Introduction

Most developed countries have recently passed legislation to increase retirement ages, in order to ensure the financial sustainability of social security systems. However, whether delaying retirement would improve the sustainability of health and social security programmes is still a matter of debate, given the potentially negative impact of such a policy on the health of the population. It may be that workers’ health, especially for those who have been in strenuous occupations, deteriorates both physically and mentally, generating increases in health care expenditure; in this case, retirement may reduce the amount of work-related stress and strain and provides individuals with more leisure time that can be used to invest in their health (e.g., physical activity). For example, Gorry et al. [1] claim that policies increasing eligibility ages may have hidden costs due to a negative impact on individuals’ health whose costs already represent a financial burden for public health care programs. If instead people are engaged in fulfilling jobs, within a smart and health promoting workplace, work may be a better guarantee of preserving individuals’ health than retirement: in fact, workers have incentives to invest in their health in order to maintain their income. Likewise, retirement might negatively affect health when it leads to social isolation and a diminished sense of purpose [2]. Therefore, under these different perspectives, increasing retirement ages may have additional benefits besides reducing the cost of pensions. As we will show in next section, the literature has tried to distinguish empirically between the two scenarios but findings vary widely depending on different methodological choices.

There is evidence about the importance of health behaviours such as not smoking, moderate alcohol consumption and physical activity, as well as weight control, to reduce mortality and improve functional capacity, among middle-aged and elderly adults [3–5]. Promoting healthy lifestyles has therefore been one of the policy strategies that international organisations and national governments have pursued to influence individual behaviours. Examples of such policies are information campaigns about risk factors, health education and ad hoc incentives through taxation, regulations (e.g., labelling rules or smoking bans) or nudging [6, 7]. These interventions are targeted mainly at younger generations, who are considered to be less aware of health risks [8]. However, although elderly people may be better informed, they are less prone to change their lifestyle; they have had more time to develop habits and may be particularly set in their ways (see [9], with regard to food expenditure, for instance), suggesting that such policies will have less effect on them than on younger individuals.

According to [10], nevertheless, large behavioural changes may occur after retirement, which is almost always a remarkable life event, as a consequence of changes in terms of time discounting, incomes or beliefs about the future. For this reason, we focus on the role of retirement in shaping lifestyles in later life. By examining behavioural adjustments upon retirement, rather than health outcomes, we can shed more light on the mechanisms that could explain previous mixed findings on the impact of retirement on health. We will analyse smoking, alcohol consumption and low engagement in physical activity, which are three modifiable risk factors contributing to more than a quarter of the disease burden in developed countries, according to the World Health Organization [11].Footnote 1 We will also estimate the causal effect of retirement on health care utilization as measured by visits to a general practitioner and consultations with a specialist during the last 12 months.Footnote 2

Given this background, we attempt to answer the following questions. Do individuals change their lifestyle upon retirement? Who are those more likely to invest in their health by pursuing healthy behaviours after retirement? The latter information can be useful for targeting purposes when designing policies relating to people in later life.

Our paper makes two new contributions to the empirical literature on the effects of retirement on individuals’ health behaviour.

First, we analyse retirement and health behaviours in Europe within a multi-country framework. We therefore do not focus on one specific country as other studies have done, but we analyse changes in health behaviours using harmonised individual panel data drawn from the Survey of Health Ageing and Retirement in Europe (SHARE), a survey that offers the possibility of comparing several European countries using nationally representative samples of the population aged 50+. Our identification strategy therefore relies not only on gender and year of retirement differences in eligibility criteria but also exploits the heterogeneity among countries. Furthermore, this multi-country framework allows us to investigate the role of different institutional settings (i.e. different types of health care system) on post-retirement health behaviours.

Second, we investigate heterogeneity in retirement effects, exploiting very detailed objective and subjective individual information, especially about job characteristics and early-life conditions, never considered before in the literature. In this way, we are able to highlight some underlying mechanisms that may explain individuals’ health investments upon retirement.

