Impact of the Fukushima nuclear accident on obesity of children in Japan (2008–2014)

Accepted Manuscript Title: Impact of the Fukushima nuclear accident on obesity of children in Japan (2008–2014) Author: Eiji Yamamura PII: DOI: Reference:

S1570-677X(16)00002-2 http://dx.doi.org/doi:10.1016/j.ehb.2016.01.001 EHB 558

To appear in:

Economics and Human Biology

Received date: Revised date: Accepted date:

26-9-2015 5-1-2016 7-1-2016

Please cite this article as: Yamamura, E.,Impact of the Fukushima nuclear accident on obesity of children in Japan (2008ndash2014), Economics and Human Biology (2016), http://dx.doi.org/10.1016/j.ehb.2016.01.001 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ip t cr

us

Highlights We studied the effect of the 2011 Fukushima nuclear accident on childhood obesity. A difference-in-differences approach was used.

an

For cohorts aged 5–7, obesity was higher in damaged areas than in other areas.

For cohorts aged 5–7, the influence of the accident persisted, even after 3 years.

Ac c

ep te

d

M

The increase in obesity after the accident was greater in Fukushima than elsewhere.

Page 1 of 39

ip t cr

Eiji Yamamura*

an

Department of Economics, Seinan Gakuin University, Fukuoka, Japan

us

Impact of the Fukushima nuclear accident on obesity of children in Japan (2008–2014)

ep te

d

M

*Corresponding author. Department of Economics, Seinan Gakuin University, 6-2-92 Nishijin, Sawara-ku, Fukuoka 814-8511, Japan. Tel.: +81 92 823 4543; fax: +81 92 823 2506. E-mail address: [email protected] (E. Yamamura).

Ac c

ABSTRACT This study used prefecture-level panel data from Japan for the period 2008–2014 to investigate the influence of the 2011 Fukushima nuclear accident on the body mass index (BMI) z-score and obesity rates of children over time. I adopted a difference-in-differences approach and found the following: (1) for the cohort aged 5–7 years in 2010, the BMI z-score and obesity rates in disaster-affected areas were higher than in other areas, although this was not observed for the other cohorts; (2) for the cohort aged 5–7 years in 2010, the influence of the accident persisted even after 3 years; and (3) the differences in the BMI z-score and obesity rate before and after the accident were greater for Fukushima Prefecture than for the other affected areas (Iwate and Miyagi prefectures). I infer that health-conscious parents, whose children had lower BMIs, may have moved from Fukushima, thereby increasing the BMI z-score of the child population living in Fukushima by around 0.05 for the cohort aged 5–7 years. The enforced reduction in physical activity increased the BMI z-score of children living in Fukushima by around 0.19 for that cohort. JEL classification: I18; H12. Keywords: Fukushima Nuclear accident Body mass index Obesity

Page 2 of 39

ip t cr us an M 1

Ac c

ep te

d

1. Introduction On March 11, 2011, Japan was struck by a large-scale natural disaster, combining both an earthquake and tsunami. These damaged the Fukushima Daiichi Nuclear Power Plant, located on the coast of Fukushima Prefecture in northeast Japan. A level-7 nuclear disaster rating was assigned—a level previously reached only once, in the Chernobyl disaster. Inevitably, Fukushima’s residents had to deal with the danger of radiation exposure, and a number of people left the area (Japan Times, 2012; Matsuda et al., 2015). The Great East Japan Earthquake and Fukushima nuclear accident had a substantial impact on economic conditions (Ando and Kimura, 2012; Hayashi, 2012) and happiness levels (Uchida et al., 2014; Rehdanz et al., 2015; Yamamura et al., 2015) in Japan. According to media reports, the nuclear accident led to “a lack of physical exercise and stress stemming from prolonged living in shelters and restrictions on playing outside” (Yomiuri, 2012). Consequently, “an alarming trend toward obesity has been found among children in Fukushima Prefecture, which has the highest rate of obese children in every age group between 5 and 9 years old” (Yomiuri, 2012)1. Studies that have assessed other nuclear accidents, such as Three Mile Island and Chernobyl, provide evidence that nuclear accidents have both short- and long-term detrimental effects on human lives2. However, little is known about how and to what extent the Fukushima accident has affected body mass index (BMI) z-scores and changes in obesity rates. Overweight children are thought to have a higher risk of developing various diseases in later life. This raises the possibility that the Fukushima accident has indirectly influenced residents’ health status through rising obesity. It is therefore important to assess the effect of the Fukushima accident on children’s physical In addition to Fukushima Prefecture, children in Miyagi and Iwate prefectures had high obesity levels (Yomiuri, 2012). The Chernobyl accident was found to reduce happiness levels (Danzer and Weisshaar, 2009) and the performance of the labor market in the Ukraine (Lehmann and Wadsworth, 2011). That accident also exerted effects on other European countries. For instance, Germans were found to be more likely to worry about the environment after the Chernobyl disaster (Berger, 2010). In Sweden, students born in regions exposed to higher levels of Chernobyl radiation fallout produced poorer performances at secondary school (Almond et al., 2009). Other major disasters have been shown to influence the outcomes of elections and policies in the United States (Eisensee and Strӧmberg, 2007; Kahn, 2007). 2

Page 3 of 39

ip t cr

us

condition. The present study draws on prefecture-level panel data covering 2008–2014 and uses a difference-in-differences approach to assess the long-term effect of the nuclear accident on the BMI z-score and obesity rate of children in Japan. Section 2 of this paper provides an overview of the data and empirical method. Section 3 presents and discusses the major findings. The final section offers conclusions.

Ac c

ep te

d

M

an

2. Data and methods 2.1. Data The Japanese Ministry of Education, Culture, Sports, Science and Technology conducts the School Basic Survey across the country annually. This survey collects data on the height, weight, and obesity rate of school-aged children. Japan comprises 47 prefectures, and the ministry releases the average heights and weights for each. Height and weight data are further broken down for boys and girls between the ages of 5 and 17 years. This study adopted a difference-in-differences method using data from the period 2008–2014. The sample areas (those affected in the 2011 disaster) were Fukushima, Iwate, and Miyagi prefectures because the Great East Japan Earthquake directly affected them3. Data from those three prefectures were not collected in 2011 because of the impact of the disaster. To assess the effect of the Fukushima accident on young children over time, I conducted my estimations using cohort groups. The composition of the dataset used in this study appears in Table 1. I used BMI as a measure of overweight. That measure is flawed because it ignores the distinction between fat and muscle (Burkhauser and Cawley, 2008), and so physical activity that builds muscle and burns fat will have an ambiguous effect on BMI. However, Cawley et al. (2013) argued that this factor is less relevant in the case of elementary school students, whose physical activity involves less muscle-building exercise and who are less prone than adolescents are to adding muscle mass. Younger children are more likely to be influenced by their social circumstances than older ones, so I focused on children aged under 10 years. I divided the data used in this analysis into three cohort groups according to the children’s age in 2010: 5–7, 8–10, and 11–17 years. For example, the same cohort included children who were 7 years old in 2010 and 9, 10, and 11 years old in 2012, 2013, and 2014, respectively. Separate data were obtainable for boys and girls. The sample sizes differed slightly among the cohorts because of data availability. I calculated the average BMI values, which reflect the average level of overweight and obese children in each cohort, but may have been influenced by outliers in the sample. To control for this phenomenon, some studies have used the BMI z-score, which is standardized by age-specific BMI averages (Cawley et al., 2013; Inokuchi et al., 2011). Therefore, I also calculated the BMI z-score based on the mean BMI values and standard deviations for each age (Inokuchi et al., 2007). The method of calculation for the BMI z-score appears in Table 2. As shown in Table 3, the BMI z-score—unlike the BMI—does not increase with children’s age. Using the BMI z-score therefore allows a comparison of the obesity rates among different age-groups. I also assessed the obesity rate to determine the robustness of the BMI z-score data. The obesity rate is defined in Table 2. My focus in this study was on the BMI z-score and obesity rates from 2008 to 2014, and I compared them between disaster-affected and other areas. 3

It should be noted that some areas in Fukushima, Iwate, and Miyagi prefectures were probably unaffected by the disaster. Fukushima Prefecture, for example, consists of three subregions separated by mountain ranges. The nuclear accident occurred only in the coastal area, and other subregions were relatively isolated. Because of data limitations, subregional data on weight, height, and obesity were not available for the present study. Caution is therefore required in interpreting the results of this paper.

