Retirement age and Mental Health: Postponing retirement may improve mental health (Under review of Health Economics)

Transcription

1 Retirement age and Mental Health: Postponing retirement may improve mental health (Under review of Health Economics) KEY WORDS: retirement age; mental health; instrument variable; 1983 Social Security Amendments 1

2 Retirement age and Mental Health: Postponing retirement may improve mental health Zhaoxue Ci ABSTRACT With aging populations, policy makers are attempting to encourage people to work longer in order to sustain the financial stability of social security systems. However, the effect of postponing retirement on mental health remains uncertain. The US 1983 Social Security Amendments raised social benefit age by two months each year for the cohorts who were born after The amendments provided a unique opportunity to assess the causal effect of retirement age on mental health, free of biases caused by endogeneity between retirement age and mental health. This paper instrumented retirement age with the amendments and controlled for the effects of self-assessed health and previous mental health conditions. Using data from the 1992 to 2012 waves of the US Health and Retirement Survey, the results indicated that a slight gradual raise in retirement age was beneficial to retirees mental health. This paper provided several alternative mechanisms to explain how retiring later can improve mental health, including income effect, social capital, and inherent personality. KEY WORDS: retirement age; mental health; instrument variable; 1983 Social Security Amendments 2

3 1. INTRODUCTION Given increasing life expectancy and aging populations, many countries have been revisiting social security policies to encourage people to retire later (Gustman & Steinmeier, 1985; Mastrobuoni, 2009; Staubli & Zweimüller, 2013). Countries such as Australia, Norway, and the US have raised the benefit age in order to stimulate seniors to work longer. It is worth noting that the success of related policy changes partly depends on the potential effect of retirement age on health, particularly mental health. Suppose that working longer imposes a mental burden on workers, then health expenditures would increase. The increased health expenditures will offset, in part, the benefit of those policies in relieving the fiscal difficulties of social security systems. Existing studies have indicated that retirement has a significant impact on mental health, but these studies have generated conflicting conclusions. Some have claimed that retirement reduces stress, and therefore improves mental health (Midanik, Soghikian, Ransom, & Tekawa, 1995; Drentea, 2002; Latif, 2011; Eibich, 2015). Nevertheless, others have argued that retirement is associated with a higher possibility of mental illness due to significant lifestyle changes (Mein, Martikainen, Hemingway, Stansfeld, & Marmot, 2003; Dave, Rashad, & Spasojevic, 2008; Mckenzie, Gunasekara, Richardson, & Carter, 2014; Vo, et al., 2015). Yet, very few studies have focused on the effect of retirement age on mental health in particular. Some studies have reported a positive association between retirement age and mental health (Gill, et al., 2006; Sahlgren, 2013), but limited studies have identified a causal effect. The major challenge in identifying the causal effect of retirement age on mental health lies in endogeneity: retirement age might affect mental health, but people may choose retirement ages based on their mental health wellbeing. 3

4 The existing literature has adopted an instrumental variables (IV) approach to address endogeneity (Dave, Rashad, & Spasojevic, 2008; Calvo, Sarkisian, & Tamborini, 2013; Hagen, 2016). All of them concluded that a higher retirement age leads to better mental health. Dave, Rashad, and Spasojevic (2008) used a spouse s retirement status as a predictor of a respondent s retirement status. Yet, it is possible that a spouse chose to retire because of their own health condition or the respondent s mental health condition. Calvo, Sarkisian, and Tamborini (2013) instrumented retirement age with a benefit age increase and an early retirement offer from the employers. However, the incentive of offering an early retirement could be directly related to an employee s mental health. In addition, this study did not account for the potential link between mental health and self-assessed health (SAH). Hagen (2016) employed benefit age change in Sweden as the IV for retirement age. The Swedish reform was applied to local government workers, whereas Hagen used private sector workers as the control group. It is possible that Hagen s results were biased by the working sector difference. In our study, the causal effect of retirement age on mental health was examined using the US 1983 Social Security Amendments as the IV for retirement age, controlling for SAH and the previous mental health conditions. Moreover, after mandatory retirement being abolished in many countries, the concept of and types of retirement have become more complicated. Yet, most previous studies focused on only one type of retirement, which was usually defined based on either labor force status or benefit age. Actually, many seniors who claimed themselves as retired or who had reached the benefit age were still working. To the best of our knowledge, our study is the first one to reflect the new trend of retirement by further segregating the retired population into different types considering seniors self-perception of retirement. 4

