Predictors of Perceptions of Involuntary Retirement

Transcription

1 The Gerontologist Vol. 45, No. 1, Copyright 2005 by The Gerontological Society of America Predictors of Perceptions of Involuntary Retirement Purpose: Retirement is often treated as a voluntary transition, yet selected circumstances can restrict choice in retirement decision processes. We investigated conditions under which retirees perceive their retirement as forced rather than wanted. Methods: Analyses relied on Waves 1 4 of the Health and Retirement Survey (N = 1,160; 572 men and 588 women). Logistic regression models estimated the effects of background factors, choice and restricted choice conditions, and retirement contexts on perceptions of forced retirement. Results: Nearly one third of older workers perceived their retirement as forced. Such forced retirement reflects restricted choice through health limitations, job displacement, and care obligations. Other predictors include marital status, race, assets, benefits, job tenure, and off-time retirement. Implications: Future research should establish personal and policy implications of forced retirement. Programs are needed to help older workers forced into retirement find alternative employment opportunities and to reduce the conditions leading to forced retirement. Key Words: Retirement, Choice, Context, Health, Work, Caregiving This study was funded by The National Institute on Aging under Grant R01 AG13180 (Maximiliane E. Szinovacz, principal investigator). The analyses rely on data from the Health and Retirement Study (HRS) public release and imputed data files (as of December 2003). Selected variables were taken from the data set compiled by RAND (first version). The HRS is managed by the University of Michigan. Detailed information on the data set is available from their Web site ( Address correspondence to Maximiliane E. Szinovacz, Glennan Center for Geriatrics and Gerontology, Eastern Virginia Medical School, Hofheimer Hall, Suite 201, 825 Fairfax Avenue, Norfolk, VA szinovme@evms.edu 1 Glennan Center for Geriatrics and Gerontology, Eastern Virginia Medical School, Norfolk. 2 Polisher Research Institute (formerly Philadelphia Geriatric Center), North Wales, PA. Maximiliane E. Szinovacz, PhD, 1 and Adam Davey, PhD 2 Despite the abolishment of mandatory retirement, retirement transitions are not always voluntary. Earlier research based on the Health and Retirement Surveys (HRS) indicates that a noteworthy proportion of retirees perceive their retirement as forced (Shultz, Morton, & Weckerle, 1998). Such perceptions have been linked to poorer adaptation to the retirement transition (Gallo, Bradley, Siegel, & Kasl, 2000; Shultz et al., 1998), but little is known about the conditions (other than health and unemployment) leading to perceptions of involuntary retirement. Indeed, the organizational and economic literatures model retirement as a voluntary and employee-driven transition (Hanish & Hulin, 1990; Hatcher, 2003). A previous study using crosssectional data from the first HRS wave showed influences of health, financial concerns, and postretirement leisure interests on perceived involuntary retirement (Shultz et al., 1998). Other studies linked unexpected retirement to health, retiree benefits, age, human capital, and industry (Dwyer & Hu, 2000). We examine factors predicting perceptions of socalled forced retirement among retirees by using four waves of the HRS. Theoretical Background and Hypotheses Retirement decisions derive from two factors: choice and motivation (Barnes-Farrell, 2003; Flippen & Tienda, 2000; Szinovacz, 2003). Choice refers to circumstances, mainly disability and labor-market obstacles (Quinn & Burkhauser, 1990), that eliminate older workers ability to remain employed or in their preretirement job; that is, they have no alternative but to retire. Motivation constitutes workers inclination to retire and is typically defined by the benefits costs ratio of retiring. The motivation to retire will be high if benefits (e.g., expected pensions, time for leisure or family activities, decreased job-related stress) outweigh the costs of retiring (e.g., loss or reduction of benefits in cases of early retirement, loss of a valued job, loss of contacts with co-workers). Past research has focused on economic benefits or costs associated with retirement decision processes (Leonesio, 1996), although other factors such as health, spouse s employment, or care obligations have been considered as well (Gustman & Steinmeier, 2002; Honig, 1998). When there is no 36 The Gerontologist

2 Figure 1. Theoretical model of predictors of forced retirement perceptions. choice, motivation becomes irrelevant and retirement decisions are no longer subject to cost benefit considerations (Quinn & Burkhauser). The voluntariness of retirement refers to retirees perceptions of whether retirement was voluntary or involuntary (Beehr, 1986). It derives from choice, motivation, and workers perceived control over the retirement decision. Perceived control differs from choice and indicates whether workers attribute the retirement decision foremost to their own initiative or to environmental contingencies (Heckhausen & Schulz, 1995). For example, some disabled workers may still view their retirement as voluntary because it was initiated by them rather than by their employers, whereas workers leaving employment as a result of caregiving duties may feel retirement was involuntary because a relative s illness is a condition outside their control. Thus, high costs of remaining in the labor force, especially when they derive from conditions beyond the individual s control, can lead to perceptions of involuntary retirement even when the individual had the choice to remain employed. In contrast, lack of choice is neither necessary nor sufficient for perceptions of involuntary retirement, although no choice restrictions likely promote such perceptions. Explanations of voluntariness therefore have to include conditions related to choice in retirement decisions and characteristics related to the