Retirement Pathways’ Effect on Physical and Mental Health: Evidence from Japan

Abstract

Objectives: Few studies have used rigorous causal analysis to examine the difference in health effects between abrupt and gradual transitions from full-time employment to full retirement. Methods: Data were obtained from 10,253 individuals aged 50 years and older from the 2005–2019 Longitudinal Survey of Middle-aged and Older Adults conducted in Japan. An instrumental variable approach was used to determine the causal relationship between retirement pathways and health by exploiting the public pension eligibility age as the instrument. Results: Abruptly retired individuals were more likely to have physical limitation and less likely to have depression than those not yet retired. In contrast, those who fully retired through a gradual transition were less likely to have physical limitation and more likely to have depression than those not yet retired. Discussion: Retirement pathways have different effects on physical and mental health, providing important implication for labor and health policies in aging societies.

Retirement is a complex transition from full-time employment (Cahill et al., 2006, 2015, 2018; Hayward & Grady, 1990). Some older workers transition from full-time employment to full retirement via part-time employment or self-employment (Bloemen et al., 2016; Brunello & Langella, 2013), whereas others fully retire directly from full-time employment (Blau & Shvydko, 2011; Ruhm, 1990). Since retirement has become increasingly heterogeneous in recent decades due to prolonged working life (Carr et al., 2021; Fasang, 2010, 2012), its socio-economic impact has drawn substantial interest.

However, little is known about how different retirement pathways affect older adults’ health. Conventional wisdom suggests that a gradual transition to full retirement is more beneficial than an abrupt transition (De Vaus et al., 2007). Continuity theory also suggests that a gradual transition to full retirement promotes one’s health by reducing the gap between pre- and post-retirement life (Atchley, 1989). Consequently, if gradual withdrawal from full-time employment has a better effect on health than abrupt withdrawal, politically promoting a gradual retirement (i.e., working with less time and effort after full-time employment as a retirement process; Cahill & Quinn, 2020) could alleviate the financial burden on the medical and long-term care systems managed by the pay-as-you-go system.

Some studies have been conducted to uncover the effects of retirement pathways on health (e.g., Calvo et al., 2022; De Vaus et al., 2007; Munnell et al., 2021). However, these studies were unable to determine the causal effects of retirement pathways. Several studies suggested that retirement and health are endogenously determined (Bound et al., 1999). Specifically, health is a strong predictor of retirement, whereas retirement is a determinant of subsequent health, and confounders such as unobservable health shocks might affect both retirement and health. This retirement-health relationship, called the endogeneity problem, leads to inconsistent estimates of the effects of retirement on health. Another limitation of previous studies is that there has been little comparison between the effects of retirement pathways on physical and mental health. The effects of retirement may differ according to health indicators, leading to different implications.

Therefore, this study contributes to the literature in two ways: first, by estimating the causal relationship between retirement pathways and health outcomes. To this end, this study performed instrumental variable (IV) estimations that exploit, as the instrument, the changing public pension eligibility age caused by pension system reform in Japan. This exogenous shock affects retirement behavior but does not directly affect health; hence, the IV estimation can identify the causal path from retirement to health. Understanding the causal effect of retirement pathways on health has important scientific and policy implications. Second, this study contributes to the literature by comparing the effects of retirement pathways on physical and mental health. Specifically, physical limitation and depression were used as health measures.

Background

Theoretical Framework

This study first considered the effect of leaving full-time employment on health using the health capital model (Grossman, 1972). Leaving full-time employment substantially decreases the opportunity costs of investing in health; therefore, it is significantly easier for individuals to dedicate more time and effort to promoting and maintaining health. On the other hand, individuals are not compelled to maintain good health after leaving full-time employment because they do not have to maintain high productivity. This change can lead to a decrease in health investment and, ultimately, health deterioration. Consequently, it is difficult to predict whether retirees’ health is better or worse than that of those who have not yet retired.

Following this, this study considered how different retirement pathways affect health. Continuity theory suggests that a gradual transition benefits health by easing the transition between pre- and post-retirement life (Atchley, 1989; Kim & Feldman, 2000). Retirement causes a reduction in physical and mental activity and a loss of social networks or sense of purpose, leading to poor health. Gradually retiring individuals can mitigate these sudden shocks by engaging in other health-promoting activities or networks prior to full retirement, exploiting the flexibility of working part-time, or self-employment (Zhan et al., 2009). Gradually retiring individuals thus experience fewer adverse effects; hence, they would generally be healthier upon full retirement than those who retired abruptly. Consequently, this study hypothesized that full retirement through a gradual transition is more advantageous for health than full retirement through an abrupt transition. However, as the health capital model suggests, it is difficult to predict whether fully retired individuals would exhibit better health than continuously working individuals.