Page 4 of 39

ip t cr

Ac c

ep te

d

M

an

us

Fig. 1 shows how the difference between the average BMI z-score in disaster-affected and other areas changed from 2008 to 2014 for each cohort. Throughout the study period, the difference in the BMI z-score was consistently greater than 0. This implies that the average BMI z-score of children in disaster-affected areas was higher than in other areas. For all cohorts, the difference decreased consistently from 2008 to 2010. After the accident, the difference in the BMI z-score increased consistently from 2010 to 2014 for the cohort aged 5–7 years. This was in contrast to the other cohorts: for the cohorts aged 8–10 and 11–17 years, the difference in the BMI z-score was higher in 2013 than in 2010. The difference in BMI z-score declined from 2013 to 2014 for the cohorts aged 8–10 and 11–17. In 2014, that difference amounted to around 0.18, which was lower than preaccident levels. Fig. 2 displays the difference in obesity rates for each cohort. It is evident that the differences in obesity rate after the accident were greater than those preaccident for the cohort aged 5–7 years. However, the difference in obesity rate in 2008 and 2009 for the other cohorts was almost the same as in 2013 and 2014. The obesity rate and BMI thus presented a similar pattern during the study period. By comparing the disaster-affected and other areas for each cohort, Table 4a, b, c displays the difference in BMI z-scores for children before and after the 2011 Fukushima accident. Using the sample for the cohort aged 5–7 years in 2010, Table 4a shows that after the accident, the average BMI z-score was significantly higher in children in both the affected and other areas. The difference between the periods for affected areas was 0.26, which was greater than for other areas (0.17). Before the accident, the difference in the BMI z-score of children in affected areas was higher than in other areas. Fukushima, Iwate, and Miyagi prefectures are considered rural areas. The above result is consistent with the finding that children’s BMI z-scores are likely to be higher in rural than in urban areas (Yamamura, 2012a). After the accident, for the cohort aged 5–7 years, the average BMI z-score of children living in affected areas was higher (by 0.27) than for children in other areas. Table 4b, c shows the figures for older children. There were statistically significant differences between the affected and other areas for both periods. The BMI z-scores after the accident were not significantly higher than before the accident for the cohorts aged 8–10 and 11–17 years. This suggests that the accident exerted no effect on the BMI z-scores for the older cohorts. For the robustness check, I used an alternative proxy for the degree of obesity. Table 5a, b, c presents the difference in obesity rates before and after the 2011 Fukushima accident by comparing the disaster-affected area with other areas. The results are similar to those obtained using the BMI z-scores. Table 5a indicates that for the cohort aged 5–7 years, there was a significant increase in obesity after the Fukushima accident in both the affected and other areas; however, the effect was more pronounced in the affected area than in the other areas. 2.2. Econometric framework to examine the disaster impact Table 2 presents the definition of the variables used in the estimation and the mean values and standard deviation for each cohort group. From Table 2, it is evident that BMI increased with the children’s age. After standardizing the BMI by age, the values of the BMI z-score were 0.53, 0.67, and 0.62 for the cohorts aged 5–7, 8–10, and 11–17 years, respectively. Hence, the relationship between the BMI z-score and age was not linear. The estimated function therefore takes the following form4:

4

The difference-in-differences method is explained in Ashenfelter and Card (1985) and Angrist and Pischke (2009). With this method, outcomes are observed for two groups over two time periods. One of the groups (treatment group) is exposed to a treatment in the second period but not in the first. The second group (control group)

Page 5 of 39

ip t cr

us

BMI z-score (or Obesity rate) itga = α1Affected area i * 2012 year dummy t + α2 Affected area i * 2013 year dummy t + α3Affected area i * 2014 year dummy t + α4Damaged area i + α52012 year dummy t + α62013 year dummy t + α72014 year dummy t + α8Maleg + Y’ itgaBitga +ea + ki + u itga,

Ac c

ep te

d

M

an

where BMI z-score (or Obesity rate) itga represents the dependent variable in prefecture i, year t, sex g, and cohort a. To control for the effects of years, I included year dummies; the reference group was that before the accident (2008, 2009, and 2010), reflecting conditions before the 2011 Fukushima accident. Using data for 2008–2014 to determine the effect of the Fukushima accident on the BMI z-score (or obesity rate) of the children, I compared changes in BMI z-scores (or obesity rate) after the accident between disaster-affected and other areas. I adopted a difference-in-differences approach to examine the impact of the disaster on children’s BMI z-scores (or obesity rate). The interaction terms between Affected area and year dummies are key variables to examine the changing effect of the accident over time. The interaction terms indicate how and to what extent the BMI z-score or obesity rate was greater than in the base period (2008–2010). If the BMI z-score increased after the Fukushima accident and the effect was observed in 2012, the coefficient of Affected area i * 2012 year dummy t would be positive. As well as the nuclear accident, the three study prefectures were affected by the earthquake itself and tsunami. Therefore, it is possible that any observed effects on obesity could have derived from the tsunami and earthquake, rather than the nuclear accident. Because of its proximity to the nuclear plant, Fukushima was most affected by the accident. Therefore, I divided the affected areas into “Fukushima” and “other affected areas.” The key variables here are the cross terms between Fukushima and year dummies (for example, Fukushima i * 2012 year dummy t) and those between Other affected area and year dummies. It was expected that their coefficients would be positive. If the influence on obesity were mainly from the accident, then the coefficients of the cross terms would be greater for Fukushima than Other affected area. I included Male to represent sex differences. The vectors of the control variables (including unemployment rate, per capita income, rate of expenditure for cooked food, and expenditure for food) are denoted by Y itga. Per capita income and unemployment rate capture economic conditions. BMI z-scores and obesity rates depend on calorific intake, but the related data could not be obtained for this study. The Japan Statistics Bureau defines cooked food as various fast food options that are high in calories. Therefore, I included cooked (percentage of food expenditure spent on cooked food) to capture calorie intake. I supplemented this with snack (percentage of food expenditure spent on snack food). The coefficients of cooked and snack were therefore expected to be positive. The regression parameters are denoted by α; B is the vector of the regression parameters for the control variables; ea represents cohort effects, which are controlled by including cohort dummies; ki represents time-invariant prefecture effects, which are controlled by the fixed-effects model; and the error term is denoted by u itga. 2.3. Econometric framework to examine outward migration

is not exposed to a treatment during either period. The key variables are the treatment-group dummy, the second-period dummy, and a cross term between the treatment-group and the second-period dummies. The cross term is the key variable used to examine the effect of the treatment.