propensity to perceive retirement as involuntary, contingent on costs and benefits of retiring and perceived control over the retirement transition. On the basis of these insights, we propose a theoretical model of perceptions of involuntary retirement (Figure 1) that includes background factors, no choice and restricted choice, and retirement contexts that relate to propensities. We assume these propensities to derive from control over life course experiences and events that influence retirement transitions, from retirement contexts, and from age norms and retirement plans. We assume choice factors to override or mediate the influence of propensity factors. Background. Background factors serve as controls in our analyses. We thus present only a brief justification for their inclusion but no hypotheses. Demographic factors such as gender, race, marital status, and fertility influence work histories and retirement decisions. Earlier studies documented pervasive variations in retirement transition processes by gender (Dwyer & Mitchell, 1999; Flippen & Tienda, 2000; Ruhm, 1996) and race or ethnicity (Burr, Massagli, Mutchler, & Pienta, 1996; Couch, 1998; Flippen & Tienda). Many of these influences are indirect, as they are tied to gender or racial or ethnic differences in work histories, health, or human capital (Bound, Schoenbaum, & Waidmann, 1996; Flippen & Tienda; Hayward, Friedman, & Chen, 1996). Marital status has been linked to retirement decisions both directly and indirectly through the Vol. 45, No. 1,

3 influence of selected spousal characteristics. Nonmarried men tend to exit the labor force later or under more adverse circumstances than married men, whereas the opposite holds for women (Brown, Fukunaga, & Wicker, 1996; Williamson & Mc- Namara, 2003). Furthermore, many couples aim for joint retirement, so that one spouse s continued employment reduces the other spouse s motivation to retire (Blau, 1998; Gustman & Steinmeier, 2002; Szinovacz & DeViney, 2000). Human capital and financial status can be tied to retirement transitions directly and indirectly through their impact on work histories and health. Lower educated individuals have more unstable work histories that can lead to unemployment in later life (Flippen & Tienda, 2000; Hipple, 1999). They are also more prone to health problems as they age (Hayward, Friedman, & Chen, 1998). Preretirement financial status defines workers need to maintain employment. Workers anticipating a satisfactory retirement income plan to retire earlier (Adams, 1999), and wealth has been linked to earlier (Beehr, Glazer, Nielson, & Farmer, 2000) and more planned retirement (Dwyer & Hu, 2000). Work contexts can influence perceptions of involuntary retirement either through their association with work-related rewards and costs or their relationship to employment opportunities. Some occupations provide intrinsic rewards that lower the motivation to retire (Hayward et al., 1998). However, such occupations typically involve lengthy careers that render involuntary retirement less likely. Other industries and occupations are more prone to cutbacks and thus affect choice through job displacement (Couch, 1998; Hipple, 1999). Industries and occupations involving strenuous or stressful work can undermine health and ultimately lead to forced retirement (Herzog, House, & Morgan, 1991; Zimmerman, Mitchell, Wister, & Gutman, 2000). Similarly, retirement benefits (such as pension coverage and health insurance) and protection of older workers have been linked to employment in specific industries as well as to firm size and unionization (Fronstin, 1999). Choice Over the Retirement Transition. Two conditions can prompt involuntary retirement under conditions of no choice: disability and job displacement. There is considerable evidence linking retirement transitions to health limitations (Bound, Schoenbaum, Stinebrickner, & Waidmann, 1999; Dwyer & Mitchell, 1999; Hayward et al., 1998). Although some workers have disabilities that allow them to continue work in physically or mentally less demanding jobs, many others are too disabled to continue any kind of employment (Bound et al., 1999). Similarly, job displacement can occasion involuntary retirement (Chan & Stevens, 2002). Some displaced workers find employment elsewhere, but especially older workers are often unable to secure alternative employment opportunities (Hipple, 1999). Our first and second hypotheses are as follows: We expect that both health limitations and job displacement lead to increased perceptions of forced retirement. Some conditions restrict choice though still permitting some flexibility in retirement timing. Foremost among such conditions are roles that conflict with employment. Past research indicates that heavy family care obligations can prompt workers to leave the labor force or to retire (Hayward et al., 1998; Szinovacz & DeViney, 2000; Zimmerman et al., 2000). However, care sometimes involves financial obligations that entice workers to remain in the labor force (Blau, 1998; Ruhm, 1996). To separate these opposing forces, we find it essential to include both family financial and care obligations in the same models. Our third hypothesis is as follows: Because care obligations are typically brought about by events beyond the individual s control, we propose that they will lead to increased perceptions of forced retirement. Retirement contexts impinge on workers motivation to leave their jobs or the labor force and thus define the cost benefit ratio of retirement. We consider three sets of cost benefit factors that may influence perceptions of involuntary retirement: work conditions, benefits, and retirement expectations and timing. Work demands can influence older workers inclination to continue working altogether or to remain in their main career jobs. Older workers may have difficulty performing physically demanding tasks and may be unwilling to stay in stressful or dissatisfying jobs (Williamson, Rinehart, & Black, 1992; Zimmerman et al., 2000). Our fourth hypothesis is as follows: We hypothesize that