A MICRO-LEVEL ANALYSIS OF RECENT INCREASES IN LABOR FORCE PARTICIPATION AMONG OLDER WORKERS Kevin E. Cahill, Michael D. Giandrea, and Joseph F. Quinn* CRR WP 2008-8 Released: February 2008 Draft Submitted: January 2008 Center for Retirement Research at Boston College Hovey House 140 Commonwealth Avenue Chestnut Hill, MA 02467 Tel: 617-552-1762 Fax: 617-552-0191 http://www.bc.edu/crr * Kevin E. Cahill is an economic consultant at Analysis Group in Boston. Michael D. Giandrea is a research economist at the U.S. Bureau of Labor Statistics. Joseph F. Quinn is a Professor of Economics at Boston College. The research reported herein was performed pursuant to a grant from the U.S. Social Security Administration (SSA) funded as part of the Retirement Research Consortium. The opinions and conclusions expressed are solely those of the authors and should not be construed as representing the opinions or policy of SSA, any agency of the Federal Government, Analysis Group, the U.S. Bureau of Labor Statistics, or Boston College. © 2008, by Kevin E. Cahill, Michael D. Giandrea, and Joseph F. Quinn. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source. About the Center for Retirement Research The Center for Retirement Research at Boston College, part of a consortium that includes parallel centers at the University of Michigan and the National Bureau of Economic Research, was established in 1998 through a grant from the Social Security Administration. The Center’s mission is to produce first-class research and forge a strong link between the academic community and decision makers in the public and private sectors around an issue of critical importance to the nation’s future. To achieve this mission, the Center sponsors a wide variety of research projects, transmits new findings to a broad audience, trains new scholars, and broadens access to valuable data sources. Center for Retirement Research at Boston College Hovey House 140 Commonwealth Avenue Chestnut Hill, MA 02467 phone: 617-552-1762 fax: 617-552-0191 e-mail: crr@bc.edu www.bc.edu/crr Affiliated Institutions: American Enterprise Institute The Brookings Institution Center for Strategic and International Studies Massachusetts Institute of Technology Syracuse University Urban Institute Abstract Aggregate data reveal a sizable increase in labor force participation rates since 2000 among American workers on the cusp of retirement, reverting back to levels for older men not seen since the 1970s. While these aggregate numbers are useful in that they document overall trends, they do not elucidate the reasons behind workers’ decisions. The Health and Retirement Study (HRS), a nationally-representative, longitudinal survey of older Americans that spans 1992 to 2004, provides micro-level data regarding these retirement trends. Moreover, the HRS contains detailed information about the types of jobs older Americans are taking (e.g., full-time versus part-time, self-employed versus wage-and-salary, low-paying versus high-paying, blue collar versus white collar). This study capitalizes on the richness of the HRS data and explores labor force determinants and outcomes of older Americans, with an emphasis on retirees' choices in recent years. We present a cross-sectional and longitudinal description of the financial, health, and employment situation of older Americans. We then explore retirement determinants using multinomial logistic regression to model gradual retirement and logistic and OLS regression to model the work-leisure (whether to work) and hours intensity (how much to work) decisions of older workers. Evidence suggests that the majority of older Americans retire gradually, in stages, and that younger retirees continue to respond to financial incentives just as their predecessors did. In addition, the retirement decisions of younger and middle-aged retirees appear similar in the face of macro-level changes in the early part of this decade. I. Introduction The 2001 recession in the United States was unique in that older workers experienced increases in labor force participation rates while other workers’ rates followed the typical pattern during a recession and declined.1 Older workers’ choices during this recessionary period were even more notable because their decisions reversed a broader trend of ever-earlier retirements that bottomed out in the mid-80s.2 In addition, today’s retirees are changing the way older workers exit the labor force. Traditional one- time, permanent retirements appear to be the exception rather than the rule, as older workers increasingly change jobs later in life or reenter the labor force after “retiring.”3 Why are so many of today’s older Americans breaking from the traditional retirement pattern? The answer to the pro-work mindset of many of today’s older Americans is likely a reflection of many factors, for both labor supply and demand. People are living longer, are healthier, and have higher levels of formal education compared to earlier generations. Jobs are also less physically demanding than in the past, as the economy shifts away from manufacturing occupations towards service ones. At the same time, a strong labor market, like that of the 1990s and mid-2000s, provides older workers with many job opportunities. These changes have allowed older workers to remain productive well beyond their late 50s and early 60s. Many of the financial incentives surrounding retirement have changed as well. Defined-benefit pension plans that offer a set annuity payment upon retirement are less common in today’s private sector and many existing defined-benefit plans are being converted to cash balance plans or replaced with defined-contribution plans managed by the worker.4 Social Security, the bedrock of financial security late in life, is facing financial strain and will likely provide lower replacement rates than in the past.5 Finally, private saving, the third pillar of retirement income, is currently near record low rates.6 As a result, today’s retirees have experienced a general shift towards a “do-it-yourself” 1 Eschtruth and Gemus (2002). 2 Quinn (1999), Purcell (2006). 3 Cahill, Giandrea, and Quinn (2006). 4 Cahill and Soto (2003). 5 Munnell (2003). 