Page 6 of 39

ip t cr

an

us

According to reports of Japan’s National Police Agency, 18,404 people died or became untraceable in Fukushima, Iwate, and Miyagi prefectures as a consequence of the Great East Japan disaster. The nuclear accident most probably gave people an incentive to leave the area, thereby reducing its population—with respect to both adults and children. To assess the change in the population, I used the following:

Ac c

ep te

d

M

In the affected area for each cohort by age in 2010, t is the number of years that have passed since the accident. For this calculation, I used the raw number of students of each school age for each year from the Report on School Basic Survey (20082014). Changes in the child population in the affected areas were caused by both deaths and migration. The reference year was 2010, the year before the accident. Compared with the reference year, the index indicates a decrease in the number of students after the accident. The left-hand columns of Table 6 present the data for the cohorts aged 5–7 and 8–10 years during the period 2010–2014. In Japan, compulsory education comprises elementary and junior high school, with students aged 6–15 years. I was therefore unable to obtain data for the student population aged over 15 years. Thus, I was unable to calculate the decrease in 2014 for the cohort aged 11 years in 2010, and I do not report changes for the cohort aged 11–17 years. I turn now to Table 6a, b, c. Table 6a shows data for the entire affected area, including Fukushima, Iwate, and Miyagi prefectures. A definite decrease in the population was evident. In 2011, the decrease was 0.48% for the cohort aged 5–7 years and 0.47% for the cohort aged 8–10 years. For the cohort aged 5–7 years, the decrease in 2011 was far greater than in later years, which implies that there was no persisting impact of the accident. However, the difference between the years was small for the cohort aged 8–10 years. After I divided the sample into Fukushima and other affected areas, a clear difference between them became apparent. For Fukushima, the population decrease for the cohort aged 5–7 years was greater than that for the cohort aged 8–10 years. The population decreased steadily over time for the cohorts aged 5–7 and 8–10 years. This implies that the children who were relocated to other prefectures later returned to Fukushima. In contrast to Fukushima, the rates were negative for the cohort aged 5–7 years, but positive for that aged 8–10 years in the case of other affected areas. The above population decrease included children who left the area as well as a number of deaths, and those proportions may have differed in different areas. Therefore, I divided that decrease into “dead and missing” and “migration.” I calculated the percentage of dead and missing as follows:

Unfortunately, because of a lack of data, I was unable to calculate that rate for each cohort. Therefore, I assumed that the dead and missing rate would be the same for all cohorts within a prefecture. The rate of dead and missing people for all affected areas was 0.32; however, that figure broke down to 0.09 for Fukushima and 0.45 for other affected areas. This suggests that even though the nuclear accident occurred in Fukushima, the death rate was much lower there than elsewhere. The implication is that the tsunami and earthquake likely caused more deaths than the nuclear accident.

Page 7 of 39

ip t cr

M

an

us

I defined the outward migration rate as the population decrease minus the number of dead and missing individuals. The greater the value of the outward migration rate, the greater is the number of children relocated. Table 6 presents changes in the outward migration rate; that rate was positive for both cohorts in Fukushima. The rate for the cohort aged 5–7 years was approximately twofold that for the cohort aged 8–10. The outward migration rate in 2011 was 1.92 for children aged 5–7 years and 0.92 for those aged 8–10 and those values steadily diminished over time to 0.52 and 0.33, respectively, in 2014. In other affected areas, the outward migration rate was negative for children aged 8–10 in 2011 and 2012. (The absolute values can be interpreted as the migration rate.) The outward migration rate became positive in 2013 and 2014. The Fukushima accident may have caused residents in the affected area to migrate to safer regions (Yamamura et al., 2014). For example, if better-educated parents with lower-BMI children tended to migrate to unaffected areas, the levels of obesity in the more affected areas would increase. This means it is necessary to take into account the selective migration of health-conscious parents. The estimated function takes the following form: z-score BMI (or Obesity rate) itga = β1Decrease ita * Affected area i + β2Decrease ita + α8Maleg + Y’ itgaCitga +ea + ki + u itga.

ep te

d

In this equation, the reference year is 2010, where Decrease is 0. The decrease can be calculated for each age cohort a, in each prefecture i, and in each year t5. There were no available data to assess the changes in the numbers of boys and girls6. The cross term Decrease * Affected area is a key variable to examine how the effect of Decrease on obesity differed from that in other areas. If health-conscious parents in affected areas decided to move elsewhere, the cross term would be positive. Apart from the covering period and key variables, the control variables and estimation model (fixed-effects model) are the same as the function indicated in section 2.3.

Ac c

3. Results and discussion 3.1. Impact of the disaster Before interpreting the results of the regression estimates, I checked the validity of the difference-in-differences method. The procedure I used to identify the treatment effect in the difference-in-differences estimators was the parallel-paths assumption. According to that assumption, the average change in outcome for the treated group would have been the same as that for the control group if there is no treatment. Testing for systematic pre-trends has become a standard means for checking the validity of this method (Mora and Reggio, 2012; 2015)7. The condition for that test is that the

5

Detailed migration rate could not be obtained for each age cohort. Accordingly, outward migration rate was not added as the set of independent variables. The number of students for each sex in each cohort was available for 2010–2013, but not for 2014. 7 The method is simply described as follows: The equation 6

Page 8 of 39

ip t cr

Ac c

ep te

d

M

an

us

data have to include more than one pretreatment period. In the present study, there are three periods before the accident (2008, 2009, and 2010) and three after it (2012, 2013, and 2014). The P value results of the Wald test for the parallel paths appear in Table 7 and suggest that I cannot reject common pretreatment dynamics, and so the parallel-paths assumption holds. Tables 8 and 9 present the estimation results for the fixed-effects model. In each table, columns 1 and 4 indicate the results for the cohort aged 5–7, columns 2 and 5 those for the cohort aged 8–10, and columns 3 and 6 those for the cohort aged 11–17 years. Columns 1–3 comprise the results of key variables related to the cross terms between affected area and year dummies. In columns 4–6, the key variables are the cross terms between year dummies and either Fukushima or other affected area. The positive and significant values in column 1 of Table 8 show that the effect of the Fukushima accident persisted throughout the 2012–2014 period for younger children. The other columns suggest that the accident had different effects on obesity in different age cohorts. An increase in obesity was observed in younger children, though there was a decrease in older children. Column 4 of Table 8 indicates that the accident appeared to have exerted a greater effect on the BMI z-scores of children aged 5–7 years living in Fukushima than in other affected areas; however, children in other affected areas were also more likely to become obese after the event. For children aged 8–10 years, the impact of the accident was evident in Fukushima; it was smaller for children aged 5–7 years, and the effect was not observed in other affected areas. Therefore, the effect on the BMI z-score was greater for younger children living nearer to the Fukushima nuclear plant. The accident appears to have decreased the BMI z-scores for children aged 11–17 years. Male is significant and positive only in columns 3 and 6 in Table 8. This indicates that in the cohort aged 11–17 years, boys have higher BMI z-scores than girls do; however, this difference was not observed in younger cohorts. This suggests that older boys’ physical activity tended to involve more muscle-building exercise, which increases their BMI z-scores above that of their female peers. As a whole, the results in Table 9 are very similar to those in Table 8. This indicates that the findings are robust when an alternative index of obesity (obesity rate) is used as the dependent variable. In summary, the results suggest that the accident had an effect on the obesity of children in the cohort of 5–7 years; that effect endured over time, and it was greater in Fukushima than in other affected areas. From this, I can conclude that the effect derived from the nuclear accident, rather than the tsunami and earthquake. 3.2. Effect of outward migration

is the vector of the control variables. is the year dummy for year t. D is the treatment indicator. Treatment began after the last pretreatment period. So the post-treatment period is . Pretreatment trends are equal between treated elements and controls only when . Hence, the test of the null hypothesis of common pretreatment trends is:. The test statistic is the Wald test of the joint significance of all pretreatment in the above equation.