Our baseline estimates show that the probability of being inactive or not doing any vigorous physical activity decreases after retirement. We then provide evidence about individual heterogeneous effects on health behaviours upon retirement related especially to early-life conditions, education (we find stronger effects for individuals with higher education) and job characteristics, underlining the importance of the relief from work-related strain and time constraints as a barrier to engaging in regular physical activity.

These findings provide important information for the design of policies aiming to promote healthy lifestyles in later life, by identifying those who are potential target individuals, and which factors may affect their behaviour. Our results also suggest that the retirement and pre-retirement period may well offer a suitable opportunity to provide support for adopting a healthy lifestyle later in life. However, current policies, concerned mainly with the sustainability of social security systems, are progressively increasing retirement eligibility ages. This stresses the importance of policies promoting healthy lifestyles well before the end of the working life in order to anticipate the benefits deriving from individuals’ health investments.

The paper is organised as follows. The next section presents a Literature review, followed by a section on Data and some descriptive statistics. The Empirical strategy is then described, followed by Results and Conclusions.

Literature review

In recent decades, the economic literature has investigated the relationship between health and retirement, but findings have been ambiguous, for various reasons. Some authors found, on the basis of physical or mental health indicators, that retirement helps to preserve good health (e.g., [13–19], while others estimated a negative or nil effect of retirement on health (e.g., [20–23]).Footnote 3 Mixed findings can be explained by different outcomes or empirical strategies used, as well as by the existence of several competing channels, such as lifestyles and access to health care, through which retirement affects health.Footnote 4

In particular, according to Dave et al. [20] and Behncke [22], on the one hand, retirement could have a negative impact on health because of a decrease in work-related physical exercise, loss of ambition, or lower engagement in social or intellectual activities, accelerating the decline in health due to ageing. On the other hand, retirement provides individuals with less job-related stress and more leisure time; in addition, retirement may even increase investment in health since the retired have a lower marginal value of time, reducing the cost of health investment. For example, Bound and Waidmann [14], drawing on the standard Grossman’s model of demand for health [25], highlight that, since non-work time increases after retirement, we would expect that individuals spend more time investing in their health, especially in activities that are time-intensive (e.g., time spent in health-promoting behaviours). As the authors point out, because of different job characteristics, these effects vary from one individual to another: some may experience positive effects, others negative or no effects of retirement on health.

Understanding the effect of retirement on individuals’ health is quite important in order to fully assess the welfare and budgetary consequences of policies that increase retirement ages. Such policies might reduce retirement benefits and increase tax revenue through longer working lives, enhancing the financial sustainability of social security systems (as shown by [20]). Conversely, according to other studies, the same policies can produce indirect second-order effects in terms of health care utilization and related costs depending on their impact on individuals’ health [1 and 17]. However, analysing the health consequences of retirement is not an easy task because the retirement decision is endogenous. For example, several studies have shown that people who experience negative shocks to health disproportionately select into retirement (e.g. [26]).

In this paper, we focus on health behaviours rather than health since lifestyles may play a key role in explaining health upon retirement. Some studies [27–29] have investigated behavioural changes in later life but consider retirement as exogenous. However, endogeneity issues have to be considered also when analysing the relationship between retirement and health behaviour.

To our knowledge, there are four studies that have specifically considered retirement and health behaviours accounting for an endogeneity bias. Looking at US data, drawn from the Health and Retirement Study (HRS), Insler [17] used an instrumental variables strategy based on individuals’ predicted probability of working past ages 62 years and 65 years reported in the period in which they entered the sample, and found that retirement positively affects health through a reduction in smoking and an increase in exercise. Using the same dataset, Kämpfen and Maurer [30] provide instrumental variables estimates based on early and statutory retirement ages, showing that, when individuals retire, they increase physical exercise, meeting the federal government’s 2008 Physical Activity Guidelines. Within a regression discontinuity framework, Eibich [31] found that, in Germany, retirement affects negatively smoking and outpatient care utilization, positively sleep duration, engagement in activities and alcohol consumption. Zhao et al. [32] used data from the Health and Retirement Survey, a longitudinal study conducted by the National Institute of Population and Social Security (IPSS) in Japan to show that, on retirement, individuals significantly reduce their level of smoking and are more likely to exercise.Footnote 5