demanding or stressful work conditions will lead to increased perceptions of forced retirement. Retirement decisions are further influenced by benefit considerations (Gruber & Wise, 1999). Social Security and pension eligibility requirements provide windows of opportunity for retirement transitions and penalize workers who exit outside those windows (Social Security Administration, 2003). Other benefits such as health insurance coverage or early retirement incentives also may constrain choice (Beehr et al., 2000; Blau & Gilleskie, 2003; Fronstin, 1999). Retirement involving reduced benefits probably occurs foremost under conditions of no choice. Otherwise, workers will decide to retire when benefits are optimal. Our fifth hypothesis is as follows: This suggests that benefit receipt will enhance perceptions of voluntary retirement once choice factors are controlled. Life transitions including retirement are subject to norms about the appropriate timing of such transitions (Settersten & Hagestad, 1996). Even though norms guiding retirement timing have become more flexible (Henretta, 1997), retirement 38 The Gerontologist

4 before the normative age may give rise to perceptions of involuntary retirement. Our sixth hypothesis is this: We thus expect that perceptions of involuntary retirement will decrease with retirement age. Retirement may also be perceived as involuntary if it was off time or unplanned from the retiree s perspective. Therefore, our seventh hypothesis is as follows: We hypothesize that perceptions of forced retirement will prevail among those respondents who retired earlier than planned. Methods Sample Our analyses use HRS data (Juster & Suzman, 1995). We pooled individuals who retired between Waves 1 and 2 ( ), Waves 2 and 3 ( ), or Waves 3 and 4 ( ) to achieve a sufficient number of cases for the analyses. Because our dependent variable is the perception of forced retirement, our subsample consists of retirees. We first identified all HRS individuals aged 51 and older who were employed 10 hr or more at baseline and did not define themselves as retirees. We then selected individuals who self-defined as completely or partly retired at Waves 2, 3, or 4 and simultaneously left the labor force. We excluded Time 1 nonemployed or retired respondents because we lack baseline data for these individuals and individuals working fewer than 10 hr at Time 1 because this population may be underemployed or already in postretirement bridge jobs. We also excluded individuals for whom selected missing variables could not be imputed (n = 41 men and 37 women). Respondents who left the labor force but did not selfidentify as retirees were not asked whether their labor force exits were voluntary or desired. We also had to exclude these individuals from the analyses. The final sample consisted of 572 men and 588 women. Some of these men and women come from the same couple (those in which both spouses retired between waves) but others do not. To avoid nonindependence, we performed all analyses separately by gender. To reduce potential attrition bias, we used a logistic regression model to predict the probability that an individual would drop out of a wave, given a wide variety of variables (e.g., age, gender, health, wellbeing, socioeconomic status, or having died between waves). We divided original sampling weights by the probability of nonresponse (i.e., Heckman s k) and used these new weights in all analyses. Results did not depend on the particular weighting strategy used, because our models also include most of the important variables predicting nonresponse. We adjust for potential bias that is due to missing data through the multiple imputation of missing data (Little & Rubin, 1989; Schafer, 1997). We preserved data already imputed by HRS or RAND. Vol. 45, No. 1, Measures The dependent variable is the perception of forced retirement. Retirees were asked the following question: Was retirement something you wanted to do or something you felt you were forced into? The answer categories were forced, wanted to, and part forced, part wanted. The last category was chosen by few respondents (n = 51 men and 62 women) because it was not implied in the question. Sensitivity analyses revealed that the part wanted part forced group differs from retirees who claim only forced and from retirees who claim only wanted retirement, and it should not be combined with either category. We therefore focused on forced (=1) and wanted (=0) retirement, excluding the middle group. When more HRS individuals have retired and the middle category reaches a larger size, multinomial analyses should be conducted to explore which conditions give rise to perceptions of partly forced retirement. Demographic Background. We performed all analyses separately by gender. We subdivided race or ethnicity into three dummy variables (Black, Hispanic, or other, with Whites as the reference). For marital status we created three dummy variables (continuously not married, started a new relationship between waves, and ended a relationship), with those continuously in the same relationship as the reference. We treated nonmarried individuals with partners as married. We dummy-coded fertility into those who have any children (1) versus those who do not (0). Human Capital and Finances. The variables we considered under this category were education, income and assets, spouse s employment, and dependents. We coded education in years. For financial status we included total net household assets at baseline, change in assets between waves, and earnings at baseline. We recoded baseline assets into $1,000 groups and truncated them ( 100,000 = 100,000; 1,000,000 = 1,000,000). We also recoded change in assets into $1,000 groups (truncated at 6 $500,000). We logged earnings. Change in household income between waves had no effect on perceptions of forced retirement, so we dropped this variable from the final models. We include two variables assessing financial dependents: any children aged 18 or younger in the household (1=yes, 0=no) and provision of at least one half of any person s financial support (1 = yes; 0 = no), both assessed at Time 2. We coded spouses employment status as 1 = not employed and 0 = employed at the time of the respondent s retirement. Work Context. We initially considered industry, occupation, firm size, and whether the respondent was covered by a union or employee-association

