1.1 The Japanese health insurance scheme

Health insurance coverage is an important determinant of access to health care. Over the past decade, numerous medical and economic studies have shown the effects of coverage on health care use and outcomes (Sommers et al., Reference Sommers, Gawande and Baicker2017). This research suggests that expanded coverage significantly increases access to care and use of both preventative and curative care, yielding significant positive benefits to health. We investigate whether expanded health insurance coverage might also increase access to care and improve the health of individuals in areas affected by natural disasters. Specifically, we use a triple differences (DDD) model to examine the effects of health insurance coverage on health care use and outcomes following the GEJE. To assess how copayment affects health care use, we exploit a discontinuous decrease in the copayment rate for medical care at the age of 70 years in the Japanese public health insurance system.

The Japanese public health insurance scheme is a backbone of the social security system for the elderly, together with public pension and long-term care insurance.Footnote ¹ The universal coverage in public pension and health insurance is a main characteristic of the Japanese social security system, and both schemes adopted the social insurance system, financed by social insurance premiums and supplemented by tax revenue in the form of subsidies. Since the 1990s, measures against the declining birthrate in addition to the aging society have surfaced as an important policy issue, and pension and health insurance system reforms have been implemented. Concerns regarding fiscal burden have been continuously raised with various proposals on policy reforms (National Institute of Population and Social Security Research, 2019). However, such fiscal burdens often do not fully account for its beneficial health effects. We examine health effects of such institutional provision and provide additional insights to evaluating governmental old-age support programs. As previously noted, the copayment rate varies across age under the Japanese public health insurance scheme that provides standard medical treatment under the uniform medical fee schedule. We exploit this policy variation in copayment rates. Specifically, the copayment rate in 2011, when the GEJE occurred, was 30% for those aged 7–69 years, 20% for those aged 6 and under, and 10% for those ages 70 or above.

1.2 Effects of the Great East Japan Earthquake

Previous studies have examined the effects of the 2011 GEJE on labor markets (Higuchi et al., Reference Higuchi, Inui, Hosoi, Takabe and Kawakami2012), risk preference (Naoi et al., Reference Naoi, Seko and Ishino2012; Jiang et al., Reference Jiang, Asai and Moridaira2013; Hanaoka et al., Reference Hanaoka, Shigeoka and Watanabe2018), macroeconomic outcomes (Ando and Kimura, Reference Ando and Kimura2012; Hosoe, Reference Hosoe2014), and psychological happiness, and well-being (Yamamura et al., Reference Yamamura, Tsutsui, Yamane, Yamane and Powdthavee2015; Sone et al., Reference Sone, Nakaya, Sugawara, Tomata, Watanabe and Tsuji2016; Sugano, Reference Sugano2016; Okuyama and Inaba, Reference Okuyama and Inaba2017). Several studies have also examined the health effects of the disaster (Nishigori et al., Reference Nishigori, Sugawara, Obara, Nishigori, Sato, Sugiyama, Okamura and Yaegashi2014; Yonekura et al., Reference Yonekura, Takeda, Shetty and Yamaguchi2014; Tomata et al., Reference Tomata, Suzuki, Kawado, Yamada, Murakami, Naka, Shibata, Ojima, Hashimoto and Tsuji2015; Tsuboya et al., Reference Tsuboya, Aida, Hikichi, Subramanian, Kondo, Osaka and Kawachi2016, Reference Tsuboya, Aida, Hikichi, Subramanian, Kondo, Osaka and Kawachi2017; Murakami et al., Reference Murakami, Sugawara, Tomata, Sugiyama, Kaiho, Tanji and Tsuji2017).

Most of these studies focused on examining the health of those in disaster-afflicted areas. For example, Yonekura et al. (Reference Yonekura, Takeda, Shetty and Yamaguchi2014) examined salivary cortisol levels of 63 GEJE victims and found that 22% could be classified as having ‘high depression.’ Nishigori et al. (Reference Nishigori, Sugawara, Obara, Nishigori, Sato, Sugiyama, Okamura and Yaegashi2014) also reported 21% of postnatal women recruited from 26 hospitals and clinics in the coastal area of the Miyagi prefecture had depression, which was much higher than the previously reported prevalence rate of 13.9%. Using data from the Japan Gerontological Evaluation Study, Tsuboya et al. (Reference Tsuboya, Aida, Hikichi, Subramanian, Kondo, Osaka and Kawachi2016, Reference Tsuboya, Aida, Hikichi, Subramanian, Kondo, Osaka and Kawachi2017) found that house loss and disruption of access to medicine after the disaster were associated with increased depressive symptoms and reduced ability to perform instrumental activities of daily living among those living in the affected area in Miyagi prefecture. Using 2011 and 2013 health check-up records in Miyagi prefecture, Murakami et al. (Reference Murakami, Sugawara, Tomata, Sugiyama, Kaiho, Tanji and Tsuji2017) found that disaster victims who had moved to temporary or rental housing had the highest risk of negative health effects resulting from alcohol consumption. These studies did not control for the common effect of the GEJE and unobserved heterogeneity across regions, potentially leading to overestimation. An exception is Tomata et al. (Reference Tomata, Suzuki, Kawado, Yamada, Murakami, Naka, Shibata, Ojima, Hashimoto and Tsuji2015), who used monthly municipal insurer-aggregate panel data from January 2011 to January 2014 and found that disabilities were more prevalent in disaster areas than elsewhere. However, the authors did not control for heterogeneity among individuals.

