Return Migration and Decontamination after the Fukushima Nuclear Power Plant Accidents

Transcription

1 Return Migration and Decontamination after the 2011 Fukushima Nuclear Power Plant Accidents Shingo Nagamatsu Faculty of Societal Safety Sciences, Kansai University Adam Rose Sol Price School of Public Policy, University of Southern California Jonathan Eyer Sol Price School of Public Policy, University of Southern California December 3,

2 Return Migration and Decontamination after the 2011 Fukushima Nuclear Power Plant Accidents by Shingo Nagamatsu, Adam Rose, and Jonathan Eyer Abstract: Return migration is a key to community recovery from many disasters. Japanese governments have conducted radiation decontamination efforts in the Exclusion Zone designated after the 2011 Fukushima nuclear disaster in order to encourage return migration. Little is known, however, about the factors that influence post-disaster migrants to return, and, if people are relatively unresponsive to decontamination, then the costs of promoting recovery may exceed the benefits. We exploit a unique survey of Fukushima evacuees to estimate the factors that influence their decision to return after a disaster. Location-specific capital characteristics, such as housing tenure and the extent of property damage, are strong factors for intent to return. We also found that middle income and lower-high income group members are discouraged about returning home and likely to defer their decisions, and that high-income group members are inclined not to return. The radiation dose rate of the home location was a very significant factor for intent to return, but its marginal effect on was minor. Our simulation analysis found that the number of returnees encouraged by this decontamination was 14,367, less than 10% of the total evacuees, while decontamination cost per a returnee was 3.0 million USD. This result implies that the government could have improved the well-being of evacuees at a lower cost by policies other than decontamination. Key words: population displacement, return migration, Fukushima nuclear disaster, radiation decontamination, logit analysis JEL Classifications: R10, O15, H12, H84 Acknowledgements: The research presented in this paper was sponsored by a grant from the Japan Foundation Center for Global Partnership. The authors wish to thank the participants of the USC Bedrosian Center Symposium on Population Migration and Repatriation Following Major Disasters for their helpful comments on the paper and to thank Dylan Coyle for his valuable research assistance. 2

3 1. Introduction Major disasters often cause a mass migration of the people who live in the directly affected area. The home community suffers from this population loss in several ways, including an immediate significant drop in disposable income and spending and subsequent work opportunity loss, which can result in still further out-migration. In such cases, return migration, or repatriation, is thus typically regarded to be an essential condition for community recovery. However, population return has not been an easy undertaking in recent disasters, such as the Great Hanshin Awaji (Kobe) Earthquake of 1995 and Hurricane Katrina of The cascading earthquake and tsunami disasters in Japan in 2011 caused depopulation and recovery issues throughout the region (see, e.g., Nagamatsu, 2017), but these issues were most acute in Fukushima, which was contaminated by radioactive materials from the ensuing nuclear power plant accident within its borders. The objectives of this study are two-fold. First, we will explore what factors affect a Fukushima migrant s intention to return by an econometric analysis of a survey of Fukushima evacuees. Second, based on the empirical results, we evaluate the decontamination policy for repatriation of evacuees that was formulated by the Japanese national government. 2. Literature on return migration Return migration after large disasters has been an issue of interest to many scholars in various fields (see, e.g., an early review by Hunter, 2005). Hurricane Katrina disaster of 2005 and its slow recovery process triggered several important studies on the issue. Elliott and Pais (2006) investigated the effect of race and class on the intent to return to New Orleans by using Gallup poll data on Katrina survivors. They found that home ownership increased the likelihood of returning, and household income before the disaster decreased it. Fussell et al. (2010) investigated the effect of race on returning after Katrina and found that black residents were more likely to return to their original (flooded) home than white residents. However, Reinhardt (2015) insisted that race does not have a direct impact on the intent to return, but rather works through political trust. From an economic perspective, a theoretical model of return migration was proposed by Paxson and Rouse (2008). They built on the concept of location specific capital (DeVanzo, 1981) and found that such variables as housing tenure and living in homes of relatives/friends had a significant effect on people s intent to return, but that flood exposure is the single most important factor. Landry et al. (2007) estimated the willingness to pay (WTP) of Katrina migrants to return home by modeling return probability as a function of wage differentials between their home (New Orleans) and host region (Houston). They found the differential was inversely significant as expected, with an estimated WTP 3