6 National Income and Product Accounts. approach to retirement, and are now in charge of their retirement finances more than at any time in the post-war era. While these changes will undoubtedly impact retirement patterns in the long run, they do not, in and of themselves, explain why labor force participation rates among older workers jumped so abruptly in the early part of this decade. For insight regarding this question, we examine how long-term changes have made retirees vulnerable to short-run market forces. Perhaps it took a shock in the financial markets, such as the 2001 recession, to uncover the impact of the “do-it-yourself” approach. Seen this way, the key to understanding workers’ retirement decisions in recent years is to understand the interaction between long-run incentives and short-term market fluctuations. This interaction may explain why the early retirement trend subsided in the mid-80s and 90s. This interaction may also explain why increases in labor force participation among older workers subsided most recently as the economy improved. Under this hypothesis, we might expect the timing of retirement to be cyclical, as workers’ expectations and plans are continuously updated in response to the changing state of their financial position. This is a fundamental shift from the past. Previously, the timing of retirement was largely immune to changes in market conditions, as investment risk was borne by the federal government and an individual’s employer. With the advent of 401(k)s and with the extension of Social Security’s Normal Retirement Age (NRA) from 65 to 66, and eventually to 67, a worker’s retirement benefit now depends on the current state of the market as cyclical effects help determine the stock of retirement assets. One possible implication, going forward, is that more older workers can be expected to postpone retirement or reenter the labor market during a recession, and then retreat from the labor force during a boom. Aggregate data on work force participation are consistent with this explanation; however, micro-level data are required to examine this hypothesis. We analyze data from the Health and Retirement Study (HRS), a nationally-representative dataset of two cohorts of older Americans.7 Detailed information on demographics, health status, work history, income, wealth, and more are available for each respondent, making the HRS an ideal data set for this study. 7 Juster and Suzman (1995). 2 This paper is structured as follows. Section II provides some background on retirement trends and exit patterns, with a focus on recent developments. Sections III and IV outline the estimation strategy for how we plan to identify the key determinants of retirement outcomes, especially in recent years, and the data we use for our analyses. We present our results in Section V and comment on the implications of our research in Section VI. We conclude that future retirees are likely to be much more flexible with respect to their eventual labor force exit compared to retirees in the past. II. Background Labor force participation rates among older workers have risen in recent years, both among men, whose rates were relatively constant from the mid-1980s through much of the 1990s, and among women, whose rates have increased steadily since the mid- 1980s. Men aged 60 to 64 years experienced an increase in labor force participation from 53 percent in 1995 to 57 percent in 2004, a 7 percent increase over the decade. Men aged 65 to 69 exhibited an even larger increase over the same time period, from 27 percent to 33 percent. The story is similar for women, but begins at a younger age. Women aged 55 to 59 had an increase in labor force participation from 60 percent to 65 percent, while those aged 60 to 64 and aged 65 to 69 experienced increases of 18 percent and 29 percent, respectively.8 Several recent studies have noted this trend, as well as the diverse patterns by which older workers exit the labor force. Cahill et al. (2006), for example, examined transitions from career jobs and found that more than one-half of older workers with full- time career jobs worked on another job (a “bridge job”) prior to complete labor force withdrawal. Moreover, not only were bridge jobs common among older cohorts of workers, but diverse patterns were even seen among the youngest of retirees (i.e., those age 57-62 in 2004). Maestas (2004), Quinn (1999), and Ruhm (1990) have also examined the transition that workers take when exiting the labor force. They have described an increasingly “blurred” retirement process where employees gradually 8 U.S. Bureau of Labor Statistics, 2005. 3 transition from career jobs to a retirement that sometimes includes reentry into the labor market before a permanent exit.9 Research and anecdotal evidence provide insight into the key determinants of these trends. The retirement literature has shown that financial incentives and demographic and socio-economic characteristics matter.10 The importance of other factors has also been examined. For example, the long-term shift from defined-benefit pension plans that provide a regular benefit amount in retirement to employee-controlled defined-contribution plans has been shown to influence the timing of retirement.11 Cyclical economy-wide factors, such as stock market performance, have recently been proposed as an important determinant of labor force participation. Eschtruth and Gemus (2002) outlined the way older workers increased participation in the labor force during the most recent recession. Gardner and Orszag (2003) took a micro-level approach based on a survey of almost 4,500 individuals aged 50-64 in the U.K. and found that one-quarter planned to delay their expected retirement date because of the stock market decline of 2000. Using a micro-level approach, Coile and Levine (2006) investigated the relationship between changes in stock market valuations and retirement activity using the HRS, the Current Population Survey (CPS), and the Survey of Consumer Finances (SCF). They estimated reduced form equations to determine whether the growth of stock market values in the late 1990s and the subsequent decrease in the early 2000s impacted retirement and labor force re-entry behavior of older workers. The authors found that rates of retirement and re-entry were not significantly different between older workers with larger stock portfolios and those without. Coronado and Perozek (2003) used HRS data to investigate the effects of wealth on leisure decisions later in life. Somewhat contrary to Coile and Levine, they found that the unexpected stock market gains of the late 1990s led workers who held stocks to retire, on average, about seven months earlier than non-stockholders. 