5 The remainder of this paper is organized as follows: section 2 describes the study setting and data; section 3 provides the empirical methodology; results are presented in section 4; a discussion of possible mechanisms and potential limitations follow in section 5; and section 6 concludes the paper. 2. STUDY SETTING and DATA The US 1983 Social Security Amendments When it was established in 1935, the US social security system set the benefit age as 65. At that time, American s life expectancy was about 60 years at birth, and the expectation was to live to 72 at 60 (Wiatrowski, 2001). As life expectancy increased dramatically, the aging population imposed financial difficulties on the US social security system. To address the financial problems and encourage individuals to retire later, in 1983 the Congress signed the amendments to gradually increase the eligible age for full benefits from age 65 to age 66 in 2009 and age 67 in 2017 (Kollmann, 2000). Starting from 2000, the benefit age for full benefit increases 2 months each year for cohorts born after 1937 until it reaches 67. Besides, the Americans can collect a Age 62 Benefit at 62 with a roughly 20% reduction in their monthly benefit before they reach the full benefit age. As the 1983 amendments were enacted, the monthly benefit reduction associated with the Age 62 Benefit changed accordingly. The detailed full benefit age and the Age 62 benefit for the affected cohorts are illustrated in Table Validity of instrument variable First of all, substantial literature has documented that the benefit age has a significant impact on individuals retirement age decision (Andrew, 1998; Börsch-Supan & Schnabel, 1998; Neumark & Powers, 2005; Coile & Gruber, 2007; Mastrobuoni, 2009; Bongsang, Adam, & Perelman, 2012), and raising the benefit age could raise the average retirement age (Fields, 1984; Gustman & 5

6 Steinmeier, 1985; Barbara & Axel, 2004). In particular, Mastrobuoni (2009) found that the average retirement age of the affected cohorts of the 1983 amendments had increased one month every year. Secondly, the 1983 amendments are exogenous to mental health. This policy change is subject to the year of birth regardless of mental health conditions. People born after 1937 were selected as the affected cohorts for financial purpose rather than a health or mental health concern. Therefore, the amendments could be expected to trigger retirement decision changes, but not directly affect mental health, making it a valid IV for retirement age. Table 1: Full Retirement and Age 62 Benefit by Year of Birth Year of Birth Full Retirement benefit Age At age 62, the retirement benefit would be reduced by 1937 or earlier % and 2 months 20.83% and 4 months 21.67% and 6 months 22.50% and 8 months 23.33% and 10 months 24.17% % and 2 months 25.83% and 4 months 26.67% and 6 months 27.50% and 8 months 28.33% and 10 months 29.17% 1960 and later % Note. Data source: The United States Social Security Administration 2.2 Data Data from wave 1 to wave 11 ( ) of the US Health and Retirement Study (HRS) were used in this study. HRS is a longitudinal biennial panel survey including 37,317 individuals and their spouses that are nationally representative of individuals aged 51 and above (RAND Health and Retirement Study Longitudinal File, public use dataset, ). We extracted the data 6

7 on the mental health indicator, SAH, retirement age, current ages at the survey, gender, race, education, marital status, veteran status, and asset level. The study sample was selected according to the following four criteria. First, Americans are eligible to collect social security benefits as early as 62, and as late as 70. The sample was restricted to the respondents aged between 62 and 70 because the amendments would affect these respondents most directly. Considering age might be related to mental health, restricting the age range also helped to reduce the measurement bias. Secondly, only those respondents who retired between 62 and 66 were included to ensure the amendments were relevant. As mentioned above, the Americans can start to claim their benefit at 62. Meanwhile, in 2012 (the last wave of HRS) the full benefit age for the youngest cohorts who became eligible for the full benefit was 66 (see Table 1). That is, if the respondents chose to retire before 62 or after 66, their decisions seemed to be irrelevant to the amendments. Third, the respondents who applied for Social Security Disability Insurance and/or Supplemental Security Income were excluded. Finally, respondents who reported a poor health condition or experienced health deterioration were excluded. The last two criteria were employed to reduce the bias caused by the potential link between physical health and mental health, and also to mitigate the concern that some individuals might use self-evaluated health condition as a justification for retirement Define retirement and two samples There are at least two notable trends we should be aware of when discussing retirement. First, the US banned mandatory retirement age in most industries in From then on, most US people have been able to choose their retirement timing voluntarily. Second, according to HRS, a significant portion of respondents who reported themselves as retired and/or reached the benefit age were still working. Nowadays, retirement is a mixture of benefit eligibility, labor force status, 7