We contribute to this literature in two ways. First, we analyse retirement and health behaviours in Europe within a multi-country framework. Second, we investigate in greater detail the heterogeneous effects of retirement on health behaviours linked to individuals’ characteristics, considering also objective and subjective information about job characteristics and early-life conditions.Footnote 6

Data

We use data drawn from SHARE, a multi-disciplinary survey that collects information on individuals aged 50 or over, plus their partner, regardless of age. The first wave of SHARE took place in 2004/2005 and involved eleven European countries. Other countries have been added in the following waves but in this paper we select only those that participated in all SHARE regular waves from 2004 to 2012––the first (2004/2005), second (2006/2007) and fourth (2011/2012) wave––to exploit the longitudinal dimension of the survey: Austria, Belgium, Denmark, France, Germany, Italy, the Netherlands, Spain, Sweden and Switzerland.Footnote 7 The third wave, called SHARELIFE, collects retrospective information, e.g. early-life conditions, that we will use to investigate heterogeneous effects related to retirement. We select individuals who self-report being retired from work or employed/self-employed and whose age is between 45 years and 85 years,Footnote 8 with no missing information about employment status, gender, education, age, marital status, number of grandchildren and health behaviours defined according to three dimensions: smoking, physical inactivity and alcohol consumption.

Smoking is a dummy variable that acquires value 1 if the individual currently smokes, and 0 otherwise. Engagement in activities is captured by two dummies: No activities, which takes value 1 if the person reports never or almost never practising any activity requiring either a moderate or substantial level of energy; No vigorous activities, which equals 1 if the respondent reports never or almost never taking part in sports or vigorous activities; this distinction can be suggestive of physical exercise intensity. Regarding alcohol consumption, since the questions have been changed over time, we are able to exploit only information about drinking frequency for all waves; we therefore define a variable Drink every day, which takes value 1 if the person reports drinking alcohol almost every day.Footnote 9

We also consider two measures of health care use: the Number of visits to the general practitioner and a 0–1 dummy for having consulted a specialist in the last 12 months (Visits to the specialist).

A key variable in our analysis is retirement. We define as retired those individuals who self-declare to be retired from work. Retirement is considered an absorbing state: no transitions from retirement back to work are therefore observed. Since some respondents may report being retired simply because they left their main job, even though they are still working, we also use a narrower definition of retirement which combines self-reported employment status and information about paid work during the last four weeks before the interview (see Appendix 3, Table 6).Footnote 10

Table 1 presents the summary statistics of health behaviour variables, socio-economic and demographic covariates.

Table 1 Summary statistics
Full size table

In Figs. 1, 2, 3, 4, 5 and 6, we illustrate the relationship between health behaviours and age, distinguishing between higher (tertiary) and lower (secondary or primary) education levels,Footnote 11 pooling data from wave 1 to wave 4.

Fig. 1
figure 1

Proportion of smokers, by age and education level

Full size image
Fig. 2
figure 2

Proportion of individuals not practising any activity, by age and education level

Full size image
Fig. 3
figure 3

Proportion of individuals not practising any vigorous activity, by age and education level

Full size image
Fig. 4
figure 4

Proportion of individuals drinking every day, by age and education level

Full size image
Fig. 5
figure 5

Number of visits to the general practitioner, by age and education level

Full size image
Fig. 6
figure 6

Proportion of individuals having had consultations with specialists, by age and education level

Full size image

Figure 1 shows the (unweighted sample) proportion of smokers by age for individuals with higher and lower education respectively: among the latter, we can see a general negative association between smoking and age (possibly due to selective mortality, as argued in [35], but no marked changes can be noticed around typical retirement ages (e.g., 65 years).

