., Bourguignon, F. (Eds.), 2000. Handbook of Income Distribution. North-Holland, Amsterdam.Berger, E.M., 2010. The Chernobyl disaster, concern about the environment, and life satisfaction. Kyklos 63, 1–8.Carroll, N., Frijters, P., Shields, M.A., 2009. Quantifying the costs of drought: new evidence from life satisfaction data. J. Popul. Econ. 22, 445–461.Clark, A.E., Frijters, P., Shields, M.A., 2008. Relative income, happiness and utility: an explanation for the Easterlin Paradox and other puzzles. J. Econ. Lit.

46, 95–144.Di Tella, R., MacCulloch, R., Oswald, A., 2001. Preferences over inflation and unemployment: evidence from surveys of happiness. Am. Econ. Rev. 91, 335–341.Dolan, P., Peasgood, T., White, M., 2008. Do we really know what makes us happy? A review of the economic literature on the factors associated with

subjective well-being. J. Econ. Psychol. 29, 94–122.Ferreira, S., Moro, M., 2010. On the use of subjective well-being data for environmental valuation. Environ. Resour. Econ. 46, 249–273.Frey, B.S., Stutzer, A., 2000. Happiness, economy and institutions. Econ. J. 110, 918–938.Frey, B.S., Stutzer, A., 2002. What can economists learn from happiness research? J. Econ. Lit. 40, 402–435.Frey, B.S., Luechinger, S., Stutzer, A., 2009. The life satisfaction approach to valuing public goods: the case of terrorism. Public Choice 138, 317–345.Goebel, J., Krekel, C., Tiefenbach, T., Ziebarth, N.R., 2013. Natural Disaster, Policy Action, and Mental Well-Being: The Case of Fukushima. SOEP Papers on

Multidisciplinary Panel Data Research., pp. 599.Helliwell, J.F., Wang, S., 2012. The state of world happiness. In: Helliwell, J., Layard, R., Sachs, J. (Eds.), World Happiness Report. United Nations, New York.Kahneman, D., Wakker, P.P., Sarin, R., 1997. Back to Bentham? Explorations of experienced utility. Quart. J. Econ. 112, 375–405.Kimball, M., Levy, H., Ohtake, F., Tsutsui, Y., 2006. Unhappiness After Hurricane Katrina. NBER Working Paper 12062. Cambridge.Levinson, A., 2012. Valuing public goods using happiness data: the case of air quality. J. Public Econ. 96 (9–10), 869–880.Luechinger, S., 2009. Valuing air quality using the life satisfaction approach. Econ. J. 119, 482–515.Luechinger, S., Raschky, P.A., 2009. Valuing flood disasters using the life satisfaction approach. J. Public Econ. 93, 620–633.

MacKerron, G., 2012. Happiness economics from 35 000 feet. J. Econ. Surv. 26, 705–735.MacKerron, G., Mourato, S., 2009. Life satisfaction and air quality in London. Ecol. Econ. 68, 1441–1453.Maddison, D., Rehdanz, K., 2011. The impact of climate on life satisfaction. Ecol. Econ. 70, 2437–2449.Metcalfe, R., Powdthavee, N., Dolan, P., 2011. Destruction and distress: using a quasi-experiment to show the effects of the September 11 attacks on mental

well-being in the United Kingdom. Econ. J. 121, F81–F103.

M

O

RS

S

T

UUVVWWWWWY

K. Rehdanz et al. / Journal of Economic Behavior & Organization 116 (2015) 500–517 517

urray, T., Maddison, D., Rehdanz, R., 2013. Do geographical variations in climate influence life satisfaction? Clim. Change Econ. 4 (1), http://dx.doi.org/10.1142/S2010007813500048.

htake, F., Yamada, K., 2013. Appraising the Unhappiness due to the Great East Japan Earthquake: Evidence from Weekly Panel Data on Subjective Well-Being. GCOE Discussion Paper No. 310. Graduate School of Business, Osaka University.

ehdanz, K., Maddison, D., 2005. Climate and happiness. Ecol. Econ. 52, 111–125.eko, M., Teruyama, H., Yamamoto, I., Higuchi, Y., Keio/Kyoto Joint Global Center of Excellence Program (Eds.), 2012. Higashi Nihon Daishinsai ga kakei ni

Ataeta Eikyo (The Impacts of East Japan Earthquake on Household). Keio University Press, Tokyo.eko, M., Teruyama, H., Yamamoto, I., Higuchi, Y., Keio/Kyoto Joint Global Center of Excellence Program (Eds.), 2013. Kakei Paneru deta kara mita shijou no

shitu (Market Quality in Household Panel Survey Data). Keio University Press, Tokyo.ov, W., Diener, E., 2007. Culture and subjective well-weing. In: Kitayama, S., Cohen, D. (Eds.), Handbook of Cultural Psychology. Guilford, New York, pp.

691–713.chida, Y., Kitayama, S., 2009. Happiness and unhappiness in East and West: themes and variations. Emotion 9, 441–456.chida, Y., Ogihara, Y., 2012. Personal or interpersonal construal of happiness: a cultural psychological perspective. Int. J. Wellbeing 2 (4), 354–369.an Praag, B., Baarsma, B., 2005. Using Happiness Surveys to Value Intangibles: The Case of Airport Noise. Economic Journal 52, 111–125.an Praag, B.M.S., Ferrer-i-Carbonell, A., 2008. Happiness Quantified: A Satisfaction Calculus Approach. Oxford University Press.elsch, H., 2002. Preferences over Prosperity and Pollution: Environmental Valuation Based on Happiness Surveys. Kyklos 55, 473–494.

elsch, H., 2006. Environment and Happiness: Valuation of Air Pollution Using Life Satisfaction Data. Ecol. Econ. 58, 801–813.elsch, H., 2008a. The Social Costs of Civil Conflict: Evidence from Surveys of Happiness. Kyklos 61, 320–340.elsch, H., 2008b. The Welfare Costs of Corruption. Applied Economics 40, 1839–1849.elsch, H., Ferreira, S., 2014. Environment, well-being, and experienced preference. Int. Rev. Environ. Resour. Econ. 7 (3–4), 205–239.

amane, F., Ohgaki, H., Asano, K., 2013. The Immediate Impact of the Fukushima Daiichi Accident on Local Property Values. Risk Anal. 33, 2023–2040.