8 lifestyle transitions, and more; whereas, previous studies failed to consider this complication. To better reflect the reality, this study proposed two definitions of retirement and constructed two samples the full sample and the restricted sample accordingly. In HRS, there are two questions that could be used to define retirement. One is whether retired, the other is labor force status. The first one reflects the respondent s self-perception on whether she/he retired or not and could be used as our first definition of retirement (i.e. self-perceived as having retired but might still be active on the labor market), while the labor force status could be used to construct our second definition of retirement. We used the information collected from whether retired to construct the full sample. HRS asked respondents whether retired without providing a definition of retirement. If the answer was completely or partially retired, the retirement year was asked. Full sample therefore included all respondents who reported themselves as retired as long as they were able to provide the year of retirement (6,712 observations). In the full sample, many respondents were working or looking for a job in spite of retirement status. Retirement for them was more of a lifestyle transition based on self-judgment. Alternatively, the traditional definition of retirement refers to exiting the labor force completely. Based on that, the restricted sample was made up of the respondents who were not working nor looking for a job and mentioned retired (4,593 observations) when asked about their labor force status. Because the restricted sample excluded retirees who were still working, the sample size of the restricted sample was smaller than the full sample. This paper took advantage of 1983 amendments as a natural experiment. The amendments affected almost all Americans born after 1937, but not those born before For this reason, the treatment and control groups were formed utilizing the threshold of Following Mastrobuoni (2009), 8

9 the individuals born between 1938 and 1941 were categorized as the treatment group, and the individuals born between 1928 and 1937 were categorized as the control group. We can expect an average retirement age jump due to the amendments, but we do not expect notable alterations in mental health for individuals born in those specific years (Neumark & Powers, 2005; Calvo, Sarkisian, & Tamborini, 2013). The full sample was comprised of 1,985 treated observations and 4,727 controlled observations. The restricted sample was comprised of 1,311 treated observations and 3,282 controlled observations Measure mental health HRS measured mental health with a Center for Epidemiologic Studies Depression scale (CES-D). The CES-D was derived from six questions about negative feelings and two questions about positive feelings. The respondents evaluated frequencies of their feelings from 0 (rarely or none) to 3 (most or almost all the time). The sum of the six negative-feeling frequencies minus the two positive-feeling frequencies resulted in the CES-D. The total score ranged from 0 to 8, with a higher score indicating a worse mental health condition. 3. METHODOLOGY 3.1 The econometric model The following ordinary least square (OLS) specification was employed as a baseline model: (1) MH i = a + b r i + c X i The independent variable mental health (MH) is measured by the CES-D. r i represents the retirement age of respondent i. X is a vector of covariates including age, gender, race, education, marital status, veteran status, self-reported health, previous self-reported health, and previous mental health. Parameter b illustrates the correlation between retirement age and mental health. 9