Figures 2 and 3 show the proportion of inactive individuals, i.e. those who do not practise any activity (Fig. 2) or any vigorous activity (Fig. 3), by age. The two graphs highlight a positive association with age, but it is notable that, among highly educated individuals, there is a spike in the proportion of inactive people at age 56 years when looking at activities requiring a moderate level of energy, and a decrease at age 65 in terms of vigorous activities. Among lower educated individuals, the proportion of inactive people increases at 55 years when considering vigorous activities. Figure 4 shows the proportion of individuals, by age, who drink alcohol almost every day, revealing a slight increase after age 60 years for both highly educated and less well educated people. Figures 5 and 6 show the average number of visits to the general practitioner, and the proportion of individuals who have had at least one consultation with specialists in the last year, by age and education level: significant increases in the average number of visits to the general practitioner are seen after age 68 years, for both highly educated and less well educated individuals, and the figure for those who have had consultations with a specialist increases significantly after the age of 70 years for less well educated individuals.Footnote 12

The figures provide a first descriptive evidence of possible changes in health behaviours around retirement age. In the next section, we will explain the empirical strategy used to identify the causal effect of retirement on health behaviours.

Empirical strategy

The effect of retirement on health behaviours

This study aims to discover whether individuals change their health behaviours upon retirement. To this end, we propose the following specification:

$$y_{it} = \, \alpha_{1} {\text{retired}}_{it} + \, X_{it} \beta \, + \, u_{it}$$
(1)
$$u_{it} = \, \mu_{i} + \, \varepsilon_{it} ,$$
(2)

where y it is the outcome of interest (i.e., the health behaviour variable), X it is a vector of individual characteristics (e.g., age, gender, marital status, educational level, etc.); the error term u it can be decomposed into unobserved time-invariant heterogeneity (µ i) and an idiosyncratic error term (ε it). We are interested in α 1, the coefficient associated with retired. Standard ordinary least squares (OLS) estimates of α 1 yield unbiased results if the orthogonality condition is satisfied (i.e. retirement should not be correlated with the error term); however, this is unlikely to hold. As pointed out in the literature (e.g., [13–16]), when assessing the role of retirement on health, endogeneity issues have to be taken into account. The same applies to health behaviours, since retirement is a choice that individuals make for several unobservable reasons that could also affect lifestyles. To control for observed and unobserved time-invariant individual heterogeneity, we estimate individual fixed-effects (FE) panel data models.Footnote 13

Using FE models allows us to account for observable characteristics (such as gender, country, birth cohort and educational attainment) that do not vary over time and may be important sources of bias,Footnote 14 as well as for unobserved time-invariant factors that could confound our estimates. However, controlling for time-invariant characteristics is not enough to permit causal interpretations, since we need to account also for time-varying individual unobserved factors and reverse causality: health behaviours, also through their interaction with health conditions, may induce retirement. We overcome this problem by adopting an instrumental variable (IV) approach. We exploit the information about changes in eligibility rules for early retirement and old-age pension across several European countries and over time as instruments for retirement (see Appendix 1 for a detailed description).Footnote 15 Using changes in pension eligibility rules as instruments for retirement is a widespread methodological choice in the literature (see, for instance, [37, 38] and [16]).Footnote 16

We run FE two-stage least squares (FE-2SLS), our preferred specification, to estimate the effect of retirement on health behaviours; however, for completeness we report also OLS, FE and pooled two-stage least squares (2SLS) estimates. In the FE-2SLS specification, since we exploit the within-individual variability, to be able to identify the effect of retirement, we need a sufficient number of respondents who switch from employment to retirement. In our sample, we have 1999 transitions into retirement.Footnote 17

The relevance of our instruments can be tested directly by looking at F-statistics for the excluded instruments ([39] and critical values for weak identification [40]; see the section Results).Footnote 18 The validity assumption, which requires that the instruments affect health behaviours only through retirement (and can be therefore excluded from the structural equation) is supported by the fact that changes in eligibility rules arguably represent a source of exogenous variability in social security regulations that are unlikely to have a direct effect on our outcomes.

