The retirement effect on mental health in Europe during 2006-2015: Evidence of Ashenfelter’s dip

501 The retirement effect on mental health in Europe during 2006-2015: Evidence of Ashenfelter’s dip Thang Vo Duyen Tran University of Economics Ho Chi Minh City Abstract Since agein

501

The retirement effect on mental health in Europe during 2006-2015:

Evidence of Ashenfelter’s dip

Thang Vo Duyen Tran

University of Economics Ho Chi Minh City

Abstract

Since ageing raises concerns over the economic efficiency of rising pension age, the impact of retirement

on various aspects of life is on the focus of Euro-pean countries’ policies Using panel data from the Survey

of Health, Ageing and Retirement in Europe (SHARE), this study investigates the effect of retirement on mental health measured by the EURO-D scale (12 levels of depression) Age above specific-country eligible pension age is used as an instrumental variable for retirement status in the fixed effect model to remedy the potential endogeneity bias This study is the first effort to capture the mental health effect in anticipation of retirement, a phenomenon called ‘Ashenfelter’s dip’ or ‘pre-programme dip’ This study also compares short term effects and long term effects of retirement, which is rarely done before Different impacts of reasons for retirement categorized into three groups are also analyzed in this study

The study indicates that retirees feel less depressed than people who remain in the labor force When the age above pension age of individuals is included to pre-dict retirement behavior, the results confirm an analogous effect of retirement on mental health In terms of reasons for retirement, retiring due to positive circum-stances and aspirational motivations reduce depression remarkably, while there is no evidence to confirm that retiring by negative circumstances affect one’s mental health

The study finds a similar effect for people who are expected to retire in the next two years, but this is not the case for people who know they will retire in the next four years The potential retirees seem to adjust their lifestyles in response to future retirement Two years after retirement, the effect is reverted, but after four years the results are not conclusive Retirees may adapt to their new life completely and the effect

of retirement is no longer important

Keywords: SHARE, retirement, mental health, panel data, instrumental variable, Ashenfelter dip

1 Introduction

For the last few decades, population ageing has been on the rise to be one of the biggest challenges for politicians and scientists all over the world, especially in Western countries (Butterworth et al., 2006) Particularly in Europe, the rate of elderly accounted for 20.3% in 2000, which was the highest in comparison

to other areas’ The projection is that by 2050 there will be over one-third of European population are individuals aged over 60 (see table 1) Furthermore, Eurostat (2015) reports that within the last 50 years, life expectancy at birth in the EU-28 has increased nearly 10 years and European countries are most likely

to have the highest life expectancy in the world

502

Since the number of senior citizens is believed to be growing gradually, the popula-tion of retirees is also increasing as a result, which, in turn, raises concern not only for governments but also for work organizations and individuals in general Because, with the number of retirees now enlarging over time, governments have to take into account the sustainability of social welfare, enterprises have to assure the productivity of elderly employees, and people are more likely to feel uncertain about their future when they get retired (Desmette et al., 2015) Therefore, it is essential to understand the correlation between retirement and mental health for developing employment policies (Butterworth et al., 2006; Zhu, 2016) as well as individuals’ life satisfaction

2 Literature review

In terms of changes in the life of retirees, previous studies which attempt to investi-gate the correlation between retirement and mental health has shown conflicting results due to different strategies employed (Zhu, 2016) For instance, the studies of Salokan-gas et al (1991), Mein et al (2003), Charles et al (2004) and Mojon-Azzi et al (2007) all find a positive impact of retiring on the mental health of senior citizens to such

an extent Similar results are also revealed in Jokela et al (2010)’s research, which takes in to account the age at retirement, reasons for retirement and length of time spent in the retirement through a longitudinal data set In Europe, several studies have found evidence exposing that retiring could possibly impact the mental health of the elderly (Belloni et al., 2016; Borsch-Supan and Schuth, 2014; Heller-Sahlgren, 2017) How-ever, the influence may vary depending upon short- and long-term (Bianchini et al., 2015; Heller-Sahlgren, 2017), men and women (Belloni et al., 2016), as well as differ-ent geographical areas (Aichberger

et al., 2010)

