Introduction In many countries population aging is contributing to - - PDF document

SLIDE 1

1

Title: Health trajectories into retirement: a 12 year follow-up study based on HRS data

Michael Boissonneault1,2 and Joop de Beer1

Abstract

Objectives Retirement is a complex process that unfolds over many years. Changes in health that occur alongside this process possibly have an important influence on it, however, this question has mostly been studied using panel data representing individuals at two points in time. We study the impact of health on retirement by modelling individual trajectories of health that are based on data representing individuals at up to 8 points in time, and by linking these trajectories to one of five retirement pathways that account for the complexity of the retirement process. Methods We follow 3,495 Health and Retirement Survey (HRS) participants over an average of 12 years between 1992 and 2014. Health is measured using an index based on health conditions and mental health. We use a retirement typology that comprises the retirement pathways (1) Sustained work until age 66 (2) Crisp transitions into retirement (3) Reversal of the retirement process (4) Gradual retirement and (5) Blurred retirement. Latent class growth analysis is used to model individual health trajectories. Results Four distinct health trajectories for men (Persistently good (21%); Deteriorating (51%); Persistently poor (25%); and improving (3%)) and five for women (Persistently good (24%); Progressive decline (40%); Accelerated decline (19%); Persistently poor (10%); and Improving (7%)) are identified. Non-whites and people with lower education are more likely to be assigned the Persistently poor health trajectory and the Blurred retirement pathway. Persistent poor health leads more often to blurred retirement pathways while persistent good health leads less often to such retirement outcomes. The health trajectory determines the retirement pathway to a greater extent among women than among men. Conclusions Our results show the complex interplay between change in health and transition into retirement, where different health trajectories seem to trigger different retirement patterns.

1 Netherlands Interdisciplinary Demographic Institute (NIDI-KNAW)/University of Groningen, The Netherlands 2 Corresponding author: P.O. Box 11650, NL-2502 AR The Hague, The Netherlands. boissonneault@nidi.nl. ++ 31

(0)70 3565239

SLIDE 2

2

Introduction In many countries population aging is contributing to increases in the share of non-working to working people (United Nations 2015). This poses concerns about the capacity to maintain economic growth (Bloom et al. 2010) and the sustainability of public pensions (Bongaarts 2004). One policy response consists in encouraging longer economically active lives (OECD 2011). This is being done by increasing the normal retirement age, restricting access to early retirement,

• r by providing financial incentives for delaying retirement (OECD 2015). However, older

workers more often have health complaints which force them to retire before the state pension age (Ilmarinen 2001). Such forced retirements form a hurdle towards longer economically active lives and induce public spending through disability benefits programs (OECD 2010). As a result, it is important to have a better understanding of how declining health forces older workers to leave the workforce early. Retirement is not viewed anymore as a clear-cut, once-in-a-lifetime transition, but rather as a complex process that takes place over several years (Schultz and Olsen 2013). Consequently, some referred to retirement as a phase (Denton & Spencer 2009; Borland 2005). The retirement phase starts with the end of some “relatively permanent pattern of labor market activity” (Borland p. 1), or of a “career job” (Feldman 1994), and ends up with the entry into full and permanent retirement, or death. Although these two events can directly follow each other, the retirement phase often includes many intermediate segments. Such segments include bridge- employment (Cahill 2013) or “unretirement” (Maestas 2010), for example. Empirical evidence showed that, in the United States, retirement patterns that include such intermediate segments are quite common: between one-third and two-thirds of older American workers ever experience bridge-jobs (Cahill 2006; Cahill 2013), and up to one fourth “unretire” (Maestas 2010). Others have identified the presence of “blurred” segments inside the retirement phase, for example when someone first becomes unemployed and then truly retires, making it difficult to assess when the transition into retirement actually occurred (Mutchler 1999). In sum, properly studying retirement supposes taking into account that retirement may be ambiguous and reversible. Considering the phenomenon over a longer time frame may thereby help providing a more accurate picture.

SLIDE 3

3

Health is usually considered as the second most important predictor of retirement, after economic

• incentives. It is convenient to view health as part of the work ability framework (Ilmarinen

2001). Work ability is “a process of human resources in relation to work” (Ilmarinen 2001, p. 549). An important component of such human resources is a person’s health. As people age and their health deteriorates, the human resources that can be taped into in order to answer work demands also decline. One way that older workers can cope with declining resources is by adjusting their work. This can be done for example by adjusting the amount of hours worked or by changing tasks. For the older worker, such changes are often viewed as some pre-retirement