10 However, the estimation of b cannot provide a causal inference because of endogeneity. There are two potential sources of endogeneity: uncontrolled confounders, and reverse causality. Uncontrolled confounders refer to unobservable factors which might influence both retirement age and mental health. Dave et al. (2008) mentioned that a health shock and a poor self-reported health could cause endogeneity in the analysis of retirement effect on health. To address this concern, those respondents who reported a poor health condition, experienced a health deterioration, or received disability pensions were dropped as discussed in Section 2.2. Reverse causality means that although the retirement age may affect mental health, it is also possible that people might choose to retire earlier because of a worse mental health, or vice versa. With reverse causality, the effect of retirement age on mental health could be overestimated. This study utilized the 1983 amendments to instrument retirement age in order to overcome reverse causality. As discussed in Section 2.1, the 1983 amendment is a valid instrument for retirement age because it affects retirement age directly, but it affects mental health only through retirement age. Besides the endogeneity concern, mental health might vary by age. Thus, we included age-fixed effects analysis in addition to the basic IV approach. We estimated the following first and second stage equations: (2) r i = a 0 + a 1 Z + a 2 X i + a 3 V it + M m=1 a 4m A m + ϵ it (3) MH it = b 0 + b 1 r i + b 2 X i + b 3 V it + M m=1 b 4m A m + ε it In the first stage, Z represents the instrument variable. Z is equal to 1 if respondent i was born after 1937, and zero otherwise. There are a number of M age groups, and A m is a dummy variable for age group m. V it is a vector of time-dependent covariates including marital status, asset level, SAH, 10

11 mental health two years previous 1, and SAH two years previous. X i is a vector of time-invariant characteristics including gender, race, year of education, and veteran status. ϵ it is the random error term. In the second stage, r i is the estimated retirement age got from the first stage. 3.2 Cluster on individuals HRS collected data by interviewing each individual every two years, so there are multiple observations for each individual. We thus controlled for clustering on each individual. The coefficients of clustered and un-clustered regressions should be the same, but the standard errors can be different. Equations for un-clustered and clustered variance estimators are: (4) V = (X X) 1 N [ i=1 (ε i x i ) (ε i x i )] (X X) 1 (5) V cluster = (X X) 1 J j=1 μ μj (X X) 1, μ = ε j j jth cluster x i i For simplifying, the time subscription is dropped. ε i is the error term for the i th observation, and N is the total number of observations. μ j is the within-cluster error term for the j th cluster, and J is the total number of clusters. In the full sample, 3,006 clusters were generated; and in the restricted sample, 2,242 clusters were generated. 4. RESULTS 4.1 Descriptive statistics The descriptive statistics are presented in Table 2 for the full sample and Table 3 for the restricted sample. The average retirement age of the treatment group was 0.15 years and 0.20 years later than the control group in the full sample and the restricted sample, respectively. The average mental health was better in the treatment groups of both samples. 1 HRS was implemented biennially. This is why previous SAH and mental health here were collected two years ahead. 11

12 As shown in Table 2 and Table 3, the statistics of the treatment groups and the control groups were similar. The average ages in different groups were just above 66. In the full sample, the treatment group had a slightly higher level of education, a better current SAH, and previous mental health; although, they had a worse previous SAH, marital status, and fewer assets. The only difference in the restricted sample was the treatment group had more average assets ($99,116.99) compared to the control group ($95,009.27). The overall racial makeup of the control and the treatment groups was almost identical. In comparison with the treatment group, the control groups had less males and more veterans in both samples. 4.2 Regression Results OLS estimation results The OLS results are shown in column (1) of Table 4 and Table 5: in the full sample the insignificant correlation between retirement age and mental health was in the range of [ 0.001,0.025]; in the restricted sample, the insignificant correlation was in the range of [ 0.007,0.027]. However, due to the endogeneity concerns discussed previously, these results were not compelling. We then utilized the instrument variable to identify the causal effect of retirement age on mental health IV estimation results Consistent with the literature, first stage estimation results (column (2) of Table 4 and Table 5) demonstrate that the 1983 amendments raised the average retirement age. Specifically, given a 99% confidence interval, in the full sample the average retirement age was raised by years, and in the restricted sample the average retirement age was raised by years. That is, on average the cohorts born between 1938 and 1941 retired 1.6 months and 1.9 months later relative to the cohorts born between 1928 and 1937 in the full sample and restricted sample, respectively. Why does the retirement age increase in the full sample less than in the restricted sample? Maybe it is 12