4 of $3,954. Yun and Waldorf (2016) presented a micro level migration-income model, based on American Community Survey data of migrants affected by Hurricanes Katrina and Rita, and found that the forced migrants incurred an income loss, as well as disaster damage, which they denoted as double victimization. Due to their relative infrequence, migration behavior in response to nuclear accidents has not been investigated nealry as much as other types of disasters. World Bank (2002), a study of the migration from the contaminated region in Belarus after the Chelnobyl accident reported a large population decrease, especially among younger residents. Rowland (2003) identified population increase in Narovlya Rayon in Belarus during , much of which lies in exclusion zone. This implies increased return migration to the Rayon evacuation zone in However, the study did not identify the factors behind the return migration trend. Goldhaber et al. (1983) determined the effect on residential mobility and population composition after the Three Mile Island nulcear accident of This study found demographic attributes of people who moved into the area were not different from those who had moved out. Several studies of the Fukushima nuclear accident have focused on return migration. Orita et al. (2013) investigated behavior of residents in Kawauchi village, only part of which was in the Exclusion Zone. They found that the radiation dose rate and people s anxiety about radiation were independently associated with the intent to return. Munro and Managi (2017) also investigated the radiation effect on people s intention to return using a logit analysis based on personal inverviews of 520 evacuees in Fukushima and Miyagi prefectures. They found that the intent to return is only weakly responsive to changes in ambient radiation levels, which implies that decontamination policy may have only a limited impact on eventual return migration. They also found that high-income individuals and those without children were more likely to return. 3. Fukushima Daiichi nuclear power plant accident of Overview of the event The 2011 Fukushima Daiichi nuclear power plant accident, which was ranked as a seven, the highest score on the International Nuclear and Radioactive Event Scale (INES), was one of the most serious nuclear disasters in human history. On March 11, 2011, a tsunami generated by a M9.1 earthquake flooded the basement of the plant s turbine buildings in which diesel generators and switchboards were located. This made the plant unable to provide electricity to cool the reactors and caused core meltdowns in reactors Nos. 1, 2, and 4, releasing significant amounts of radionuclides to the 4

5 atmosphere. Following the emergency, an evacuation order was announced by the national government of Japan to the residents living in a 3km radius area from the plant, but the area was subsequently expanded as the accident was proven to be more serious. On April 21, after 40 days of the tsunami, the national government established several zones to control radiation. The Evacuation Zone (highlighted in blue in Figure 1), in which no one except designated persons were allowed to stay or visit, was set within 20km radius from the plant. A Deliberate Evacuation Zone (navy in Figure 1) was established beyond the 20 km evacuation zone to include areas where the projected dose criterion of 20 msv/year might be exceeded. Although the evacuation order had not been announced then, the residents who lived in this area were mandated to move out within a month of the establishment of this zone. Finally an Evacuation Prepared Area in Case of Emergency (orange in Figure 1) was established to warn the people who live there to evacuate immediately if the situation at the plant were to become worse. Designation of this area was cancelled on September 30, 2011, as the condition of the damaged reactors became stable. Instead, as some areas outside the aforementioned evacuation zones were proven to have a projected dose rate above 20 msv/year sporadically even outside the zones, the national government designated them as Specific Spots Recommended for Evacuation (IAEA, 2015). The Evacuation Zone and Deliberate Evacuation Zone had been strictly delineated under government control. In March 2012, these areas were restructured into the following three zones depending on the radiation level: 1) Difficult to Return zone, in which the current annual cumulative radiation dose rate exceeds 50 msv/year and may exceed 20 msv/year even in 5 years after the accident; 2) Restricted Habitation, in which the annual cumulative dose rate exceeds 20 msv/year; and 3) Preparing to Lift Evacuation Order, in which the annual cumulative dose rate is under 20mv/year (Ministry of Environment 2018). We henceforth term these areas as the Exclusion Zone in this study, because evacuation from these areas was mandatory. However, there were many other people who evacuated from areas outside the Exclusion Zone (hereafter Non-Exclusion Zone ), who primarily wished to protect themselves from the risk of radiation exposure. The national government, therefore, has distinguished these two types of evacuees; mandatory evacuees as the former and voluntary evacuees as the latter, and provided different categories of support programs. As of June 2012, approximately 164,000 people who had lived in Fukushima Prefecture evacuated at the peak. As of October 2014, 81,000 were estimated to evacuate from the Exclusion Zone (mandatory evacuees), and the other 83,000 were from the Non-Exclusion Zone (voluntary evacuees) (Ministry of Environment, 2018). 5

6 Figure 1. Fukushima Exclusion Zones as of April 22, Source: IAEA (2015) Decontamination and recovery Recovery from the nuclear disaster was a difficult issue that required the consideration of several factors. The first factor was uncertainty of the recovery process. Many evacuees did not know when and how much the contaminated area would be cleaned up. People did not even know if they could go back home in some highly contaminated areas. Partly because of the uncertainty, a significant number of people made up their mind not to return and to start life anew in the place to which they evacuated. Some other people did not make a proactive decision, but found it difficult to return because they had rooted themselves in the community to which they evacuated. Decontamination was the most important issues in every place contaminated by the radionuclides, both the exclusion and non-exclusion zone. It was regarded as the most fundamental activity to eliminate the health risk. In particular, reducing the radiation dose rate in the Exclusion Zone was a necessary condition for lifting the evacuation order, repatriating the evacuees, and recovering peoples livelihoods in the zone. The national government legislated an Act on Special Measures concerning decontamination (Ministry of Environment, 2015) in January Under this legal framework, the national government decided a basic policy of decontamination on August