Page 9 of 39

ip t cr

d

M

an

us

Tables 10 and 11 show how population changes influenced the BMI z-scores and obesity rate. Overall, outward migration from the affected areas had no effect (Table 10). However, if the affected area is divided into Fukushima and other areas, there is a change in the results, as seen in columns 3 and 4 of Table 10. These findings suggest that the population decrease raised the BMI z-score in Fukushima, but not in other affected areas. That may be because health-conscious parents, who therefore had healthier children with a lower BMI, moved away from Fukushima after the nuclear accident, but they not do so from other affected areas. Thus, the health-conscious parents were responding directly to concerns about the degree of exposure to nuclear leaks. As noted in section 2.3, the rate of dead and missing people in Fukushima was far smaller than in other affected areas. This also indicates that a selective migration by health-conscious parents was triggered by the nuclear accident, not by the natural disasters, even though the natural disasters caused a larger number of deaths. For the two cohort groups aged 5–7 and 8–10 years, a 1% decrease in the child population led to an increase in the BMI z-score of 0.058 and 0.089, respectively. Similar effects were evident for the obesity rate. The outward migration effect can be calculated as follows:

ep te

I calculated the effects of outward migration on the BMI z-score and obesity rate from figures in Table 6, and they appear in Table 12a, b. The outward migration effect was seen only in Fukushima, and so the results for only that area are presented in Table 12. The effects of the disaster can be broken down into outward migration and other effects. In this paper, I assume that other effects are chiefly the effect of lack of physical activity, which can therefore be calculated as:

Ac c

The outward migration effect is not observed for Affected areas and Other affected areas. Accordingly, in Table 12a, b, the coefficients in Tables 8 and 9 are reported as the effect of lack of physical activity when they are statistically significant. Table 12a shows that for the cohort aged 5–7 years, the effect of lack of physical activity increased over time for both Fukushima and other affected areas, whereas the outward migration effect decreased from 2012 to 2014. Lack of physical activity exerted about twice the effect on the BMI z-score in Fukushima as it did in other affected areas. Compared with the cohort aged 8–10 years, lack of physical activity had an effect four or five times greater on the BMI z-score for the cohort aged 5–7 years. For the cohort aged 8–10 years, the outward migration effect was almost the same as the effect of lack of physical activity. Table 12b shows that the results were similar when obesity rate was the dependent variable, which means that the results are robust. It seems likely that residents who left Fukushima gradually returned over time, and so the outward migration effect declined. However, the lack of physical activity continued, and it had a persistent, growing effect for younger children. 3.3. Discussion

Page 10 of 39

ip t cr

Ac c

ep te

d

M

an

us

The results of this study suggest that the Fukushima accident increased the BMI z-scores for children younger than 10 years and decreased them for children aged 11 years and over. Because physical activity does not build muscle, this may suggest that a decrease in physical activity increased fat in young children. Among older children, physical activity may lead to increased muscle bulk, burning fat, and therefore increased BMI. In other words, physical activity produces opposite effects on BMI in younger and older children. The overall result of physical activity depends on whether the effect of building muscle is greater than that of burning fat. Reduction in physical activity reduces muscle and increases fat. My results show that the former effect was larger than the latter, resulting in a reduction in the BMI z-score. One possible explanation for the reduction in obesity in children aged 11–17 years is the influence on teenage lifestyles of Western beauty standards, which strongly emphasize thinness; that influence has spread to other countries via the mass media (Vignerova et al., 2007). In Poland, for example, adolescent girls tend to desire to be thin because of the influx of American culture (Chrzanowska et al., 2007). If that desire to be thin also applies to Japanese teenagers, the nuclear accident may have increased social pressure in that regard. The accident naturally led to insufficient exercise, which resulted in the increased weight of teenagers who did not care about their body shape. This inference suggests that BMI z-score is unlikely to be a source of measurement error and may be taken as reliable in this paper. Another potential explanation for the different results between the younger and older cohorts is as follows. In the older cohorts, less outdoor physical activity may have been compensated by more indoor physical activity toward maintaining physical condition. However, it is necessary to develop particular strategies in effectively using smaller spaces, such as gymnasiums, to burn calories. Younger children are less likely to develop such strategies: they are less willing to engage in indoor physical activities, partly because they lack the motivation to do so, which results in their BMI z-score not being at the proper level. Studies on the Chernobyl disaster have not indicated any strong, significant effects on physical activity (Lehmann and Wadsworth, 2011). However, the Soviet Union could control information about the accident, and so people who lived near the Chernobyl nuclear plant lacked the necessary information to change their activity levels. This may be one of the reasons for that accident not having had any obvious physical effect on BMI. By contrast, people in Japan can obtain information from multiple sources, including the Internet. The Japanese government was unable to control information about the accident, and so people’s physical activity changed accordingly8. Cawley et al. (2013) indicated that time spent on physical education in elementary school reduces the BMI z-score. The findings of the present study are consistent with those of Cawley et al. (2013), although I did not directly examine the effects of the time spent on physical activity. Cawley et al. (2013) used instruments to control for the endogeneity bias of the time spent on physical activity and provided strong evidence for doing so. One strength of the present study is that the natural disaster and nuclear accident can be considered an exogenous event. Although tsunamis and earthquakes do occur in northeastern Japan, the unobserved time-invariant feature was controlled in the fixed-effects estimates. Hence, the accident was unlikely to have caused any endogeneity bias—even though I did not use any instruments to control for that. I believe that my results support the argument that outdoor exercise was restricted for children living in Fukushima after the nuclear accident. Consequently, younger elementary school children were unable to burn calories. This outcome has a greater effect on younger children than on older 8

In various countries, sufficient information is provided by the free media for peer views to play an important role in forming individual opinions about nuclear energy (Yamamura, 2012b).

Page 11 of 39

ip t cr

us

ones. One policy implication of my findings is the importance of promoting physical exercise for younger children to help them develop good habits and maintain proper weight and good health in the future.

Ac c

ep te

d

M

an

4. Conclusions The Great East Japan Earthquake and associated tsunami affected people living in parts of northeastern Japan, particularly Fukushima, Iwate, and Miyagi prefectures. It caused a massive radiation leak from the Fukushima nuclear plant. The radiation leaks influenced human behavior because exposure to radiation has a detrimental effect on health. After the earthquake, in the affected areas, schools and parents prevented children from playing outside because of the risk of radiation exposure (Yomiuri, 2012). A decrease in outdoor exercise is thought to reduce calorie burning, which causes an increase in the BMI z-score and obesity (Yomiuri, 2012). It was likely that the impact of the accident on BMI z-scores and obesity rates would differ between affected areas and other areas in Japan. By employing a difference-in-differences approach, this study investigated the long-term impact of the Fukushima accident on children’s BMI z-scores. Using prefecture-level panel data from Japan for the period 2008–2014, I investigated how the 2011 Fukushima nuclear accident affected the BMI z-score (and obesity rate) of children aged 5–7 years in 2010; I also observed whether the effect changed over time. I used a difference-in-differences approach to show that for the cohort aged 5–7 years, the BMI z-score and obesity rates in the disaster-affected areas were higher than in other areas; however, this trend was not evident in the other cohorts. For the cohort aged 5–7 years, the influence of the accident persisted even after 3 years. For the cohort aged 8–10 years, this tendency occurred in Fukushima Prefecture, but not in other affected areas. The impact of the accident was also greater for the cohort aged 5–7 years than for that aged 8–10 years. I infer that health-conscious parents, whose children had a lower BMI, probably moved out of Fukushima, thereby increasing the levels of obesity seen in Fukushima. However, this effect was far smaller than that of a lack of physical activity. These findings suggest that restrictions placed on outdoor exercise as a result of the nuclear accident in Fukushima prevented younger elementary school children from burning calories. Such children developed inactivity habits, which led to persistently higher BMI z-scores than before the accident. That was not the case for older children. Accordingly, it would appear to be important for young children to obtain sufficient physical exercise to maintain a proper weight and enjoy good health in their future lives. However, lack of physical activity is only one possible explanation for the relationship between the accident and obesity. It is necessary to identify other factors that increase obesity. I used average prefectural obesity rates in this estimation, although Fukushima, Iwate, and Miyagi prefectures contain areas that were unaffected by the disaster. This method possibly caused estimation bias because of the aggregate-level data. In this paper, I did not use individual-level data and so could not examine the data more closely. I also did not examine a proxy for the degree of physical activity, although I would argue that the increase in obesity was probably due to reduced physical activity. To check that this is indeed the case, it would be necessary to employ individual-level panel data. Future studies should use individual-level data and control for access to healthy food and time spent by parents with children because these factors change children’s activity levels (Wronka and Pawlinska-Chmar, 2007; Yaniv et al., 2009). Because of the lack of data, those factors were not considered in the present study, which may have caused omitted-variables bias. Furthermore, the development of body size differs substantially by sex as children become older (Yamamura, 2012a). Performing a study analysis separately for boys and girls would be a valuable future task.