• phase. As a result, the succession of segments that characterize the retirement process could in

part reflect changes in human resources that takes place with age. However, most studies are ill- designed to study this question because health and retirement are defined on the base of data representing individuals at usually one or two points in time only (Burdorf 2012; Beehr and Bowling 2013; Amick 2015). Most studies so far focused on one stage of the retirement process only. They focused for example on single transitions from full time work into early retirement (Alavinia et al. 2008; Pit et al. 2010, Karpansalo 2004; Leijten 2015; Pietilainen 2011; Robroek 2013; Schuring 2013; Van den Berg 2010), from retirement back to work (i.e. unretirement) (Maestas 2010; Cahill 2013), from career jobs to bridge-jobs (Cahill et al. 2013; Kerr & Amstrong-Stassen 2011) or from full time work to part-time work or partial retirement (Cahill et al. 2013). The same studies often considered health at one point in time only, either immediately before (Karpansalo 2004; Leijten 2015; Pietilainen 2011; Robroek 2013; Schuring 2013; Van den Berg 2010) or after some retirement transition (Alavinia et al. 2008; Pit et al. 2010). We note, however, some exceptions, mainly from the economic literature. Some have considered economic activity over a period of more than 2 years, paying attention to sequences of labor activity and ordering individuals into “crisp” and “blurred” retirement patterns (Mutchler et al. 1997). Others have considered health at two points in time, distinguishing between “contemporaneous” and “lagged” health (Disney 2006; Jones 2010), or considering “health shocks”, which is the difference in values of health between two points in time (Erdogan-Cifti et al. 2008; Riphahn 1999). One study considered health at three points in time, which allowed to account for the rate at which health deteriorates prior to some retirement transition (Bound et al. 1999). These studies allowed to postulate that change in health plays some role in determining retirement transitions, over and beyond health as

SLIDE 4

4

measured at one point in time. Also, they allowed to postulate that the effect of health on retirement may differ depending on the segment of retirement that is being considered. This paper extends this knowledge by expanding the time horizon considered to study the link between health and the retirement process. We characterize retirement and health based on information taken at up to 8 points in time, stretching out over an average of 12 years. By doing so, we account for 5 distinct retirement pathways, which we link to individual, latent trajectories

• f health. This in turn allows to examine whether individuals with different underlying rates of

change in health adopt different strategies in front of retirement. We focus on the United States, a country with comparatively important heterogeneity in retirement patterns. This exercise can prove instructive in the changing European retirement landscape, where a destandardization of the retirement process is taking place (Kanabar 2012). Methods Data We make use of the RAND HRS data file, an easy to use longitudinal data set based on the HRS data (Health and Retirement Study 2017; RAND HRS Data 2016). The HRS is sponsored by the National Institute on Aging (grant number NIA U01AG009740) and is conducted by the University of Michigan. HRS is a longitudinal survey representative of the American population age 50 and older with biannual data collection since 1992. Data contain in-depth information on the health, work and retirement of each selected participant and their spouse. From the original data set we select participants with sustained participation to HRS from their latest interview prior to reaching age 55 until their first interview after reaching age 65. We ignore information gathered before and after these two time points in order to avoid selection that could bias the representivity of the retirement outcomes. Respondents with missing information on labor force status at one or more waves, as well as respondents who are not working at least part time at any wave, are excluded. The final sample comprises a total of 3,495

• respondents. Most are observed 7 times (89.1%), the rest 6 (5.7%) or 8 (5.3%) times.

SLIDE 5

5

Measures Health To identify trajectories of health, we compute for each participant a health index that sums up their number of health conditions and their score on the Center for Epidemiologic Studies Depression Scale (CES-D). We choose these two measures because they were shown to strongly predict different forms of early transitions into retirement (Van Rijn 2013) while being less subject to measurement error than self-assessed health or ability to work (Dwyer 1999). For the presence of health conditions, HRS respondents are asked about whether they have any of the following: high blood pressure, diabetes, any form of cancer, lung disease, heart disease, stroke and arthritis. We attribute a value of one to each item of the list and sum them up for each

• participant. The CES-D scale is a screening test for depression and depressive disorder. HRS

uses the short, 8 items-version with 0 meaning having no depression symptom and 8 meaning having the maximum number of depression symptoms. The scale was described elsewhere (Rahdloff 1977). As shown in Table 1, both measures have a similar prevalence in the sample used, although the number of health conditions seem to grow faster over time than the CES-D score.

Table 1 Sample description for the health variables Mean value (standard deviation)

• no. valid
• bs.

