Unemployment affects future working conditions and job security negatively, thus reducing life satisfaction after reemployment. These employment-related scars of unemployment should not matter anymore when a person has retired. Using German panel data, we analyze unemployed persons’ transition into retirement to test whether unemployment leaves scars beyond working life and thus for reasons that are not employment-related. We find that involuntary unemployment between the last job and retirement causes a loss in life satisfaction after retirement. People who influenced or even initiated unemployment, by contrast, show no scarring. The scarring effect goes beyond what can be explained by the income loss originating from reduced pensions. It shows up independently of whether the unemployment spell directly before retirement was the only experience of unemployment in a person’s career, or whether she had also experienced unemployment at earlier times. We do not find evidence that early retirement or involuntary retirement are the reasons why formerly unemployed retirees display unemployment scarring.
Unemployment, Scarring, and Well-being
Involuntary unemployment affects people’s lives in several ways. The immediate loss of income, alleviated by the gain in leisure time, only partly explains the effect of unemployment on subjective well-being (Blanchflower and Oswald 2004; Winkelmann and Winkelmann 1998). Deviating from the social norm to work, which negatively affects social identity, explains why the pain of unemployment goes far beyond that caused by the loss of income (Clark 2003; Hetschko et al. 2014; Schöb 2013; Stutzer and Lalive 2004; Van Hoorn and Maseland 2013). An important issue in comprehensively assessing the cost of unemployment is whether these effects vanish when people overcome joblessness, or whether they persist in reemployment or retirement. In short, does unemployment leave permanent scars?
Unemployment might be scarring for several reasons. A job loss diminishes future labor market prospects and thereby still affects unemployed people negatively when they are reemployed. Joblessness causes a loss of human capital (Schwerdt et al. 2010) and sends worsened productivity signals (Kroft et al. 2013). A new job is therefore often characterized by higher job insecurity and thus higher income volatility, lower wages, and worse working conditions than the previous job (e.g., Arulampalam et al. 2001; Brand 2006; Dieckhoff 2011; Eliason and Storrie 2006). These employment-related scars can explain why life satisfaction does not fully recover after reemployment (Clark et al. 2001; Young 2012). Even decades later, early career unemployment seems to have a detrimental effect on well-being (Bell and Blanchflower 2011; Daly and Delaney 2013), and each further job loss lowers life satisfaction in subsequent employment and unemployment spells (a process referred to as “sensitization”; Luhmann and Eid 2009).
Besides employment-related scars, the negative well-being effect of past unemployment could stem from life domains other than working life. We provide novel evidence that unemployment leaves such non–employment-related scars. To separate employment-related from non–employment-related scars, we analyze transitions of employed and unemployed persons to retirement, using large-scale German panel data. Observing people before and after retiring is suited for identifying non–employment-related scars because retirees should not suffer from employment-related scars unless they intend to return to the workforce. In our sample, retirement entries are nearly always definite in the sense that retirees neither report intentions to return to work nor actually return to the labor market later on. The German public pension system enables retirees to continue living under material conditions similar to those before retirement (Börsch-Supan and Schnabel 1998). Their future income path is determined by their pensions and savings so that employment-related income volatility has vanished. Insecurity about future employment prospects, which is one pathway of employment-related unemployment scarring (Knabe and Rätzel 2011; Lange 2013), disappears after retirement.
Non–employment-related scars may persist after retirement. Some of them are of a monetary nature. People with previous unemployment experience save less and contribute less to the public pension system than continuously employed people and hence receive lower pensions after retirement and possess less wealth. They thus have fewer consumption opportunities and enjoy lower utility. Other non–employment-related scars are of a nonmonetary nature. People with previous unemployment experience may look back in anger when assessing their life. For example, having deviated from the social norm to work in the past may lead to a shaken self-image. Retirees with previous unemployment experience might also suffer from the fact that they would have had higher retirement incomes if they had not lost part of it because of events they could not influence (e.g., involuntary unemployment). In that case, dissatisfaction stems not from their actual consumption level but rather from the deviation from their aspired level, which might also be seen as a nonmonetary effect. To the extent that factors causing monetary scarring can be observed and controlled for, the transition to retirement can be exploited to identify the non–employment-related scar and to break it down into a monetary and a nonmonetary component.