Adversely, there are also studies on the topic providing evidence to the contrary Buxton et al (2005) with a total of 1,875 respondents in a cross-sectional analysis shows that early retirees were more likely to have generalised anxiety disorders and depressive disorders The findings support the idea of increasing retirement age in Britain up to

Falba et al (2008) employ Poisson regression for 4,241 observations extracted from the Health and Retirement Study (HRS) and suggest that working longer and retiring earlier than expected each may compromise psychological well-being Another study with data from HRS of Calvo et al (2012) shows that early retirements, those occurring prior to traditional and legal retirement age, worsen health This study, however, does not find any disadvantages associated with late retirements Therefore, they suggest that raising the retirement age may decrease the subjective health of retirees because the group of early retirements has been enlarged

At the same time, the impact of retirement on health may lead to differential out-comes in one single study One example is the work of Johnston et al (2009), which shows robust evidence that in the short-run, retirement increases individual’s sense of well being and mental health, but not necessarily their physical health Then the au-thors argue that government expenditure for health would not be significantly affected by increasing the official retirement age Recently, Zhu (2016) employs a panel data set of the Household, Income and Labour Dynamics in Australia (HILDA) Survey and finds that retirement status has positive and significant effects on women’s self reported health, physical and mental health outcomes

A study of Byles et al (2016) confirms that retirement is associated with psychological distress among men

in Australia How-ever, this study does not find any association between retirement and psychological distress among women These findings support the provision of flexible employment options for older adults

Source: United Nations (2015)

2.1 Impact mechanism of retirement on mental health

Since there are still conflicting results revealed by previous studies on this topic, contrasting mechanisms through which retirement affect the mental health of individu-als have also been discussed in different aspects On the one hand, retirement seems to have a positive impact on the well-being of people

by three ways Firstly, Belloni et al (2016), Hessel (2016), Zon et al (2016) and Stenholm et al (2017) argue that once elder employees start to enroll in retirement, the relief from stress caused by working and/or the precarious working environment could improve their mental health to such an ex-tent Secondly, retirees are believed to have more leisure time than people who remain in the labor force, therefore, they are more likely to get engaged in physical activities including exercises, which could advance their health remarkably (Evenson et al., 2002; Zon et al., 2016) At the same time, it is discussed in the study of Myllyntausta et al (2017) that retirees seem to sleep better due to the fact that they have more unoccupied hours after they stop working According to Zon et al (2016), having more time once retired also means that senior citizens are able to have more social engagement, which could be an advantage for their functional health Lastly, social capital and networks of elderly have actually been investigated in many research (e.g Gannon et al (2014), Liu et al (2016), O’Doherty et al (2017)), and is regarded as one of the main impact mechanisms that retirement affect mental health through Because retirees are likely to have more time to find new and voluntary contacts (Heller-Sahlgren, 2017)

On the other hand, retirement, which is sometimes refereed as a “stressful event” with major changes for a certain number of people (Ekerdt, 1987; Salokangas et al., 1991), could have a negative influence on the mental health of retirees through different channels To begin with, although the increasing stock of social capital and enlarging network is believed to improve the mental health of most individuals, to other people, stop working could worsen their mental health, because of the decreased bonds with their former colleagues (Stenholm et al., 2017), or the loss of social contacts and par-ticipations related to work (Zon et al., 2016) Moreover, since having occupations are often considered as a basic role of an individual in societies, people who lose that when they get retired are supposed to have less self-respect and feel isolated, which worsens their mental health as a result (Hessel, 2016) In terms of financial issues, retirement particularly leads to a decrease of the regular income In most cases, this affects the financial insecurity of individuals, especially those who have fewer economic resources when they are retired (Heller-Sahlgren, 2017; Zon et al., 2016) Above all, adapting to retirement requires people to have changes in not only the frequency but also the inten-sity of work-related activities (Grundy et al., 1999) As a consequence, the adjustments in their lifestyle are accounted for a worse health outcomes, which includes depression (Dave

et al., 2006)