First wave Last wave Men Health conditions 1486 0.62 (0.92) 1.44 (1.32) CES-D 1386 0.68 (1.38) 0.93 (1.55) Women Health conditions 2009 0.78 (1.05) 1.34 (1.46) CES-D 1990 0.98 (1.69) 1.2 (1.80)

SLIDE 6

6

Retirement The Rand-HRS data summarizes labor force status at the moment of the interview in a single

• variable. The variable is constructed using several questions taken from the HRS, notably about

self-assessed labor force and retirement statuses. Because it uses answers to multiple questions, this variable gives more precise and more consistent information than responses to any other single labor force attainment question taken from the HRS. Details about its construction are available elsewhere (Bugliari et al. 2016). The Rand-HRS labor force variable contains seven categories: 1.Works Full time, 2.Works part- time, 3.Unemployed, 4.Partly retired, 5.Retired, 6.Disabled, 7.Not in the labor force. Based on this variable, we construct a retirement typology. Based on previous literature (Beehr 2014; Schultz and Olsen 2013; Denton and Spencer 2009), we found that a five categories typology rendered best the complexity of retirement. The construction of each category is illustrated in Figure 1. The first category is labelled “Sustained work” and is made up of respondents who are working at their last observation and who are either working or unemployed at each preceding

• wave. The second category, “Crisp retirement”, comprises respondents who are not working at

all at the last wave and who previously made a direct transition from work to retirement. The third category is labelled “Reversed retirement” and comprises respondents who make a transition from work to retirement (complete or partial) or inactivity (disability or out of the labor force), and who then make a transition to work again at any point during follow up. The fourth category is called “Gradual retirement”. It is made up of the respondents who make a transition from work to partial retirement, and eventually to any non-working status (retirement, unemployment, disability, out of the labor force) or of the respondents who make a transition from work to retirement and then to partial retirement. Finally, the fifth category comprises atypical modes of entry into retirement and is labelled as “Blurred retirement”. It comprises respondents that make a transition from work to unemployment or inactivity (disability or out of the labor force) and who do not work again during follow-up. For clarity, we note that we distinguish between full and part-time work only if part-time work follows full time work and eventually leads to retirement or is sustained until the last wave (i.e. gradual retirement). Working part-time is otherwise simply seen as “working”. We disregard phases of unemployment if they are followed by phases of full or part-time work. Phases of unemployment that lead to or follow retirement are considered as part of the retirement process.

SLIDE 7

7 W = Working (part or full-time) R = Retired U = Unemployed I = Inactive (disabled, out of the labor force) P = Partly retired / Working part time T 1 T 2 T 3 Sutained work W W W Crisp retirement W R R / U / I Reversed retirement W R / I / P W Gradual retirement W P R / U / I / P Idem. R P Blurred retirement W U / I R/ U / I / P Figure 1 Retirement typology. Three distinct phases are considered. During the first phase (T 1), everyone working. The second phase (T 2) is the transition phase. The third phase (T 3) is the one that leads past age 65. Each phase corresponds to either one or many waves in the survey.

We note that it is possible that we omit some short-lived labor force transitions since we use information collected in average two years apart. However, the HRS data offers the possibility of constructing individual labor force histories that stretch out over more than a decade, something that data sets with observations closer apart (e.g. labor force surveys) do not usually offer. The main implication is that we may be underestimating the proportion of blurred, reversed and gradual retirement pathways. Statistical analysis We estimate trajectories of health using latent class growth analysis (LCGA) (Nagin and Land 1993). LCGA is an extension to growth mixture modeling (GMM). Contrary to GMM, which assumes that one longitudinal trajectory fits the whole population, LCGA allows to identify different trajectories that describe different latent processes inside a population. LCGA is based

• n the maximum likelihood method. We use the R package lcmm, which provides a series of

functions that allow to perform latent class growth analysis (Proust-Lima et al. 2015). More specifically, we used the function lcmm for estimating a latent class growth model based on

• rdinal longitudinal outcomes. Due to their distinct labor force participation patterns, separate

models for men and women were estimated. Models based on different link functions (linear, splines, beta) and different number of latent classes (limited to 5 to facilitate interpretation) were

• estimated. We choose for men and women the best fitting model based on the Bayesian

SLIDE 8

8

information criterion (BIC). We use the function predictY to obtain for each year of age the marginal predictions of belonging to different classes with the 95% bounds. Other papers that uses LCGA in relation to health and retirement include Serra et al. (2017), De Wind et al. (2017) and Laaksonen et al. (2012). As a measure of retirement timing, we note the age respondents were last observed working and average the values over all respondents that belong to each specific class of health trajectory and each retirement pathway. We use logistic models to test for significant association between each class of health trajectory and each retirement pathway, and to test for significant differences according to race or ethnicity (non-hispanic white vs. other) and (self-reported) years of education. Results Health trajectories The best fitting models according to the BIC contain 4 trajectories for men and 5 for women. The trajectories are best modelled according to a spline with 9 equidistant nodes. We refer the reader to the appendix for each model’s BIC score and other metrics. For men, the trajectories are labeled persistently poor, improving, deteriorating and persistently good; for women, they are labeled persistently poor, improving, progressive decline, accelerated decline and persistently good (Figure 2). More than half of the men in our sample are assigned to the deteriorating trajectory, while around one fourth are assigned to the persistently poor trajectory and a little less than one fifth to the persistently good trajectory. Only a small minority (2.9%) were assigned to the improving trajectory. For women, the most important trajectory is the progressive decline

• ne, with about two fifth of the sample. A little bit less than one fourth of the sample was

assigned to the persistently good trajectory, while a bit less than one fifth of the sample was assigned to the accelerated decline trajectory. One tenth are part of the persistently poor trajectory and about 7% is part of the improving one.