Empirical Findings

Several studies have investigated the relationship between retirement pathways and health outcomes, using methods that do not sufficiently address the endogeneity problem. Previous studies provided mixed findings regarding the comparison between abrupt and gradual transitions. Some studies have found retirement through a gradual transition to be more beneficial to health than abrupt retirement, with a slower decline in functional ability (Azar et al., 2019), fewer chronic diseases (Calvo et al., 2022), and better self-rated health (De Vaus et al., 2007). However, other studies found retirement through a gradual transition to be associated with better self-rated health but more chronic diseases than abrupt retirement (Madero-Cabib et al., 2020), with no statistical difference concerning to depression (Calvo et al., 2009), mortality (Scotti, 2022), or self-rated health (McDonough et al., 2017).

Regarding the comparison between retirement pathways and continued employment, previous studies found abrupt retirement to be worse than continued employment concerning health (Calvo et al., 2022; Munnell et al., 2021; Wels & Takami, 2021; Zhan et al., 2009), with the exception of no statistical difference (McDonough et al., 2017). For example, abrupt retirement was negatively associated with mental health (Wels & Takami, 2021; Zhan et al., 2009) and functional limitations (Zhan et al., 2009). However, the findings on retirement through a gradual transition relative to continued employment are mixed: negative association with depression (Munnell et al., 2021) and no statistical association with self-rated health (McDonough et al., 2017).

In summary, the relationships between retirement pathways and health remains inconclusive. Although some studies used propensity scores or lagged models to determine causal relationships, the endogeneity problem has not yet been fully addressed. Consequently, this study aimed to reveal the rigorous causal effects of retirement pathways on health.

Institutional Settings in Japan

Mandatory retirement is common in Japan. As of 2015, 92.6% of companies exceeding 30 employees enforced mandatory retirement, which in most cases (80.5%) was at age 60, with few companies enforcing retirement at ages 61–64 (2.6%) and age 65+ (16.9%; Ministry of Health, Labour, and Welfare [MHLW, hereafter], 2015b). Full-time regular employees—those who work full-time with an open-ended employment contract—commonly retire at the age of 60, while some can continue their jobs until the age of 65 or older.

Japan is also undergoing public pension reform. Public pension in Japan comprises a two-tiered benefit system. Employees are first eligible for earnings-related benefits (partial benefits) and then for flat-rate benefits (full benefits) in the study cohort. To ensure the financial sustainability of the public pension system, the Japanese government incrementally increased the age of pension eligibility from 60 to 65 years. Table 1 shows the birth cohorts used in this study and their pension benefits eligibility ages. The eligibility age for full pension benefits varies across birth cohorts. The eligibility age for partial benefits also differs across birth cohorts; however, the variation is not significant. As the table shows, different cohorts face different eligibility ages for pension benefits. Consequently, individuals of the same age face different incentives to withdraw from the labor force. Moreover, the different incentives depend on the policy change, which is unrelated to individuals’ work/retirement decisions. Therefore, like an experiment, this public pension reform provides an exogenous shock to retirement behavior.

Table 1.

Eligibility Age for Pension Benefits.

Birth cohort in the study	Men, women (MAP)		Women (EP)
	Flat-rate benefits (full)	Earnings-related benefits (partial)	Flat-rate benefits (full)	Earnings-related benefits (partial)
04/02/1945–04/01/1946	63	60	60	60
04/02/1946–04/01/1947	63	60	61	60
04/02/1947–04/01/1948	64	60	61	60
04/02/1948–04/01/1949	64	60	62	60
04/02/1949–04/01/1950	65	60	62	60
04/02/1950–04/01/1952	65	60	63	60
04/02/1952–04/01/1953	65	60	64	60
04/02/1953–04/01/1954	65	61	64	60
04/02/1954–04/01/1955	65	61	65	60
04/02/1955–	65	62	65	60

Note. MAP = mutual aid pension (for employees in the public sector and private schools); EP = employer pension (for employees in the private companies).

Source. Material for the Social Security Council in the MHLW (October 11, 2011). https://www.mhlw.go.jp/stf/shingi/2r9852000001r5uy-att/2r9852000001r5zf.pdf (Accessed February 7, 2023).

However, this increase in pension eligibility age has not been matched by changes in companies’ mandatory retirement age, which is typically set at 60 years, causing most employees in their early 60s to no longer be eligible for full public pension benefits. To bridge this growing gap between the mandatory retirement age and full pension eligibility, the government revised the Act on Sustainability of Employment of Older Persons in 2006, requiring companies to guarantee employment until the eligibility age for full benefits depending on their employees’ desires. To this end, companies were required to implement one of the following three measures: (1) raise the mandatory retirement age; (2) introduce a continuous employment system allowing employees to remain employed after mandatory retirement; or (3) abolish the mandatory retirement system.

This revision in the law had no requirement for the type of employment offered to employees and did not stipulate the percentage of pre-retirement salary to be paid. Consequently, most companies (81.7%) have implemented option (2) (MHLW, 2015a). Approximately 90% of companies with continuous employment systems have chosen to re-employ mandatorily retired employees; among re-employed workers, nearly half of them work in a lower position and with less responsibility than before (National Personnel Authority, 2018), that is to say, they work in non-standard jobs (e.g., part-time, temporary, or contract work).