9 Mutchler, Burr, Pienta, and Massagli (1997). 10 For example, see Gruber and Madrian (1995), Mutchler et al. (1997), Quinn (1999), and Coile (2004). 11 Munnell, Cahill, Jivan (2003); Friedberg and Webb (2005). 4 This paper builds upon the micro-level research and concentrates on the determinants of the labor market participation decisions of older workers using two cohorts of retirees spanning 1992 to 2004. III. Methodology The purpose of this analysis is to assess how the shift towards a do-it-yourself approach to retirement has altered decisions regarding the timing of labor force withdrawal, bridge job behavior, and labor force reentry. In particular, we examine the influence of specific components of the do-it-yourself approach, such as the role of private pensions, health insurance, and savings. We hypothesize that workers are now more sensitive to short-term market conditions as the size of potential retirees’ non- annuitized retirement nest eggs rises and falls with financial markets. To evaluate this hypothesis, we first make a distinction between two underlying causes for differences in outcomes – changes in determinants and changes in impacts. As retirement inputs change, such as the switch from defined-benefit to defined-contribution plans over the past two decades, retirement outcomes change. The underlying cause of the change (i.e., inputs) does not necessarily imply a change in retirement decision making. A more subtle change, one that is addressed in part here by examining cohort differences, is whether behavioral changes have taken place, and if known determinants of retirement now influence retirement decisions differently than in the past. Our sample consists of the HRS Core, born between 1931 and 1941, and the HRS War Babies, born between 1942 and 1947, with multiple data points over time within each cohort. We model the retirement decision in three ways, each making use of the longitudinal nature of the HRS and each providing a different insight into the choice between work and retirement. One way to model workforce decisions is to make use of the fact that each respondent contributes multiple observations, one from each survey. This approach allows us to control for individual-specific factors and time effects. Using person-year observations, we model retirement as a series of decisions. The respondent first decides whether to stay working and, if so, whether to remain on a FTC job or to transition to a bridge job. He or she then decides how many hours to work. For the purposes of our 5 analysis, we examine hours worked conditional on being on a FTC job or a bridge job at the time of the interview. That is, we do not model job type and hours worked as being jointly determined.12 We begin by pooling the data for each cohort and transforming the dataset into one with person-year observations. We then model the work-leisure decision later in life among those with work experience since age 49 as follows: * Rit = + 1 Xi + 2 X it + 3 RETit + WarBaby i + 4 Y 5 t (1) + 6( WarBaby i *Yt )+ i + it . * where i stands for individual and t stands for time. The latent variable, Rit , determines the observed choice. Rit indicates the actual outcome and is equal to 1 if individual i is working at time t, and is equal to zero otherwise. Xi and Xit represent, respectively, time invariant and time-varying vectors of characteristics believed to be significant determinants of the retirement process, such as age, health status, marriage status, and other demographic characteristics. RETit is a set of retirement incentive variables associated with private pensions and health insurance plans. WarBabyi is a binary variable equal to one if the individual respondent is among the War Babies cohort. Yt is a series of year indicator variables (or, alternatively, macro-economic measures, such as the unemployment rate) that are intended to account for economy-wide factors. The error term consists of two parts, an individual-specific component, vi , assumed to be uncorrelated with the set of explanatory variables, and a white noise error component, it. The interaction terms between the War Baby indicator and the time dummies are a key interest for this study. These interactions capture the extent to which the work decisions of the War Babies differed systematically by year from those of the Core. We hypothesize that the Core sample, those on the cusp of retirement, might have been more vulnerable to the macro events, such as the stock market decline (i.e., 6 0 implies cross-cohort differences). We estimate equation (1) three ways, using a linear probability model with and without fixed effects and using a logistic model. 12 See Hay, Leu, and Rohrer (1987) for a discussion of ordinary least squares and sample selection models; also, Manning, Duan, and Rogers (1987). 6 The next model is based on the multinomial regression model presented in Cahill, et al. (2006) in which gradual retirement is examined among a set of respondents who were on a FTC job at the time of the first interview. Gradual retirement from full-time career employment is measured using a three-way indicator of labor force withdrawal, equal to zero (still on a full-time career job), one (transitioned to a bridge job), or two (direct exit from the labor force). The model is structured as follows: R3* = + 1 X i + 2 X it 1 + 3 RETit 1 + it it (2) R3* is the latent variable that determines the observed choice; R3it indicates the actual it outcome, equal to zero, one, or two, as noted above. Xi, Xit, and RETit are as defined above for equation (1). The model differs from the one presented in Cahill et al. (2006) mainly because of the time at which the independent variables are measured. Instead of using a wave-one baseline measure, time-varying determinants are defined as of the time period just prior to the transition. For example, if a respondent in the HRS Core sample moves from a FTC job to a bridge job in the fourth wave (1998), the independent variables are measured as of the third wave (1996). This prior-to-transition designation is an improvement over the previous model specification because changes in status since the first interview are incorporated into the model, which are particularly important for the Core sample given the potential for a 12-year gap between