5 contract. Preliminary analyses revealed that only a few industries and none of the occupation categories were significantly related to perceptions of forced retirement, and effects of industry varied by gender. We thus retained employment in mining construction (this is one category in the HRS), public administration, or sales for men and in agriculture forestry fishing (one category in the HRS) for women, with others comprising the reference category. Firm size was not significant and we dropped it from the final models. Union coverage was coded 1 = yes and 0 = no. Choice. Choice is defined by health and job loss. For health, we included number of limitations in activities of daily living (ADLs) at Time 1, change in ADLs between waves, change in self-reported health between waves (self-reported health at Time 1 had no influence on perceptions of forced retirement and was dropped from the final models). We measured selfrated health with a single item: Would you say your health is..? (1 = poor to 5 = excellent). We assessed ADL limitations for 16 tasks such as walking several blocks, walking across a room, getting up from a chair, or getting in and out of bed; we coded them as 1 (any difficulty) or 0 (no difficulty). Scores could range from 0 to 16. We assessed changes in health with change scores. For self-rated health, positive change scores indicate increases in self-rated health; for ADLs, positive change scores reflect increases in limitations. We also included whether respondents left their last job because of health reasons (1 = yes; 0 = no). Job loss was derived from the question why respondents left their last job. Those indicating that they were laid off or let go or that their business closed were coded 1, those who left for other reasons served as the reference (0). Restricted Choice. We consider care obligations as factors restricting choice over the retirement transition. We assessed care for parents or parentsin-law and care for grandchildren by asking respondents whether they provided such care since the last wave, and, if so, how many hours they spent caregiving. Few respondents were involved in parent care, so we used a dummy variable (1 = provides care; 0 = no care). For grandchildren, we used four dummy variables: respondents with dependent (aged 18 or younger) grandchildren in the household (1 = yes; 0 = no); respondents without grandchildren in the household providing care between 1 and 400 hr; those without grandchildren in the household who spent over 400 hr caregiving; and those without grandchildren. Those with grandchildren who provided no care served as reference. Lacking direct questions about care for spouses, we use spouses ADLs at Time 1 (coded the same way as for respondents). Change in spouses ADLs had no effect on perceptions of forced retirement, and we dropped it in the final models. The final variable included under restricted choice is whether respondents left their last job because of care obligations for family members (1 = yes; 0 = no). Only 8 men reported this reason for leaving their job, so we included this variable only in models for women. Retirement Contexts. We measured work demands and attachment with three variables: years worked in the last job (coded in years), whether respondents left their last job because they found other work or just quit (1 = yes; 0 = no), and whether the last job was stressful. We derived job stress from two items: My job requires me to do more difficult things than it used to and My job involves a lot of stress (1 = strongly disagree; 4 = strongly agree). We combined these two items on the basis of preliminary factor analyses using a larger set of questions pertaining to job characteristics. Scores could range from 2 to 8. The second set of variables refers to benefits, including own and spouse s pension coverage and health insurance. We assessed pension coverage with three dummy variables: had a pension at Time 1 and did not gain additional pension income after retirement; covered by a pension plan at Time 1 but did not receive a pension after retirement; and started to receive a pension after retirement. Those not covered by a pension plan and not receiving a pension after retirement serve as the reference. We also included whether the spouse was covered by a pension plan at Time 1 (1 = yes; 0 = no). We distinguished between those covered at their last jobs by health insurance that would not continue into retirement and those covered by health insurance that also covers retirees. Respondents without health insurance served as the reference. Whether respondents were covered by spouses health insurance, regardless of their own, and employer retirement incentives had no effect on perceptions of forced retirement, and we did not retain them in the final models. The final set of retirement context variables pertains to the timing of retirement. We included respondents age at Time 1 and whether their retirement was earlier than expected (based on the discrepancy between expected and actual retirement age; 1 = earlier and 0 = at expected time or later). Controls. Control variables included the wave from which responses were drawn and whether respondents self-defined as completely or partly retired (partly = 1; completely = 0). Means and standard deviations for all variables are shown in Table 1. Analyses We based analyses on logistic regressions, weighted to adjust for nonresponse and attrition. 40 The Gerontologist