Applying a difference-in-difference (DiD) approach, we follow workers from their final employment spell into retirement using data of the German Socio-Economic Panel study (SOEP). A treatment group consists of persons who experience involuntary unemployment between their final employment spell and retirement. We thereby place special emphasis on workers who lose their jobs because of a plant closure, given that this most likely represents an exogenous entry into unemployment. Persons in the control group are continuously employed before retiring. Our analyses indicate that unemployment on the eve of retirement leaves a scar on workers’ well-being that persists after retirement and that cannot be explained by changes in income. Unemployment seems to reduce well-being beyond working life for nonmonetary reasons.
Our results complement the aforementioned research on the well-being effects of unemployment. We also contribute to the understanding of long-term determinants of well-being over the life course (Frijters et al. 2014; Layard et al. 2014) and add to the literature on the well-being effects of retiring (e.g., Bender 2012; Calvo et al. 2009; Nikolova and Graham 2014).1 One of this literature’s central findings is that people who are forced to retire out of employment suffer from the transition, whereas voluntary retirement does not alter the well-being of employees (e.g., Bonsang and Klein 2012). In contrast, we address retirement out of unemployment.
We build on a previous analysis (Hetschko et al. 2014) in which we showed that unemployed persons’ life satisfaction increases when they retire (for supporting evidence, see also Ponomarenko et al. 2019; Wetzel et al. 2016). Because retiring from unemployment is not accompanied by substantial changes in material and time resources, the positive effect on life satisfaction is attributed to the change in social identity, which restores compliance with the social norm to work. When unemployed, people deviate from that norm, whereas not working is the norm for retirees. The finding suggests that norm deviance is responsible for a significant share of the detrimental effect of unemployment on life satisfaction. The present study addresses a different question. Whereas our previous work analyzed how life satisfaction changes when transitioning from unemployment to retirement, in this study, we examine how a last unemployment episode affects the life satisfaction of retirees relative to their life satisfaction when they were employed. In the process, we clarify whether retirement restores formerly unemployed workers’ life satisfaction fully, or whether it leaves scars.
As a byproduct of our previous analysis, we found that average life satisfaction between a last episode of employment and retirement does not change in a small subgroup of people who retire from unemployment (Hetschko et al. 2014:157–158). Here, we go far beyond this simple mean comparison. Taking account of the heterogeneity and endogeneity of unemployment, we find evidence for unemployment scarring if the job loss was involuntary. Moreover, we compare the evolution of life satisfaction for unemployed retirees with a control group of formerly employed retirees, thus accounting for a pure retirement effect on well-being. We consider previous unemployment experience, control for sociodemographic and job characteristics, and examine various points in time before unemployment and after retiring. Persons who had been unemployed for more than one year before retiring are now also included in the analysis.
Our analysis relies on 25 waves (1991–2015) of the SOEP, a representative survey of the population in Germany (Wagner et al. 2007).2 Each year, approximately 20,000 individuals from about 11,000 households are interviewed and provide information on their well-being, income, employment status, education, health, and so forth. The great advantage of the SOEP is its panel structure, which allows us to compare the living conditions and subjective well-being of the same person before and after unemployment and retirement.
During the period considered in our study, people in Germany could receive pensions when they reached the statutory retirement age3 and fulfilled additional conditions concerning pension contributions. Early retirement was possible at age 63, and if the person was female or unemployed, even at age 60. Monthly pensions were lowered by 0.3 % for every month a person retired before reaching the statutory retirement age. Those who retired because of unemployment (Altersrente wegen Arbeitslosigkeit; § 237 SGB VI) were eligible for pensions if they had been unemployed for at least 52 weeks since age 58.5 and had been insured for at least 15 years in the public pensions system (Mindestversicherungszeit). As of 1992, they additionally need to have contributed for 8 of the last 10 years before retirement (Pflichtbeitragszeit). Furthermore, the early retirement age for the unemployed is gradually increasing for people born after 1941 and converging with that of employees (Lühning 2006).
Retirees in Germany are allowed to earn labor income. For most of our investigation period, early retirement pensions were not reduced unless earnings exceeded a certain threshold; in 2015, this threshold was 450 euros per month, except for two months per year in which up to 900 euros could be earned. Beyond that, it was possible to keep two-thirds, one-half, or one-third of the pension if post-retirement monthly labor income did not exceed, respectively, 13 %, 19 %, or 25 % of the former monthly labor income. Unlike early retirees, people who have reached the statutory retirement age can fully keep their pension irrespectively of the amount of additional labor earnings. According to the German Federal Statistical Office, only 1 in 20 Germans aged 65–74 was employed in 2006. The share doubled from 2006 (5 %) to 2016 (11 %). Still, the majority of retirees in this group rely mainly on pensions (Destatis 2017), suggesting that many employed retirees work in part-time or casual jobs.