the initial interview and first transition. We estimate equation (2) separately for the Core and War Babies samples. This allows us to examine the extent to which differences exist by cohort. Given the decision to remain working, we then examine hours worked on the full- time career job (i.e., prior to transition) and hours worked on the bridge job (i.e., post transition) among those who have had a FTC job since age 49. The hours equation is as follows: H it = + 1 X i + 2 X it + 3 RETit + 4 WarBabyi + 5 Yt (3) + 6 (WarBabyi *Y ) + v + it . t i The dependent variable, H it , represents annual hours worked, conditional on either working on an FTC job or a bridge job. Other variables are as defined above. Separate equations are estimated for FTC hours and for bridge job hours, with time indicators 7 measured as of the current wave of HRS data; a respondent’s hours decision in any given wave is based on the incentives that exist at that time and the respondent’s choices made up to that point. The interaction terms, as in equation (1) above, allow us to test whether cohort and time differences exist with respect to the work and work-intensity decision, respectively. IV. Data The Health and Retirement Study (HRS) is a nationally-representative, longitudinal survey of older Americans that began in 1992. The survey now spans a dozen years from 1992 to 2004 and includes those born between 1931 and 1947, among others, and provides micro-level information regarding labor force decisions. Moreover, the HRS provides detailed information on the demographic and economic characteristics of older Americans and the types of jobs they hold (full-time versus part-time, self- employed versus wage-and-salary, low-paying versus high-paying, blue collar versus white collar, etc.). The original HRS Core set of respondents consisted of about 12,600 persons from approximately 7,600 households with respondents 51 to 61 years old in 1992 (born between 1931 and 1941), and their spouses. Respondents were first interviewed in 1992 and follow-up interviews have been conducted every two years. Attrition across waves ranged from 3 to 10 percent, so that after six years, about 85 percent of the Core sample remained, and after twelve years about 75 percent remained. The HRS was expanded dramatically in 1998 (Wave 4) with the addition of the War Babies cohort. The War Babies were born between 1942 and 1947 (age 51 to 56 in 1998) and like the Core, they are interviewed every two years. This paper focuses on labor force exit and retirement patterns, and we therefore exclude respondents with no work experience after age 49. The large majority of both HRS men and women, however, have work experience later in life, as shown in Table 1. Just over 90 percent of men in both the Core and War Babies samples have worked since age 49. Work experience is somewhat lower among women with 77 and 87 percent of the Core and War Babies, respectively, having work experience after age 49. 8 For the analysis of gradual retirement, we make an additional restriction based on whether these HRS respondents had a FTC job since age 49. The longitudinal nature of the HRS allows us to do this, since the questionnaire from the initial interview asked about a respondent’s current job and all previous jobs that had lasted five years or more. If a respondent was not working at the time of the Wave 1 interview, he or she was asked about the most recent job held, if any. Information on short-term jobs in the past lasting less than five years was not available, unless the respondent was not working at the time of the first interview and tenure on the last job held was shorter than five years. For the purposes of this paper, we define a full-time career (FTC) job as one that requires at least 1,600 hours per year (“full time”) and that lasts ten or more years (“career”). Therefore, a bridge job is employment following a FTC job that does not meet both of these requirements. We find that over 70 percent of men in both samples had a FTC job since age 49, while 45 and 51 percent of women in the Core and War Babies samples, respectively, had a FTC job since age 49. Some of our analyses require the samples to be restricted to those on FTC jobs at the time of the first interview. Among the Core sample, 52 percent of men and 37 percent of women were on a FTC job in 1992. A substantially larger percentage of War Babies were on a FTC job in 1998, with 70 percent of men and 50 percent of women reporting that they were employed in a FTC job at the time of their first interview. V. Results Retirement Outcomes Two key outcomes of interest are work status and the prevalence of bridge jobs. Each of these outcomes is examined over time, from 1992 through 2004. Work status in each survey year among the HRS Core and War Babies is shown in Figure 1. Not surprisingly, older retirees were less likely to be working in each year. The relevant comparison, however, is how each of the cohorts compare over time. The work status of the younger Core group of men resembled that of the War Babies in 1998, with a participation rate of nearly 90 percent at ages 51 to 56. Six years later, about three 9 quarters of each group was still working. Among men, it did not appear as though there were substantial cohort differences with respect to work status. The story is different among women, although not dramatically so. More women in the War Babies cohort were working at the time of the first interview compared to the younger Core cohort. The difference continued six years later, and had grown slightly. Perhaps more interestingly, the older Core women had a rapid decline in work status in the first six years of the survey, which then leveled off and remained fairly stable between 2002 and 2004. These descriptive statistics for the older Core women provide some evidence that cohort differences and year effects may be important when examining work status in the pooled sample. The second outcome of interest focuses on the way older workers exit the labor force. We measure gradual exits from the labor force using the bridge job concept mentioned earlier. Table 2 describes bridge job prevalence as of 2004, stratified by work status and gender. By 2004, about 50 percent of Core men were either currently working on a bridge job or were currently not working, but had worked on a bridge job prior to exiting the labor force. A similar percentage