6 Table 1. Means and Standard Deviations of Dependent and Independent Variables, by Gender Variable Men (n ¼ 567) Women (n ¼ 588) M SD M SD Dependent variable Perceived retirement was forced Demographic background Black Hispanic Other race Not married New partner Marriage ended Has children Partner not employed Human capital and finances Education Net assets at Time Earnings at Time Change in assets Has financial dependents Dependent children in household Work context Industry mining/ construction Industry public administration Industry sales Industry agriculture/ forestry/fishing Covered by union contract No-choice factors No. of ADLs at Time Change in ADLs Change in selfreported health Stopped work for health reasons Lost last job Care obligations Has dependent grandchildren in household Care for parents at Time Care for grandchildren, 400 hr Care for grandchildren 400 hr Has no grandchildren Spouse s ADLs at Time Stopped work for family care reasons Vol. 45, No. 1, Variable Table 1. (Continued) Men (n ¼ 567) Women (n ¼ 588) M SD M SD Retirement contexts Started pension receipt Had pension at Time 1; no new pension Enrolled in pension plan; no new pension Covered by health insurance while working Covered by health insurance in retirement Covered by spouse s health insurance Job was stressful Quit last job Spouse enrolled in pension plan at Time Years worked in last job Timing Age at Time Perceived early retirement Controls Wave Wave Partly retired Notes: Information is based on analyses from Waves 1 4 of the Health and Retirement Surveys. Model 1 excludes all no-choice factors, Model 2 excludes job displacement. Model 3 excludes health-related no-choice factors, and Model 4 includes all no-choice factors. Reference categories are as follows: race, White; marital status, married with employed partner; industry, other industry; pensions, has no pension; grandchildren for men, no grandchildren in household; grandchildren to women; provides no care to grandchildren; date from Wave 4; otherwise reference is 0 (no). ADL = limitations in activities of daily living. The HRS uses a complex survey sampling design (e.g., randomly selecting geographic regions and then sampling individuals within them). This design artificially reduces the standard errors. We adjusted for sampling design effects (partitioning variance within and between primary sampling units) to produce valid standard error estimates. As implemented in Stata (2003), these estimates are also robust to minor violations of the heteroskedasticity assumption. Our analytic strategy derives from the theoretical model shown in Figure 1 as well as from considerations concerning parsimony. We estimated initial models (not shown) that included all baseline

7 variables and, separately, each set of the restricted choice and retirement context predictors. In each case, we estimated models that excluded and included health or job displacement. We retained significant or near-significant predictors for the final estimates presented in Tables 2 and 3. Eliminated variables did not influence effects of other predictors. As in the initial analyses, we ran models in Tables 2 and 3 separately by gender. We also presented results with and without the health and job displacement variables to determine whether predictors influence perceptions of forced retirement independently or through their associations with health limitations or job displacement. Results Results are presented in Tables 2 (men) and 3 (women). We first address specific hypotheses and then discuss the influence of background variables and controls. Hypotheses 1 and 2 refer to the effects of choice factors on perceptions of forced retirement. The analyses support both hypotheses. Health problems and job displacement are positively associated with perceptions of forced retirement. Among men, we find significant independent effects of ADLs, changes in ADLs, and self-reported health, as well as stopping work for health reasons (Table 2). For women, the main health predictor is stopping work for health reasons. Changes in ADLs are significant when job displacement is not in the model (Model 2) but are no longer significant once job displacement is controlled (Model 4). We expected care obligations to increase perceptions of forced retirement (Hypothesis 3). Preliminary analyses (not shown) indicated that effects of care obligations differ by gender. Tables 2 and 3 reflect these gender differences and include only care obligations with a significant or near significant (p,.10) effect. Men with dependent grandchildren in the household and those who cared for their parents in the time period prior to the Time 2 interview are more prone to view retirement as forced. Both effects are stronger when health and job displacement are in the equation (Model 4). This suggests that care obligations explain perceptions of forced retirement once the variability attributable to choice factors is removed. Among women, three care variables relate significantly to perceptions of forced retirement: care for grandchildren outside the household, spouse s ADLs, and stopping work for family care reasons. Contrary to our hypothesis, women who spent 400 hr or more caring for grandchildren are less likely to perceive their retirement as forced, whereas women whose spouses have more ADLs and those stopping work for family care reasons are more likely to view their retirement as forced. Stopping work for family care reasons is only significant in Model 4, attributable to a confounding effect of negative relationships among stopping work for family reasons and stopping work for either health reasons or as a result of job displacement. Thus, Hypothesis 3 receives partial support. Our hypothesis that negative work conditions promote perceptions of forced retirement, Hypothesis 4, receives only very limited support. None of the work conditions predicted perceptions of forced retirement for men, and we did not include these variables in the final models. Women s perceptions of forced retirement are positively related to stressful work conditions but only when health factors are not in the model (Table 3, Models 1 and 3). Women are less likely to perceive retirement as forced the longer they worked in their last job. Benefit receipt is tied to perceptions of forced retirement (Hypothesis 5), but not always