7 The policy established a basic framework of decontamination: 1) the national government was responsible for decontamination of the Exclusion Zone, and for providing technical and financial support for municipal governments to implement decontamination plan, 2) decontamination in the areas with additional radiation dose below 20mS/year (the Preparing to Lift Evacuation Order area) was required to lower the dose rate to less than 1mSv/year in a long term. Decontamination in more contaminated areas did not have a target dose rate, but was to aim at decreasing the restricted zone gradually and promptly. As of March 23, 2018, the scale of the decontamination work undertaken by the national government and completed was: 23,000 houses, farmlands of about 87,000 ha, forests of 7,800 ha, and roads of 1,500 ha (Ministry of Environment, 2018). The decontamination work, on the other hand, has generated a serious issue of radioactive waste disposal because of the work itself. The Japanese government decided to construct an interim storage facility for preserving the waste in Futaba and Okuma towns, both of which were included in the Difficult to Return zone, promising that the waste would be disposed finally outside Fukushima after about 30 years, expecting that the radioactivity of the waste would reduce to half through natural decay. However, construction of the facility itself was not an easy task. For example, the national government has been obtaining ownership of the site, which needs to have contract with 2,360 landowners. However, only 64.4% of them have been made by August Table 1 shows the decontamination cost shown in the budget of Japanese governments, both national and local. All costs are transformed in USD term, assuming 1 USD=112 JPY as of March The total cost of decontamination was estimated to be about 29.0 billion USD by March 2017, the end of fiscal year. The money actually spent was about 23.4 billion USD. In addition, the national government is estimating 14.3 billion USD (16 trillion JPY) for constructing the interim storage facility. The total cost of treatment and storage of contaminated soil was estimated to be 43.3 billion USD (Ministry of Economy, Trade, and Industry (METI) 2016). (Table 1 to be shown here) 4. The Model and Data 4.1 Conceptual framework. The majority of the literature on post-disaster migration deals with displacement and resettlement, with lesser contributions focused on population return. Recently, Rose et al. (2018) outlined a framework for the economic analysis of repatriation, acknowledging that economics is only one of the several important dimensions of the issue. The framework identifies the many key factors relating to background conditions at the home and new locations, characteristics of the 7

8 resettlement area, and push and pull factors affecting the decision to return. The factors extend beyond the narrow economic realm to include socioeconomic and demographic variables. The vast majority of the variables have previously been identified in the literature, and many of them have been found to be statistically significant in case studies. The framework, however, is the most comprehensive compilation of these factors to date and provides a valuable analytical grouping, or typology, of them. The framework also examines how the various factors influence the well-being of disaster evacuees in their resettlement locations and original home locations, where conditions have likely changed dramatically since they left. It also serves as a useful guide for data collection. Finally, it provides a foundation for devising policy remedies that link a reduction in vulnerability to improved well-being by taking into account the realities of the situation, including individual motivations, broader societal objectives, constraints, and obstacles to policy design, implementation and effectiveness. We utilize the framework as the underpinning of our econometric analysis below. It helps us identify key explanatory variables and a priori hypotheses about the direction of their influence on return migration. The framework also broadens the analysis to distinguish factors affecting both the decision to return and not to return. Finally, the policy analysis aspect of the framework provides a broader perspective on current policy in terms of not just pursuing the narrow goal of repatriation for its own sake, but rather pursuing the broader goal of maximizing the well-being of evacuees in relation to both remaining in the resettlement area or returning home. Ideally, we would already have applied the framework to the collection of appropriate data, but rather than wait for this to be done, we have applied it to a valuable recent survey of resettlement associated with the Fukushima disaster. We acknowledge this data set does not include all the variables identified by the repatriation framework. However, the framework enabled us to omit consideration of extraneous ones to our task. 4.2 Formal Model The return decision is not necessarily dichotomous. Many migrants from the affected areas have to decide their future location under uncertainty of recovery processes and the surrounding destination environment. In such cases, deferring the return decision could be a rational response. In this sense, return and not return decision are not necessarily two sides of a coin. Therefore, we should assume the decision processes are asymmetric between returning and not returning. To reveal such complexity, we employ a multi-nominal logit model further with a simple logit model being used in most previous studies. Suppose that an individual has three options on the intention to return, denoted as J. We define that J = 1 if the individual will return to his/her home, J = 2 if the individual will not return to his/her home, and J = 3 if the individual has not decided yet, about 8