The SOEP requires people to indicate their current labor market status. People are regarded as unemployed if they state to be not employed and registered as unemployed. We consider only those people who indicated part-time employment and full-time employment as employed. Workfare participants and people who report to be in marginal or irregular employment4 are not considered as employed. Moreover, we regard people as employed only if they work at least three hours per weekday and if the employment spell lasts longer than one year.
To identify transitions to retirement, we use monthly self-reported information about a subject’s employment status over the course of the previous year. We exclude people who are neither employed nor unemployed in the month preceding the first month about which they report to be retired, are younger than 55 when retiring, or return to employment later on. Accordingly, people might work in addition to being mainly retired. With these restrictions, we identify 1,243 transitions from unemployment to retirement and 2,276 transitions from employment to retirement (basic sample). A vast majority (93 %) of these people already state directly after retirement that they are not going to return to work in the future. As a robustness check, we later exclude retirees who report positive working hours or state a willingness to return to the workforce.
Data Used and Sample Size
Subjective well-being is measured by the question, “How satisfied are you with your life, all things considered?” Answers can range from 0 to 10. To examine whether formerly unemployed retirees reach the level of well-being they reported before they lost their jobs, we track their life satisfaction back to the time when job loss took place. To this end, we need to observe these people continuously from their final spell of employment onward, while they are unemployed, and when they finally retire. In the case of retirement out of unemployment, we additionally restrict our sample to workers who are also observed in employment for at least two SOEP interviews before they are identified as unemployed right before retirement. To ensure comparability with formerly employed retirees, the employment before unemployment must have been a regular one. Three-fifths (60 %) of the transitions of unemployed workers into retirement cannot be considered further because they do not meet these requirements. Further observations are lost as we identify the reason for the termination of employment. We later focus on unintended job losses (first and foremost plant closure, but also other dismissals) to circumvent the endogeneity of other entries into unemployment (e.g., resignation, end of a fixed-term contract, or mutually agreed termination). Because these reasons were not ascertained consistently before 1991 as well as in 1998 and 1999, we do not take into account respondents who, in these years, reported that their job had ended since the last interview.
Information on lifetime unemployment experience is obtained from workers’ employment biographies. These biographies are constructed using (1) a one-time biographical questionnaire in which SOEP respondents, inter alia, reported their entire previous employment history, and (2) information from annual survey responses. Data on disposable household income are self-reported by the household heads; the SOEP provides imputed values for nonresponding households. We calculate equivalent income for each person by dividing their real net household income (at 2011 prices) by the weighted sum of household members using the modified OECD scale (1 for the first adult, 0.5 for every additional person aged 14 or older, and 0.3 for every person younger than 14).
We also use information about the presence of children and people in need of care in the household, as well as about age, sex, years of education, relationship status, and homeownership as a proxy for household wealth. Health is taken into account by overnight stays in hospital as well as disability status. Job characteristics include tenure and part-time employment. Data on job autonomy are generated based on workers’ concrete occupation, position, and responsibility in carrying out tasks. Because of the data requirements, our basic sample shrinks to 2,149 transitions into retirement: 1,838 people retire from employment, and 311 retire from unemployment. Table 1 describes the characteristics of the two groups, whereby formerly unemployed retirees are further distinguished by the reason for job loss (55 after a plant closure, 148 because of other kinds of employer-initiated dismissals, and 108 for other reasons). Particularly the characteristics of the last job differ between people who are continuously employed before retirement and those who retire from unemployment.
Scars From Unemployment at the End of Working Life
We follow workers on their late career path into retirement. A final period of unemployment on the eve of retirement is considered the treatment that might scar beyond working life. Its potential effect is estimated by a DiD approach. We compare the change in life satisfaction from the last employment spell to the first interview after retiring between workers who retire from unemployment and those who retire from employment. The control group of employees enables us to account for a potential pure retirement effect on life satisfaction, which could otherwise be misinterpreted as a scarring effect of unemployment. Because the DiD compares within-person changes of life satisfaction, we automatically control for time-invariant unobservable individual heterogeneity.
Figure 1 illustrates the average life satisfaction of workers who are either employed before retirement or who retire from involuntary unemployment. Life satisfaction substantially drops some time before workers lose their jobs because of anticipation of the imminent job loss and consequential feelings of job insecurity (Lüchinger et al. 2010). Figure 1 confirms the positive impact of retiring from any unemployment spell found by Hetschko et al. (2014) for the subgroup of unemployed people who were displaced. This subgroup does not reach the level of life satisfaction shown in the second-to-last year they were employed. Retiring employees’ life satisfaction remains comparatively constant throughout the whole transition process. Finally, Fig. 1 indicates a common trend in the life satisfaction of formerly unemployed and formerly employed retirees before the former lost their jobs.