is observed among the women. A non- trivial portion of both men and women were still working on a FTC job (13 and 17 percent, respectively), so the eventual incidence of bridge job behavior will be even higher. If we assume those still on FTC jobs will leave their jobs in a way resembling those who have already left, then about 60 percent of men and women with FTC jobs in their work history will have taken on bridge jobs prior to retiring. Retirement Determinants The retirement literature has identified key demographic, socio-economic, and financial retirement incentives that influence the retirement decision. In this section, we explore how the two outcome variables of interest, work status and bridge job behavior, compare with respect to these predictors, again, by cohort and by gender. We examine work status and hours worked among all respondents, and bridge job status among those who were on a full-time career job as of the first interview. Tables 3a shows that 38 percent of the Core men and 76 percent of the male War Babies were still working in 2004. Among those with FTC jobs in Wave 1, about 16 10 percent of HRS Core men were still on that FTC job in 2004 and, among those who left, about 55 percent had taken a bridge job. More than one half of the HRS War Babies with FTC jobs at the time of the first interview were still on the FTC job, and about two-thirds of those who left took a bridge job. Not surprisingly, men were more likely to be working in 2004 if they were younger, reported being in better health, or had dependent children. Bridge job status was more common among younger retirees who moved off of a FTC job for both the Core and War Babies samples. Bridge jobs were also more common among those who reported better health status and among those who had a college degree. These determinants appear to have influenced the HRS Core and War Babies in similar ways. The general story is similar for women, albeit at different levels, and with cross- cohort differences more pronounced than among men. For example, as shown in Table 3b, Core women with dependent children were 16 percentage points more likely to be working than the overall average (55 percent vs. 39 percent). Among the War Babies, women with dependent children were six percentage points less likely than others to be working in 2004 (65 percent vs. 71 percent). Core women with dependent children were also much more likely to take on a bridge job relative to the overall average than were those women with children in the War Babies cohort. Differences by cohort and gender also exist with respect to economic characteristics, such as health insurance status, pension status, wage, and occupation. We discuss a few of these variables here and refer the reader to Table 3c (men) and Table 3d (women) for complete details. The impact of a defined-benefit pension plan was substantial among Core men, as those with defined-benefit plans were less likely to be working and less likely to be on an FTC job in 2004 compared to the overall average. These pension plans often have specific incentives incorporated into their benefit structures that induce individuals to leave their jobs at specific early retirement ages, so the result is plausible. Interestingly, these effects were not seen among the War Babies, perhaps because this group has not yet reached the pivotal ages for early retirement within these plans. A second point that we highlight is a “u-shaped” relationship between both wage and bridge job status and between occupational status and bridge job status. One 11 explanation for this pattern is that those who were fairly well off (high wage or white collar, highly skilled) may have taken a bridge job to “try something new” or for “a change” – not out of necessity, while those who were struggling financially may have taken a bridge job because they had no other choice. Those in the middle of the distribution might not have been as influenced by these factors. The occupational variables appear to support this relationship as well, as those in white-collar, highly- skilled occupations and those in blue-collar, not-highly-skilled occupations were the most likely to take on bridge jobs. The result holds for both cohorts, although the relationship was stronger among the Core sample. The relationships discussed above hold for HRS Core women as well, although as before, cross-cohort differences exist. Core women with defined-benefit pension plans were less likely to be working in 2004 compared to other workers, and the “u-shaped” relationship for occupational status held as well. In contrast, female War Babies with defined-benefit plans were more likely than others to be working and the “u-shaped” occupational status relationship looks instead like the left-hand side of a “u-shaped” curve. Bridge job status was highest among those in white-collar, highly-skilled occupations and lowest among those in blue-collar, not-highly-skilled positions. Multivariate Analysis We now examine how the retirement determinants described above (e.g., health status, financial incentives) affect the outcome variables of interest in a multivariate setting. We begin with the work-leisure decision (i.e., whether to work) using logistic regression and then examine retirement as a three-way process (i.e., whether to remain working full-time, take on a bridge job, or exit the labor force) using a multinomial logistic regression model based on respondents who were on a full-time career job at the time of the first survey. We then examine the work intensity decision (i.e., how much to work). Specification #1: Logistic Regression Model of the Work-Leisure Decision We analyze the work-leisure decision and the work intensity decision separately. The separate analyses provide the flexibility necessary to utilize the longitudinal nature of 12 the HRS, by constructing a dataset of person-year observations. We restrict our sample to those who have had work experience since age 49 to ensure that the work-leisure decisions are for those who have had work experience later in life. Tables 4a and 4b (men and women, respectively) present marginal effects from a logistic regression of work status in each Wave.13 As expected, men were more likely to be working if they were younger and in excellent or very good health, if they had higher levels of formal education or dependent children, or if they were self-employed