SLIDE 9

9 Figure 2 Trajectories of health prior to retirement age. Health is measured based on an index made up of the sum of the sum of health conditions and the score on the CES-D scale (short version). The solid line indicates the estimated trajectories, the dashed lines the 95% confidence intervals, and the dotted lines the

• bserved mean values.

Correlates of the health trajectories Table 2 sums up the respondents’ characteristics according to the health trajectories. Among men, respondents assigned to the persistently poor trajectory have a significantly lower share of whites, while those assigned to the declining and persistently good trajectories have significantly

Index value

SLIDE 10

10

higher shares of whites than the rest of the sample. Men assigned to the persistently good and declining trajectories have in average more years of education, while those assigned to the persistently poor trajectory have in average fewer. We find a lower share of whites among women assigned to the persistently poor and improving trajectories, but a bigger one among women assigned to the accelerated decline trajectory. Also, women assigned to the persistently good trajectory have in average more years of education while women assigned to the persistently poor trajectory have fewer. Correlates of the retirement pathways As shown in Table 3, no retirement pathway constitutes the norm in our sample. Among men, the most common pathway is the crisp retirement pathway, with somewhat more than one fourth

• f the sample. The sustained work, reversed retirement and gradual retirement pathways all have

similar proportions and each make up around one fifth of the sample. About 8% of men experienced a blurred retirement. Whites are underrepresented among the blurred retirement

• pathway. This group also has fewer years of education in average, while the sustained work

group has in average more years of education. By construction, men who kept working throughout the whole of the follow up were last observed working in average at age 66. Also unsurprisingly, men who reverse their retirement pathway were last observed working at a relatively high mean age of about 63 years old. Men who experience a crisp, gradual or blurred retirement were last observed working at ages comprised between 58.4 and 60.1 years old. Differences between each of these ages are statistically significant. Among women, the reversed retirement pathway is the most common one, comprising almost 30% of the sample. Almost one fourth were assigned to the Crisp retirement pathway, and about

• ne fifth to the gradual one. About one woman out of eight experienced sustained work or a

blurred retirement. Whites are underrepresented among the blurred retirement pathway, while they are overrepresented among the gradual retirement pathway. Women among the sustained work and gradual retirement pathways have in average more years of education. Women who were assigned to the blurred retirement pathway have in average fewer years of education. Women who kept working beyond age 65 have the highest mean age (65.9) , followed by those who reversed their retirement pathway (62.6), went through the crisp retirement pathway (59.5), went through gradual retirement (58.6) and experience da blurred retirement (57.4). Each group

SLIDE 11

11

is statistically different from the rest of the sample in terms of the average age when last

• bserved working.

Table 2 Respondents’ characteristics according to health trajectories Subjects Non-Hispanic White Race or Ethnic Group Education no. (%) no. (%) Years SE Men 1471 (100.0) 1178 (80.1) 13.1 (3.1) Persistently poor 364 (24.7) 253*** (69.5) 11.8*** (3.4) Improving 42 (2.9) 31 (73.8) 13.2 (2.4) Declining 752 (51.1) 625*** (83.1) 13.3* (3.0) Persistently good 313 (21.3) 269** (85.9) 14.1*** (2.5) Women 2005 (100.0) 1208 (75.1) 12.9 (2.7) Persistently poor 208 (10.4) 131*** (63.0) 11.7*** (2.9) Improving 143 (7.1) 96* (67.1) 12.6 (2.5) Progressive decline 804 (40.1) 621 (77.2) 12.7 (2.8) Accelerated decline 371 (18.5) 298* (80.3) 13.0 (2.6) Persistently good 479 (23.9) 360 (75.2) 13.6*** (2.5) * Significant at the 0.05 level ** Significant at the 0.01 level *** Significant at the 0.001 level Table 3 Respondents’ characteristics according to retirement pathways Subjects Non-Hispanic White Race or Ethnic Group Education Age last worked