Page 12 of 39

ip t cr us an M

Ac c

ep te

d

5. References Almond, D., Edlund, L., Palme, M., 2009. Chernobyl’s subclinical legacy: pre-natal exposure to radioactive fallout and school outcomes in Sweden. Quarterly Journal of Economics 124(4), 1729-1772. Ando, M., Kimura, F., 2012. How did the Japanese exports respond to two crisis in the international production networks? The global financial crisis and the Great East Japan earthquake. Asian Economic Journal 26(3), 261-287. Angrist, J.D., Pischke, J.S. 2009. Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton University Press: Princeton. Ashenfelter, O., Card, D., 1985. Using the longitudinal structure of earnings to estimate the effects of training programs. Review of Economics and Statistics, 67, 648-660. Berger, E.M. 2010. The Chernobyl disaster, concern about the environment and life satisfaction. Kyklos 63, 118. Burkhauser, R.V. and Cawley, J., 2008. Beyond BMI: the value of more accurate measure of fatness and obesity in social science research. Journal of Health Economics 27(2), 519-529. Cawley, J., Fisvold, D. and Meyerhoefer, C., 2013. The impact of physical education on obesity among elementary school children. Journal of Health Economics 32(4), 743-755. Chrzanowska, M., Koziel, S. and Ulijaszek, A.J. 2007. Changes in BMI and the prevalence of overweight and obesity in children and adolescents in Cracow, Poland, 1971–2000. Economics and Human Biology, 5, 370-378. Danzer A., Weisshaar, N., 2009. The long run consequences of the Chernobyl catastrophe on subjective well-being and its set-point. Evidence from two Ukrainian data sets. Working Paper, WPEG Conference. Eisensee, T., Strömberg, D., 2007. News droughts, news floods, and U.S. disaster relief. Quarterly Journal of Economics 122(2), 693-728. Hayashi, T., 2012. Japan’s post-disaster economic reconstruction: from Kobe to Tohoku. Asian Economic Journal 26(3), 189-210. Inokuchi, M., Matsuo, N., Anzo, M., Hasegawa, T., 2007. Body mass index reference values (mean and SD) for Japanese children. Acta Paediatrica 96(11), 1674-1676.

Page 13 of 39

ip t cr

Ac c

ep te

d

M

an

us

Inokuchi, M., Matsuo, N., Takayama, J.I., Hasegawa, T., 2011. BMI z-score is the optimal measure of annual adiposity change in elementary school children. Annals of Human Biology 38(6), 747-751. Japan Times, 2012. The Japan Times Special Report. 3.11 one year on: a chronicle of Japan’s road to recovery. Tokyo: The Japan Times. Kahn, M., 2007. Environmental disasters as risk regulation catalysts? The role of Bhopal, Chernobyl, Exxon Valdez, love canal, and Three Mile Island in shaping U.S. environmental law. Journal of Risk and Uncertainty 35(1), 17-43. Lehmann, H., Wadsworth, J., 2011. The impact of Chernobyl on health and labour market performance. Journal of Health Economics 30(5), 843-857. Matsuda, Y., Kawasaki, K., Hashimoto, S., Tsuga, T., Kino, T., Yoshida, C., Oshiro, S., Fukuda, K., Eguchi, T. ., 2015. White paper of evacuation from the nuclear accident (in Japanese). (Gempatsu Hinan Hakusho. Jinbun Shoin Publishing: Tokyo. Mora, R., Reggio, I., 2012. Treatment effect identification using alternative parallel assumptions. Economics Working Papers we1233, Universidad Carlos III de Madrid, Departamento de Economía. Mora, R., Reggio, I., 2015. dqd: A command for treatment effect estimation under alternative assumptions. Stata Journal 15(3), 796-808. Rehdanz, K., Welsch, H., Narita, D., Okubo, T., 2015. Well-being effects of a major natural disaster: the case of Fukushima. Journal of Economic Behavior & Organization 116, 500-517. Uchida, Y., Takahashi, Y., Kawahara, K., 2014. Changes in hedonic and eudaimonic well-being after a severe nationwide disaster: the case of the Great East Japan Earthquake. Journal of Happiness Studies 15(1), 207-221. Vignerova, J., Humenikova, L., Brabec, M., Riedlova, L. and Blaha, P. 2007. Long-term changes in body weight, BMI and adiposity rebound among children and adolescents in the Czech Republic. Economics and Human Biology, 5, 409-425. Wronka, I., Pawlinska-Chmara, R. 2007. Childcare, height and BMI among female Polish university students, 2005. Economics and Human Biology 5 (3), 435-442. Yamamura, E., 2012a. Influence of body image in urbanized areas: differences in long-term changes in teenage body mass index between boys and girls in Japan. Journal of Bioeconomics 14(3), 243-256. Yamamura, E., 2012b. Effect of free media on views regarding nuclear energy after the Fukushima accident. Kyklos 65(1), 132-141. Yamamura, E., Tsutsui, Y., Yamane, C., Yamane, S., 2014. Effect of major disasters on geographical mobility intentions: the case of the Fukushima nuclear accident. ISER Discussion Paper 0903, Institute of Social and Economic Research, Osaka University, Osaka. Yamamura, E., Tsutsui, Y., Yamane, C., Yamane, S., Powdthavee, N. 2015. Trust and happiness: comparative study before and after the Great East Japan Earthquake. Social Indicators Research, forthcoming. Yaniv, G., Rosin, O., Tobol, Y. 2009. The effect of the fat tax and the thin subsidy. Journal of Public Economics 93(5-6), 823-830. Yomiuri, 2012. Obesity rising for Fukushima kids / 5- to 9-year-olds show highest rates of obese children across the nation. Yomiuri, December 27, 2012.

Page 14 of 39

ip t cr us

0.29 0.27

an

0.25 0.23

0.19

Age 11_17

2008

2009

ep te

d

0.17 0.15 2007

Age 8_10

M

0.21

Age 5_7

2010

2011

2012

2013

2014

2015

Year

Ac c

Fig. 1. Difference in average BMI z-score between children from the affected area and other areas for each cohort. Note: Cohorts denote the children’s ages in 2010. The difference in average BMI was calculated using the following formula for each year: Average value of BMI z-score for children from affected areas – Average value of BMI z-score for children from other areas.

Page 15 of 39

ip t cr us

4 3.5

an

3

% 2.5

M

2

1 2007

2008

2009

2010

2011

2012

2013

2014

Age 8_10 Age 11_17

2015

ep te

Year

d

1.5

Age 5_7

Ac c

Fig. 2. Difference in obesity rate between children from affected areas and other areas for each cohort (%). Note: Cohorts denote the children’s ages in 2010.

Page 16 of 39

ip t cr us

Table 1 Data structure, showing the number of data units. Prefectures Yearsa (2008–2014)

Cohortsb

Observations

3

1,410

47

6

Aged 8–10 years in 2010

47

6

2

3

1,692

Aged years 11–17 years in 2010

47

6

2

6

3,384

ep te

d

M

an

Aged 5–7 years in 2010

Sex (male and female) 2

Ac c

Note: Prefectures × Years × Sex × Cohorts = Total observations. a. Data for 2011 were unavailable. b. In the group aged 5–7 years in 2010, data were unavailable for 2008 and 2009 for the cohort aged 5 years in 2010 and for 2008 for the cohort aged 6 years in 2010. In the group aged 11–17 years in 2010, data were unavailable for 2012, 2013, and 2014 for the cohort aged 17 years in 2010 and so cannot be used to check differences before and after the disaster. Therefore, the cohort aged 17 years in 2010 was excluded. Data were also unavailable for 2013 and 2014 for the cohort aged 16 years in 2010, and for 2014 for the cohort aged 15 years in 2010.