13 because when retirement is defined upon self-evaluation rather than labor force status, the choice of retirement age is less affected by exogeneous policy changes. Table 2. Descriptive Statistics- Full Sample Treatment group Mean SD Min Max N Retirement age ,985 CES-D ,985 Age ,985 Gender ,985 Education years ,985 Race ,985 SAH ,985 Veteran ,985 Asset (2010 USD) 90, , ,969,550 1,985 Marital ,985 CES-D_ ,985 SAH_ ,985 Control group Mean SD Min Max N Retirement age ,727 CES-D ,727 Age ,727 Gender ,727 Education years ,727 Race ,727 SAH ,727 Veteran ,727 Asset (2010 USD) 97, , e+07 4,727 Marital ,727 CESD_ ,727 SAH_ ,727 Note. Mental health becomes worse as CES-D increases. Gender=1 for males, and =0 for female. Race=1 for White people, =2 for Black people, and =3 for others. Veteran=1 for veterans, and =0 otherwise. Selfassessed health becomes better as SAH increases. Asset is converted into 2010-dollar value utilizing historical CPI data in the US. Marital=1 if married or partnered, =2 if separated or divorced, =3 if widowed or never married. 13

14 Table 3. Descriptive Statistics- Restricted Sample Treatment group Mean SD Min Max N Retirement age ,311 CES-D ,311 Age ,311 Gender ,311 Education years ,311 Race ,311 SAH ,311 Veteran ,311 Asset (2010 USD) 99, , ,608,013 1,311 Marital ,311 CESD_ ,311 Health_ ,311 Control group Mean SD Min Max N Retirement age ,282 CES-D ,282 Age ,282 Gender ,282 Education years ,282 Race ,282 SAH ,282 Veteran ,282 Asset (2010 USD) 95, , e+07 3,282 Marital ,282 CESD_ ,282 Health_ ,282 Note. Mental health becomes worse as CES-D increases. Gender=1 for males, and =0 for female. Race=1 for White people, =2 for Black people, and =3 for others. Veteran=1 for veterans, and =0 otherwise. Selfassessed health becomes better as SAH increases. Asset is converted into 2010-dollar value utilizing historical CPI data in the US. Marital=1 if married or partnered, =2 if separated or divorced, =3 if widowed or never married. Column (3) of Table 4 and Table 5 provides the IV estimation results. We found that the treatment groups had better mental health outcomes compared with the control groups. In the full sample, retiring 1.9 months later reduced retirees CES-D by points, and in the restricted sample, retiring 1.6 months later reduced retirees CES-D by points. Equivalently, when retirement 14

15 was defined as exiting the labor force completely, postponing retirement by one-month could improve retirees mental health by 5.14%; when retirement was defined upon self-evaluation, postponing retirement by one-month could improve mental health by 5.59%. Besides the major results, four covariates were found to influence mental health significantly in both samples: marital status, SAH, asset level, and previous mental health. As expected, mental health benefited from a more harmonious marriage, a better SAH, and more assets. Here, the interesting implication is that the current mental health was positively associated with the previous mental health. This suggests that mental health outcomes build up over time. Table 4. Effect of postponing retirement on mental health - Full sample OLS estimation (1) 1st stage estimation (2) GMM estimation- Robust unclustered (3) GMM estimation- Robust clustered (4) Mental health Retirement age Mental health Mental health Retirement age 0.012(0.013) **(0.337) *(0.430) Z 0.135***(0.034) Marital 0.222***(0.024) 0.108***(0.023) 0.309***(0.050) 0.309***(0.063) Education ***(0.006) 0.040***(0.006) (0.016) (0.021) Race 0.140***(0.036) **(0.034) 0.076(0.054) 0.076(0.068) Gender 0.136***(0.043) **(0.041) 0.072(0.061) 0.072(0.079) Asset (2010 USD) -3.51e-08(4.72e-08) -7.63e-08*(4.48e-08) -9.69e-08**(4.27e-08) -9.69e-08**(5.31e-08) SAH ***(0.021) 0.030(0.020) ***(0.029) ***(0.031) Veteran (0.045) *(0.043) **(0.063) **(0.083) CES-D_ ***(0.012) (0.011) 0.246***(0.018) 0.246***(0.021) SAH_ (0.021) (0.020) (0.027) (0.028) F-statistic/Chi Age-fixed Yes Yes Yes Yes Observations 6,712 6,712 6,712 6,712 Note. Z = 1 if a respondent was born after 1937, and zero otherwise. *p<0.1, **p<0.05, ***p<