504

2.2 Limitations

While studies have investigated this topic on many aspects with different analyzing methods, it is true

to say that there are still limitations existing First, in addition to re-search heterogeneity, there are potential measurement problems, including unmeasured variables, level of analysis, and use of inappropriate measures Second, the unavail-ability of data across time and place as well as across either economic or mental health measures highlights a clear need for research that has longitudinal data and larger sam-ple size (Hanisch, 1999) Third, as pointed out in the research of Salokangas et al (1991) and Mein et al (2003), future studies examining this issue should take into ac-count the conditions of retirement, the distinction

of people, then clarify the problems in the process of adapting to retirement Because the differences in mental adaptation of each person in their retirement term might not only be caused by the fact that they stop working but also by a combination of earlier life conditions, social-economic sta-tus and retiring motivation Also, in order to analyze changes in the health of retired people that probably caused by retirement, studies need to be prospective and include the control group Without these two factors, many studies dealing with this issue seem to be problematic

Above all that, the most important limitation noticed from previous studies is that potential effects of retirement on elderly workers’ well-being before the actual retire-ment might have been underestimated (Hessel, 2016) In other words, since retirement seems to be a predictable event for most people, employees who are approaching retire-ment time could possibly already experience changes in their mental health However, to our knowledge, whether the impact of retirement exists preceding to the retirement still has not been confirmed by any of the former research This “pre-impact” is pretty similar to the “pre-programme - dip” first presented in the paper of Ashenfelter (1985), which is then often refereed as the

“Ashenfelter-dip” to generally describe the decline of the outcomes prior to the actual participating of individuals in a particular programme Furthermore, investigating the same topic, Coe and Lindeboom (2008) discuss that the influence of retirement on (mental-)health may not take place right at the time of retire-ment Instead, the well-being of the elderly stands the chance to be improved or worsen even before that (Hessel, 2016)

3 Contribution

In this context, fixed-effect models are employed to investigate the impact of retire-ment on mental health and, to control the potential reverse effect, an instrumental vari-able is additionally included in the models The instrument variable used in this study will be discussed specifically in section 3.3 Moreover, since motivations of retirement decision may influence the well-being of retirees when they stop working (Robinson et al., 2010), reasons for retirement are taken into account as explanatory variables The results indicate that retirees seem to feel less depressed than people who remain in the labor force to such an extent Furthermore, when the age above pension age of individuals is included in the models as an instrumental variable to predict retirement behavior, results show an analogous impact of retirement on mental health In terms of reasons for retirement, retiring due to positive circumstances and aspirational moti-vations might reduce depression remarkably, while there is no evidence to confirm that retiring by negative circumstances could affect one’s mental health

Last but most importantly, the effect of potential retirement and being retired in the last 2 years on mental health of individuals is also revealed in our research Specifically, employees who are going to retire within the next 2 years seem to have their mental well-being improved already before the retirement

4 Background

With the population of 738 million people (United Nations, 2015) - which seems to be a minority in comparison with the world population (7.349 billion) and other continents’ such as Asia (4.393 billion) or Africa (1.186 billion) - Europe, however, has always been considered as the area where most of the biggest economies in the world are According to OECD (2017), in 2015, GDP at current prices and PPPs of the Euro area was 13,627.9 billion USD, which was only lower than that of the United States (18,036.6 billion USD) Regarding countries individually, Germany, the UK and France were economies with highest GDP

in 2016, following the US, China and Japan (World Bank, 2017)

In the period from 2004 to 2015, economies in Europe experienced many changes, including both positive and negative ones These changes could be represented by the fluctuation of GDP growth, the unemployment rate and, most related to our topic, the changes in retirement policies of European countries Firstly, GDP growth of the coun-tries considered in our study is combined from World Bank data and illustrated in graph 1 with time periods divided by SHARE waves99 As can be seen from the bar chart, GDP growth of the countries varied remarkably over waves For most nations, GDP growth seemed to increase from wave 1 to wave 2, before declining sharply in the next 2 waves, where the most significant changes are observed, especially in Spain and Italy - the two countries that experienced notable negative GDP growth

Figure 1: GDP growth of European countries over waves

Secondly, using the same data combining method, unemployment rate of the 9 coun-tries is indicated

in diagram 2 Similarly to GDP growth, there was no certain trends in the unemployment rate of the