Under these settings, most older employees, depending on their mandatory retirement age and pension eligibility age, choose to transition from a job to full retirement or leave a job to work in a non-standard job position for a few years, and then retire.

Method

Data

This study used data from 15 waves (2005–2019) of the Longitudinal Survey of Middle-aged and Older Adults (LSMOA), a nationally representative longitudinal survey in Japan conducted annually by the MHLW since 2005. The sample comprised men and women aged 50–59 years (born in the period 1945–1955), who were randomly selected through a two-stage sampling procedure in 2005. Self-administered questionnaires were delivered to respondents by an enumerator from the first to the fifth survey and subsequent questionnaires were delivered by mail to those who responded in the past two years. The survey collected information on work status, health status, health behaviors, and other characteristics. The number of respondents was 34,240 (response rate: 83.8%) in the first wave and decreased via attrition to 19,930 in the 15th wave (response rate: 95.3%). Consequently, the continuity rate from the first to the 15th wave was 58.2%. No new respondents were added after the first wave.

Based on this study’s aim, the sample was restricted as follows: first, full-time regular employees in the first wave were included (13,895 individuals remained). This significant reduction from the initial sample was caused by the fact that women in their 50s either did not work or worked as non-standard workers. Second, to accurately capture the retirement pathways, continuous observations from the first wave were used (13,895 individuals and 150,652 person-waves observations remained). Third, respondents who re-entered the labor force after retirement were excluded because the nature of their retirement process was blurred (Calvo et al., 2009; De Vaus et al., 2007; 11,206 individuals and 113,813 person-waves observations remained). Fourth, respondents who retired before the age of 60 were excluded (10,583 individuals and 107,098 person-waves observations remained). Fifth, respondents with missing health measure values were excluded (10,452 individuals and 100,899 person-waves observations remained). Finally, respondents with missing covariate values were excluded (10,253 individuals and 96,297 person-waves observations remained).

Health Measures

Health measures included physical limitation and depression. Physical limitation was a dichotomous variable that took the value of 1 if respondents had any difficulty in the following ten activities and zero otherwise: walking, getting out of bed or up from the floor, sitting down and standing up from a chair, putting on and taking off clothes, washing hands and face, eating, toileting, bathing, climbing and descending stairs, and carrying purchases.

Depression scores were generated from the scores obtained on the Japanese version of the Kessler Psychological Distress Scale (K6). The K6 consists of six questions: “During the past 30 days, about how often did you…” “…feel nervous?” “…feel hopeless?” “… feel restless or fidgety?” “… feel so depressed that nothing could cheer you up?” “… feel that everything was an effort?” and “… feel worthless?” The respondents rated each question from zero (never) to four (all of the time), and the total score ranged from zero (good) to 24 (bad). The dichotomous variable indicating depression symptoms took the value of 1 if the total score was five or higher, and zero if the total score was less than five (Sakurai et al., 2011).

Retirement Pathways

The LSMOA asked respondents about their engagement in paid work and type of work. Respondents with no paid work at a certain survey point were defined as fully retired (De Vaus et al., 2007; McDonough et al., 2017; Wels & Takami, 2021). Furthermore, this study classified the type of work as follows: “Executive of company” and “Regular staff” as full-time regular employment; “Self-employed worker” and “At-home-employee” as self-employment; or “Part-time worker,” “Dispatched worker,” “Contract and entrusted employee,” and “Others” as non-standard job.

Consequently, there were four work statuses for each period: (1) full-time regular jobs; (2) non-standard jobs; (3) self-employed; and (4) fully retired. Two retirement pathway dummies were created for this study. First, the abruptly retired dummy took the value of 1 when respondents directly transitioned from full-time regular employment to full retirement, as long as they remained retired, and zero otherwise. That is, they did not have non-standard jobs or self-employment before full retirement. Second, the gradually retired dummy took the value of 1 when respondents transitioned from non-standard jobs or self-employment to full retirement, as long as they remained retired, and zero otherwise. That is, they fully retired from full-time employment through a gradual transition. Gradual retirement often refers to a transition from full-time career employment to part-time employment or self-employment (Cahill & Quinn, 2020), however, the “gradually retired” dummy is used to capture the effect of full retirement through a gradual transition, following Calvo et al. (2009).

Respondents who were continuously working as full-time regular, non-standard, or self-employed were classified as working and used as a reference category in the estimation. A binary retirement dummy (fully retired or not) was also used in the estimation for comparison with the results of the retirement pathways.

Covariates

Several time-variant covariates were used in the estimation to control for the possible effects of confounders between retirement pathways and health: age, age squared, marital status, living alone, moderate exercise, drinking, smoking, health checks, hobbies, and social activities. Age and age squared were used to control for the quadratic aging effect on health rather than the linear aging effect. Marital status took the value of 1 if respondents had a spouse and zero otherwise. Living alone took the value of 1 if respondents had no one living with them and zero otherwise. Moderate exercise took the value of 1 if respondents engaged in medium-intensity (producing slight shortness of breath) exercise and zero otherwise. Drinking took the value of 1 if respondents drank alcohol at least a few times in a month and zero if they drank less than that. Smoking took the value of 1 if respondents smoked and zero if they did not. Health checks took the value of 1 if respondents had a health check during the past year and zero if they did not. Hobbies took the value of 1 if respondents engaged in hobbies and cultural activities during the past year and zero if they did not. Social activities took the value of 1 if respondents engaged in community and social participation activities during the past year and zero if they did not.