16 Table 5. Effect of postponing retirement on mental health - Restricted sample OLS estimation (1) 1st stage estimation (2) GMM estimation- Robust unclustered (3) GMM estimation- Robust clustered (4) Mental health Retirement age Mental health Mental health Retirement age 0.011(0.016) **(0.348) **(0.432) Z 0.158***(0.042) Marital 0.211***(0.029) 0.109***(0.026) 0.304***(0.054) 0.304***(0.068) Education ***(0.008) 0.043***(0.007) (0.018) (0.022) Race 0.166***(0.044) (0.041) 0.121**(0.059) 0.121(0.074) Gender 0.181**(0.053) *(0.049) 0.112*(0.071) 0.112(0.089) Asset (2010 USD) -5.41e-08(6.23e-08) -7.89e-08(5.79e-08) -1.10e-07**(5.30e-08) -1.10e-07*(6.45e-08) SAH ***(0.025) 0.035(0.023) ***(0.034) ***(0.037) Veteran (0.057) **(0.052) (0.080) (0.102) CES-D_ ***(0.014) (0.013) 0.202***(0.020) 0.202***(0.022) SAH_ **(0.025) ***(0.023) -0.05(0.042) -0.05(0.046) F-statistic/Chi Age-fixed effects Yes Yes Yes Yes Observations 4,593 4,593 4,593 4,593 Note. Z = 1 if a respondent was born after 1937, and zero otherwise. *p<0.1, **p<0.05, ***p< Comparison of standard errors According to column (4) of Table 4 and Table 5, the clustered standard errors were consistently larger than the un-clustered standard errors. The sum of correlations between the unobserved and the observed was more variable when the observations of the same individual over time (11 waves) were analysed as a whole. That is, a positive correlation existed within each cluster. The positive correlation might, in part, emerged from mental health. The IV estimation has implied that previous mental health conditions had lasting influence on current mental health. The positive correlation among mental health outcomes for the same individual over time might increase the standard errors. 4.3 Sensitivity Analysis Multiple sensitivity analyses were conducted to test the robustness of the main findings. First, people older than 70 were included in both samples. Second, samples were extended to including 16

17 people who retired between 62 and 70. Third, we dropped the control variables which did not have statistically significant influence on mental health. We also ran a matching estimation to testify our results. The matching estimator confirmed that the cohorts who were affected by the 1983 amendments had better mental health outcomes. Overall, the causal effect of retiring later on mental health remained robust. 5. DISCUSSIONS Our empirical evidence raised an important question: what are the potential mechanisms through which postponing retirement improved mental health? There is no foregone conclusion about the mechanism. To some extent, it is because of the lack of a well-accepted theory explaining the relationship between retirement age and mental health. This study proposed several mechanisms to understand why retiring later can lead to enhanced mental health outcomes amongst seniors. The first potential pathway is income. Evidence of the positive correlation between income and mental health dates back to the 1980s (Kessler & McRae JR., 1982; Kahn, Wise, Kennedy, & Kawachi, 2000). In our study, the full and the restricted samples consistently showed positive relationship between asset level and mental health. Working longer helps to accumulate more wealth, and thus reduces retirees financial pressure. On the other hand, the Crossman model (1972) implies that if an individual has more income, he or she will be able to invest more in health. As a result, health, in particular mental health can be promoted. Social capital is another possible pathway. Social capital describes the quality and quantity of social networks. Substantial literature documented the positive connection between social capital and mental health (McKenzie, Whitley, & Weich, 2002; Silva, McKenzie, Harpham, & Huttly, 2005; Calvo, Sarkisian, & Tamborini, 2013; Forsman, Herberts, Nyqvist, Wahlbeck, & Schierenbeck, 2013; Sahlgren, 2013). Retirement is associated with losses of work-based social 17