Page 17 of 39

ip t cr

2012 year dummy 2013 year dummy 2014 year dummy Affected area Fukushima Other affected Ln(Income) c

0.62 (0.19)

16.4 (0.87) 6.61 (0.89)

17.7 (1.38) 8.48 (2.64)

20.1 (1.33) 9.90 (2.68)

---

---

---

1 if 2013, otherwise 0

---

---

---

1 if 2014, otherwise 0

---

---

---

1 if Fukushima, Iwate, or Miyagi prefectures, otherwise 0 1 if Fukushima prefecture, otherwise 0 1 if Iwate or Miyagi prefectures, otherwise 0 Log of per capita income (million yen)

------7.90 (0.14)

------7.90 (0.14)

------7.91 (0.14)

M

0.67 (0.17)

d

Obesity rate a

0.53 (0.26)

Obesity rate (%) A child is considered obese if the following value is larger than 20: (𝑊𝑒𝑖𝑔ℎ𝑡 − 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑤𝑒𝑖𝑔ℎ𝑡 𝑖𝑛 𝑒𝑎𝑐ℎ ℎ𝑒𝑖𝑔ℎ𝑡) × 100 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑤𝑒𝑖𝑔ℎ𝑡 𝑖𝑛 𝑒𝑎𝑐ℎ ℎ𝑒𝑖𝑔ℎ𝑡 Obesity rate is the percentage of obese children in each age group. 1 if 2012, otherwise 0

ep te

BMI a

z-score of body mass index, calculated by using the formula: 𝑀𝑒𝑎𝑠𝑢𝑟𝑒𝑑 𝑣𝑎𝑙𝑢𝑒 − 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑣𝑎𝑙𝑢𝑒 𝑖𝑛 𝑡ℎ𝑒 𝑟𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑡ℎ𝑒 𝑟𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 Body mass index

Ac c

BMI Z score a, b

an

us

Table 2 Basic statistics of variables used in the estimation and the mean values and standard deviations (SD) according to each subsample. Variables Definition (1) (2) (3) Ages 5–7 Ages 8–10 Ages 11–17 years in years in years in 2010 2010 2010 Mean (SD) Mean (SD) Mean (SD)

Page 18 of 39

ip t cr an

Male Decrease f,

Expenditure on snack food as a percentage of total food spend (%) (Annual expenditure on snack food per household /annual total expenditure on food per household) Expenditure on cooked food as a percentage of total food spend (%) 11.5 (Annual expenditure on cooked food per household (0.90) /annual total expenditure on food per household) 1 if male, otherwise 0 --Change in population of children (%). 0.13 [1 − {𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑡𝑢𝑑𝑒𝑛𝑡𝑠 𝑖𝑛 (2010 + 𝑡 𝑦𝑒𝑎𝑟)⁄𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑠𝑡𝑢𝑑𝑒𝑛𝑡𝑠 𝑖𝑛 (0.61) 2010} × 100]

M

Cooked e

4.41 (0.95) 8.99 (0.59)

4.38 (0.95) 8.95 (0.59)

4.31 (0.98) 8.91 (0.60)

11.4 (0.91)

11.3 (0.92)

--0.38 (1.07)

-----

d

Snack e

us

Unemployment d Unemployment rate (%)

Ac c

ep te

Note: Values in parentheses are SD. Sample of Decrease covers the period 2010–2014. Sources of data are: a. Children’s heights and weights from the Report on School Basic Survey, which is available from the Ministry of Education website, Culture, Sports, Science and Technology: http://www.e-stat.go.jp/SG1/estat/List.do?bid=000001044483&cycode=0 (accessed May 10, 2015). b. Mean values and SDs for Japanese BMIs from Inokuchi et al. (2007). The Japanese Ministry of International Trade and Industry conducted the cross-sectional survey on body sizes of Japanese people of various ages in 19781981. The survey involved 14,209 boys and 13,968 girls aged 1.518.5 years. Using the data, Inokuchi et al (2007) calculated mean values and SDs for each year of age from 5 to 17 years, respectively. The BMI z-score can be calculated for each age; for example, the BMI z-score for 5-year-old children in 2010 are calculated based on the BMI for 5-year-old children in 2010 and the mean value and SD of BMI for children aged 5 years in 19781981. Unfortunately, there are no data on mean values and SDs of BMI for children in more recent years. c. Per capita income data from the Ministry of Internal Affairs and Communications–Statistics Bureau, Director-General for Policy Planning & Statistical Research and Training Institute website: http://www.e-stat.go.jp/SG1/estat/List.do?bid=000001036889&cycode=0 (accessed May 10, 2015). d. Unemployment rates were obtained from the Ministry of Internal Affairs and Communications–Statistics Bureau, Director-General for Policy Planning & Statistical Research and Training Institute website: http://www.stat.go.jp/data/roudou/pref/index.htm (accessed May 10, 2015).

Page 19 of 39

ip t cr

Ac c

ep te

d

M

an

us

e. Percentage of expenditure on cooked food, snack food, and food were obtained from the Ministry of Internal Affairs and Communications–Statistics Bureau, Director-General for Policy Planning & Statistical Research and Training Institute website: http://www.e-stat.go.jp/SG1/estat/List.do?lid=000001064772 (accessed May 10, 2015). f. Numbers of children from the Report on School Basic Survey, available from the Ministry of Education, Culture, Sports, Science and Technology website: http://www.e-stat.go.jp/SG1/estat/List.do?bid=000001061932&cycode=0, (accessed September 1, 2015).

Page 20 of 39

Table 3 Comparison between BMI and BMI z-score by age. BMI z-score

15.5 15.7 16.0 16.5 17.0 17.6 18.2 19.1 19.6 20.2 21.0 21.3 21.5

0.12 0.44 0.57 0.72 0.75 0.67 0.68 0.76 0.62 0.60 0.67 0.56 0.51

us

cr

ip t

BMI

Ac ce pt e

d

M

an

Age (years) 5 6 7 8 9 10 11 12 13 14 15 16 17

Page 21 of 39

Table 4a Mean difference test for the BMI z-score between children from the affected area and other areas for the cohort aged 5–7 years in 2010. Preaccident After Difference in Absolute (1) accident BMI z-score t-values (2) (2)−(1) Affected (I) 0.55 0.82 0.26 3.89*** 0.55

0.17

Difference in BMI z-score (I)−(II) Absolute t-values

0.18

0.27

3.07***

8.86***

Ac ce pt e

d

M

an

us

Note: *** significant at the 1% level.

10.3***

ip t

0.37

cr

Other (II)

Page 22 of 39

Table 4b Mean difference test for the BMI z-score between children from the affected area and other areas for the cohort aged 8–10 years in 2010. Preaccident After Difference in Absolute (1) accident BMI z-score t-values (2) (2)−(1) Affected (I) 0.91 0.88 −0.02 0.68 0.64

Difference in BMI z-score (I)−(II) Absolute t-values

0.20

0.24

5.20***

13.5***

Ac ce pt e

d

M

an

us

Note: *** significant at the 1% level.