Thus, based on retirement eligibility criteria among countries, over time and between genders, we define as instruments two zero–one dummies indicating whether the individual is eligible or not either for early (EligibleER) or statutory (normal) retirement (EligibleSR), respectively.

For binary outcomes, we specify a linear probability modelFootnote 19 where we control for marital status (having a partner), education, age, age squared, household net wealth quartile dummiesFootnote 20 and the number of grandchildren (to account for grandparenting effects). The same set of covariates is used when looking at the continuous variable Number of visits to the general practitioner.

Coe and Zamarro [16] and Zamarro et al. [43], looking at the effect of retirement on health using SHARE data, noticed that panel attrition may be a problem, because people in poor health due to unhealthy behaviours are more likely to exit the panel, and this may lead to invalid inference. We have performed a robustness analysis (see Appendix 4) following [44], showing that attrition is not an issue in our case.

Heterogeneous effects

We investigated in greater detail heterogeneity in retirement effects related to gender, education, early-life conditions, household net wealth and job features. To this end, we estimated our models separately for males and females, highly (Isced5_6) and less well (Isced0_4) educated individuals. The sample was also split according to an indicator of early-life conditions, Few books, representing the presence of fewer than 25 books at the parental home at age ten; this information, collected in SHARELIFE, can be considered a proxy for parental education and economic status during childhood.Footnote 21 We consider heterogeneity related to wealth by providing estimates for individuals having household net wealth below or above a country-specific yearly median value. Finally, to understand whether job characteristics play a crucial role in explaining how individuals change their behaviours upon retirement, we exploit work quality and job information collected in SHARELIFE and regular waves (first, second and fourth).

Retirement may indeed be beneficial for those working in physically demanding and stressful occupations, based on the evidence that working in manual jobs negatively affects health (see for instance [46]) and may induce people to adopt unhealthy behaviours such as smoking. In SHARE, a battery of work quality questions is asked, differing between SHARELIFE and regular waves. In order to make use of comparable information available in all waves, we take account of two specific questions related to strenuousness and time pressure. Work quality indicators are related to the main job for retired individuals, and to the last job for those still working.Footnote 22 Respondents are asked whether the job was/is physically demanding and whether it exerted/exerts heavy time pressure.Footnote 23 Based on the answers, we consider separately those individuals who agree (or strongly agree) with the statement and those who disagree (or strongly disagree). To support the evidence based on self-reported job characteristics, which may suffer from differences in reporting style (see for instance [48, 49]) or justification bias, we classify individuals as either blue/white collar or low/high skilled workers,Footnote 24 using job descriptions provided by the respondent. The related question in the SHARE questionnaire is able to capture mainly the first digit of the International Standard Classification of Occupations (ISCO-88 code).Footnote 25

Results

In Table 2, we report pooled 2SLS and fixed-effect 2SLS estimates (our preferred specification) for each health behaviour considered as an outcome; for comparison, we also report pooled OLS and fixed effects specifications. The estimated standard errors are robust to clustering at the country and cohort level.

Table 2 The effect of retirement on health behaviours
Full size table

Table 2 column 1 (OLS estimates) represents only a partial (not significant) association between retirement and smoking. Column 2 (FE estimates) shows that, when we account for time-invariant heterogeneity, transiting into retirement is associated with a higher probability of quitting smoking. Columns 3 and 4 report 2SLS and FE-2SLS estimates respectively: when we account for the endogeneity of retirement, we find no statistically significant effects on the probability of smoking. In Table 2, we also report selected first-stage coefficients, showing the relevance and strength of our instruments: the coefficients of being eligible for early and statutory retirement are always highly significant (at the 1 % level) and the F-statisticsFootnote 26 on the excluded instruments are well above ten [39], and the critical values for weak identification testing [40]. As in previous studies [52, 37, 38, 16], our results therefore confirm that eligibility rules are important determinants of retirement decisions.