99 See Borsch-Supan, Brandt, et al (2013) for more specific information on interview time of each wave

506

countries Instead, the rate was likely to fluctuate from wave 1 to wave 6 The highest proportion could be easily noticed in Spain, with the rate being over 25% in wave 4, while the lowest rate was found in Switzerland As for other economies, Sweden, Italy and France all reached the peak of unemployment rate around 7% to 12% in either wave 5 or 6 Whereas, Austria and Belgium were likely to maintain their unemployment rate over waves, except for a slight decrease in wave 4 Next, Germany seems to be the most standout country in the graph, since it is the only economy that had the unemployment rate slumping gradually during the period

Consider now to the unusual events that happened in European countries from 2004 to 2015, it could

be said that the European debt crisis since 2009 have made a significant impact on European economies, especially those in European Union Retrospectively, the crisis could be traced back to the global financial crisis from 2007 The Great Crisis from the US is believed to have extended to European countries, where

it began with Greece before spreading out to other countries in the Eurozone, mostly Portugal, Ireland, Italy and Spain (usually described as the PIIGS countries) During the crisis, unemployment rate was found

to increase in many countries, consequently, tax revenues declined while transfer payment grew sharply Moreover, many governments in the crisis also had to bail out the banking systems, which made the public debt increase even larger (Moro, 2014) Measuring the impact of the crisis on European countries, Eurostat (2017) reports that GDP of the EU-28 plunged by 4.4% in 2009 and 0.5% in 2012 However, the period of 4 years from 2013 witnessed a continue growth in real GDP until it reached 1.9% in 2016

To understand thoroughly the circumstances under which individuals in our sam-ple data were, it is essential to track the retirement policies reformed in Europe at the time Since increasing life expectancy and population aging have been determinated as potential pressure on social welfare of European governments (Desmette et al., 2015), many countries, especially OECD ones, have been following a common path that is to increase normal and early retirement age, as well as tighten the generosity of the pen-sion system (OECD, 2015e) Specially, in 2011 the reform of raising retirement age was performed in Italy to frame the equalization between the two genders (OECD, 2015d) While in Denmark, from 2014, pension age was planned to be increased from 65 to 67 years by 2022 for people who were born after July

1955 (OECD, 2015b) Tax rates on the pension are particularly set higher for early retirees in Belgium: rather than 16.5% as previously, it is now 20% for retirement at age 60 and 18% at age 62 OECD (2015a) In 2014, pension income tax was also increased in Sweden, while pension benefits for Spain elderly was decided to

be adjusted in the future depending on the contributions of individuals Furthermore, in Austria, the

507

penalty for early retiring was raised from 4.2% to 5.1%, whereas pensions below 1,200 EUR were “frozen”

in France from April 2014 Lastly, in 2015, the insurance for old-age, survivors and disability was reduced

to approximately 9% in Germany

Figure 3: Depression level of European countries over waves

3 Data and Methods

3.1 Data source

This study uses data from the Survey of Health, Ageing and Retirement in Eu-rope (SHARE)100 (see Borsch-Supan, Brandt, et al (2013) for Data Resource Profile) SHARE is a longitudinal, multidisciplinary and cross-national survey, which aims to collect data of health, socio-economic status along with social and family networks of non-institutionalized people aged over 50 in 21 European countries and Israel101 Those countries have such divergent institutional conditions that the sampling has to be de-signed differently for each of them, it ranges from simple random selection of house-holds to complicated multistage Households are selected for the survey if it includes at least one member who was born before

1955, is a native speaker of the country and not living abroad at the time

The interviews in SHARE are conducted with Computer Assisted Personal Inter-view (CAPI) and a paper-and-pencil questionnaire CAPI questionnaires contain 20 modules, which covers different aspects

of an individual such as demographics, social networks, children, physical and mental health, behavioral risks, cognitive function, health care, employment and pensions, grip strength, walking speed, peek flow, social support, financial transfers, housing, household income, assets, activities, expectations, social and physical activities, and consumption Information that is more sensitive, like social and psychological well-being, religiosity and political affiliation, is collected by paper-and-pencil questionnaire

100 The SHARE data collection has been primarily funded by the European Commission through FP5 (QLK6-CT-2001-00360), FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812) and FP7 (SHARE-PREP: No.211909, SHARE-LEAP: No.227822, SHARE M4: No.261982) Additional funding from the German Ministry of Education and Research, the Max Planck Society for the Advancement of Science, the U.S National Institute on Aging (U01 AG09740-13S2, P01 AG005842, P01 AG08291, P30 AG12815, R21 AG025169, Y1-AG-4553-01, IAG BSR06-11, OGHA 04-064, HHSN271201300071C) and from various national funding sources is gratefully acknowledged