or earned more per hour. Men were less likely to be working if they reported being in fair or poor health. Men were also less likely to be working if their spouse was in excellent or very good health and more likely to be working if their spouse was in fair or poor health. Older workers responded to key retirement incentives as well, as expected. If health insurance was portable in retirement, that is, if a worker did not lose his health insurance if he stopped working at that job, then he was more likely to stop working, all else equal. Men with defined benefit pension plans on their jobs were also less likely to be working, a result that is consistent with the age-specific early retirement incentives incorporated in such plans. Having a defined contribution plan had a negative influence on work relative to not having a pension plan, although the impact of defined contribution plans was much weaker than that of defined benefit plans. A focus of our analysis is the interaction between the War Babies indicator and macroeconomic factors. Overall, the War Babies were more likely than the Core respondents to be working at every age and their probability of remaining in the labor force declined consistently over the survey years. When these two variables are interacted, however, we find that the differences between the Core sample and the War Babies were insignificant in 2002 and in 2004. Perhaps the stock market collapse in 2000 influenced Core respondents’ decisions to return to the labor force and, therefore, their work decisions started to resemble those of the War Babies. The results of the analysis for the HRS female respondents are similar to those of the male sample. We keep them separate here, though, because of potential differences in work intensity on FTC jobs or bridge jobs, as discussed below. We also highlight that 13 We also perform this estimation two additional ways, using a linear probability model and a linear probability model with fixed effects. We obtain similar results using all three methods. 13 the same time effects are seen among the women as with the men. The work decisions of the Core females no longer differed from those of the War Babies after 2000, while they did differ before. Again, like their male counterparts, the work-leisure decisions of the female Core respondents relative to the War Babies appear to have been influenced by macroeconomic factors. Specification #2: Multinomial Regression Model of Gradual Retirement We next consider a multinomial logistic regression specification where workers in FTC jobs are faced with the choice of remaining on their FTC job (no transition), leaving their FTC job for another job (bridge job transition), and leaving their FTC job for no job (retirement transition). Separate models are estimated for men and for women, and for the Core and War Babies respondents. We find that Core men were more likely to remain on their FTC job rather than leave the labor force if they were younger at the time of the first interview, had a dependent child, had health insurance, had a higher wage on the FTC job, were white collar, were self-employed, or owned a home (Table 5a). Men were less likely to remain on a FTC job if they had a defined-benefit pension plan. Health status was not a significant determinant of remaining on a FTC job per se, but was a significant determinant of how an individual made a transition, via a bridge job or a direct exit. Those in excellent or very good health were much more likely than those in good health to take a bridge job while those in fair or poor health were much less likely to do so. Bridge job transitions were also more prevalent among those who were younger, college educated, married and without health insurance, and was less prevalent among those with a defined-benefit pension plans or with lower wages on the full-time career job. Men whose spouses were working were also more likely to take a bridge job, suggesting that, transitions may be jointly determined among spouses. The experience among the male War Babies was, overall, similar to their Core counterparts (Table 5b). Health status, health insurance, pensions, self-employment and wages were all significant determinants of retirement transitions. Some differences across these cohorts were found. Health insurance portability, for example, was a significant predictor of leaving a FTC job among the male War Babies, which is intuitive. 14 If a respondent was able to maintain his health insurance after leaving his FTC job, he may be more likely to make a transition. Spouse’s health status was also a significant determinant of transitions among the male War Babies, perhaps additional evidence that transitions may be jointly determined among spouses. The direction is interesting. Those with a spouse in excellent or very good health were more likely than others to leave a FTC job and take a bridge job. The factors that influenced FTC and bridge job behavior among men were also significant determinants of retirement transitions among the Core women (Table 5c). Among other factors such as the influence of health insurance and pensions, spouse’s work status had a significant influence on the retirement transitions of Core women, both in terms of transitions away from FTC employment and transitions onto bridge jobs. Another interesting finding is that, while wage was a significant driver of retirement transitions among men, wealth appeared to be a more significant determinant for the Core women. Some differences across cohorts existed among the female samples (Table 5d). For example, female War Babies were more likely to remain on a FTC job and less likely to take a bridge job if they had a dependent child, a finding that is not significant among the Core women. Factors such as health insurance and pensions, in contrast, were only marginally significant among the female War Babies. It is not clear why this was the case, although we suspect the marginal significance associated with economic characteristics among the female War Babies may be a function of the sample size of the cohort. Specification #3: OLS Model of Work Intensity Given the decision to work, respondents then decide how many hours to work. This work intensity decision can be quite complicated, with decisions about job type and hours worked being jointly determined. For the purposes of our analysis, however, we simplify the decision