9 his/her repatriation decision. In the multi-nominal logit model, the probability of individual i choosing option J can be given by the following formula; exp (x i β J ) P(Return i = J) = 1 3, J = 1, 2, and 3 J=1 exp ( x i β J ) where x i denotes an explanatory vector of variables of individual i, and β J denotes a vector of parameters relating to the intent to return option J. We include three types of variables in x ij : individual attributes, damages, and radiation exposure. 5. Data Most of the data we use are from the Alternative Dispute Resolution Center for Nuclear Accident Disaster Compensation (ADR Center) dataset, which was established after the accident by the Japan Ministry of Education, Culture, Sports, Science, and Technology (MEXT). The dataset was created from a survey-based study conducted by the ADR Center. Its primary purpose was to identify whose, what type of, and how much loss was incurred due to the evacuation in order to provide a basic fact for remediation of disputes on compensation issues. Including both economic and emotional. However, the broad set of variables collected is well suited to the range of questions we plan to address in our study. The questionnaire was distributed to: 1) households who had lived in Exclusion Zones and who had registered their residential relocation address regarding themselves as evacuees; or 2) households who had lived in other area and who had registered the residential address where they relocate at national government, regarding themselves as evacuee. The survey was conducted from March 1 to 19, The number of target households was 41,754, and 10,082, respectively, of which 24.1% overall responded. This is the only survey to date that can capture the path of the refugees from the Fukushima nuclear power plant accident. The radiation dose rate at the original location is an average value of dose rates across all grids that cover the entire jurisdictional area of a municipality, computed from the dataset of MEXT, USDOE, and NRA and released in 2012 (JAEA, 2012). The reason for this refinement is that the ADR dataset does not identify the location where the respondent was before and after the accident more precisely than the municipal level. The variable distance is a measurement of how far the respondents have migrated away from their hometown, which is measured as the length between city halls of the original municipality and the current one, calculated from x-y data by the authors. We expect that the distance from the original municipality would negatively affect the intent to return because longer distances are likely to result in higher moving and resettlement costs. However, one suspects that distance from home is endogenous because people who intended to return soon would try to reduce return cost and 9

10 therefore would stay as near as possible to their home. Including distance as an explanatory variable could thus cause a significant bias on the estimated value of coefficients. One possible solution to the endogeneity issue is introducing instrumental variables that can predict distance but have no effect on the individual choice of returning except through the effect of distance. However, most of the variables we have, such as migration cost and radiation dose rate at the current municipality, are unlikely to pass the exclusion restriction. We therefore introduce the historical migration trend from the exclusion zone to the municipality before the disaster, from 2005 to 2010, as an instrumental variable for distance. The rationale is as follows: the historical migration pattern is determined by several factors, but distance is undoubtedly a key factor. Fewer people migrate to the area if it is more distant from home. Thus, distance of two locations are is correlated to the past migration pattern to some degree. On the other hand, the historical migration pattern is exogenous to people s decision on migration following the disaster. This is a similar method to that was used by Munro and Managi (2017), instrumenting current location by the average value of others who migrated from the same location. However, as they admit, there is some possibility that the intent to return is conditional on friends and neighbors, thus potentially biasing their results. Our method does not include such a possibility because we used the past migration pattern, which is independent from the disaster. We include several additional explanatory variables in our statistical analysis. The choice of variables emanates from the conceptual framework we developed for repatriation analysis. At the same time, our choices are limited by the data collected in the ADR survey. Descriptive statistics of the variables and their sources are presented in Table 2. Our dataset is not balanced because not every respondent answered every questions in the ADR survey. The remainder of the variables not elaborated on above are self-explanatory. (Table 2 to be shown here) 5. Factors that affect intent to return Logit analysis We begin with the simple logit model to compare our result to existing studies. We created a dummy dependent variable that takes the value of 1 if a person wishes to return (J = 1) and 0 if a person has not made a decision yet or has decided not to return (J = 2 or J = 3). The estimated results of this logit model are shown in Table 2. The columns shown as Eq.1 is the base model, whereas Eq.2 introduced cross-terms between personal attributes and radiation dose rate following Munro and Managi (2017). Eq. 3 employed second-stage estimation using number of migrants from home before the disaster as an instrumental variable for distance. Finally, samples are split into two 10

11 groups: those who lived in the Exclusion Zone (Eq. 4) and those who evacuated from Non Exclusion Zone (Eq.5). (Table 3 to be shown here) Estimation results of Eq.1, 2, and 3 shows the factors that encourage migrants intention to return. In regard to personal attributes, male migrants, migrants whose age is over 60, and the variable reflecting migrants who have an elder family member are found to be likely to have a higher intent to return. On the other hand, migrants who have children has a negative impact on the intent to return. However, contrary to the result of Munro and Managi (2017), the cross-terms with Radiation show low significance. Housing tenure encourages the migrants intent to return, and this result is consistent with many existing studies (Elliott and Pais 2006, Landry et al. 2007, Paxson and Rouse 2008, Groen and Polivka 2010, Reinhardt 2015, Yun and Waldorf 2016). In regard to housing damage, house collapse and house inundation have a negative impact on returning, which was also indicated in many existing studies (Elliott and Pais 2006, Paxson and Rouse 2008, Reinhardt 2015). Interestingly, house contamination does not have any impact on intentions to return. This is not surprising, however, because people are anxious about the radiation dose of not only their houses but also those of their neighbors. This view is supported by the results that radiation dose rate at the origin is negatively significant. In regard to livelihood of the migrants, our preliminary analysis included several dummy variables to see if the industry in which the migrant had been working is influential for the return migration decision. Among them, Sales and Restaurants was the only significant sector. The result shows that working in this sector has a negative impact on return migration, probably because the industry is heavily dependent on local population, and not likely to recover unless evacuee repatriate. However, this impact is not obvious when we split the sample (Eq.4 and 5). The variable distance has a negative impact on intention to return in Eq.1, 2, and 3. However, the absolute value of the coefficient in Eq. 3 is far larger than that of Eq 1 and 2, which implies that the endogeneity biases of the coefficients in Eq. 1 and 2 are corrected by the instrumental variable in Eq. 3. The coefficient of the exclusion zone are very intuitive -- it affects intentions to return negatively. Coefficients of Income dummy variables above 4 million JPY are negative, which implies that wealthier people are less likely to plan on returning than those making below 2 million JPY per year. Point estimates generally become more negative as income rises, though the effects are not generally statistically different from one another. Existing studies do not necessarily have a consensus on how income affects a person s intent to return. For example, Landry, et al. (2007) found a positive 11