We use data from a comprehensive panel survey, the Japanese Study for Aging and Retirement (JSTAR), to examine the effect of the catastrophic natural disaster on afflicted individuals' health and their demand for medical and long-term care and then investigate the protective health effects of health insurance against disaster. Using the JSTAR has great advantages for this purpose. First, since the survey region includes Sendai, which is near the GEJE epicenter (Figure 1), and other areas not affected by the earthquake, we can estimate the average treatment effect of the GEJE by using a difference-in-differences (DD) model. Second, rich data from the JSTAR enables a comprehensive study of the health effects of GEJE.

Figure 1.

The JSTAR study regions and their maximum seismic intensity scales of the GEJE.

Notes: Maximum seismic intensity scales (see Appendix 1) of the GEJE are in parentheses.

We find that the disaster had a large negative impact on individuals' health. However, we also find that use of medical and long-term formal care after the earthquake helped limit this negative impact. Moreover, we find that a discontinuous decline in the copayment rate for medical care at age 70 in the public health insurance system had a protective impact on health and that the reduction in copayment did not increase health care utilization among those who were affected by the GEJE, suggesting a substitution effect.

The rest of this paper is organized as follows. Section 2 provides a discussion of our data and measures. In Section 3, we present our estimation strategy and present the estimation results of the DD effect of the 2011 GEJE on health. Our empirical results on the effect of health insurance policies for the elderly affected by the GEJE are reported in Section 4. Finally, concluding remarks are provided in Section 5.

3.1 Estimation strategy

Because the GEJE disaster is a localized natural disaster, we use a DD approach to examine its effects on health. Previous studies estimated the effect of the GEJE disaster as the average change in the health and medical care utilization in the afflicted area. Specifically, their estimates can be written as E[Y _i1 − Y _i0|Treat _i = 1], where Y_it is health status or medical care use of individual i at period t, and Treat is an indicator variable equal to one for the afflicted area and zero elsewhere. This difference includes the effect of both the GEJE disaster and other common macroeconomic conditions, including those discussed in Section 1. To identify the true effect of the GEJE disaster, we use the average difference in the non-afflicted area as the counterfactual, which is denoted asE[Y _i1 − Y _i0|Treat _i = 0]. Because this difference only includes national common shocks around the GEJE disaster, we can identify the average treatment effect of the GEJE disaster alone by taking the DD: DD = E[Y _i1 − Y _i0|Treat _i = 1] − E[Y _i1 − Y _i0|Treat _i = 0]. The key assumption for identification in the DD model is that the trends of health in both groups would be similar without treatment. As shown in Figure 2, every health indicator for both groups, except for symptoms of depression and poor memory function, has a similar trend from 2007 to 2009 before diverging from 2009 to 2011.

Our specified DD model is:

(1)$$Y_{it} = \alpha _0 + \alpha _1After_t + \alpha _2Treat_i + \alpha _3( {After_t \times Treat_i} ) + {\boldsymbol x}_{it}{\boldsymbol \alpha }_{\vector x} + {\boldsymbol z}_{it}{\boldsymbol \alpha }_{\boldsymbol z} + c_i + u_{it}$$

Y_it is an individual i's health indicators, with higher values representing worse health conditions in year t. After_t is equal to 1 for the sample in 2011, representing the period after the GEJE, and zero otherwise. Treat_i is equal to one for a Sendai resident and zero otherwise. α₃ is a DD estimator representing the average treatment effect of the GEJE disaster on health and we expect α₃ to be positive if it had an adverse effect on health. x_it includes individual i's demographic and socioeconomic status, including gender, age, marital status, yearly gross household income, amount of household financial assets, as well as indicators of having family dependents and owning real estate (housing or land). z_it refers to community characteristics which include prefectural or municipal level (see Table 2) macro characteristics, such as medical institution density, nursing care facility density, unemployment rate, and financial capability index, which would also affect health and care utilization. c is an individual fixed effect and u is an error term. We take the first-difference (FD) to eliminate individual effect c and time-invariant variables of x (Wooldridge, Reference Wooldridge2010) and estimate equation (2) using ordinary least squares (OLSs).

(2)$$\Delta Y_{it} = \alpha _1\Delta After_t + \alpha _3( {\Delta After_t \times Treat_i} ) + \Delta {\boldsymbol x}_{{\vector it}}{\boldsymbol \alpha }_{\vector x} + \Delta {\boldsymbol z}_{{\vector it}}{\boldsymbol \alpha }_{\vector z} + \Delta u_{it}$$

In this model, the OLS estimator of α₁ represents the year effect and α₃ represents the average ‘treatment’ effect of the GEJE disaster on health. In addition, we estimate clustered robust standard errors that allow correlated residuals within municipalities to accommodate any serial correlation (Bertrand et al., Reference Bertrand, Duflo and Mullainathan2004).