5.52***

ip t

0.70

cr

−0.06

Other (II)

Page 23 of 39

Table 4c Mean difference test for the BMI z-score between children from the affected area and other areas for the cohort aged 11–17 years in 2010. Preaccident After Difference in Absolute (1) accident BMI z-score t-values (2) (2)−(1) Affected (I) 0.82 0.83 0.01 0.21 0.61

Difference in BMI z-score (I)−(II) Absolute t-values

0.20

0.21

6.92***

14.9***

Ac ce pt e

d

M

an

us

Note: *** significant at the 1% level

0.68

ip t

0.61

cr

−0.01

Other (II)

Page 24 of 39

Table 5a Mean difference test for obesity rate (%) between children from the affected area and other areas for the cohort aged 5–7 years in 2010. Preaccident After Difference in Absolute (1) accident obesity rate t-values (2) (2)−(1) Affected (I) 6.38 10.4 4.09 4.31*** 6.91

2.49

Difference in obesity rate (I)−(II) Absolute t-values

1.96

3.56

3.85***

9.35***

Ac ce pt e

d

M

an

us

Note: *** significant at the 1% level.

12.4***

ip t

4.41

cr

Other (II)

Page 25 of 39

Table 5b Mean difference test for obesity rate (%) between children from the affected area and other areas for the cohort aged 8–10 years in 2010. Preaccident After Difference in Absolute (1) accident obesity rate t-values (2) (2)−(1) Affected (I) 11.1 11.4 0.36 0.52 8.24

Difference in obesity rate (I)−(II) Absolute t-values

2.63

3.23

4.43***

11.6***

Ac ce pt e

d

M

an

us

Note: *** significant at the 1% level.

1.42

ip t

8.48

cr

−0.24

Other (II)

Page 26 of 39

Table 5c Mean difference test for obesity rate (%) between children from the affected area and other areas for the cohort aged 11–17 years in 2010. Preaccident After Difference in Absolute (1) accident obesity rate t-values (2) (2)−(1) Affected (I) 12.2 12.8 0.58 1.38 9.72

Difference in obesity rate (I)−(II) Absolute t-values

2.57

3.08

6.28***

15.2***

Ac ce pt e

d

M

an

us

Note: *** significant at the 1% level

0.69

ip t

9.64

cr

−0.08

Other (II)

Page 27 of 39

Ac ce pt e

d

M

an

us

cr

ip t

Table 6a Changes in child population and outward migration rate (%) in the affected area. Decrease Outward migration rate Aged 5–7 Aged 8–10 Aged 5–7 Aged 8–10 years in years in years in 2010 years in 2010 2010 2010 2010 0 0 0 0 2011 0.48 0.47 0.16 0.15 2012 0.16 0.52 −0.16 0.20 2013 0.02 0.59 −0.30 0.27 2014 0.11 0.52 −0.21 0.20

Page 28 of 39

Ac ce pt e

d

M

an

us

cr

ip t

Table 6b Changes in child population and outward migration rate (%) for Fukushima. Decrease Outward migration rate Aged 5–7 Aged 8–10 Aged 5–7 Aged 8–10 years in years in years in 2010 years in 2010 2010 2010 2010 0 0 0 0 2011 2.24 1.24 1.92 0.92 2012 1.67 1.04 1.35 0.72 2013 1.20 0.81 0.88 0.49 2014 0.84 0.65 0.52 0.33

Page 29 of 39

Ac ce pt e

d

M

an

us

cr

ip t

Table 6c Changes in child population and outward migration rate (%) for other affected areas. Decrease Outward migration rate Aged 5–7 Aged 8–10 Aged 5–7 Aged 8–10 years in years in years in 2010 years in 2010 2010 2010 2010 0 0 0 0 2011 −0.41 0.08 −0.73 −0.24 2012 −0.59 0.27 −0.91 −0.05 2013 −0.57 0.47 −0.89 0.15 2014 −0.25 0.45 −0.57 0.13 Note: Outward migration rate is “decrease − death rate.” The death rate was the number of deaths (and missing people) in each prefecture / total population in each affected prefecture in 2010. Deaths (and missing people) data used in this paper were obtained from the National Police Agency of Japan website, http://www.npa.go.jp/archive/keibi/biki/higaijokyo.pdf (accessed September 1, 2015). Missing people are considered those who were untraceable after the Great East Japan earthquake, but whose bodies have not been found. They comprise about 10% of the total number of dead and missing. Population data used in this paper were obtained from the of the Statistics Bureau, Ministry of Internal Affairs and Communications website: http://www.stat.go.jp/data/kokusei/2010/index.htm (accessed September 1, 2015).

Page 30 of 39

Table 7 Pretrend test. (1) (2) (3) Aged 5–7 years Aged 8–10 years Aged 11–17 years in 2010 in 2010 in 2010 BMI z-score 0.73 0.51 0.63 Obesity rate 0.67 0.64 0.40 Note: Sample period is from 2008 to 2014. Treatment period is from 2012 to 2014. Numbers in the table show P-value.

Ac ce pt e

d

M

an

us

cr

ip t

H0: Common predynamics

Page 31 of 39

ip t cr us

(3) Age 11–17 years in 2010 −0.06*** (−8.97) 0.02 (0.65) −0.03** (−2.42)

Ac c

ep te

d

M

an

Table 8 Determinants of the BMI z-score (fixed-effects model). (1) (2) Aged 5–7 Aged 8–10 years in 2010 years in 2010 Affected area *2012 0.08* 0.02 year dummy (1.92 (0.561 Affected area *2013 0.10*** −0.01 year dummy (3.14) (−0.22) Affected area *2014 0.09*** −0.05 year dummy (4.70) (−1.30) Fukushima *2012 year dummy Fukushima *2013 year dummy Fukushima *2014 year dummy Other affected *2012 year dummy Other affected *2013 year dummy Other affected *2014 year dummy 2008 year dummy 2009 year dummy 2010 year dummy 2012 year dummy

0.17*** (14.8) 0.31*** (27.3) 0.57***

0.07*** (4.17) 0.08*** (6.56) 0.05**

−0.11*** (−11.5) −0.10*** (−18.6) −0.18***

(4) Aged 5–7 years in 2010

(5) Aged 8–10 years in 2010

(6) Age 11–17 years in 2010

0.15*** (10.4) 0.18*** (9.73) 0.13*** (6.28) 0.05 (1.02) 0.06*** (6.94) 0.07*** (5.86)

0.07*** (7.97 0.06*** (4.67) 0.03* (1.70) 0.006 (0.10) −0.03 (−1.02) −0.09** (−2.20)

−0.06*** (−8.23) 0.11*** (10.8) −0.04** (−2.55) −0.06*** (−7.98) −0.02** (−2.20) −0.03* (−1.70)

0.17*** (14.7) 0.31*** (26.9) 0.57***

0.06*** (4.06) 0.08*** (6.37) 0.05**

−0.11*** (−11.4) −0.11*** (−18.6) −0.18***

Page 32 of 39

ip t cr

(30.2) (2.27) (−9.88) (30.5) (2.39) (−9.69) 0.60*** 0.04* −0.21*** 0.60*** 0.04** −0.21*** (35.4) (2.01) (−12.5) (35.2) (2.10) (−12.4) 2014 year dummy 0.58*** 0.003** −0.23*** 0.58*** 0.006 −0.23*** (37.5) (0.16) (−13.9) (37.1) (0.32) (−13.8) Ln (Income) −0.11 −0.02 0.19* −0.09 −0.14 0.18* (−1.05) (−1.62) (1.82) (−0.83) (−1.03) (1.73) Unemployment −0.004 −0.02 0.01* −0.003 −0.02 0.01* (−0.46) (−1.62) (1.91) (−0.37) (−1.55) (1.89) Cooked food 0.01 −0.001 −0.001 0.01 −0.003 −0.003 (1.64) (−0.03) (−0.37) (1.24) (−0.59) (−0.67) Snack food −0.001 0.002 0.003 0.0006 0.002 0.003 (−0.09) (0.20) (0.42) (0.01) (0.20) (0.44) Male −0.01 0.001 0.14*** −0.01 0.001 0.14*** (−0.69) (0.11) (20.2) (−0.68) (0.11) (20.9) Observations 1410 1692 3384 1410 1692 3384 R-squared 0.70 0.10 0.43 0.70 0.10 0.44 Note: Cohort dummies were considered as time-invariant variables and controlled via a fixed-effects model. Prefecture dummies were included, but their results are not reported. Numbers in parentheses are t-statistics calculated using robust standard errors clustered at the prefecture level. *, **, and *** indicate significance at the 10%, 5%, and 1% levels.