18 capital. The reduced social capital, in turn, hurts mental health. Retiring later could enhance workbased social capital, hence benefits mental health. Besides the work-based social capital, the spouse can be considered as one type of family-based social capital. Our first stage estimation results suggest that people who had worse marriage relations or had no partners were more likely to retire later. That is, those people would love to invest more time in the work-based social capital in order to compensate the losses of the family-based social capital. It is not surprising that in general the seniors who had a partner were found to have better mental health conditions compared to the seniors without a partner. The third possible pathway is the difference in inherent personalities. For instance, individuals can be classified into two types: work-lovers and leisure-lovers. We can anticipate that later retirement can improve work-lovers mental health more than leisure-lovers, or even impair leisure lovers mental health. Our estimated retirement age effect on mental health was an average of the effects from work-lovers and leisure-lovers. We identified a significantly positive effect because both types benefited from retiring later to different extents, or because the positive effect on the worklovers dominated the ambiguous effect on the leisure-lovers. In our econometric models, the inherent personalities can be hardly measured explicitly. But, the inherent personalities were reflected implicitly by the increased standard errors and the reduced confidence level of the effects of retirement age on mental health when controlling for clustering on each person. Our study has several limitations. First, the CES-D is designed to measure depression, but depression is only a narrow view of mental health disorders. Regarding this, the HRS team defended themselves by claiming depression is the most common psychiatric disorder among elderly, so the HRS team chose to focus on it in particular (Steffick, 2000, p.3). Among the various depressive indicators used in psychiatric and psychological research CES-D was chosen because 18

19 CES-D has been widespread. Moreover, CES-D and many other indicators used similar questions. (Steffick, 2000, p.4). CES-D has been utilized by vast amount of literature focusing on mental health (Steffick, 2000; Lindeboom, Portrait, & Berg, 2002; Dave, Rashad, & Spasojevic, 2008). From this point of view, CES-D is not perfect, but it fits this research. Second, although the 1983 amendments did not declare any significant benefit change, we cannot conclude that there was absolutely no benefit adjustments overtime given the complexity of social security systems. Our results might be biased by the potential benefit adjustments. However, our results based on the available information are reliable and progressive. 6. CONCLUSIONS This study found that a one-month retirement age raise had a modest positive effect on mental health. It should be highlighted though that in our study setting, retirement age was raised gradually and slightly. The 1983 amendments were designed to raise benefit age by two months each year, and it was announced well ahead to the public. This is crucial because it takes time for people to adapt to policy change and subsequent life changes. If the change is massive and unexpected, its effect on mental health might be uncertain. Our results confirmed that mental health builds up over time. Previous mental health had a significant impact on current mental health. This empirical evidence thus implies that as it requires effort to build and maintain physical health, so it is with mental health. Moreover, given the lack of discussion on mechanisms in literature, this study provided three possible pathways through which retiring later improved mental health. Future studies could consider constructing a formal theoretical model to incorporate these pathways and better motivate empirical works in this area. 19

20 Our results implied that policies aimed at raising retirement age may not only improve the financial difficulties of the social security systems, but also ameliorate retirees mental health. Meanwhile, our findings on mental health highlight the importance for people to maintain mental health over time. 20