12 relationship between income and intent to return from an econometric analysis of Hurricane Katrina migrants. They explain that higher income enables people to have better resources to make the return trip, are more likely to own homes in areas less likely to have been flooded, and have better resources to rebuild in the event that their home has been damaged. The study by Reinhardt (2015) yielded similar results. More notably, Munro and Managi (2017) found this was true for Fukushima evacuees. Our result here is opposite to these studies, rather consistent with Elliott and Pais (2006), which found a negative correlation between higher income and intent to return. Their explanation is that lower income affords them less opportunity to pursue options elsewhere. We next split the samples into two categories, one for those who migrated from Exclusion Zone (Eq. 4) and the other from Non-exclusion zone (Eq. 5). The people in these two types of samples should have different attitudes towards radiation, because the samples in the former are mandatory evacuees, while those of the later are voluntary. From these facts, we expect that the return decision process is simpler in the Non-Exclusion Zone, because their primary concern would be radiation dose rate. Our result is consistent with those expectations. Coefficients of dummies for males is not significant in the Non-Exclusion Zone. The self-employed variable also lost its significance. Interestingly, income dummies are all insignificant. In addition, distance lost its significance at the 10% level. We could expect that the intent to return is affected by the radiation dose rate more in the Non-exclusion zone than the Exclusion Zone. For the sake of this comparison, we estimated the marginal effects on radiation dose rate for each estimation. The results are consistent with this expectation. The marginal effect of the radiation dose rate is in the Exclusion Zone and in the Non-Exclusion Zone. This result is also consistent of the estimation of Munro and Managi (2017). Their estimated coefficients are between to , which are in-between our Exclusion Zone and Non-Exclusion Zone estimates. This is because their analysis did not split the sample between these two zones Multi-nominal Logit analysis The results of multi-nominal logit analysis are shown in Table 3. The columns noted R show the model of return decision ( will return ) vs. deferring decision ( not decided yet ), while the NR column presents the not returning decision ( will not return ) vs. deferring decision ( not decided yet. ) (Table 4 to be shown here) The estimation results of Eq. 6 reveal an interesting contrast in the decision-making processes between returning and not returning. Coefficients of housing tenure, house collapse, Exclusion 12

13 Zone, and radiation dose rate are significant in both R and NR, and have opposite signs. This means that these factors have a symmetric effect on returning and not returning decisions. For example, a person with housing tenure who incurs no housing collapsing damage, who lived outside of the Exclusion Zone, and whose original location is under lower radiation dose rate is likely to return, and is less likely not to return. We term these factors as strong factors on returning. On the other hand, some other variables such as male, elderly, house inundation, and sales and restaurant are significant only in the returning decision. This implies that being male, no inundation damage to the house, not working in the sales and restaurant sector, and proximity of current location encourage people return home. However, these factors do not discourage them to make a decision not to return. We term these as weak return factors. Interestingly, the coefficient of age over 60 is positive in both the returning and not-returning decisions. This implies that a household headed by an elderly person is inclined to decide definitively either to return or not to return, perhaps because of a stronger desire for settling down as early as possible and living the rest of their life in peace. It is also interesting to note the influence of the coefficients of the income dummies. The coefficients of JPY 10M and higher are not significant in the return decision, but are significant in the not-returning decision. This implies that an Upper-High income (JPY 10M and higher) encourages people to decide not to return rather than to defer the decision (a weak factor for not to return), while Lower-Middle, Upper-Middle, and Lower-High incomes (JPY 4M to 6M, 6M to 8M, and 8M to 10M) encourage people to defer the decision These income groups have relatively greater options for their future than Lower-income groups, and therefore that pushes people to defer the decision. In that sense, these income groups have enough resources to cope with the uncertainty of the recovery process they may face. However, because Upper-High income households (JPY 10M and higher) have enough resources to re-establish life in a new location, they are more inclined not to return. This implies that monetary transfers, such as compensation from TEPCO and life support assistance from governments, discouraged the migrants to return home, or even encouraged them to settle down anywhere else. We again split the sample between the Exclusion zone (Eq. 7) and Non-Exclusion Zone (Eq.8). We found that house collapse has different effects between these zones -- in the Exclusion Zone, house collapse discourages people to return, but in the Non-Exclusion Zone, it encourages people to make the decision not to return. The instrumented variable distance was a weak factor in the return decision. However, it was also a weak factor in the not to return decision in the Non-Exclusion Zone. As we observed in the previous analysis, income dummy variables have no significant impact on people s decision in the Non-Exclusion Zone. Finally, the radiation dose rate became only a weak factor for returning in Non-Exclusion Zone. This implies that decontamination is effective to encourage some voluntary evacuees to return, but has no impact on not returning decision of voluntary evacuees. 13