The levels of life satisfaction differ between the two groups during the last career episode of employment. This difference relates to the reason for job loss (see Table 1). People who are about to lose their jobs because of a plant closure report a similar level of well-being as people who continue to work until retirement. The life satisfaction of people dismissed for other reasons differs from that of employees. Plant closures are thus the best possible way to approach causal effects of unemployment (Kassenboehmer and Haisken-De New 2009). We report separate results for this particular reason for job loss whenever we are able to rely on 50 or more observations.
Multivariate Analyses to Identify Nonmonetary Effects
Because of the anticipation of unemployment indicated by Fig. 1, we consider the second-to-last SOEP interview given in employment as the pretreatment reference point for the unemployed. Regarding the employed, the reference point is the last SOEP interview before retirement. Afterward, the reference point is the first SOEP interview for both groups. Hence, the time between the last observation in employment and the first observation in retirement differs between the two groups. We address this issue in the Sensitivity Analysis section.
We assume that life satisfaction at the second-to-last SOEP interview before unemployment is not yet influenced by approaching unemployment or any related circumstance. In fact, Fig. 1 seems to confirm this view, given that people show similar levels of life satisfaction until that point in time. To address this issue further, we shift our reference point back in time as part of the sensitivity checks. Moreover, we will round off our analysis by using a panel regression with individual fixed effects, which makes the pretreatment reference point in time flexible.
By considering the different reasons for unemployment, the OLS estimation allows us to explore whether the endogeneity of voluntary entries into unemployment accounts for differences in the life satisfaction changes between unemployed and employed retirees. Two drawbacks, however, are that (1) our overall treatment group of all previously unemployed retirees is not homogenous, and (2) the treatment is not exogenous when unemployment is self-chosen. After revealing differences dependent on the reason for unemployment, we therefore turn to a matching approach with a more homogenous and more exogenous treatment. Here, the reasons for unemployment before retirement are restricted to (1) plant closures and other dismissals, and (2) plant closures only. The matching is performed using entropy balancing (EB), which reweights the control group rendering it as comparable as possible to the treatment group across all SD and JC (as measured at the second-to-last interview before unemployment). The EB algorithm assigns weights to all control group observations that differ as little as possible from the base weights (one over the number of control observations), upon the condition that the mean and the variance of all characteristics adjust to the treatment group observations (Hainmueller and Xu 2013; for previous applications of EB, see, e.g., Marcus 2013; Preuß and Hennecke 2018; Stanton and Thomas 2016).
As Table S1 (in the online appendix) illustrates for our sample, EB guarantees the statistical balance of the conditioning variables in the treatment and control groups. In contrast, propensity score–based reweighting techniques involve selecting variables to improve the balance, which can imply neglecting theoretically important factors of influence. EB is hence less arbitrary than reweighting based on propensity scores. We nevertheless check the robustness of our results regarding the selected matching technique by performing propensity score reweighting as well, using the same sets of SD and JC as conditioning variables.
After reweighting, we estimate the scarring effect, as before, as the difference in the within-person change in life satisfaction from the last spell of employment to the first year of retirement between people who retire from unemployment and employment, respectively. We again control for the parallel relative change of income in order to separate monetary from nonmonetary scarring, and for the year of retirement. These variables are not measured at the pretreatment reference point in time and therefore are not part of the matching but instead are part of the subsequent regression.
Table 2 displays our main results, starting with the regression that separately includes the reasons for unemployment. As column 1 displays, the treatment variables for plant closures and dismissals show statistically significant and large negative effects, unlike the more endogenous reasons of unemployment. These three coefficients hardly change when we control for the relative income change between the final employment spell and the first year of retirement as well as for job characteristics and sociodemographic characteristics (columns 2 and 3). A scarring effect also shows up when we restrict the treatment group to involuntary unemployment while reweighting the control group using EB (column 4), and when we consider only plant closures as a trigger of job loss and reweight the control group accordingly (column 5).6 Hence, whether we use a multiple regression or EB to account for pretreatment levels of SD and JC does not affect our results.