101 Wave 1 started with 11 European countries and Israel, other countries (Czech Republic, Poland, Ireland, Luxembourg, Hungary, Portugal, Slovenia, Estonia, and Croatia) were added in later waves

individual i at time t 𝑋⃗it is a combination of control variables which represent demographic background (age, marital status, number of children and years of education) as well as health background (numbers of chronic diseases, visiting hospital and BMI) Furthermore, factors that could possibly affect the well-being of individuals such as limitations with daily activities and the frequency

of playing sports are also included in 𝑋⃗it Next, ui is unobserved heterogeneity by time

Table 2: Country-specific pension age

Source: OECD (2015c), OECD (2015d), and OECD (2015e)

However, the coefficient in FE models is estimated with the assumption that it does not correlate with Retireit (retirement decision) This, in turn, is believed to be violated easily for different reasons (Zhu, 2016)

In addition, Eibich (2015) points out that it is crucial to concern the endogeneity of retirement, which could

be caused by bias in omitted variables and justification, as well as reverse causality Therefore, following Belloni et al (2016), Coe and Zamarro (2011), Heller-Sahlgren (2017), and Zhu (2016), Fix Effects Instrumental Variable (FE-IV) estimation is then applied in this study to control unobservable factors by time variance and reverse causal impact

Retireit = θInstrumentit + λ𝑋⃗it + ui + ɛit (2)

Equation 2 is the first stage of the FE-IV models, in which, Instrument it is the instrument for Retire it,

defined as Instrument it=I(Ageit ≥ Agep

t ) Where I is the indicator function, Ageit is the age of individual i at time t and Agep

t is the country-specific pension age by OECD (2015e) as in table 2 I takes the value of Ageit

− Aget if the condition is true, and 0 otherwise The second stage in the FE-IV estimation could be described

as in equation 3, which is similar to equation 1 in which Retire it is the predict value of retirement status

from the first stage function (2)

MHit = α + β1𝑅𝑒𝑡𝑖𝑟𝑒̂ it + β2𝑋⃗it + ui + єit (3)

The coefficient 1 in the Equation 4 represents the average impact of retirement on mental health in the year of retirement This average impact may includes impact from the current retirement and past retirement Therefore, we seperate the impact of past retirement from the impact of current retirement by

509

adding to the Equation 4 an dummy (β’1𝑅𝑒𝑡𝑖𝑟𝑒 ̂ past) indicating whether individuals have already retired

in previous waves So β’1 is the impact of past retirement

MHit = α + β1𝑅𝑒𝑡𝑖𝑟𝑒̂ it + β’1𝑅𝑒𝑡𝑖𝑟𝑒̂ past + β2𝑋⃗it + ui + єit (4)

In addtion, the retirement event is predictable, and therefore the potential impact of retirement decision

on mental health might not coincide with the exact timing of retire-ment (Coe and Linderboom, 2008) and (Philipp 2015) Mental health might improve or

worsen in anticipation of retirement, close to the so called ‘Ashenfelter’s dip’ or ‘pre-programme dip’ (Ashenfelter and Card, 1985) and (Heckman 1999)102 The current study attempts to capture this potential effect by checking whether the level of mental health changes among non-retired and will-retired in two years and in four years before the actual retirement The model in this case is as follows:

MHit = α + β1𝑅𝑒𝑡𝑖𝑟𝑒̂ it + β’’1𝑅𝑒𝑡𝑖𝑟𝑒̂ future + β2𝑋⃗it + ui + єit (5)

Where Retiref uture denotes whether an individual will retire in the next two years (or in the next four

years for the longer period) or not So β’’ 1 is the impact of ‘Ashenfelter’s dip’ The control variables (𝑋⃗it) are chosen in the same year with mental health indexes

Next, to test the consistence of results from the fixed-effect models, we employ the propensity score matching (PSM) approach formalized by Rosenbaum et al (1983) This method, additionally, is believed

to control the impact of cofounding covariates effectively (Brenna et al., 2015) In our study, PSM was used

to identify control and treatment groups based on potential characteristics that could possibly affect mental health of an individual Similarly to the FE models, the observables include gender, country, marital status, number of children, years of education, numbers of chronic diseases, BMI and daily activities