as we predicted. Men tend to perceive retirement as forced if they did not get additional pension benefits upon retirement. They are less likely to perceive their retirement as forced when they are covered by their spouse s health insurance or when their spouse was enrolled in a pension plan at Time 1. Women s perceptions of forced retirement are not affected by pensions. They are less likely to perceive their retirement as forced if they were covered by health insurance while working or after retirement. Age per se affects only men s perceptions of forced retirement (Hypothesis 6), and this becomes nonsignificant once job displacement and health factors are included (Table 2, Model 4), indicating that the age effect can be attributed to the combined effects of job loss and poor health. There is clear support for our final hypothesis, Hypothesis 7, that perceptions of forced retirement reflect lack of correspondence between planned and actual timing of retirement for men and women. Diverse background factors also affect perceptions of forced retirement. We find significant influences of marital relationship and fertility factors as well as some race differences. African American men are less likely than Whites to perceive their retirement as forced, whereas the opposite holds for Black women. Unmarried individuals of both genders and men who remarried between waves are less prone to view their retirement as forced. In contrast, women whose marriages ended between waves are more likely to perceive their retirement as forced. Having children reduces perceptions of forced retirement among men and women. Concerning human capital and finances, data suggest a tendency toward perceptions of voluntary retirement among those with greater educational or financial assets. Men are less likely to view retirement as forced the better educated they are and the higher their earnings, net assets, and increase in assets between waves. However, most of these relationships become nonsignificant (excepting baseline net assets) once health and job loss are in the 42 The Gerontologist

8 Table 2. Logistic Regression Models of Forced Retirement for Men (n = 567) Variable Model 1 Model 2 Model 3 Model 4 Demographic background Black 1.243** 1.303* 1.463** 2.007** Hispanic Other race Not married ** * New partner * * Marriage ended Has children 1.121** 1.540** 1.232** 1.702** Partner not employed Human capital and finances Education 0.130** * Net assets at Time ** 0.002* 0.003** 0.002* Earnings at Time ** ** Change in assets 0.002** * Dependent children in household * Has financial dependents ** Work context Industry mining/construction 0.743* Industry public administration 1.937* Industry sales 1.076* * Covered by union contract 0.653* * No-choice factors No. of ADLs at Time ** 0.399** Change in ADLs 0.273** 0.274* Change in self-reported health 0.338** 0.447** Stopped work for health reasons 2.304** 3.273** Lost last job 1.721** 3.148** Care obligations Has dependent grandchildren in household 1.132* 1.620* 1.287* 2.344* Care for parents at Time * Retirement contexts Started pension receipt Had pension at Time 1; no new pension 1.459* 1.219* 1.671* 1.629** Enrolled in pension plan; no new pension Covered by health insurance while working Covered by health insurance in retirement Covered by spouse s health insurance * * Spouse enrolled in pension plan at Time * * Years worked in last job Timing Age at Time ** 0.145** 0.231** Perceived early retirement 1.136** 0.904** 1.103** 0.767* Controls Wave Wave Partly retired ** Constant ** 9.355** ** F 2.888* 5.154** 2.56* 4.512* ndf ddf Notes: Information is based on analyses from Waves 1 4 of the Health and Retirement Surveys. Model 1 excludes all no-choice factors, Model 2 excludes job displacement, Model 3 excludes health-related no-choice factors, and Model 4 includes all no-choice factors. Reference categories are as follows: race, White; marital status, married with employed partner; industry, other industry; pensions, has no pension; grandchildren, no grandchildren in household; date from Wave 4; otherwise reference is 0 (no). ADL = limitations in activities of daily living. *p,.05; ** p,.01. Vol. 45, No. 1,

9 Table 3. Logistic Regression Models of Forced Retirement for Women (n = 588) Variable Model 1 Model 2 Model 3 Model 4 Demographic background Black 0.684* Hispanic Other race Not married 1.146** 1.732** 1.118** 2.039** New partner Marriage ended ** ** Has children * * Partner not employed 0.916* 1.071** 0.888* 1.076** Human capital and finances Education Net assets at Time ** 0.001* 0.002** 0.002* Earnings at Time Change in assets * Dependent children in household Has financial dependents Work context Industry agriculture/forestry/fishing Covered by union contract No-choice factors No. of ADLs at Time Change in ADLs 0.160* Change in self-reported health Stopped work for health reasons 2.057** 3.074** Lost last job 1.169** 2.621** Care obligations Care for grandchildren, 400 hr Care for grandchildren 400 hr 1.261* 1.685* 1.254* 1.736* Has no grandchildren Spouse s ADLs at Time * * Stopped work for family care reasons * Retirement contexts Started pension receipt Had pension at Time 1; no new pension Enrolled in pension plan; no new pension Covered by health insurance while working 0.710* * Covered by health insurance in retirement 1.163** 1.134** 1.175** 1.323** Job was stressful 0.225* * Quit last job Years worked in last job 0.072** 0.067** 0.071** 0.061* Timing Age at Time Perceived early retirement 1.254** 1.068** 1.196** 0.823* Controls Wave * * Wave * * Partly retired 0.695* 1.165** Constant F 5.464** 8.714** 5.666** 10.94** ndf ddf Notes: Information is based on analyses from Waves 1 4 of the Health and Retirement Surveys. Model 1 excludes all no-choice factors, Model 2 excludes job displacement, Model 3 excludes health-related no-choice factors, and Model 4 includes all no-choice factors. Reference categories are as follows: race, White; marital status, married with employed partner; industry, other industry; pensions, has no pension; grandchildren, provides no care to grandchildren; date from Wave 4; otherwise refernce is 0 (no). ADL = limitations in activities of daily living. * p,.05; ** p, The Gerontologist