21 REFERENCES Andrew, S. (1998, November 1). New evidence on pensions, social security, and the timing of retirement. Journal of Public Economics, 70(2), Börsch-Supan, A., & Schnabel, R. (1998, May). Social security and declining labor force participation in Germany. The American Econmic Review, 88(2), Barbara, B., & Axel, B.-S. (2004). Pension reform in Germany: the impact on retirement decisions. FinanzArchiv: Public Finance Analysis, 60(3), Bongsang, E., Adam, S., & Perelman, S. (2012, May). Does retirement affect cognitive functioning? Journal of health Economics, 31(3), Calvo, E., Sarkisian, N., & Tamborini, C. R. (2013). Causal effects of retirement timing on subjective physical and emotional health. Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 68(1), Coile, C., & Gruber, J. (2007, May). Future social security entitlements and the retirement decision. Review of Economics and Statistics, 89(2), Dave, D., Rashad, I., & Spasojevic, J. (2008). The effect of retirement on phsical and mental health outcomes. Southern Economic Journal, 75(2), Drentea, P. (2002). Retirement and mental health. Jounral of Aging and Health, 14(2), Eibich, P. (2015). Understanding the effect of retirement on health: Mechanisms and heterogeneity. Journal of Health Economics, 43, Fields, G. S. (1984). The effects of social security reforms on retirement ages ans retirement incomes. Journal of Public Economics, 25, Forsman, A. K., Herberts, C., Nyqvist, F., Wahlbeck, K., & Schierenbeck, I. (2013). Understanding the role of social capital for mental wellbeing among older adults. Ageing & Society, 33, Gill, S. C., Butterworth, P., Rodgers, B., Anstey, K. J., Villamil, E., & Melzer, D. (2006). Mental health and the timing of Men's retirement. Social Psychiatry and Psychiatric Epidemiology, 41, Gustman, A. L., & Steinmeier, T. L. (1985). The 1983 social security reforms and labor supply adjustments of older individuals in the long run. Journal of Labor Economics, 3(2), Hagen, J. (2016). What are the health effects of postponing retirement? An instrumental variable approach. IFAU - Institute for Evaluation of Labour Market and Education Policy, Working Paper Series(2016:11). Kahn, R. S., Wise, P. H., Kennedy, B. P., & Kawachi, I. (2000). State income inequality, household income, and maternal mental and physical health: cross sectional national survey. BMJ: British Medical Journal, 321(7272), Kessler, R. C., & McRae JR., J. A. (1982). The effect of wives' employment on the mental health of married men and women. American Sociological Review, 47(2), Kollmann, G. (2000). Social Security: Summary of Major Changes in the Cash Benefits Program. Legislative History. Social Security. Latif, E. (2011). The impact of retirement on psychological well-being in Canada. The Journal of Socio-Economics, 40,

22 Lindeboom, M., Portrait, F., & Berg, G. J. (2002). An econometric analysis of the mental-health e ects of major events in the life of older individuals. Health Economics, 11, Mastrobuoni, G. (2009, December). Labor supply effects of the recent social security benefit cuts: Empirical estimates using cohort discontinuities. Journal of Public Economics, 93(11-12), McKenzie, K., Whitley, R., & Weich, S. (2002, October). Socail capital and mental health. The British Journal of Psychiatry, 181(4), Mckenzie, S. K., Gunasekara, F. I., Richardson, K., & Carter, K. (2014). Do changes in socioeconomic factors lead to changes in mental health? Findings from three waves of a population based panel study. Journal of epidemiology and community health, 68(3), Mein, G., Martikainen, P., Hemingway, H., Stansfeld, S., & Marmot, M. (2003). Is retirement good or bad for mental and physical health functioning? Whitehall II longitudinal study of civil servants. Journal of Epidemiol Community Health, 57, Midanik, L. T., Soghikian, K., Ransom, L. J., & Tekawa, I. S. (1995). The effect of retirmeent on mental health and health behaviors: the Kaiser permanente retirement study. Journal of Gerontology: Social Sciences, 508(1), S59-S61. Neumark, D., & Powers, E. T. (2005). The effect of the SSI program on labor supply: improved evidence from social security administrative files. Social Security Bulletin, 65(3), RAND Health and Retirement Study Longitudinal File, public use dataset. ( ). Produced and distributed by the University of Michigan with funding from the National Institute on Aging (grant number NIA U01AG009740). Ann Arbor, MI. Sahlgren, G. H. (2013). Work longer, live healthier: the relationship between economic activity, health and government policy. London: The Institute of Economics Affairs. Silva, M. J., McKenzie, K., Harpham, T., & Huttly, S. R. (2005). Social capital and mental illness: a systematic review. Journal of Epidemiol Community Health, 59, Staubli, S., & Zweimüller, J. (2013). Does raising the early retirement age increase employment of older workers? Journal of Public Economics, 108, Steffick, D. E. (2000). HRS/AHEAD documentation report: documentation of affective functioning measures in the Health and Retirement Study. Ann Arbor: Survey Research Center, University of Michigan. Vo, K., Forder, P. M., Tavener, M., Rodgers, B., Banks, E., Bauman, A., & Byles, J. E. (2015). Retirement, age, gender and mental health: findings from the 45 and Up Study. Aging & Mental Health, 19(7), Wiatrowski, W. J. (2001, April). Changing retirement age: ups and downs. Monthly Labor Review, 124(4),