Ac c

ep te

d

M

an

us

2013 year dummy

Page 33 of 39

ip t cr us

(3) Age 11–17 years in 2010 −0.56** (−2.04) 0.31 (0.77) 0.31 (1.57)

Ac c

ep te

d

M

an

Table 9 Determinants of obesity rate (fixed-effects model). (1) (2) Aged 5–7 Aged 8–10 years in 2010 years in 2010 Affected area *2012 1.76** 0.70 year dummy (2.19) (1.21) Affected area *2013 2.08*** 0.01 year dummy (3.11) (0.02) Affected area *2014 2.33*** −0.23 year dummy (4.12) (−0.36) Fukushima *2012 year dummy Fukushima *2013 year dummy Fukushima *2014 year dummy Other affected *2012 year dummy Other affected *2013 year dummy Other affected *2014 year dummy 2008 dummy 2009 year dummy 2010 year dummy 2012 year dummy

1.32*** (6.73) 2.69*** (13.9) 5.29***

1.20*** (5.77) 2.00*** (11.6) 3.03***

−0.73*** (−4.24) −0.68*** (−6.18) −0.92***

(4) Aged 5–7 years in 2010

(5) Aged 8–10 years in 2010

(6) Age 11–17 years in 2010

3.75*** (20.3) 3.80*** (15.3) 3.60*** (13.5) 0.79*** (3.64) 1.30*** (5.80) 1.77*** (3.64)

1.25*** (7.95) 1.39*** (6.89) 1.24*** (4.16) 0.45 (0.55) −0.61 (−1.49) −0.90* (−1.79)

−0.68*** (−5.32) 1.32*** (7.71) 0.61** (2.16) −0.49 (−1.30) −0.16 (−1.42) 0.17 (0.85)

1.27*** (6.56) 2.66*** (13.7) 5.34***

1.16*** (5.73) 1.98*** (11.3) 3.10***

−0.74*** (−4.21) −0.68*** (−6.18) −0.90***

Page 34 of 39

ip t cr

(17.2) (9.70) (−3.53) (17.3) (9.76) (−3.40) 6.40*** 2.90*** −0.85*** 6.43*** 2.94*** −0.84*** (23.8) (11.3) (−3.60) (23.7) (11.4) (−3.50) 2014 year dummy 7.10*** 2.31*** −0.36 7.14*** 2.36*** −0.35 (25.8) (10.2) (−1.64) (25.9) (10.5) (−1.56) Ln (Income) 2.17 −0.98 2.16 2.80 −0.21 2.28 (1.25) (−0.51) (1.31) (1.63) (−0.11) (1.32) Unemployment 0.10 −0.18 0.11 0.12 −0.16 0.11 (0.55) (−1.01) (0.81) (0.72) (−0.92) (0.81) Cooked food 0.17 0.10 −0.12 0.11 0.05 −0.14* (1.25) (0.98) (−1.62) (0.76) (0.47) (−1.82) Snack food 0.05 0.16 −0.01 0.08 0.16 −0.01 (0.41) (0.96) (−0.15) (0.64) (1.01) (−0.17) Male 0.88*** 1.20*** 2.21*** 0.88*** 1.20*** 2.21*** (7.36) (13.3) (22.3) (7.35) (13.3) (22.3) Observations 1410 1692 3384 1410 1692 3384 R-squared 0.73 0.40 0.31 0.73 0.40 0.31 Note: Cohort dummies were considered as time-invariant variables and controlled via a fixed-effects model. Prefecture dummies were included, but their results are not reported. Numbers in parentheses are t-statistics calculated using robust standard errors clustered at the prefecture level. *, **, and *** indicate significance at the 10%, 5%, and 1% levels.

Ac c

ep te

d

M

an

us

2013 year dummy

Page 35 of 39

Ac ce pt e

d

M

an

us

cr

ip t

Table 10 Determinants of the BMI z-score (fixed-effects model). (1) (2) (3) (4) Aged 5–7 Aged 8–10 Aged 5–7 Aged 8–10 years in 2010 years in 2010 years in 2010 years in 2010 Affected area* 0.004 −0.002 Decrease (0.14) (−0.85) Fukushima* 0.058*** 0.089*** Decrease (3.22) (9.23) Other affected areas* 0.025 0.054 Decrease (1.14) (1.35) Decrease 0.0001 0.002 0.0003 −0.001 (0.02) (0.06) (0.05) (−0.74) Observations 1128 1410 1128 1410 R-squared 0.46 0.12 0.46 0.13 Note: Control variables reported in Table 8 are also included. Cohort dummies were considered as time-invariant variables and controlled via a fixed-effects model. Prefecture dummies were included, but these results are not reported. Numbers in parentheses are t-statistics calculated using robust standard errors clustered at the prefecture level. *, **, and *** indicate significance at the 10%, 5%, and 1% levels.

Page 36 of 39

Ac ce pt e

d

M

an

us

cr

ip t

Table 11 Determinants of obesity rate (fixed-effects model). (1) (2) (3) (4) Aged 5–7 Aged 8–10 Aged 5–7 Aged 8–10 years in 2010 years in 2010 years in 2010 years in 2010 Affected area* 0.21 0.42 Decrease (0.30) (0.98) Fukushima* 1.16*** 1.58*** Decrease (3.94) (11.7) Other affected areas* 0.33 0.05 Decrease (0.56) (1.23) Decrease 0.23 0.04 0.23 0.05 (1.35) (0.98) (1.37) (1.03) Observations 1128 1410 1128 1410 R-squared 0.45 0.12 0.45 0.24 Note: Control variables reported in Table 8 are also included. Cohort dummies were considered as time-invariant variables and controlled via a fixed-effects model. Prefecture dummies were included, but these results are not reported. Numbers in parentheses are t-statistics calculated using robust standard errors clustered at the prefecture level. *, **, and *** indicate significance at the 10%, 5%, and 1% levels.

Page 37 of 39

Table 12a Decomposition of the accident effect on the BMI z-score in Fukushima.

0.10

----

----

0.09

----

----

0.07 0.13 0.10 ----

0.08 0.05 0.03 ----

0.01 0.02 0.00 ----

0.06

----

0.07

----

ip t

(2) Aged 8–10 years in 2010 Lack of Outward physical migration activity -------

M

an

us

cr

-------

0.06 0.04 0.03 ----

----

----

----

----

Ac ce pt e

d

Affected area*2012 Affected area*2013 Affected area*2014 Fukushima*2012 Fukushima*2013 Fukushima*2014 Other affected areas*2012 Other affected areas*2013 Other affected areas*2014

(1) Aged 5–7 years in 2010 Lack of Outward physical migration activity 0.08 ----

Page 38 of 39

ip t cr (2) Aged 8–10 years in 2010 Lack of physical Outward migration activity ------------------−0.19 1.44 0.62 0.77 0.72 0.52 ----------

----

----

----

Ac c

ep te

d

M

an

us

Table 12b Decomposition of the accident effect on obesity rate (%) in Fukushima. (1) Aged 5–7 years in 2010 Lack of physical Outward migration activity Affected area*2012 1.76 ---Affected area*2013 2.08 ---Affected area*2014 2.33 ---Fukushima*2012 2.18 1.57 Fukushima*2013 2.78 1.02 Fukushima*2014 3.00 0.60 Other affected 0.79 ---areas*2012 Other affected 1.30 ---areas*2013 Other affected 1.77 ---areas*2014

Page 39 of 39