10 model (Table 2, Models 1 and 4). This suggests that human capital and finances act primarily through effects on health and job displacement. Only women s net assets at baseline predict perceptions of forced retirement, and this effect holds in all models (Table 3, Models 1 4). Partner s employment has no influence on men s perceptions of forced retirement, whereas women with nonemployed spouses are less likely to perceive their retirement as forced. The existence of financial dependents reduces men s perceptions of forced retirement once job loss and health are included (Table 2, Model 4). Most of the work context variables (industry, occupation) had no influence in preliminary analyses. Among women, neither employment in agriculture nor coverage by a union contract is significant in the final models. However, the effect of agriculture is quite strong and only lacks significance because very few women were employed in agriculture. Men employed in mining and sales are more likely and those in public administration less likely to perceive their retirement as forced. Union coverage seems to exert some protective effect against forced retirement for men. None of these effects retain significance once job loss and health are controlled. Self-definition as partly retired and retirement wave have little influence on men s perceptions of forced retirement. Partly retired women tend to view their retirement as forced but only when health and job loss are not controlled. In contrast, retirement wave remains significant in Model 4, suggesting that earlier female retirees as a group were more likely to experience forced retirement. Vol. 45, No. 1, Discussion Our discussion focuses on the seven tested hypotheses. In line with the first two hypotheses and previous research (Bound et al., 1999; Chan & Stevens, 2002; Hipple, 1999), we find that leaving work as a result of illness or job displacement is a primary predictor for perceptions of forced retirement. Other health measures predicted forced retirement only for men, suggesting that some men with health limitations do not attribute their retirement to poor health but nevertheless perceive the resulting transition as forced. The data provide some support that care obligations can restrict choice over the retirement decision (Hypothesis 3). Among a small group of men, forced retirement is precipitated by grandchild or parent care, whereas women s retirement seems more responsive to general family care obligations (stopping work for family care reasons) and to spouse s illness. However, the effect for spouse s ADLs becomes nonsignificant when women s health variables are in the model, a finding that is partially attributable to comorbidity between spouses. Women s health also may be directly affected by spousal care (Yee & Schulz, 2000), and our data show that spouse s ADLs and change in women s self-reported health are negatively correlated (r =.14, p,.01). In contrast, women providing substantial care to grandchildren outside the household tend to view their retirement as voluntary. This finding contrasts with research indicating stresses associated with grandchild care (Lee, Colditz, Berkman, & Kawachi, 2003). One reason for this discrepancy is probably the amount of care provided by our population compared with that addressed in earlier studies. Hypotheses 4 and 5 addressed the importance of cost and benefit factors associated with work and retirement. We found very limited evidence for effects of work conditions. Beyond their association with health-related retirement (job stress among women), negative work conditions do not appear to play a significant role in perceptions of forced retirement. Effects of benefit factors were likewise more complex than anticipated. Men were more likely to perceive their retirement as forced if they did not receive additional pension benefits in retirement. These men may have been in bridge jobs they either lost or found undesirable. In contrast, men covered by spouse s health insurance or whose spouses had a pension plan were less likely to perceive their retirement as forced, suggesting that spouses benefits offer men greater leeway in their retirement transitions. For women, years in the last job and being covered by health insurance at work or into retirement reduce perceptions of forced retirement. Apparently, long-term employment with the same employer and in a benefit-friendly environment protects women against job loss and provides a desirable retirement context that reduces feelings of forced retirement. Our last two hypotheses refer to retirement timing, both chronologically (relative to social norms) and personally (in terms of expected retirement age). Men who retired at a younger age were more likely to see their retirement as forced, but these effects were completely accounted for by variables reflecting lack of choice. Personal timing effects are much stronger. Men and women retiring earlier than expected were more than twice as likely as on-time retirees to perceive their retirement as forced. This demonstrates that control over life transitions and the implementation of retirement plans contribute to positive retirement experiences. We also expected race and ethnicity to affect perceptions of involuntary retirement, primarily through association with health problems or job loss. However, African American men were more likely than Whites to view their retirement as voluntary, even when effects of health or job displacement were controlled. One explanation relates to the filter pattern in the HRS. Forced retirement was only asked of individuals self-defining as retired. Past research has shown that African