The effect of a change of income in the matched sample seems to be different from the unmatched sample, although both are insignificant. Considering people’s age at the last spell of employment in the matching exercise heavily stretches the weights because that age differs strongly between formerly employed and formerly unemployed retirees. Fortunately, including or excluding people with very high weights does not affect the scarring effects, but it does modify the effect of the relative change in income. In the next section, we present a sensitivity check that adjusts the time span from employment to retirement for the two groups of retirees and in the process reduces the difference in age. This resolves the issue of very high weights and yields an effect of the income change similar to that of the unmatched sample.
Our findings point to large nonmonetary scarring effects of involuntary job loss, whereas unemployment that was more likely voluntarily chosen does not reduce well-being beyond retirement. This is the main reason why in our prior research (Hetschko et al. 2014) we did not detect scarring effects based on our simple comparison of average life satisfaction of all unemployed people from before the job loss to after retirement. When it comes to involuntary unemployment, about one-half of the average effect of losing work persists after retirement (Kassenboehmer and Haisken-DeNew 2009). Nonmonetary unemployment scarring thus also comes close to one-half of the short-term effect of widowhood on life satisfaction (Clark et al. 2008) and has a magnitude similar to that of the short-term effect of marriage (Qari 2014). Because we use the first year of retirement as reference point here, there might be adaptation over time, just as in the cases of widowhood and marriage. We return to that topic later in the article.
Next, we examine whether the magnitude of the negative impact of previous unemployment on retirees’ well-being depends on the frequency and length of unemployment experience (sensitization). We pool the two groups with employer-initiated dismissals (plant closure and dismissal). The combined treatment group is then split into a group of people for whom the unemployment spell before retirement was the first in their life and a group that has had previous unemployment experience (see Table 3, column 1). The corresponding effects do not differ significantly.
In addition, we distinguish between employed people with and without unemployment experience over the life course and find neither a sizable nor a statistically significant difference between them in the change in life satisfaction when retiring. This, however, does not mean that previous unemployment experience before the final spell could not leave scars in workers’ well-being. If previous unemployment episodes left scars, these would reduce the well-being already at the pretreatment point in time, so that they would not be detected by our DiD identification strategy. We then also interact retiring from unemployment with the length of the final unemployment spell (column 2) and the overall length of all unemployment spells (column 3). Apparently, the length of unemployment and lifelong unemployment experience hardly aggravate the scarring effect of retiring from unemployment. Contrary to Luhmann and Eid (2009), we do not find evidence for sensitization.
For our sensitivity checks, we use the main specification of the pure regression analysis (Table 2, column 3) as well as the two matching tests with subsequent regressions (Table 2, columns 4 and 5). In some cases, the number of observations of people who lost their jobs because of a plant closure falls short of 50, so we do not report separate results for this kind of job loss anymore. Moreover, we test for the role of previous unemployment experience by applying the sensitivity check to the first specification presented in Table 3 (column 1).
We consider additional controls for personality traits, employing the so-called five factor model (openness to new experience, neuroticism, conscientiousness, agreeableness, and extraversion). Each trait is measured using the mean value of three behavioral self-assessments, which respondents perform on seven-point scales (Richter et al. 2017:46–49). Given that these data are available for only 2005, 2009, and 2013, we transfer the information to observations of the same person in other years, assuming temporal stability over up to 10 years (Cobb-Clark and Schurer 2012). The data from 2005 describe personality traits from 1996 to 2006 in our analysis, information from the 2009 wave is transferred to the years from 2007 to 2010, and the 2013 information is assigned to the years from 2011 onward. Accordingly, we lose all observations of people who did not participate in 2005, 2009, or 2013, or retired prior to 1996. The corresponding estimations suggest statistically significant scarring effects of a similar magnitude as before (Table S2, online appendix).
Sometimes employers have to make severance payments to workers that they want to dismiss in order to meet the requirements of employment protection legislation. This renders the parting more voluntary from the employee’s point of view because it is not clear whether they would have had to leave otherwise. The severance payment may compensate them for both the monetary and nonmonetary losses of well-being that they experience because of joblessness. Including these cases in our sample thus biases our results toward 0, given that we are interested in the effect of involuntary unemployment. To shed light on this conjecture, we test whether people who receive such payments show smaller scarring effects. To this end, we interact involuntary unemployment before retirement with receipt of a severance payment (yes/no) in our regression. Nearly two-fifths (38 %) of retirees from involuntary unemployment received such a payment. They show weaker scarring effects than people who did not benefit from severance pay (difference significant at the 5 % level; see also Table S2 in the online appendix).