With regard to engagement in activities (Table 2, columns 5–8), we find a significant effect in the pooled OLS regression (column 5), where retirement is associated with a reduction in the probability of being inactive, while no significant effects are estimated in the fixed-effect model (column 6). Columns 7 and 8 of Table 2 show that, accounting for endogeneity, retirement causes a highly significant reduction in the probability of being inactive.

In columns 9–12 of Table 2, we focus on the effect of retirement on sports and vigorous activities. 2SLS estimates show that retirement causes a reduction in the probability of being inactive, in line with what we have seen when looking at activities requiring a moderate level of energy.Footnote 27

We stress that the identification strategy, with regard to FE-2SLS estimates, relies on those individuals who switch between waves from employed or self-employed to retired; therefore, we are able to estimate a short- (or medium-) rather than long-term effect of retirement on health behaviours.Footnote 28

Table 2, columns 13–16 report estimates for the probability of consuming alcohol every day. OLS and FE estimates are confirmed by FE-2SLS results: the transition into retirement causes changes in drinking behaviour, in line with the literature. Eibich’s [31], for instance, find that, in Germany, retirement causes a statistically significant increase in the probability of regular drinking and a reduction in the probability of no alcohol consumption.

In columns 17–20 of Table 2, we focus on the number of visits to a general practitioner in the last twelve months. Retirement is associated with a higher number of visits in the OLS specification, but no significant causal effects are estimated by 2SLS.

The last four columns of Table 2 show that retirement is associated with a higher probability of having contact with a specialist in the last twelve months (column 21), but the causal effect is confirmed only in the pooled 2SLS specification (column 23), while no significant effects are estimated when exploiting the within-individual variability in the data with FE-2SLS (column 24).

In Appendix 3, Table 6, we report additional robustness analysis for our 2SLS estimates. The baseline results do not change: whether we include among controls the number of chronic diseases and limitations in the basic and instrumental activities of daily living (ADLs and IADLs);Footnote 29 if we allow the non-linear age effect to be country-specific; if we exclude older individuals, aged over 75 years; if we use an alternative definition of retirement, considering as retired those individuals that not only self-report being retired but also did not do any paid work in the four weeks before the interview; or if we include also the number of children as control. We gain only a marginal significance in the FE-2SLS for the probability of smoking when accounting for country-specific non-linear age effects.

The estimates shown so far are based on pooled data from the selected ten European countries.Footnote 30 We also run FE-2SLS estimates grouping countries according to the existence of a gate-keeping system to access specialist health care services: countries with general practitioners acting as gate-keepers (Denmark, Italy, the Netherlands, Spain and Sweden), and countries without a gate-keeping system (Austria, Belgium, France, Germany and Switzerland). The aim is to investigate whether there are group-specific significant differences from our baseline results possibly related to different institutional frameworks. The estimates in Table 3 suggest the existence of differential retirement effects on some health behaviours that could be linked to the type of health care system.

Table 3 The effect of retirement on health behaviours––FE-2SLS––by type of health care system
Full size table

First, in countries with gate-keeping, individuals are significantly less likely to be inactive after retirement, and this effect still remains once Mediterranean countries (Italy and Spain), characterised by higher rates of sedentariness, are considered separately from Denmark, the Netherlands and Sweden.Footnote 31 Gate-keeping systems require the authorisation of referrals to specialists by designated primary care providers, such as the general practitioners. In these systems the role of general practitioners in nudging healthier lifestyles (including increased physical activity)––in order to prevent diseases and the use of secondary health care services––is therefore emphasized, especially where there is an involvement of primary care physicians in chronic disease management.Footnote 32 In countries where general practitioners do not act as gate-keepers, individuals access directly specialist physicians who provide secondary health care treatments and may have lower incentive to promote healthy lifestyles through counselling.