Once control and treatment groups are already chosen, the changes in mental health of retirees is estimated as in equation 6 Similar functions are also engaged in the research of Aranda (2015) and Marcus (2013)

E[Y

𝐸[𝑌1𝑖(𝑡+𝑠)− 𝑌0𝑖(𝑡+𝑠)|𝐷𝑖𝑡 = 1] = 𝐸[𝑌1𝑖(𝑡+𝑠)|𝐷𝑖𝑡 = 1] − 𝐸[𝑌0𝑖(𝑡+𝑠)|𝐷𝑖𝑡 = 1] (6)

𝐷𝑖𝑡 denotes the change in retirement status, individuals who are considered as re-tirees are categorized

as treated and take 𝐷𝑖𝑡 = 1 On the contrary, people who are untreated take 𝐷𝑖𝑡 = 0 Furthermore, 𝑌1𝑖(𝑡+𝑠)

is the depression level of individual i at time t + s (s ≥ 0), considered as after enrolling in retirement

Meanwhile, 𝑌0𝑖(𝑡+𝑠) is likely to present the mental state of the person if there had not been a change in tirement status Apparently, E[𝑌0𝑖(𝑡+𝑠)|𝐷𝑖𝑡 = 1] is the expected outcome estimation of treated population would have had if they did not retire, or before the retirement term of treated group

re-Because of the potential Ashenfelter’s dip, the difference-in-difference (DID) ap-proach is not useful in the case of retirement (***) But the PSM estimation itself can reveal if the Ashenfelter’s dip is present or not We firstly use PSM to choose the control and treatment groups at the year of retirement Then we trace back to get information of mental health of these two groups in previous waves The difference in mental health after controlling for other variables are attributed to the impact of anticipating retire-ment

102 Ashenfelter (1978) noted a potentially serious limitation of this procedure when he observed that the mean earnings of participants

in government training programmes decline in the period prior to pro-gramme entry Subsequent research finds this regularity, sometimes called ‘Ashenfelter’s dip’ or the ‘pre-programme dip’, for participants in many other training and adult education programmes (see Ashenfelter and Card, 1985, Bassi, 1983, 1984, and the comprehensive survey by Heckman et al., 1999)

to do in the pre-vious month, concentration: has difficulties when it comes to concentrating; enjoyment and tearfulness: whether they have cried at all in the last month If the answer for one of those questions is

“yes”, it is coded as “1”, vice versa, “no” is coded as “0” The scores were summed ultimately based on the answers of each person The EURO-D score ranges from 0, which represents not depressed, to 12 very depressed

3.3.2 Retirement status

The central explanatory variable used in this paper is retirement status, which takes the value “1” for retirees and “0” for other people However, according to Heller-Sahlgren (2017), there are three different ways to define retirement status First, the “other people” could be employed, unemployed people, homemakers as well as those who are permanently ill or disabled The second retirement definition, meanwhile, takes homemakers, permanently ill or disabled persons into retirees, as long as they report not doing any paid work for the last month Thirdly, the sample simply includes people who admit being retired or employed Our study follows Butterworth et al (2006) and recodes people who were employed/self-employed or unemployed as “not retired” This way of defining retiring is also argued to

be used internationally (Butterworth et al., 2006)

3.3.3 Reasons of retirement

There are ten reasons of retirement in SHARE data The interviewees were asked whether they got retired to (1) became eligible for public pension, (2) became eligible for private occupational pension, (3) became eligible for a private pension, or (4) be-cause of an early retirement offer with special incentives, (5) a redundancy, (6) own ill health, (7) ill health of relative(s)/friend(s), (8) to retire at same time as spouse

or part-ner, (9) to spend more time with family or finally, (10) to enjoy life However, we did not exploit all

of the reasons individually, because there seem to be similar motivations in some of those Instead, we follow Robinson et al (2010), who uses three subscales to measure reasons of retirement, and divides the ten reasons into three groups Reasons 8, 9 and 10 are categorized into “Aspirational Motivations”, while reasons 1 to 4 are in the group of “Positive Circumstances” The last group includes the rest of the reasons and is under the name of “Negative Circumstances”