Work After Retirement
While retired, some people continue to work to supplement their pensions. Working retirees might be exceptional and could reduce the comparability of treatment and control groups because aspects related to working life presumably no longer matter to post-retirement life satisfaction. We therefore repeat our estimations for a more homogeneous sample of people who do not work at all after retirement. The results, presented in Table S3 of the online appendix, are also fully in line with our main findings.
Time Span Between Final Spell of Employment and First Year of Retirement
The effect of aging on life satisfaction might play a part in our treatment effect: several years pass in the lives of formerly unemployed retirees between the final spell of employment and the first year of retirement, whereas the comparison group of former employees grows only one year older. Because there is no variation in this respect in the latter group, we cannot simply control for the time span between the final job and the first year of retirement. Hence, we conduct another check that synchronizes the time span from employment to retirement between treated and control observations. The reference point in time is set to the fifth-to-last SOEP interview before retirement in both groups. We require the control group to continue to be employed from the fifth-to-last to the last interview before retirement. Persons in the treatment group are employed at least at the fifth and at the fourth interview before retirement but become unemployed at some point thereafter (i.e., at the third-to-last, or second-to-last, or last interview before retirement) and stay jobless until retirement. Synchronizing time spans in this way does not change our main results (Table S3 in the online appendix).
Shifting the Reference Points in Time
Our identification strategy relies on the second-to-last year before unemployment as reference point in time, representing the final episode of employment in the career of people who later retire from unemployment. To relax this assumption, we vary the pretreatment reference point in time. Almost the same effects appear when we consider the third-to-last interview before unemployment as reference point for the treatment group (Table S4, online appendix). In contrast, they become less negative (and statistically insignificant) when we shift this reference point to the last interview before job loss, probably because of the anticipation of the event. Regarding the employed, our results are also robust regarding the choice of different possible reference points before retirement (Table S5, online appendix).
Formerly unemployed retirees in the treatment group might be negatively selected when their job losses date back a long time, but they were not able or willing to find new jobs. Some life event might prevent them from finding a new job and at the same time reduce their well-being, confounding our estimated scarring effects from unemployment. However, this issue does not seem to drive our results, given that we also obtain a strong scarring effect when we consider short-term unemployment only (of less than one year; see Table S6, online appendix).
Involuntary Retirement, Endogenous Retirement, and Early Retirement
Some workers might be able to influence the timing of retirement thanks to early retirement gateways provided by the German public pension system. Approximately three-fourths (76 %) of employees in our sample retired before the statutory retirement age, which was 65 years for most of our investigation period (see Fig. S1, online appendix). Unlike for employees, retirement is often involuntary for unemployed people in our sample. Following the subsidiarity principle, unemployed persons are no longer allowed to receive means-tested, second-tier unemployment benefits as soon as they become eligible for other sources of income, such as early retirement pensions. Thus, most commonly at the age of 60 years, they are forced to retire, which is why the share of unemployed people retiring soars at that age (Fig. S1). The scarring effect of unemployment may thus originate from the lack of voluntariness of retirement that the comparison group of employees enjoys. However, only 5 % of formerly unemployed retirees in our sample stated an intent to return to working life after retirement, which Bonsang and Klein (2012) suggested as proxy for voluntariness. To check whether involuntary retirement drives our results, we leave this small group out of the sample; we find the same scarring effects as before (Table S7, online appendix).
As another check, we compare formerly unemployed people who probably have not determined the timing of retirement with a control group of employees who have also been forced to retire. Given that both groups did not retire voluntarily, we should not see a scarring effect if involuntariness of retirement explained our results. This kind of analysis also addresses endogenous retirement because people who are forced to retire transition for an exogenous reason. For most of the employed, the statutory retirement age is the latest gateway into retirement. Even permanent contracts normally incorporate clauses according to which people have to retire as soon as they reach retiring age. Collective agreements and company agreements also include such rules. Thus, most people aged 65 in our data can no longer postpone retirement (Fig. S1). As explained earlier, unemployed people are likely to retire involuntarily from age 60 onward. These two groups form a subsample of 581 people, provided that the unemployed lost their jobs because of a plant closure or a different kind of dismissal by employer. To avoid losing too many observations due to missing values, we control for five-year classes of the year of retirement and the income change from the last employment spell to retirement only. The analysis yields a nonmonetary scarring effect of –0.294 (p = .05). Hence, whether workers can decide about the timing of retirement hardly modifies our findings.