• no. (%)

no. (%) Years (SD) Age (SD) Men 1486 (100.0) 1189 (80.2) 13.1 (3.1) 61.6 (3.8) Sustained work 304 (20.5) 252 (82.6) 13.6** (3.2) 65.9*** (0.6) Crisp retirement 422 (28.4) 346 (82.2) 13.0 (2.6) 60.1*** (3.1) Reversed retirement 306 (20.6) 242 (78.8) 13.0 (3.4) 62.9*** (3.0) Gradual retirement 328 (22.1) 262 (80.5) 13.2 (2.9) 59.5*** (3.2) Blurred retirement 126 (8.5) 89** (70.6) 12.2** (3.6) 58.4*** (3.3) Women 2009 (100.0) 1507 (75.1) 12.9 (2.8) 60.7 (4.1) Sustained work 247 (12.3) 195 (78.5) 13.5*** (2.3) 65.9*** (0.6) Crisp retirement 483 (24.0) 372 (77.2) 13.1 (2.5) 59.5*** (3.4) Reversed retirement 590 (29.4) 437 (73.9) 12.7 (2.8) 62.6*** (3.3) Gradual retirement 418 (20.8) 334** (80.1) 13.2** (2.5) 58.6*** (3.4) Blurred retirement 270 (13.4) 170*** (63.1) 11.6*** (3.3) 57.4*** (3.1) * Significant at the 0.05 level ** Significant at the 0.01 level *** Significant at the 0.001 level

SLIDE 12

12

Health trajectories according to retirement pathways Figure 3 breaks down the health trajectories according to the retirement pathway. Three significant differences can be found among men. First, compared to men within other health trajectories, men with persistently poor health take the blurred retirement pathway more often. Second, men within the deteriorating health trajectory more often retire gradually. Third, men among the persistently good health trajectory more often reversed their retirement pathway. Men within the persistently good health trajectory were last observed working in average at a higher age than the rest of the sample. There is greater variation in the association between the health trajectory and the retirement pathway among women. Women within the persistently poor trajectory are less likely to keep working until the last observation and more likely to experience a blurred retirement pathway. This last association concerns a particularly large share of women and is by large significant. Women with persistently good health experience more sustained work pathways and fewer blurred pathways into retirement compared to women experiencing other health trajectories. Women among the persistently poor trajectory, and to a lesser extent among the progressive decline trajectory, were last observed working in average at a younger age. Women among the persistently good trajectory were last observed working in average at an older age than the rest of the sample.

SLIDE 13

13 Figure 3 Distribution of retirement pathways according to the health trajectories. The values within the bars are the percentages of respondents with the retirement pathway. The figures on the right hand side of each bar are the mean age of the respondents when they were last observed working inside of each health

• trajectory. (* Significant at the 0.05 level ** Significant at the 0.01 level *** Significant at the 0.001

level)

Conclusion The present paper extended the current knowledge about the link between poor health and early retirement by following cohorts of older American workers over a period of 12 years, thereby linking changes in health with retirement pathways that workers went through. Unlike previous

** ** ** * *** *** *** ** ** ** ***

SLIDE 14

14

studies, we made full use of the longitudinal information that we disposed of. Health was modelled using LCGA, a method that allows to specify different latent classes that describes change in a specific outcome over time. Retirement was conceptualized using a retirement typology that accounted for the reversible and often ambiguous nature of retirement. Each participant was assigned to the latent class of health trajectory that best described his or her change in health over the study period, and to a single retirement pathway that took into account the labor force information collected at each wave. We checked for significant associations both between the classes of health trajectories and the different health pathways, as well as within each class or category according to race or ethnicity and years of education. To complete the analyses, we also presented a measure of age at retirement according to the different classes of health trajectory and the different retirement pathways. Blacks or Hispanics were found to be more likely to be part of more adverse health trajectories. People with persistently good health had in average more years of education, while people with persistently poor health had fewer. Similar links were found in longitudinal studies on Americans concerning race (Liang 2009) and education (Martin 2015; Tang 2014). We also found associations between retirement pathways and race, as well as between the retirement pathway and the number of years of education; most importantly, people who go through a blurred retirement pathway are more likely to be of white/non-hispanic race or ethnicity and to have fewer years of education. Mutchler (1999) found an association between blurred retirement pathways and years of education, but not with race. Blurred retirement is defined by making a transition from work to unemployment, disability or to being out of the labor market, without returning to work thereafter. This kind of pathway reflects in most cases some constraint: the person did not decide to retire (because they do not identify themselves as such) but their withdrawal from the labor market still led them to actually retire. Attention should be paid by policy makers in order to overcome such undesired outcomes among marginalized groups. The last part of our analysis consisted in studying the link between the different health trajectories and retirement pathways. The different patterns observed among men and women raise interesting questions. First, women assigned to the persistently good health trajectory had a significantly higher proportion who kept working until age 66. No such association was found among men. Second, men with persistently good health more often went back to work after a first retirement, while this was not the case for women within the same health trajectory. Third,