Estimation Strategy

The outcome equation used to estimate the effects of retirement pathways on health was as follows:

$$H_{it}={\alpha AR}_{it}+{\beta GR}_{it}+{\mathbf{X}}_{it}\mathrm{\gamma }+{\mu }_i+{\lambda }_t+{\epsilon }_{it},$$

where $H$ represents one of the health measures and the subscripts i and t represent individuals and time, respectively. $AR$ and $GR$ represent the abruptly retired and gradually retired dummies, respectively. $\mathbf{X}$ is a vector of covariates, $\mu$ is an individual fixed effect, $\lambda$ is a time-fixed effect, and $\epsilon$ is an error term. Although both dependent variables were dichotomous, a linear probability model was used for ease of interpretation (Coe & Zamarro, 2011; Oshio & Kan, 2017).

This study used a fixed effects (FE) model, exploiting the change in both dependent and independent variables within individuals. If no correlation was identified between retirement and the error term, this model removed the unobserved fixed effect such as preference, and thus obtained consistent estimates of the retirement pathways. However, this condition could be violated for two reasons: first, the bidirectional relationship between retirement pathways and health leads to a correlation between retirement and error terms; second, unobserved time-varying shocks, such as family illness, can affect respondents’ retirement decisions and health, resulting in a correlation between retirement and error terms.

Consequently, this study used a fixed effects instrumental variables (FE-IV) estimation to address this problem (Heller-Sahlgren, 2017). To obtain consistent estimates of retirement, the instruments were required to have no correlation with the error term $\epsilon$ (exogeneity condition) and to have a correlation with the retirement pathway dummies (relevant condition). The IV approach is ideal for Japanese context because the Japanese government has reformed the public pension system; thus, the pension eligibility age varies depending on the birth cohort. The politically determined pension eligibility age is not affected by individual retirement behavior but affects the retirement behavior; in other words, the pension eligibility age is exogenous to the retirement pathways. Thus, the instruments are relevant to and exogenous to retirement-health relationship.

In the first stage of the FE-IV estimation, each retirement pathway dummy was regressed on the same set of instruments and covariates using the FE linear probability model as follows (below is the case of abruptly retired dummy):

$${AR}_{it}={\mathbf{X}}_{it}\mathrm{\delta }+{\mathbf{Z}}_{it}\mathrm{\theta }+{\varphi }_i+{\eta }_t+e_{it},$$

where $\mathbf{Z}$ is a vector of instruments, $\varphi$ is an individual fixed effect, $\eta$ is a time fixed effect, and $e$ is an error term. Four instruments were used for the baseline model: (1) over partial pension eligibility age dummy, which took the value of 1 if respondents were equal to or older than the eligibility age for the earnings-related benefits and zero otherwise; (2) over full pension eligibility age dummy, which took the value of 1 if respondents were equal to or older than the eligibility age for the flat-rate benefits and zero otherwise; (3) years since partial pension eligibility age; and (4) years since full pension eligibility age.

The predicted values of retirement $\widehat{AR}$ and $\widehat{GR}$ derived from the first-stage results were used in the second-stage estimation as follows:

$$H_{it}=a{\widehat{AR}}_{it}+\beta {\widehat{GR}}_{it}+{\mathbf{X}}_{it}\mathrm{\gamma }+{\mu }_i+{\lambda }_t+{\epsilon }_{it}.$$

As variations in $\widehat{AR}$ and $\widehat{GR}$ were produced by the exogenous instruments, that is, the endogenous part of retirement pathways (caused by reverse causality or unobservable confounders) was removed, the retirement pathway estimates were consistent (see Angrist, 2006 for more details).

Abrupt retirement may occur earlier than full retirement through a gradual transition. This means that the difference between the abrupt and gradual transitions to full retirement includes the differences in timing as well as pathways, which may contaminate the effect of retirement pathways on health. To control for the retirement timing, this study performed two separate estimations following Celidoni et al. (2017). The effect of early retirement was estimated using instruments (1) and (3). This estimation could identify the effect of retirement occurring around the ages of 60–61 because the IV estimation used the information of those who retired when they became eligible for the partial pension benefits. This allowed for determining the effects of retirement by controlling for retirement timing. The second estimates the effect of normal retirement using instruments (2) and (4). This estimation could identify the effect of retirement occurring around the ages of 64–65 because the IV estimation used the information of those who retired when they became eligible for the full pension benefits.