and explore hours worked conditional on whether employment was in a FTC job or a bridge job. Table 6a reports estimated coefficients from an OLS regression of hours worked per year conditional on being on a FTC job among the sample of men who have had a FTC job since age 49. Like the work-leisure regressions, observations are person-year with time-dependent variables measured as of the survey 15 year. Table 6b then examines hours worked conditional on being on a bridge job. Therefore, the results presented in Table 6a represent hours worked pre-transition and the results presented in Table 6b represent hours worked post-transition. Tables 6c and 6d present pre- and post-transition hours, respectively, among female respondents. Hours worked on the FTC job were higher among younger men and those in excellent or very good health, as expected, since full time employment is a requirement for a full-time career job. The order of magnitude is nontrivial as well, as men over age 65 worked 133 fewer hours per year than those under age 58. Men who were in excellent or very good health worked about 25 hours more per year than those in good health, and those in fair or poor health worked about 18 hours less. Hours worked were also positively associated with pension plans and white collar occupations. Men with defined- contribution plans worked about 72 hours more per year prior to making a transition compared to those with no pension plan, and those with defined-benefit plans worked about 14 hours more. College graduates and those with portable health insurance worked fewer hours, all else equal. And, finally, while self-employment status was consistently a strong predictor of working later in life, being self-employed had no statically significant impact on the number of hours worked prior to transition. Work intensity after a transition exhibited a similar pattern as that prior to transition, albeit with higher magnitudes and some notable exceptions. Spouse’s health status influenced hours worked post transition. Men with a spouse in fair or poor health worked on average about 60 more hours per year on their bridge jobs than otherwise similar males. Those with less than a high school degree also worked more post transition, by about 85 hours per year. The largest sway in hours worked per year post transition, of 150 hours or more, was associated with health insurance status, pension status, and self-employment status. Men with portable health insurance or no health insurance worked much fewer hours than those with non-portable health insurance, while those with pensions worked much more. Interestingly, while self-employment status had no significant impact on hours worked prior to transition, those in self-employed bridge jobs worked more than 200 hours less than wage-and-salary men in their post-transition job. This finding may be indicative of men using self-employment as a method of reducing the number of hours worked as they transition to full retirement. 16 The War Baby indicator variable was not significantly different than zero either before or after transition. As before, we also interacted the War Baby indicator variable with the year dummies to determine whether any macroeconomic effects affected the War Babies differently than the Core respondents. The result was similar to that found in the work-leisure analysis – differences between the Core and War Babies vanished after the stock market decline. This blurring of the coefficients is consistent with the story that macro-economic factors brought some Core workers back into the labor force and that their work intensity decisions resembled those of the War Babies. Many of the main determinants of hours worked prior to transition among the male sample did not hold for the female sample. Most notably, perhaps, was that age and health status did not have a statistically significant impact on hours worked prior to transition. Several retirement incentives did, however, such as health insurance status and pension status, with patterns that resembled those among men. One finding of note was that while being married and having dependent children had a positive influence on hours worked among men, albeit with the latter effect not being statistically significant, these two factors had a negative impact on hours worked among women. The determinants of post-transition hours worked among women closely resembled those among the men, almost surprisingly so, especially with respect to age, college degree, health insurance and pension status, and self-employment. A similar pattern with respect to HRS Core and HRS War Baby differences also held for women. Some differences, however, warrant mention. Statistical significance was not found for own health status and spouse health status or with not being a high school graduate. And, while being married was associated with more post-transition hours worked among men, the opposite was true for women. VI. Conclusion The advent of 401(k)s in the 1980s, and their explosive growth since then, combined with an increase in Social Security’s Normal Retirement Age and low savings rates, means that today’s retirees are more vulnerable to short-run market forces than at any point in the post-war era. This shift means that the retirement income security of many individuals is dependent upon the existing state of financial markets, housing prices 17 (for those who own homes), and the economy as a whole. Older workers may therefore need to re-think their long-term retirement plans in light of short-run market conditions. Going forward, the timing of retirement may be influenced by macroeconomic factors to the extent that these affect pensions and other financial variables. In this paper we examine retirement patterns from full-time career employment using a three-way outcome measure and we examine the work-leisure decision and work intensity later in life using data on two cohorts from the Health and Retirement Study. We find that work status across cohorts was consistent over time among men while some differences exist for women, with the younger cohorts more likely to have worked longer. We also find that bridge job status continued to be common among younger retirees, as with older ones, with about two thirds of those making a transition from a full-time career job to a bridge job. While the descriptive findings suggest little that would imply stark time