SLIDE 15

15

among women, persistently good health led less often to blurred retirement pathways, but this was not the case for men. Fourth, for both men and women, the persistently poor health trajectory contained relatively more blurred retirement pathways; however, this association was much stronger and significant at a higher level among women than among men. Fifth, only men with deteriorating health were more likely to gradually retire compared to other trajectories. Lastly, persistently poor or good health had a bigger impact on the mean age at which participants were last observed working among women than among men. The differences in retirement patterns between men and women are well documented (Fischer 2015). Some also underscored different effects of health on retirement between sexes (). Here, we showed that persistently good or poor health induce different retirement patters in men and women. These differences could point towards the necessity for more men to keep on working in order to provide for their household, despite poor health. The higher prevalence of blurred retirement pathways among women could also be explained by a higher share of women who move to a housewife role later in their career, a status that men may find less suitable. An important question in the economic literature has been whether the absolute level of health,

• r rather change in health induces early retirement (French 2017). Although we did not directly

address this question, we provided interesting insights into it. Some findings were expected, while others were less so. First, we found that constant good health led more often to prolonged work until age 66 (women), and that people among this trajectory were last observed working at an older age. This finding is in accordance with prior observations that good health leads more

• ften to “on time” retirement or to sustained work past the retirement age (Fisher et al. 2016).

Also this leads less often to blurred retirement pathways, a finding that is not surprising if we consider that many people who take this route were to some degree forced to retire (Szinovacz and Bailey 2005). We also found that return to work is more likely among men within the persistently good health trajectory, a finding that is comparable to Maestas’s (2010). Other of our findings seem to contradict previous research. We found similar or slightly lower ages when last

• bserved working among people whose health had been constantly poor compared to people

whose health had been declining. The literature that we reviewed suggested a similar or bigger effect of change in health over constant poor health (Erdogan-Cifti et al. 2008; Riphahn 1999). Furthermore, although research on this topic is scarce, we naturally expected more blurred (and unexpected) retirements among people with declining health, and more crisp and gradual

SLIDE 16

16

retirements among people with constantly poor health. This would correspond to a natural adjustment to declining work ability: workers who have been struggling for a long time with an imbalance between individual resources and work will look for some early retirement options, while those for whom the imbalance is quickly disrupted may unexpectedly retire, either via unemployment or disability. Our results rather showed opposite associations. Retirements that were arguably planned (crisp and gradual retirement pathways) are more often to be found among deteriorating health trajectories, while blurred retirements are more often to be found among the persistently poor health pathway. As a result, we suggest that poor health functions more as a cumulative process, and that an accumulation of negative health outcomes can become “too much” at any given point in time and thereby induce some unplanned retirement. The deteriorating health pathways that we identified, on the other hand, might be part of a normal aging process that people who are entering old age expect. People who experience such decline in health are therefore able to adjust consequently. We conclude that changes in health, rather than health at one point in time affect the process of retirement, although in a different way than would have naturally been expected.

SLIDE 17

17

Literature Alavinia, S. M., & Burdorf, A. (2008). Unemployment and retirement and ill-health: a cross- sectional analysis across European countries. International archives of occupational and environmental health, 82(1), 39-45. Amick BC, McLeod CB, Bültmann U. Labor markets and health: an integrated life course

• perspective. Scand J Work Environ Health. 2015;41(4):325–419.

Beehr, T. A. (2014). To retire or not to retire: That is not the question. Journal of Organizational Behavior, 35(8), 1093-1108. Beehr, T.A., & Bowling (2013). Variation on a retirement theme: conceptual and opetational definitions of retirement. 42-58. The Oxford handbook of retirement. Oxford University Press. Wang, M. (Ed.). Bloom, D. E., Canning, D., & Fink, G. (2010). Implications of population ageing for economic

• growth. Oxford review of economic policy, 26(4), 583-612.

Bongaarts, J. (2004). Population aging and the rising cost of public pensions. Population and Development Review, 30(1), 1-23. Borland, J. (2005). Transitions to retirement: A review. Melbourne: Melbourne Institute of Applied Economic and Social Research. Bound, J., Schoenbaum, M., Stinebrickner, T. R., & Waidmann, T. (1999). The dynamic effects

• f health on the labor force transitions of older workers. Labour Economics, 6(2), 179-202.

Burdorf, A. (2012). The need for novel strategies to analyze the dynamic pattern of worker's health over time and the consequences for sustained employability. Scandinavian journal of work, environment & health, 38(6), 485-488. Cahill, K. E., Giandrea, M. D., & Quinn, J. F. (2006). Retirement patterns from career

• employment. The gerontologist, 46(4), 514-523.