Another implication of the aforementioned opportunity—and in many cases, obligation—for unemployed workers to retire prior to the statutory retirement age of 65 is that early retirement is more prevalent in this group than among formerly employed retirees, as confirmed by Fig. S1. If early retirement reduced well-being permanently, it could thus be an explanation for scarring effects of unemployment. To explore this aspect, we lower the age limit of our sample to 60 years, excluding extreme cases of early retirement, which should reduce the scarring effects we measure if they were caused by early retirement. Still, we do not find substantially smaller scarring effects compared with our main results (Table S8, online appendix).7 As another check, we exclude people who retire when reaching the statutory retirement age. This step would reduce the estimated unemployment scarring effects if early retirement affected life satisfaction per se, given that now formerly employed workers also retire early. Yet, setting the upper age limit of our sample to 64 years yields practically the same findings as before (see Table S8, online appendix).
Pre-trends and Adaptation: An Individual Fixed-Effects Estimation
Throughout our analysis, we assume that the control group, conditional on SD and JC variables, displays the trend in life satisfaction that the treated would have shown if they had not been treated. Reassuringly, Fig. 1 indicates that the pre-trend of life satisfaction does not differ between the two groups. To examine trends before and after the treatment, we conduct an individual fixed-effects estimation that uses lag and lead variables around an event to document life satisfaction several years before and after it (Clark and Georgellis 2013; Clark et al. 2008; Hijzen et al. 2010). The lead variables allow us to test whether the second-to-last year before unemployment is representative for pre-unemployment life satisfaction—that is, life satisfaction at that time does not differ from other pre-unemployment points in time. The lag variables enable us to investigate adaptation by comparing life satisfaction at different times after retirement with pre-unemployment life satisfaction. Finally, we shed light on the evolution of life satisfaction around retirement for formerly employed retirees.
Recall that the panel includes retirees older than 54 who are either employed or unemployed before retiring. For the fixed-effects estimation, we also restrict the age of observations of these people at earlier points in time, when they were employed or unemployed, to 50 years or more, to build a sample of comparable people approaching the end of their working life.
Empirically, we distinguish between a treatment group of retirees from unemployment because of plant closure or dismissal (U) and a control group of retirees from employment (E).8 Moreover, we include being unemployed at other points in time (m), without distinguishing between the respective reasons. If this variable equals 1, the person experiences unemployment that does not lead into retirement without intermediate employment.
The estimated β and γ coefficients have to be interpreted as deviations from the within-person average life satisfaction level when employed. Because we estimate lead effects, these reference employment episodes have to be at least four years before retirement for people who retire out of employment and at least three years before the last unemployment spell for people who retire out of unemployment.
A drawback of the fixed-effects approach is that we are not directly able to control for job characteristics, which are not observed for jobless and retired people. To overcome this problem, we use entropy balancing again and reweight the control group of people retiring from employment such that mean and variance across all SD and JC (see Table 1) of the last job equalize (as before, at time R – 1 for the controls and time U – 2 for the treated). Age and income, which are already considered by the fixed-effects regression, are not included in the set of balancing variables. The weights are then assigned to all observations of the same person. Figure 2 displays the findings for the lag and lead variables; Table S9 (online appendix) presents the results in full and allows for a comparison of the results based on the matched and on the unmatched sample.
The fixed-effects approach supports our previous findings. As the upper panel of Fig. 2 shows, life satisfaction at the second-to-last interview before unemployment does not differ from earlier points in time (U – 2). It declines over the following year (U – 1), indicating job insecurity before job loss. Unemployment leads to a strong loss of well-being (U), which is partly reversed by retiring (from U to R0). Up to retirement, the control group of employees shows no changes in life satisfaction (middle panel).
A large scarring effect shows up at R0 (lower panel). It is calculated as the difference in the change of life satisfaction from employment to retirement between formerly unemployed retirees and formerly employed retirees: βR0 – γR0 = –0.513 (p < .01). It is of similar size as the effects presented before based on the DiD calculations. The scarring effect lessens within the following year (R1) but remains substantial (–0.305, p = .024). Scarring continues to play a part when we go another year forward in time (R2). In the longer run, however, we cannot reject that there is adaptation. When all later points in time are pooled, people who retire from unemployment seem to differ less than before from people who retire from employment (–0.155, p = .213).
When it comes to retirement from employment, we deliberately refer to the unweighted control group because these results represent the average retiree from employment (first column of Table S9), whereas the reweighted control group resembles people who retire from an involuntary episode of unemployment. In line with the previous literature, we find no clear association between retirement and life satisfaction for the average retiree (Bonsang and Klein 2012; Henning et al. 2016; Luhmann et al. 2012).