Results

Descriptive Statistics

Table 2 presents the mean values from the analytical sample (person-waves observations) by retirement pathways. Regarding physical health, the proportion of working individuals with physical limitation (6.3%) was substantially less than abruptly retired (12.6%) or gradually retired (12.8%) individuals. On the other hand, regarding mental health, the ratio of reporting depression showed little difference: working (24.0%); abruptly retired (24.1%); and gradually retired (23.9%). Regarding retirement pathways, the number of observations for working was the largest (81.9%), followed by abruptly retired (10.0%), and gradually retired (8.2%). A significant number of respondents were still working during the survey period. Regarding retirement timing (not shown in Table 2), the average retirement age when the respondents fully withdrew from the labor market for abruptly and gradually retired was 61.2 and 63.9 years, respectively. Those who retired fully through a gradual transition withdrew from the labor market later than abrupt retirees.

Table 2.

Mean of Person-Waves Observations by Retirement Pathways.

	Working	Abruptly retired	Gradually retired
Health measures
Physical limitation	.063	.126	.128
Depression	.240	.241	.239
Instruments
Over partial pension eligibility age	.440	.993	.999
Over full pension eligibility age	.191	.669	.862
Years since partial pension eligibility age (0–13)	1.552	5.214	6.842
Years since full pension eligibility age (0–13)	.525	2.522	3.479
Covariates
Age (50–73)	59.3	65.3	66.9
Marital status	.870	.811	.850
Living alone	.076	.105	.088
Moderate exercise	.351	.571	.582
Drinking	.659	.529	.598
Smoking	.295	.182	.182
Health checks	.850	.643	.660
Hobbies	.657	.729	.682
Social activities	.437	.544	.528
Number of observations	78,831	9,597	7,869

Note. Abruptly retired = individuals fully retiring directly from full-time employment; gradually retired = individuals fully retiring from full-time employment through a gradual transition. Parentheses show minimum and maximum values.

Main Results

Table 3 presents the results of the binary retirement model in the top panel, and those of the retirement pathways model in the bottom panel. Each panel shows a comparison between the FE and FE-IV (second-stage results). The first-stage results for the FE-IV estimation (see Supplemental Table A1) indicated that the instruments explained retirement behavior well.

Table 3.

Effect of Retirement on health.

	Physical limitation				Depression
	FE		FE-IV		FE		FE-IV
Retirement (ref: working)
Fully retired	.026 (.004)	***	.029 (.021)		.000 (.006)		−.104 (.029)	***
F for instruments a			220.2				220.2
Retirement pathways (ref: working)
Abruptly retired	.026 (.006)	***	.076 (.025)	***	−.012 (.008)		−.181 (.038)	***
Gradually retired	.025 (.006)	***	−.058 (.040)		.011 (.007)		.041 (.056)
Sanderson–Windmeijer F for instrumentsa,b
Abruptly retired			241.2				241.2
Gradually retired			162.1				162.1
Gradually−abruptly	−.001		−.133	***	.022	**	.221	***

Note. FE = fixed effects; FE-IV = fixed effects instrumental variables; abruptly retired = individuals fully retiring directly from full-time employment; gradually retired = individuals fully retiring from full-time employment through a gradual transition; gradually−abruptly = difference in the estimates of gradually and abruptly retired dummies, and the test result of equal coefficients. Observations = 96,297; individuals = 10,253. Standard errors clustered by individuals are shown in parentheses. All regressions included age, age squared, marital status, living alone, moderate exercise, drinking, smoking, health checks, hobbies, social activities, and wave dummies.

^aF > 10 indicated that the instruments were relevant (Stock et al., 2002).

^bSanderson–Windmeijer F was used for multiple endogenous variables (Sanderson & Windmeijer, 2016).

Significance levels: ^***p < .01, ^**p < .05, ^*p < .1.

First, the results of the binary retirement model were examined. The FE and FE-IV estimates showed different results for physical and mental health. Regarding physical limitation, the FE-IV estimate showed no significant effect of full retirement, whereas the FE estimate showed a significant negative association. The FE-IV estimate showed a significant beneficial effect on depression, whereas the FE estimate showed no significant association. Causal estimation demonstrated that fully retired individuals were 10.4 percentage points less likely to have depression than working individuals.

Following this, the results of the retirement pathways model were examined. The FE and FE-IV estimates showed different results for the two health measures. Regarding physical limitation, although the FE estimates showed that both retirement pathways had a significant negative association, the FE-IV estimates demonstrated that an abrupt transition had a significant negative effect, whereas full retirement through a gradual transition had no significant effect. Causal estimation showed that abruptly retired individuals were 7.6 percentage points more likely to have physical limitation than working individuals. Therefore, individuals fully retiring through a gradual transition were 13.3 percentage points less likely to have physical limitation than those abruptly retiring. Regarding depression, although the FE estimates showed no significant association with either retirement pathway, the FE-IV estimates demonstrated that an abrupt transition had a significant beneficial effect, whereas full retirement through a gradual transition had no significant effect. Causal estimation demonstrated that abruptly retired individuals were 18.1 percentage points less likely to have depression than working individuals. Consequently, individuals fully retiring through a gradual transition were 22.1 percentage points more likely to have depression than those abruptly retiring.