Table 3 also shows that in countries with gate-keeping, individuals are more likely to drink regularly after retirement. However, as we will explain below, regular alcohol consumption cannot be simply interpreted as a signal of unhealthy behaviour.

Moreover, Table 3 suggests that different health care settings may also influence the access to outpatient care services after retirement: in countries without gate-keeping, retirees significantly reduce the number of visits to general practitioners.Footnote 33 This result, which is in line with Eibich’s [31] findings for Germany, may depend on the fact that, after retirement, the probability to be diagnosed a chronic disease increases, and therefore individuals are more prone to access directly specialists where a prior referral by the general practitioner is not required. However, this should be considered as a short-term effect, given the identification strategy we used.

The estimates in Table 3 suggest that different frameworks of European health care systems matter in shaping individuals’ health behaviours after retirement, even though the analysed effects might be short-lived. However, a complete analysis of actual determinants of these effects is rather complex and deserves further investigation.

In Table 4, we analyse heterogeneity in retirement effects by estimating the FE-2SLS model of Table 2 in subgroups defined according to gender, education, early-life condition, household net wealth and job characteristics.

Table 4 The effect of retirement on health behaviours—heterogeneous effects—FE-2SLS
Full size table

According to our estimates, heterogeneous retirement effects in smoking behaviour may be observed. In particular, we find a statistically significant (at 5 % level) negative effect for individuals classified as blue collar. For individuals with physically demanding jobs, a negative significant (at 10 % level) effect of transiting into retirement is estimated. These results are in line with those of Eibich’s [31], who looked at behavioural differences related to occupational strain.

The transition into retirement causes a significant reduction in the probability of being inactive among individuals of both sexes, with a partner, with high parental socio-economic status during childhood (“no few books”), whose job entailed time pressure, or who has been classified as white collar or highly skilled. In addition, comparing the effect of retirement on the probability of being inactive between highly educated and less well educated individuals, we can see that the point estimate for the former is larger. Table 4 shows also that retirement has a negative and significant effect on the probability of never, or almost never, practising vigorous activities among individuals of both sexes who have a partner, those with high parental socio-economic status during childhood, whose net wealth is above median, whose job was not physically demanding or was classified as white collar/highly skilled. These results are in line with Eibich’s [31] findings for Germany, with the conclusions of the systematic review conducted by Barnett et al. [58] and with some descriptive evidence [59, 45] about the role of job characteristics in determining heterogeneity of the retirement effect.Footnote 34

A significant increase in drinking behaviour (at the 5 % level) due to retirement is estimated only for male individuals, those without a partner, those with low parental socio-economic status during childhood, or whose job entailed time pressure; transiting into retirement has a significant positive effect (at the 10 % level) on the probability of drinking every day for individuals whose net wealth is below median. While smoking and inactivity are undoubtedly unhealthy behaviours, changes in alcohol drinking habits, captured by our binary indicator, cannot be clearly evaluated, since we do not have an indicator of drinking intensity for all waves. However, our result can be suggestive of a potential vulnerable sector of the population. Although previous studies suggest that regular alcohol consumption does not necessarily have a negative effect on health [61, 31], the alcohol-related burden of disease among older age groups, owing to their lower ability to handle the same levels and patterns of alcohol consumption they had had in their younger days, is an increasing public health concern [62].

Regarding health care use, we find a significant increase (at the 10 % level) on the probability of having a specialist visit only for male retirees and for those with a partner.

In general, the analysis of heterogeneity in retirement effects highlights a systematic socio-economic gradient across different dimensions, and the protective role of partnership.