Cahill, K. E., Giandrea, M. D., & Quinn, J. F. (2013). Retirement patterns and the macroeconomy, 1992–2010: The prevalence and determinants of bridge jobs, phased retirement, and reentry among three recent cohorts of older Americans. The Gerontologist, 55(3), 384-403.

SLIDE 18

18

Denton, F. T., & Spencer, B. G. (2009). What is retirement? A review and assessment of alternative concepts and measures. Canadian Journal on Aging/La revue canadienne du vieillissement, 28(1), 63-76. de Wind, A., Leijten, F. R., Hoekstra, T., Geuskens, G. A., Burdorf, A., & van der Beek, A. J. (2017). “Mental retirement?” Trajectories of work engagement preceding retirement among

• lder workers. Scandinavian journal of work, environment & health, 43(1), 34-41.

Disney, R., Emmerson, C., & Wakefield, M. (2006). Ill health and retirement in Britain: A panel data-based analysis. Journal of health economics, 25(4), 621-649. Dwyer, D. S., & Mitchell, O. S. (1999). Health problems as determinants of retirement: Are self- rated measures endogenous?. Journal of health economics, 18(2), 173-193. Erdogan-Ciftci, E., Van Doorslaer, E., & Lopez Nicolas, A. (2008). Health, financial incentives and retirement in Spain. Tinbergen Institute Discussion Paper No. 08-093/3. Available at SSRN: https://ssrn.com/abstract=1277145 or http://dx.doi.org/10.2139/ssrn.1277145 Feldman, D. C. (1994). The decision to retire early: A review and conceptualization. Academy of management review, 19(2), 285-311.

French, E., & Jones, J. B. (2011). The Effects of Health Insurance and Self‐Insurance on Retirement

• Behavior. Econometrica, 79(3), 693-732.

Health and Retirement Study, RAND HRS Data public use dataset. Produced and distributed by the University of Michigan with funding from the National Institute on Aging (grant number NIA U01AG009740). Ann Arbor, MI, (2017). Ilmarinen, J. E. (2001). Aging workers. Occupational and environmental medicine, 58(8), 546- 546. Ilmarinen, J. (2009). Work ability—a comprehensive concept for occupational health research and prevention. Scandinavian journal of work, environment & health, 1-5. Kanabar, R. (2012). Unretirement in England: An empirical perspective. Discussion Papers in Economics, (12/31).

SLIDE 19

19

Karpansalo, M., Manninen, P., Kauhanen, J., Lakka, T. A., & Salonen, J. T. (2004). Perceived health as a predictor of early retirement. Scandinavian journal of work, environment & health, 287-292. Kerr, G., & Armstrong-Stassen, M. (2011). The bridge to retirement: Older workers’ engagement in post-career entrepreneurship and wage-and-salary employment. The Journal of Entrepreneurship, 20(1), 55-76. Kingston, A., Davies, K., Collerton, J., Robinson, L., Duncan, R., Kirkwood, T. B., & Jagger, C. (2015). The enduring effect of education-socioeconomic differences in disability trajectories from age 85 years in the Newcastle 85+ Study. Archives of gerontology and geriatrics, 60(3), 405-411. Laaksonen, M., Metsä-Simola, N., Martikainen, P., Pietiläinen, O., Rahkonen, O., Gould, R., & Lahelma, E. (2012). Trajectories of mental health before and after old-age and disability retirement: a register-based study on purchases of psychotropic drugs. Scandinavian journal of work, environment & health, 409-417. Leijten, F. R., de Wind, A., van den Heuvel, S. G., Ybema, J. F., van der Beek, A. J., Robroek, S. J., & Burdorf, A. (2015). The influence of chronic health problems and work-related factors on loss of paid employment among older workers. J Epidemiol Community Health, 69(11), 1058- 1065. Liang, J., Xu, X., Bennett, J. M., Ye, W., & Quiñones, A. R. (2009). Ethnicity and changing functional health in middle and late life: A person-centered approach. Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 65(4), 470-481. Maestas, N. (2010). Back to work expectations and realizations of work after retirement. Journal

• f Human Resources, 45(3), 718-748.

Martin, L. G., Zimmer, Z., & Lee, J. (2017). Foundations of activity of daily living trajectories of

• lder Americans. The Journals of Gerontology: Series B, 72(1), 129-139.

Mutchler, J. E., Burr, J. A., Massagli, M. P., & Pienta, A. (1999). Work transitions and health in later life. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 54(5), S252-S261.

SLIDE 20

20

Nagin, D. S., & Land, K. C. (1993). Age, criminal careers, and population heterogeneity: Specification and estimation of a nonparametric, mixed Poisson model. Criminology, 31(3), 327- 362.