As mentioned in the methods section, this study confirmed the effect of retirement pathways by controlling for retirement timing. Table 4 presents the second-stage results of the FE-IV estimations. The first-stage results (see Supplemental Table A2) indicated that the first-stage mechanism was properly captured, being largely in line with the results of Celidoni et al. (2017).

Table 4.

Effect of Retirement Pathways on health by Retirement Timing.

	Physical limitation				Depression
	Early retirement		Normal retirement		Early retirement		Normal retirement
Retirement pathways (ref: working)
Abruptly retired	.070 (.026)	***	.007 (.080)		−.205 (.039)	***	−.468 (.127)	***
Gradually retired	−.052 (.044)		−.030 (.052)		.030 (.064)		.190 (.083)	**
Sanderson–Windmeijer F for instrumentsa,b
Abruptly retired	698.2		53.6		698.2		53.6
Gradually retired	457.3		117.3		457.3		117.3
Gradually–abruptly	−.122	***	−.037		.234	***	.658	***

Note. Abruptly retired = individuals fully retiring directly from full-time employment; gradually retired = individuals fully retiring from full-time employment through a gradual transition; gradually−abruptly = difference in the estimates of gradually and abruptly retired dummies, and the test result of equal coefficients. Observations = 96,297; individuals = 10,253. Standard errors clustered by individuals are shown in parentheses. All regressions included age, age squared, marital status, living alone, moderate exercise, drinking, smoking, health checks, hobbies, social activities, and wave dummies.

^aF > 10 indicated that instruments were relevant (Stock et al., 2002).

^bSanderson–Windmeijer F-statistics was used for multiple endogenous variables (Sanderson & Windmeijer, 2016).

Significance levels: ^***p < .01, ^**p < .05, ^*p < .1.

The results do not significantly differ from the results which do not control the retirement timing. Regarding physical limitation, abrupt retirement showed a significant negative effect in early withdrawal but no significant effect in normal withdrawal; full retirement through a gradual transition showed no significant effect in either early or normal retirement. Hence, in early retirement, individuals fully retiring through a gradual transition were significantly less likely to have physical limitation than those abruptly retiring. Regarding depression, abrupt retirement had a significant beneficial effect in both early and normal retirement. This study found that individuals fully retiring through a gradual transition were 19.0 percentage points more likely to experience depression in normal retirement than those working. Consequently, in early and normal retirement, individuals fully retiring through a gradual transition were more likely to have depression than those abruptly retiring.

Robustness Checks on Model Specification and Sample Selection

For the results presented in Table 4, several robustness checks were performed regarding model specification and sample selection. Figure 1 shows the results of four additional analyses. First, the model without potential mediators (moderate exercise, health checks, hobbies, and social activities) was estimated to check the sensitivity (see Figure 1, first result from the left and Supplemental Table A3). Although the estimates were slightly smaller in absolute value than those in Table 4, the results did not differ much from the main results.

Figure 1.

Estimates of retirement pathways on health by model specification and sample selection. Note. Abruptly retired = individuals fully retiring directly from full-time employment; gradually retired = individuals fully retiring from full-time employment through a gradual transition; no mediators = results based on the model that did not use mediators (moderate exercise, health checks, hobbies, and social activities); lagged health = results based on the model that added lagged health (physical limitation and depression) as covariates; no NS = results based on the sample that excluded those who were working in non-standard jobs/self-employment; re-entry = results based on the sample that included those who re-entered the labor force. The estimates of weak instruments (F < 10) are not shown. Significance levels: ^***p < .01, ^**p < .05, ^*p < .1.

Second, as this study used the FE model that exploited the change within individuals, the previous health status is not so influential but nevertheless, might have some influence. Thus, a model using lagged health status as covariates was estimated to control for the effect of health on subjects’ subsequent health and retirement behavior (see Figure 1, second result from the left and Supplemental Table A4). Given that this estimation drops the first-wave observations, the lagged health status for the first-wave observations were imputed using their current (first-wave) health status. The results did not differ much from the main results, which indicates that the effect of the previous health level is negligible.

Third, the baseline model was estimated using the subsample that excluded those who were working in non-standard jobs/self-employment (see Figure 1, second result from the right and Supplemental Table A5). As the reference category included full-time regular workers, non-standard workers, and self-employed individuals, the comparison with fully retired individuals might have been contaminated. The results were similar to the main results; however, the difference in the effects on depression between gradually retired and abruptly retired individuals in early retirement was no longer significant. As the instruments were no longer relevant for normal timing, the results were not shown in Figure 1.

Fourth, the baseline model was estimated using the extended sample that included labor force re-entrants (see Figure 1, first result from the right and Supplemental Tables A6 and A7). Re-entry of the labor was already common in Japan; hence, this was not caused by the current pension reform (Yamada & Seike, 2001). However, it is important to consider these re-entrants. This study removed those who re-entered the labor force because their retirement behavior was blurred. However, given that labor force re-entry is one form of gradual retirement (Cahill & Quinn, 2020), these re-entrants can be included in the reference category as a working group. The descriptive statistics indicated that labor force re-entrants were physically less healthy than working individuals but were healthier than fully retired individuals, and they were mentally healthier than those in other retirement pathways. Regarding the estimation results, although the estimates were larger in absolute value than those in Table 4, their implication did not differ from that of the main results.