11 Americans may reject a self-definition as retiree if labor force withdrawal occurred under adverse circumstances such as illness (Gibson, 1991). Thus involuntarily retired Black men may have been omitted from the analyses. In contrast, Black women were more prone than White women to view their retirement as forced, and this can be attributed to poor health. These results highlight race and gender differences in retirement definition processes (Ekerdt & DeViney, 1990; Szinovacz & DeViney, 1999) and the importance of considering such differences in the design of retirement surveys. The results pertaining to marital status and partner s employment were not predicted. To explain these findings, we find it important to keep in mind that our reference category consists of married retirees whose partners are employed. Thus, the negative coefficients for being unmarried and for having a nonemployed partner suggest that it is married women with employed spouses who view their retirement as forced. Past research has shown that couples prefer joint retirement and that separate retirements of spouses can often be attributed to averse retirement circumstances such as partner s illness or job displacement (Henretta, O Rand, & Chan, 1993; Szinovacz, 2003). Men with new partners perceive retirement more as voluntary once effects of health are controlled. Economic stability may enable men to remarry and also preclude unwanted retirement. Among women, loss of partner enhances perceptions of forced retirement. It is conceivable that wives left their employment to attend to dying spouses or that widowhood after retirement could give rise to more negative perceptions of the retirement transition. This study has several limitations. As is typical for research relying on large surveys, we lack in-depth information on the intricate considerations that enter retirement transitions and their perceptions. We also lack indicators for specific concepts (e.g., we use spouse s ADLs as the indicator for spousal care). Thus, our findings and interpretations require corroboration through further studies. Furthermore, the analyses pertain to perceptions of forced retirement and should not be confounded with no-choice retirement. As indicated in our theoretical framework, predictors of perceptions of forced retirement will differ from predictors of no-choice retirement. About one third of our sample of retirees perceived their retirement as forced. This result casts doubt on previous models treating retirement as a voluntary transition (Hanish & Hulin, 1990; Hatcher, 2003). If cost benefit considerations require some choice over the retirement transition (Quinn & Burkhauser, 1990), then models of retirement decisions have to differentiate between voluntary and involuntary retirees. Although health limitations and job loss account for most of these labor force exits, other circumstances, especially care obligations and the realization of retirement plans, contribute as well. This has important implications for policy makers. Planned increases in the age of Social Security eligibility are meant to entice workers to delay retirement. However, a noteworthy proportion of older workers may be unable to do so. Indeed, our data indicate that those perceiving their retirement as forced are younger than those who wanted to retire (baseline age differences are 2.0 years for men and 1.3 for women). These workers may either face particular financial hardship until reaching eligibility age or be eligible for other programs (disability or unemployment), thus reducing savings expected from changes in eligibility rules. The personal and policy impact of forced retirement requires further research as well as program and policy initiatives. Researchers should address whether and to what extent benefit considerations enter into the retirement decisions of those forced into retirement compared with those who wanted to retire. It will also be important to explore the financial circumstances of involuntary retirees, both during the time span between retirement and Social Security or pension eligibility and in the aftermath of reaching eligibility as benefits may be permanently reduced as a result of early retirement. Programs are needed to help somewhat disabled and displaced older workers find employment opportunities after forced retirement. Policies that place greater emphasis on occupational and preventative health care may protect older workers from untimely retirement and ensure that expected savings from the postponement of Social Security eligibility can be realized. References Adams, G. (1999). Career-related variables and planned retirement age: An extension of Beehr s model. Journal of Vocational Behavior, 55, Barnes-Farrell, J. L. (2003). Beyond health and wealth: Attitudinal and other influences on retirement decision-making. In G. A. Adams & T. A. Beehr (Eds.), Retirement. Reasons, processes, and results (pp ). New York: Springer. Beehr, T. A. (1986). The process of retirement: A review and recommendations for future investigation. Personnel Psychology, 39, Beehr, T. A., Glazer, S., Nielson, N. L., & Farmer, S. J. (2000). Work and nonwork predictors of employees retirement ages. Journal of Vocational Behavior, 57, Blau, D. M. (1998). Labor force dynamics of older married couples. Journal of Labor Economics, 16, Blau, D. M., & Gilleskie, D. B. (2003). The role of retiree health insurance in the employment behavior of older men (Working Paper No ). Cambridge, MA: National Bureau of Economic Research. Bound, J., Schoenbaum, M., Stinebrickner, T. R., & Waidmann, T. (1999). The dynamic effects of health on the labor force transitions of older workers. Labour Economics, 6, Bound, J., Schoenbaum, M., & Waidmann, T. (1996). Race differences in labor force attachment and disability status. The Gerontologist, 36, Brown, M. T., Fukunaga, C., & Wicker, L. (1996). Annual review, : Social class, work, and retirement behavior. Journal of Vocational Behavior, 49, Burr, J. A., Massagli, M. P., Mutchler, J. E., & Pienta, A. M. (1996). Labor force transition among older African-American and White men. Social Forces, 74, Chan, S., & Stevens, A. H. (2002). How does job loss affect the timing of retirement? (Working Paper No. 8780). Cambridge, MA: National Bureau of Economic Research. 46 The Gerontologist