So far, the heterogeneity analysis suggests, among other things, the role of a reduced occupational strain to explain part of the behavioural change due to retirement. It is, however, true also that non-work time increases after retirement, so that time constraints are no more a major barrier to time-intensive activities, such regular physical exercise. To investigate the role of time constraints in explaining the effect of retirement on physical activity, we estimate our FE-2SLS model for subgroups of individuals working/having worked always full-time or not (Table 5 based on SHARELIFE information).

Table 5 The effect of retirement on the probability of being inactive––heterogeneous effects––FE-2SLS
Full size table

The results show that significant effects (in terms of increased activity) are estimated only for the subgroup of individuals having worked/still working full-time, supporting the mechanism through which retirement provides individuals with more leisure time that can be devoted to physical exercise.

Conclusions

In this paper, we focussed on behavioural adjustments upon retirement, to shed more light on the mechanisms that could explain previous mixed findings about the impact of retirement on health.

Accounting for the endogenous choice of retirement, we were able to estimate the causal effect of retirement on smoking, drinking behaviour, engagement in activities and contacts with doctors (general practitioners and specialists).

Our baseline estimates show that the probability of being inactive or not doing any vigorous physical activity decreases with retirement: individuals provided with more leisure time change their behaviour in terms of engagement in activities; this corresponds to the so-called honeymoon phase [63, 64]. Our findings therefore underline the importance of time constraints as a major barrier to engaging in regular physical activity. Our estimates, moreover, show a significant effect of retirement on the probability of regular alcohol drinking, confirming other empirical results [31], even though this does not necessarily imply a worsening in health behaviours.

We also observe the existence of differential retirement effects by grouping countries according to the type of health care system. In particular, we find that in countries with a gate-keeping system people are significantly less likely to be inactive after retirement. This effect might suggest that the health care systems configuration plays a role in determining individuals’ health investments upon retirement, although further investigation is needed.

We also provide another innovative contribution to the literature by looking at individual heterogeneous effects of retirement not only linked to gender, education, and net wealth (as other studies have done) but also related to a larger set of objective and subjective individual information about early-life conditions and job characteristics. In particular, we find larger effects for higher educated people and for those with high parental socio-economic status during childhood, who are more likely to change lifestyles after retirement, increasing their physical activity. This is in line with the so-called ‘education gradient’ [65, 66], in which health behaviours can be seen as mediating factors through which education influences health [67]. Job characteristics also play a role in relation to physical exercise: individuals who have been classified as white collar or highly skilled increase significantly the probability of engagement in physical activities (both moderate and vigorous); those whose job entailed time pressure reduce significantly the probability of being inactive, while retirement from less physically demanding occupations increases the probability of engagement in sports or vigorous activities. We highlight also the role of time constraints as barrier to engage in regular physical activity.

Our results provide important information for the design of policies aiming to promote healthy lifestyles in later life, by identifying those who are potential target individuals and which factors may affect their behaviour. According to our study, poorly educated individuals show smaller effects regarding engagement in activities after retirement. This provides support for active ageing policies, particularly in the field of participation for that group of the population (e.g. adapted physical activity programmes responsive to older adults’ educational levels and cultural preferences; see [68–70]).

Our results also suggest that the retirement and pre-retirement period may well offer a suitable opportunity to provide support for adopting a healthy lifestyle later in life. In this respect, our findings are in line with certain general policy proposals put forward by the World Health Organization (WHO; [71]) about active ageing: ‘Provide education and learning opportunities throughout the life course; and recognize and enable the active participation of people in economic development activities, formal and informal work and voluntary activities as they age, according to their individual needs, preferences and capacities.’ Regarding physical activity, the WHO [71] suggests the importance of supporting culturally appropriate community programmes that stimulate activity, and are organised and led by older people themselves. However, evidence that strenuous physical work may hasten disabilities, preventing physical exercise, additionally requires health promotion efforts already at work aimed at providing relief from repetitive, strenuous tasks, and making adjustments to avoid unsafe physical movement.