3.3.4 Instrument variables

According to Zhu (2016), instrument variables should meet two conditions, which is being related

to the explanatory variable, and orthogonal to exogeneity condition Heller-Sahlgren (2017), whose research similarly investigates the impact of retirement on mental health, also shares this point of view on instrument variables In other words, retirement is the only “channel” that the instrument variable chosen in this paper could alter the outcomes To be more specific, we follow previous studies that employed comparable models, such as those of Belloni et al (2016), Coe and Zamarro (2011),

Next, health background combines numbers of chronic diseases, visiting hospital and BMI status variables Chronic diseases in SHARE data is coded as a count variable which ranges from 0 to 13, presenting the number of illnesses one has suffered during their life These diagnoses consist of heart diseases, hypertension, high blood choles-terol, stroke, diabetes, asthma, arthritis, osteoporosis, cancer, stomach ulcer and general disability Following the research of Aichberger et al (2010), individuals who reported less than two diseases were recoded as “0”, and “1” otherwise to calculate the effect of multi-morbidity within one person Other than that, BMI in our study is recoded from 1 to 3, for normal, over-weight and obese BMI respectively According to Alavinia et al (2008), BMI below 25 is counted as normal, BMI from 25 to 30 is overweight and that above 30 is considered obese

In the final group of control variables, we include the number of limitations in Ac-tivities of Daily Living (ADLs) and Instrumental Activities of Daily Living (iADLs), which aim to measure functional impairment (Aichberger et al., 2010) Respondents were asked to report if they have any difficulty with six of ADLs and seven of iADLs because of a physical mental, emotional or memory problem The six ADLs consist of dressing (including putting on shoes and socks), walking across a room, bathing or showering, eating, such as cutting up food, getting in or out of bed, using the toilet (including getting up or down) While the list of iADLs contains such activities like using a map to figure out how to get around in a strange place, preparing a hot meal, shopping for groceries, making telephone calls, taking medications, doing work around the house or garden, managing money (eg., paying bills and keeping track of expenses) Furthermore, we also add variables which represent the frequency of doing vigorous activities such as sports or heavy housework, and activities that require

a moderate level of energy, for instance, gardening, cleaning a car or walking

512

Table 3: Summary of retirement transition pattern

Table 4: Summary Statistics of main variables

4 Descriptive statistics

Table 3 shows the transition of retirement through waves For further purpose of analyzing, which is to balance the time spent in retirement to such an extent, retirees in wave 1 was dropped out of the sample, since it contains people who had already retired long before the interview As can be seen from the table, the number of retirees have been putting up after waves For example, from wave 1 to wave 2, there were only 274 individuals got retired, then senior citizens in the sample of SHARE kept “joining the team” until the number of retirees in wave 6 was 1,222 people It is also noteworthy that the total figure is balance in all the waves with 1,715 individuals

Next, the summary statistics of main variables divided by retirement status are illus-trated in table 4 It can be described that the average depression level, which is indicated by EUROD, is slightly higher in the group of non-retirees by 10% However, the av-erage chronic diseases of retired people is reported to be more than that of people who remained in the labor force The Body Mass Index of the two groups are 26.1 and 26.6, which could be categorized as “normal” according to Alavinia et al (2008) Addition-ally, non-retirees seem to spend more time in schools (12.9 years vs 12.5 years), and retirees is likely to have more limitations in daily activities Other than that, further descriptive statistics for country, hospital, BMI status and the frequency of vigorous activities divided by the two groups are reported in table 5 in proportion

513

Retirement Status

5 Results

The results of regression models are presented in table ?? As can be seen, the first two columns indicate results of fixed-effect models with retirement status and dummies of retired reasons as explanatory variables respectively Firstly, it is shown from the ta-ble that retirement seems to make people feel less depressed by 0.12 point The similar result has also been found in several studies, including ones that exploit the same data source (Belloni et al., 2016) and others (Charles et al., 2004; Mein et al., 2003; Mojon-Azzi et al., 2007) When reasons for retirement are considered, more specific results are interpreted in the table Firstly, people who retire by aspirational motivations (e.g to enjoy life, to retire as the same time with spouse, or to spend more