14 6. Simulation analysis of decontamination effect on intention to return Using the parameters estimated in Eq. 4 and 5, we estimated how the intent to return would change if the radiation dose rate decreased. We use the dose rate data as of October 2016, which is the latest dataset available. We assume that an evacuee will return if his/her projected probability of returning exceeds 0.5. In more formal expression, we assume individual i will return home if P(Return i = 1 dose i = dose i (2016)) 0.5, where dose i (t) denotes radiation dose rate of individual i s home at time t. Table 5 presents the simulation result for the municipalities in the Exclusion Zone, both totally and partially. From March 2012, the average dose rate decreased considerably, from 60.5% in Tamura to 77.9% in Tomioka. This is partly because of cleanup efforts of the National government, while also partly because of natural decay of radiation and weathering effects. The results of our simulation are presented in the middle columns of Table 5. The result shows that the radiation effect on peoples intention to return is very minor, from 0.6% in Tamura to 19.2% in Iitate. Overall in the Fukushima prefecture, including the Non-Exclusion Zone, the results suggest that 11.8% of migrants decided to return because of the decrease of the radiation dose rate. (Table 5 to be shown here) In the last two columns, we compare the result with factual data from census or poll as of 2016 for each municipality conducted by the Recovery Agency. The simulation fits very well in some municipalities, such as Okuma and Naraha, Tomioka, Futaba, and Iitate. However, there are large gaps between the simulation and the factual data seen in some municipalities, such as Minamisoma, Kawamata, and Katsurao. This is largely because some polls were limited to the population for whom the evacuation order was lifted immediately before the poll. This treatment eventually would result in a higher percentage of those who wish to return than the average of the population in each municipality. In this sense our simulation provides fairly reasonable results. This simulation suggests that decontamination does not increase the intention to return significantly. This implication becomes more solid if we consider the fact that the decrease of the radiation dose rate in Fukushima was not affected only by decontamination, but also by natural decay and weathering effects. From the estimates that we have derived, we can calculate the cost of decontamination for repatriating one person. As discussed earlier, the total number of evacuees is estimated as 164,000, of which 81,000 are mandatory and the other 83,000 are voluntary. Multiplying the estimated 14

15 increment percentages who wish to return ([B] - [A] in Table 5) to the number of evacuees yields the incremental estimate of the number who will return home. The results are 8,262 for the Exclusion Zone and 6,225 for Non-Exclusion Zone. So, in total, 14,487 evacuees were estimated to return because of the radiation decrease from March 2012 to March The cost for decontamination was estimated as 43.3 billion USD, including construction cost of the interim storage facility. Hence, the cost of decontamination per returning person is estimated as 3.00 million USD (336 million JPY). 7. Conclusion and policy implications Our analysis of migrants intention to return to contaminated areas of Fukushima revealed that housing tenure, house damage, and house inundation are the most significant factors will. This is consistent with the theory of location-specific capital that Paxson and Rouse (2008) developed. Income, on the other hand, is also a powerful indicator that predicts higher income is less likely to return, which is consistent with the result of Elliott and Pais (2006). Our multinominal logit analysis shows that income level works in discouraging return migration in either way; middle-income households are inclined to defer the return decision, while high-income households are inlined not to return. This result can be explained in part by uncetainty associated with how much radiation will be decreased, how much government will help them retrun or not return, and when government will lift the evacuation order. The higher income allows people to overcome some of the issues related to these uncertainties and to choose a more certain future by settling down in the place to which they evacuated. Our regression results also found out that the return decision of voluntary evacuees who migrated from the Non-Exclusion Zone is more sensitive to the radiation dose rate. This is a reflection of the structure of our samples. Mandatory evacuees included persons who would not leave unless the evacuation were mandatory, and therefore whose marginal effect of radiation would likely be very low. On the other hand, the sample of voluntary evacuees does not include such people. All of them evacuated primarily because they feared the accident and radiation, and therefore their marginal effect of radiation should be higher. In this sense, there are certain reasons why government should provide financial support to voluntary evacuees as they do to mandatory evacuees -- the term voluntary does not necessarily mean their motivation and necessity to evacuate, including their loss in economic well-being caused by evacuation, is weaker than that of mandatory evacuees. In the simulation analysis on the returning population, we found that the number of returnees encouraged by the decontamination work of the government was 14,487, or less than 10% of total evacuees. Decontamination cost per returnee was estimated as 3.0 million USD. 15