It is important to note that sample attrition could be an issue, especially in the third wave of the JSTAR, given the devastating effects of the GEJE on the Tohoku district. We examine this potential sample attrition (see Appendix 3 for further details) but do not find a significant effect. Indeed, our probit estimation result suggests that Sendai respondents had a higher participation rate than those in other regions. Nevertheless, to control for any adverse attrition bias, we employ the inverse weighting modelFootnote ⁶ (Robins et al., Reference Robins, Rotnitzky and Zhao1995; Wooldridge, Reference Wooldridge2002, Reference Wooldridge2007).

3.2 Empirical results

Table 4 shows the average treatment and time effects of the GEJE disaster (DD estimator) on health indicators. The disaster had significantly adverse effects on all health indicators. Its largest effects were on memory function, indicating that the GEJE increases the probability of having poor memory function by 15.2%. It has also significantly adverse impacts on the probabilities of having difficulties in functional health by 7.2%. We also observe modest, but significant adverse effects on chronic disease (2.8%) and depression (4.5%), but no significant effect on subjective health. However, it should be noted that the number of observations for each outcome varies due to missingness, which may potentially lead to underestimation of the impact on some indicators. As shown in Table 4, missingness is most often observed in the questions assessing depression and memory, and as those with depression and poor memory are more likely to refuse to provide an answer (Fisher et al., Reference Fisher, Hassan, Faul, Rodgers and Weir2017), our results are more likely to underestimate the effects.

Table 4.

Effects of the GEJE on health

Note: This table reports FD estimates of the effects of the GEJE and years on health outcomes.

***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively. Clustering robust standard errors allowing for correlated residuals within municipalities are in parentheses.

4.1 Effects on care utilization

Table 5 reports the effects of the GEJE disaster on medical and formal home care utilization. The results of the DD estimators indicate that the GEJE significantly increases inpatient and formal home care utilization by 2.7% and 0.6%, respectively, reflecting a greater need for these services. On the other hand, its effect on outpatient utilization is insignificant, and the utilization of dentistry is negatively influenced by the GEJE by 5.1%. We surmise this was because consumers were no longer able to visit afflicted medical facilities or because most medical staff were treating hospitalized patients and providing greater levels of inpatient care.

Table 5.

Effects of the reduction in copayment on care utilization by the DDD model

Note: This table reports FD estimates of the effect of the GEJE, the health insurance system, and years on care utilization.

***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively. Clustering robust standard errors allowing for correlated residuals within municipalities are in parentheses.

As a key parameter of interest, DDD estimators are found to be insignificant, indicating that there was no policy effect on health care utilization in afflicted areas. That is, the reduction in copayment did not increase health care utilization among those who were affected by the disaster, suggesting no income effect. These results contradict those of previous studies (Shigeoka, Reference Shigeoka2014; Fukushima, et al., Reference Fukushima, Mizuoka, Yamamoto and Iizuka2016), which may be due to the difference in sample characteristics and potential omitted variable bias. For example, Shigeoka (Reference Shigeoka2014) examined repeated cross-sectional data and estimated the policy effects without accounting for individual heterogeneities over time. Fukushima et al. (Reference Fukushima, Mizuoka, Yamamoto and Iizuka2016) used a sample that represented relatively healthy populations who have worked until the age of 70 years.

4.2. Effects on health

Table 6 shows the effects on health. The DD estimators are significant and positive for all health indicators. These results indicate that the GEJE disaster had a negative impact on individuals' health. Its effect size is especially large on memory, with the probability of poor memory increased by 40%, consistent with what was observed in the descriptive time trend analysis in Figure 2. The adverse effects of the GEJE are also observed in other health domains, but more modestly. Specifically, it significantly increases the probabilities of having limitations in functional health, showing depressive symptoms, having a chronic disease, and reporting poor subjective health by 7.8%, 3.6%, 2.9%, and 2.1%, respectively. The size of the impact, except for symptoms of depression, is larger than those presented in Table 4. This indicates that the DD estimators in Table 4 are underestimates because they include both the negative direct and indirect effects of the GEJE.

Table 6.

Effects of the Reduction on Copayment on Health by the DDD model

Note: This table reports FD estimates of the effect of the GEJE, the health insurance system, and years on health outcomes. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively. Clustering robust standard errors allowing for correlated residuals within municipalities are in parentheses.

The effects of the DDD estimators are significantly negative, except for symptoms of depression. These results indicate that copayment reduction in public health insurance reduced health deterioration caused by the GEJE with the exception of depression. Again, the largest effect is observed for memory, comprising most of the harmful effects of the GEJE. We observe modest effects of the copayment reduction on other health domains, including self-reported health, chronic diseases, and functional limitations, ranging from about 4.8 to 7.5 percentage points.