Finally, abrupt retirement and working in non-standard jobs/self-employment were compared using the subsample that excluded individuals retiring fully through a gradual transition, to see which is better: to retire completely or work a little longer but at less intensity (see also Supplemental Table A8). The results are not shown in Figure 1 because the retirement pathways dummies differ from other estimations. Abrupt retirement and working in non-standard jobs/self-employment showed no statistical difference in health effects.

Robustness checks considering heterogeneity

Next, this study examined the heterogeneity in health effects. The baseline model was estimated separately according to sex and former occupation because the effect of retirement may be heterogeneous between groups (Calvo et al., 2022; Mazzonna & Peracchi, 2017; McDonough et al., 2017). The results showed differences according to sex and occupation. The left-hand side of Figure 2 shows the results based on sex (see also Supplemental Table A9). Regarding men, abruptly retired individuals were more likely to have physical limitation and less likely to have depression, whereas individuals retiring fully through a gradual transition showed no significant effect on either health measure. Hence, those retiring fully through a gradual transition were significantly more likely to have depression than those abruptly retiring. Regarding women, abruptly retired individuals were more likely to have physical limitation but showed no significant effect on depression, whereas individuals retiring fully through a gradual transition showed a significant and beneficial effect on physical limitation but showed no significant effect on depression. Consequently, those retiring fully through a gradual transition were significantly less likely to have physical limitation than those abruptly retiring. Although the health effects of retirement pathways are different by sex, the results are consistent with the main results.

Figure 2.

Estimates of retirement pathways on health by sex and former occupation. Note. Abruptly retired = individuals fully retiring directly from full-time employment; gradually retired = individuals fully retiring from full-time employment through a gradual transition. Non-manual workers included professional, engineering, administrative, and clerical workers. Manual workers included sales, service, security, agricultural, forestry, fishery, transport, machine operation, and manufacturing process workers. The estimates of weak instruments (F < 10) are not shown. Significance levels: ^***p < .01, ^**p < .05, ^*p < .1.

The right-hand side of Figure 2 presents the estimation results based on occupation (see also Supplemental Table A10). Regarding non-manual workers (see Note in Figure 2 and Supplemental Table A10), abrupt retirement showed no significant effect on physical limitation but had a significant beneficial effect on depression, whereas full retirement through a gradual transition showed no significant effect on physical limitation but had a significant negative effect on depression. Therefore, individuals retiring fully through a gradual transition were significantly more likely to have depression than those abruptly retiring. Regarding manual workers, abruptly retired individuals were more likely to have physical limitation but less likely to have depression, whereas those retiring fully through a gradual transition showed no significant association with either health measure. The effect of abrupt retirement in normal retirement was not shown due to the F-statistics being less than 10. Consequently, individuals retiring fully through a gradual transition were significantly less likely to have physical limitation than those abruptly retiring. Although the health effects of retirement pathways are different by occupation, the results are consistent with the main results.

Finally, to see whether the subjects’ subsequent health is worse after full retirement through a gradual transition if the health status at the time of full retirement was poor, the interaction effect was examined (see Supplemental Table A11). Accordingly, this study restricted the sample to those who were fully retired, and used a normal random effects model because the health status at the time of retirement and the retirement pathways were invariant within individuals and the instruments were unable to explain the retirement decision among fully retired individuals. This estimation model substantially differs from the baseline model; thus, the following results need to be interpreted with caution. An interaction term between physical limitation/depression at the time of full retirement (=1, otherwise = 0) and gradually retired (=1, otherwise = 0) was used for the subsample composed of full retirees. The results showed that full retirement through a gradual transition had a more negative association with depression if the health status at the time of full retirement was poor, and that there was no interactive association with physical limitation. These results might provide a clue to understanding the negative effect of full retirement through a gradual transition on mental health; however, the results were contaminated by the difference between individuals.

Discussion

This study aimed to examine the causal effects of retirement pathways on the physical and mental health of older Japanese individuals. Several additional analyses confirmed the robustness of the main results. Regarding physical health, abruptly retired individuals showed a higher probability of physical limitation than working individuals. This negative effect was pronounced in manual workers but not in non-manual workers. On the other hand, although those who fully retired through a gradual transition showed only minor differences from working individuals, only women showed a lower probability of physical limitation than working individuals. Therefore, full retirement through a gradual transition was better for physical health than abrupt retirement. These results suggest that working is an important health investment for women and those who have previously engaged in physically demanding jobs. In other words, an abrupt decline in physical activity worsens physical health, whereas gradually retiring individuals avoid this abrupt change. These results largely support the hypothesis that a gradual transition to full retirement is better for health than an abrupt transition, which is in line with findings of Azar et al. (2019).