16 Rose, Eyer, and Nagamatsu (2018) have provided a policy framework of repatriation after catastrophic event, suggesting that repatriation is not necessarily the logical ending of disaster odyssey. Even though the sense of home is a strong motivation of recovery for most evacuees, the large cost of decontamination would have allowed government to consider many other options for recovery. For example, according to the result of Eq. 3, the intent to return of the migrants will decrease if they have housing tenure in their current place, business opportunities for the selfemployed, and annual income more than JPY 4m. Our analysis implies that the migrants welfare is likely higher for many than it would be if they returned home. Thus, a government policy goal of complete repatriation is not necessary if the maximization of the economic well-being of all of those affected is the end goal. Many are better off in their new locations, and others may benefit equally from other government policies, such as provision of houses, enhancing community ties, and individual support for the elderly. Moreover, these other policies may attain the goals of enhancing the well-being of evacuees at a much lower cost than would further decontamination in many geographic areas. References DaVanzo, J. (1981). Repeat migration, information costs, and location-specific capital. Population & Environment: Behavioral & Social Issues 4(1), Elliott, J. R. and J. Pais (2006). "Race, class, and Hurricane Katrina: Social differences in human responses to disaster." Social Science Research 35(2): Fussell, E., et al. (2010). "Race, socioeconomic status, and return migration to New Orleans after Hurricane Katrina." Population and Environment 31(1-3): Goldhaber, M., et al. (1983). "Moving After the Crisis: A Prospective Study of Three Mile Island Area Population Mobility." Environment and Behavior 15(1): 93. Groen, J. A. and A. E. Polivka (2010). "Going home after Hurricane Katrina: Determinants of return migration and changes in affected areas." Demography 47(4): Hunter, L. M. (2005). "Migration and Environmental Hazards." Population and Environment 26(4):

17 International Atomic Energy Agency (IAEA). (2015). The Fukushima Daiichi Accident Report by the Director General. Japan Atomic Energy Agency (JAEA)(2012) Air Dose Rate Results of the Airborne Monitoring Survey by Prefecture(Decay correction: May 31, 2012) ( accessed on Sep. 15, 2018) Landry, C. E., et al. (2007). "Going home: Evacuation-migration decisions of Hurricane Katrina survivors." Southern Economic Journal 74(2): Ministry of Environment (2018). Decontamination Projects for Radioactive Contamination Discharged by Tokyo Electric Power Company Fukushima Daiichi Nuclear Power Station Accident. (in Japanese) Ministry of Economy, Trade, and Industry(METI) 2016) Power point presentation for Annex, the 6 th meeting of the Committee for TEPCO Reform and 1F Issue. ( df, accessed in Sep. 3, 2018) Munro, A. and S. Managi (2017). "Going Back: Radiation and Intentions to Return amongst Households Evacuated after the Great Tohoku Earthquake." Economics of Disasters and Climate Change 1(1): Nagamatsu, S.(2018) Building Back Better Tohoku After the March 2011 Tsunami: Contradicting Evidence, in Vicente Santiago-Fandiño et al. (eds.) The 2011 Japan Earthquake and Tsunami: Reconstruction and Restoration, Chapter 3, pp Orita, M., et al. (2013). "Determinants of the return to hometowns after the accident at Fukushima Dai-ichi nuclear power plant: a case study for the village of Kawauchi." Radiation Protection Dosimetry 156(3): Paxson, C. and C. E. Rouse (2008). Returning to New Orleans after Hurricane Katrina. American Economic Review 98(2): Reconstruction Agency (2017). Survey report on intention to return in Exclusion Zone (in Japanese,

18 4/ikoucyousa/28ikouchousakekka_zentai.pdf, accessed on Sep. 5, 2018) Reinhardt, G. Y. (2015). "Race, Trust, and Return Migration: The Political Drivers of Post-disaster Resettlement." Political Research Quarterly 68(2): Rose, A., J. Eyer, and S. Nagamatsu (2018). Toward a Theory of Population Repatriation from Disasters. Center for Risk and Economic Analysis of Terrorism Events (CREATE), University of Southern California, Los Angeles, CA. Rowland, R. H. (2003). "Population trends in Belarus and the impact of Chernobyl, ," Eurasian Geography and Economics 44(4): World Bank (2002). Belarus Chernobyl Review. Yun, S. and B. S. Waldorf (2016). "The Day after the Disaster: Forced Migration and Income Loss after Hurricanes Katrina and Rita." Journal of Regional Science 56(3):

19 Table 1. Decontamination Cost covered by Japanese governments Unit: million USD (USD1=JPY112) Budget Amout spent (closing) National Municipal Total National Municipal Total FY2011 discretionary 140 1,713 1, ,716 1,838 FY2011 supplementary , FY2012 initial 2, ,322 2, ,190 FY2013 initial 2,633 1,812 4,445 2,582 1,782 4,364 FY2013 supplementary FY2014 initial 1,061 1,245 2,305 1,022 1,211 2,233 FY2015 initial 2,155 1,571 3,727 2,075 1,561 3,636 FY2015 supplementary FY2016 initial 2,607 2,080 4,688 1,379 2,079 3,458 FY2016 supplemantary 1,243 1,710 2, ,710 2,413 FY2017 initial 1,445 1,104 2,550 Total 14,585 14,459 29,046 10,354 13,083 23,437 Source: Ministry of Environment(2018) Decontamination Projects for Radioactive Contamination Discharged by Tokyo Electric Power Company Fukushima Daiichi Nuclear Power Station Accident, Table 3-7 on page 74 (in Japanese, traslated by the author) 19