Unemployment experienced in the past seems to cause a loss in subjective well-being, even after a person has found a new job. The persistent loss in subjective well-being from past unemployment episodes may occur, inter alia, because past unemployment worsens job characteristics and increases the fear of becoming unemployed again in the future, which in turn leads to lower expected labor income and higher expected income volatility. Beyond these kinds of scars that are related to future employment prospects, unemployment may also scar for reasons that are independent of one’s future in the labor market. To separate employment-related from non–employment-related scars, we use the transition to retirement because this life event eliminates the employment-related consequences of unemployment. Retired people no longer have to worry about future employment chances or job quality.
We find that an involuntary unemployment spell between the last job and retirement negatively affects well-being after retirement. That this effect persists when income is controlled for suggests that the non–employment-related part of the scarring effect of unemployment is at least partly of a nonmonetary nature. This finding applies to people who experienced unemployment for the first time as well as for those with previous unemployment experience. As a result, our prior finding that retirement seems to cure the nonpecuniary well-being loss of unemployment (Hetschko et al. 2014) needs to be qualified. Although we confirm the positive effect of retiring from unemployment, we show that the involuntarily unemployed continue to bear a large nonmonetary cost after the transition.
Our results confirm previous research showing long-lasting negative effects of unemployment on life satisfaction (e.g., Bell and Blanchflower 2011; Clark et al. 2001; Daly and Delaney 2013). It is also compatible with the analysis by Knabe and Rätzel (2011), who showed that a large part of the nonmonetary scarring effect stems from job prospects. We show that the scarring effects of unemployment do not concern only job characteristics but instead exceed working life. In fact, looking back on a successful career might be one aspect that contributes to individual well-being.
Finally, our results also relate to previous research emphasizing that deviating from the social norm to work and therefore being unable to match one’s preferred group identity may be an important reason why workers suffer from a job loss beyond monetary reasons. According to our results, however, the identity-restoring effect of retiring does not cure the whole negative well-being effect of a job loss. In turn, the temporary deviation from social norms while being jobless may not be the only relevant nonmonetary reason for the huge misery of the unemployed. Retrospective norm deviation in the sense that one did not live a successful working life and has not successfully concluded it also matters.
We thank the Editors and two anonymous referees for their helpful suggestions, and Tom Günther for excellent research assistance. Moreover, we are grateful for comments on a previous version to C. Katharina Spieß; Reto Odermatt; and participants of the SOEP seminar at DIW Berlin (2013), the IAAEU seminar at the University of Trier (2013), the annual conference of the German Economic Association, Hamburg (2014), and the IAB conference Labor Market Prospects of Older Workers, Nuremberg (2014). Clemens Hetschko and Ronnie Schöb acknowledge financial support by the German Science Foundation (DFG) through Project SCHO 1270/5-1.
According to Bender (2012), unemployed people suffer from retirement because they perceive the transition as more involuntary than employees. In our sensitivity analysis, we test whether this explains the scarring effects by comparing employed and unemployed workers who both were likely forced to retire.
The data are freely provided to academic users by the German Institute for Economic Research (DIW), Berlin (https://www.diw.de/en/diw_02.c.222829.en/access.html).
For most of the time, the statutory retirement age was 65 years. Since 2012, it has been gradually increased and will be 67 years from the 1964 birth cohort onward.
This group mainly comprises people on so-called mini-jobs, who currently earn no more than 450 euros per month and benefit from reduced taxes and social security contributions.
Incomes hardly vary after retirement. Most people mainly rely on the public pension, which cannot fall and increases with wage growth. If we see that a last unemployment spell reduces current monthly income by x euros after retirement, current income will be lowered by x euros in any month after retirement. The drop in current income is thus proportional to the drop in permanent income. Controlling for the change in income when entering retirement thus also captures the effect of the permanently forgone retirement income due to unemployment.
Using propensity score reweighting instead of EB, we find a scarring effect of –0.396 (p < .05) for the combined treatment group of plant closure and other dismissals and –0.486 (p < .01) for plant closures only.
Here, the EB algorithm no longer converges if we consider age at the last episode of employment as a balancing variable, so that the test relies on regressions only.
Because of low numbers of observations for the lag and lead variables, we cannot focus on unemployment due only to plant closure here.
One year dummy variable is left out of the estimation to account for the collinearity of age and the year dummy variables. A regression with age dummy variables instead of age and age squared yields practically the same results.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.