Regarding mental health, an abrupt transition to full retirement is beneficial for men and manual and non-manual workers, but not for women. In contrast, full retirement through a gradual transition largely had no effect, however, the negative effect was pronounced in non-manual workers. Hence, full retirement through a gradual transition was worse for mental health than abrupt retirement. These results do not support the hypothesis that a gradual transition to full retirement is better for health than an abrupt transition. Interestingly, this study’s results showed the opposite trend to what was hypothesized. One possible explanation is that gradual retirees do not enjoy retirement process (De Vaus et al., 2007). In other words, some gradual retirees had to work after mandatory retirement due to financial pressure (Cahill et al., 2006, 2015; Zhan et al., 2009). Moreover, older workers are discouraged because they are willing to work as full-time regular employees but are forced to work in non-standard jobs (Wels & Takami, 2021). These aspects of gradual transition to full retirement may cause mental health deterioration. In contrast, abrupt retirees were relieved from such pressure or stress, resulting in better mental health.

Work-related physical activity does not always maintain or improve the physical health of older adults (Hessel, 2016); however, as theoretically predicted, this study found that prolonged work-related activity is an effective investment in physical health, particularly for older Japanese adults. It also found that a prolonged working life negatively affects an individual’s mental health, suggesting that mental health can be susceptible to influence from retirement reasons (De Vaus et al., 2007; Zhan et al., 2009). Thus, this study identified different mechanisms for the effects of retirement pathways on physical and mental health. From a scientific viewpoint, and the second contribution of this study, it was confirmed that considering both physical and mental health in a study is important.

This study also compared the results between binary retirement and retirement pathways and found that simple retirement had no significant effect on physical limitation; however, abrupt retirement showed a significant negative effect. This result indicates that the opposite effects of retirement pathways offset each other and result in an insignificant effect of simple retirement. Consequently, the heterogeneity in retirement should be considered, particularly the difference in retirement pathways.

This study also highlighted the differences in the results between less causal (FE) and rigorous causal (FE-IV) estimations. The different results obtained in these two estimations indicated that ordinary estimation leads to inconsistent effects and that rigorous causal estimation is required for the precise policy implications. The IV estimation identified the effect of retirement pathways that occurred regardless of the previous health status and other unobservable confounders. Moreover, this study used the change in health status within individuals; therefore, the subject’s previous health did not affect the results much, as supported by the additional analysis that included lagged health as covariates. Thus, the results obtained in this study are robust.

The study participants were born during 1945–1955 and thus faced the public pension reform as explained in this paper. Owing to the increase in the eligibility age for pension benefits and the revision of the related law, the participants work longer and more frequently in non-standard jobs compared to the previous generations. Traditional lifetime employment is still being practiced in Japan; however, the participants are forced to work in non-standard jobs even if they want to work in full-time regular jobs. Thus, the protective aspect of lifetime employment (e.g., finance or motivation) has been weakened in this cohort. The identified effects of the retirement pathways on physical and mental health were found in this particular environment. After the pension reform is completed, studies should examine whether the subsequent cohorts show a similar trend; thereby deepening our understanding of the relationship between lifetime employment and the health of older adults in Japan.

However, this study had several limitations. First, a significant number of older adults were still working during the survey period. In Japan, many older workers continue to work as non-standard workers with their career employers even after mandatory retirement (Clark et al., 2015; Shimizutani & Oshio, 2010). These individuals were still in the gradual retirement process and thus have not yet become full retirees in the sample, leading to relatively weak evidence of full retirement through a gradual transition. Further investigation is required after these respondents withdraw from the labor force. Second, this study did not consider the voluntariness of retirement. The reason of retirement may affect subsequent health, although the IV approach mostly eliminates the effect. As this study already considered the pathways of retirement, there was no room to treat the reason for retirement choice in the IV estimation. Third, gradual transition to full retirement was more diverse (Cahill et al., 2006). This study did not distinguish between bridge employment (working with a new employer), phased retirement (working reduced hours with the same employer), and re-entry after full retirement because of the feasibility of the IV estimation. Future research should clarify the differences among the three components of gradual transition to full retirement. Fourth, the effects of retirement pathways are estimated as local average treatment effects (LATE; Angrist & Pishcke, 2009). That is, the LATE is the result for those who were retired because they had become eligible for pension benefits, but not the result for all retirees. Moreover, this study focused on the retirement pathways from full-time regular employment and not on the pathways from non-standard jobs. Therefore, the implications of this study cannot be generalized to the entire older adult population.

Conclusion

Despite this study’s limitations, it provides new evidence on the heterogeneity of retirement pathways: a gradual transition is possibly better for physical health but worse for mental health than an abrupt transition. This finding broadens the scientific knowledge of retirement-health relationship studies. In addition, the results assist the understanding of the possible effects of prolonged working life on medical and long-term care systems. Specifically, forcing older adults in Japan to work in non-standard jobs through mandatory retirement is detrimental to their mental health. Policy makers should consider these implications.

ORCID iD

Masaaki Mizuochi https://orcid.org/0000-0002-1143-8644

Data Availability Statement

The LSMOA was conducted based on the Statistics Act of Japan. Consequently, the micro dataset is not publicly available. This study project received official approval for the use of secondary data from the MHLW under Seitou-0713-1 as of July 13, 2021.