20 Table 2. Data description Variables Description N Mean St. Dev. Return Coded 1 if the respondent will return, 2 if not decided yet, 3 if will not return. 9, Male (dummy) Coded 1 if the respondent was male. 10, Age over 60 (dummy) Coded 1 if the respondent was 60 and over. 10, Children (dummy) Eldery (dummy) Housing tenure (dummy) House collapse (dummy) House induncation (dummy) House contamination (dummy) Coded 1 if the respondent's family have more than one children whose age was 12 and less. 10, Coded 1 if the respondent's family have more than one elder persons whose age was 70 and over. 10, Coded 1 if the respondent own house and land where he/she lived before the disaster. 9, Coded 1 if the respondent's house were totally collapsed / washed away by the earthquake / the tsunami. 9, Coded 1 if the respondent's house were inundated by the tsunami. 9, Coded 1 if the respondent's house were contaminated by radionuclides. 9, Self-employed Coded 1 if the respondent was self-employed worker. 9, Sales and restaurant Coded 1 if the respondent's business is either whole (dummy) seller, retail seller, or restaurant. 9, Employer housing Coded 1 if the respondent lives in the house provided by (dummy) his/her empoyler. 9, Distance Distance between the respondents' home municipality and current municipality. 9, Exclusion zone (1=Yes, Coded 1 if the respondent lived in the exclusion zone 0=No) before the disaster. 10, Income:JPY 2M to 4M (dummy) 9, Income:JPY 4M to 6M 9, Coded 1 if the respondent's income before the disaster Income:JPY 6M to 8M 9, was within the range. Income:JPY 8M to 10M 9, Income:JPY 10M and over (dummy) 9, Radiation dose rate at An average radiation dose rate of the municipality where original place (msv/y) the respondent lived before the disaster as of March ,

21 Table 3. Results of logit models Eq Eq.1 Eq.2 Eq.3 Eq.4 Eq.5 Sample All samples All samples All samples Exclusion zone Non Exclusion zone IV for distance Male (dummy) Age over 60 (dummy) Children (dummy) Eldery (dummy) Male * Radiation Age over 60 * Radiation Children * Radiation Eldery * Radiation Housing tenure (dummy) House collapse (dummy) House induncation (dummy) House contamination (dummy) Self-employed (dummy) Sales and restaurant (dummy) Employer housing (dummy) Distance Exclusion zone (1=Yes, 0=No) Upper Low Income (dummy): JPY 2M to 4M Lower Middle Income (dummy): JPY 4M to 6M Upper Middle Income (dummy): JPY 6M to 8M Lower High Income (dummy): JPY 8M to 10M Upper high Income (dummy): JPY 10M and higher Radiation dose rate at original place (msv/y) Constant Marginal effect of Radiation dose rate at original place Number of observation Log likelihood Psedo R-squared AIC BIC None None Number of migrators from exclusion zone before the disaster Number of migrators from exclusion zone before the disaster Number of migrators from exclusion zone before the disaster *** ** *** *** (0.060) (0.060) (0.060) (0.060) (0.060) *** *** *** *** *** (0.070) (0.070) (0.070) (0.070) (0.070) *** * ** ** (0.080) (0.080) (0.080) (0.080) (0.080) *** ** *** *** ** (0.070) (0.070) (0.070) (0.070) (0.070) (0.000) (0.000) (0.000) (0.000) *** *** *** *** *** (0.080) (0.080) (0.080) (0.080) (0.080) *** *** *** *** ** (0.200) (0.200) (0.200) (0.200) (0.200) *** *** *** * ** (0.060) (0.060) (0.060) (0.060) (0.060) (0.210) (0.210) (0.210) (0.210) (0.210) ** ** ** ** (0.080) (0.080) (0.080) (0.080) (0.080) ** ** ** * (0.160) (0.160) (0.160) (0.160) (0.160) * * * (0.170) (0.170) (0.170) (0.170) (0.170) *** *** ** * * (0.020) (0.020) (0.000) (0.000) (0.000) *** *** *** (0.080) (0.080) (0.080) * (0.090) (0.090) (0.090) (0.090) (0.090) ** ** ** *** (0.090) (0.090) (0.090) (0.090) (0.090) ** ** ** *** (0.110) (0.110) (0.110) (0.110) (0.110) *** *** *** *** (0.140) (0.140) (0.140) (0.140) (0.140) *** *** *** *** (0.150) (0.150) (0.150) (0.150) (0.150) *** *** *** *** *** (0.000) (0.000) (0.000) (0.000) (0.000) *** *** ** *** (0.110) (0.110) (0.110) (0.110) (0.110) *** *** *** *** *** (0.000) (0.000) (0.000) (0.000) (0.001) Note: *** p<.01, **p<.05, *p<.1 Robust standard errors are shown in parenthesis 21