• OECD. 2010. Sickness, Disability and Work: Breaking the barriers. OECD Publishing, Paris.
• OECD. 2011. “Linking Pensions to Life Expectancy”, in Pensions at a Glance 2011: Retirement-

income Systems in OECD and G20 Countries, OECD Publishing.

• OECD. 2015. Pensions at a Glance 2015: OECD and G20 indicators, OECD Publishing, Paris.

DOI: http://dx.doi.org/10.1787/pension_glance-2015-en Pietiläinen, O., Laaksonen, M., Rahkonen, O., & Lahelma, E. (2011). Self-rated health as a predictor of disability retirement–the contribution of ill-health and working conditions. PloS

• ne, 6(9), e25004.

Pit, S. W., Shrestha, R., Schofield, D., & Passey, M. (2010). Health problems and retirement due to ill-health among Australian retirees aged 45–64 years. Health Policy, 94(2), 175-181. Proust-Lima, C., Philipps, V., & Liquet, B. (2015). Estimation of extended mixed models using latent classes and latent processes: the R package lcmm. arXiv preprint arXiv:1503.00890. Radloff, L. S. (1977). The CES-D scale: a self-report depression scale for research in the general

• population. Applied Psychological Measurement, 1(3), 385–401.

RAND HRS Data, Version P. Produced by the RAND Center for the Study of Aging, with funding from the National Institute on Aging and the Social Security Administration. Santa Monica, CA (August 2016). Riphahn, R. T. (1999). Income and employment effects of health shocks A test case for the German welfare state. Journal of Population Economics, 12(3), 363-389. Robroek, S. J., Schuring, M., Croezen, S., Stattin, M., & Burdorf, A. (2013). Poor health, unhealthy behaviors, and unfavorable work characteristics influence pathways of exit from paid employment among older workers in Europe: a four year follow-up study. Scandinavian journal

• f work, environment & health, 125-133.

Shultz, K. S., & Olson, D. A. (2013). The changing nature of work and retirement.543-558. The Oxford handbook of retirement. Oxford University Press. Wang, M. (Ed.).

SLIDE 21

21

Schuring, M., Robroek, S. J., Otten, F. W., Arts, C. H., & Burdorf, A. (2013). The effect of ill health and socioeconomic status on labor force exit and re-employment: a prospective study with ten years follow-up in the Netherlands. Scandinavian journal of work, environment & health, 134-143. Serra, L., Gómez, L., Sanchez-Niubo, A., Delclos, G. L., & Benavides, F. G. (2017). Application

• f latent growth modeling to identify different working life trajectories: the case of the Spanish

WORKss cohort. Scandinavian journal of work, environment & health, 43(1), 42-49.

Szinovacz, M. E., & Davey, A. (2005). Predictors of perceptions of involuntary retirement. The Gerontologist, 45(1), 36-47.

Tang, F. (2014). Successful aging: multiple trajectories and population

• heterogeneity. International Journal of Social Science Studies, 2(3), 12-22.

United Nations (2015). World population ageing 2015, highlights, Department of Economic and Social Affairs, http://www.un.org/en/development/desa/population/publications/pdf/ageing/WPA2015_Highligh ts.pdf. Accessed 19 July 2016. Van den Berg, T. I., Elders, L. A., & Burdorf, A. (2010). Influence of health and work on early

• retirement. Journal of occupational and environmental medicine, 52(6), 576-583.

van Rijn, R. M., Robroek, S. J., Brouwer, S., & Burdorf, A. (2014). Influence of poor health on exit from paid employment: a systematic review. Occup Environ Med, 71(4), 295-301.

SLIDE 22

22

Appendix

Model No. Groups Log- Likelihood No. Parameters BIC Men linear 2

• 27855

8 55771 3

• 27650

11 55384 4

• 27586

14 55279 5

• 27586

17 55302 beta no convergence achieved No. nodes splines 8 2

• 15250

16 30616 8 3

• 15234

19 30606 8 4

• 15201

22 30563 8 5

• 15182

25 30547 9 2

• 15245

17 30615 9 3

• 15229

20 30604 9 4

• 15195

23 30558 9 5

• 15187

26 30564 10 2

• 15333

18 30798 10 3

• 15232

21 30617 10 4

• 15197

24 30569 10 5

• 15179

27 30554 Women linear 2

• 27855

8 55771 3

• 27650

11 55384 4

• 27586

14 55279 5 no convergence achieved beta no convergence achieved No. nodes splines 8 2

• 24680

16 49482 8 3

• 24674

19 19492 8 4

• 24608

22 19384 8 5

• 24594

25 49379 9 2

• 24554

17 49238 9 3

• 24547

20 49246 9 4

• 24480

23 49135 9 5

• 24466

26 49130 10 2

• 24583

18 49304 10 3

• 24575

21 49311 10 4

• 24507

24 49196 10 5

• 24492

27 49190