or cohort differences, the multivariate analyses shed some additional light on how the two cohorts compare. Overall, cohort differences were more pronounced among women than men, although key determinants of retirement, such as age, health status, and health insurance and pension status, influenced work decisions across all groups. We also find that cross- cohort differences in terms of work-leisure decisions and hours worked per year seem to have vanished after 2000, all else equal. One possible explanation, consistent with aggregate findings, is that the older HRS Core respondents altered their work decisions after the stock market collapse to the point where they eventually resembled their younger counterparts. It will be interesting to see how this plays out in the years to come. Another finding of note is that self-employment may be used as a mechanism by which retirees gain work flexibility later in life. Those who were self-employed were much more likely to be working in general, yet their number of hours worked on the FTC job resembled those in wage-and-salary employment. That changes on the bridge job, as those who were self-employed worked significantly fewer hours. Placing these results in the context of the overarching theme of this study, we view the shift towards “do-it-yourself” retirement as a mixed bag. On the one hand, workers have more control of their retirement assets and, as shown in this paper and others, they respond to many of the financial incentives associated with retirement by 18 working longer and by taking on bridge jobs after FTC employment. This result implies that if retirement assets are less than expected upon retirement many older workers may remain active members of the labor force well into their late 60s and 70s. On the other hand, if work later in life is not an option, because of factors such as health or inflexible work options, some retirees’ long-run well being will be vulnerable to short-term fluctuations in market conditions. What is clear is that retirement incentives have changed and these changes will likely influence the retirement decisions of older workers for years to come. With pre- emptive action by today’s middle-aged and younger workers, in the form of increased savings or more realistic work expectations, the timing of retirement may be less susceptible to short term macro-level influences. 19 References Cahill, K. E., Giandrea, M. D., & Quinn, J. F. (2006). Retirement Patterns from Career Employment. The Gerontologist, Vol. 46, No. 4, pp. 524-532. Cahill, K E., & Soto, M. (2003). How Do Cash Balance Plans Affect the Pension Landscape? Issue in Brief, No. 14 (December). Chestnut Hill, MA: Center for Retirement Research at Boston College. Coile, C. C. (2004). Retirement incentives and couples’ retirement decisions. Topics in Economic Analysis and Policy, Vol. 4, no. 1, pp 1-28. Coile, C. C., & Levine, P. B. (2006). Bulls, bears, and retirement behavior. Industrial and Labor Relations Review, Vol. 59, No. 3, pp. 408-429. Coronado, J. L., & Perozek, M. (2003). Wealth effects and the consumption of leisure: Retirement decisions during the stock market boom of the 1990s. Federal Reserve Board Finance and Economics Discussion Series Number 2003-20. Eschtruth, A. D., & Gemus, J. (2002). Are older workers responding to the bear market? Just the Facts on Retirement Issues, No. 5 (September). Chestnut Hill, MA: Center for Retirement Research at Boston College. Friedberg, L., & Webb, A. (2005). Retirement and the Evolution of Pension Structure. Journal of Human Resources, 40(2), pp. 281-308. Gardner, J., & Orszag, M. (2003). How have older workers responded to scary markets? Watson Wyatt Technical paper 2003-LS04. Retrieved Feb. 21, 2006 from Gruber, J., & Madrian, B. C. (1995). Health-insurance availability and the retirement decision. The American Economic Review, Vol. 85, No. 4, pp. 938-948. 20 Hay, J. W., Leu, R. & Rohrer, P. (1987). Ordinary Least Squares and Sample-Selection Models of Health-Care Demand: Monte Carlo Comparison. Journal of Business & Economic Statistics, Vol. 5, No. 4. (October), pp. 499-506. Juster, F. T., & Suzman, R. (1995). An Overview of the Health and Retirement Study. U Journal of Human Resources , 30 (Supplement), S7-S56. U Maestas, N. (2004). Back to work: Expectations and realizations of work after retirement. (Rand Working Paper WR-196). Manning, W. G., Duan, N. & Rogers, W. H. (1987). Monte Carlo Evidence on the Choice between Sample Selection and Two-Part Models. Journal of Econometrics, Vol. 35, No. 1-2. (May), pp. 59-82. Munnell, A. H., & Cahill, K. E., & Jivan, N. A. (2003). How has the shift to 401(k)’s affected the retirement age? Issue in Brief, No. 13 (September). Chestnut Hill, MA: Center for Retirement Research at Boston College. Munnell, A. H. (2003). The Declining Role of Social Security. Just the Facts On Retirement Issues, No. 6 (February). Chestnut Hill, MA: Center for Retirement Research at Boston College. Mutchler, J. E., Burr, J. A., Pienta, A. M., & Massagli, M. P. (1997). Pathways to labor force Exit: Work transitions and work instability. Journal of Gerontology: Social Sciences, 52B(1), S4-S12. Purcell, P. J. (2005). Older workers: Employment and retirement trends. CRS Report for Congress, Congressional Research Service. The Library of Congress. Washington, DC. 21 Quinn, J. F. (1999). Retirement patterns and bridge jobs in the 1990s. EBRI Issue Brief, No. 206 (February). Washington, DC: Employee Benefit Research Institute. 1-23. Ruhm, C. J. (1990). Bridge jobs and partial retirement. Journal of Labor Economics, 8(4): 482-501. U.S. Department of Labor. Bureau of Labor Statistics. Employment and Earnings. Washington, DC: U.S. Government Printing Office (January issues of many years). U.S. Department of Commerce - Bureau of Economic Analysis. (2006). National Income and Product Accounts. 22 Figure 1a Labor Force Participation Rates, by Wave and Cohort Men 1992 - 2004 100.0% 90.0% 80.0% 70.0% 60.0% 50.0% 40.0% 30.0% Core 51-56 20.0% Core 57-61 WB 51-56 10.0% 0.0% 1992 1994 1996 1998 2000 2002 2004 Year Source: Authors' calculations based on the Health and Retirement Study. Figure 1b Labor Force Participation Rates, by Wave and Cohort Women 1992 - 2004 90.0% 80.0% 70.0% 60.0% 50.0% 40.0% 30.0% Core 51-56 20.0% Core 57-61 WB 51-56 10.0% 0.0% 1992 1994 1996 1998 2000 2002 2004 Year Source: Authors' calculations based on the Health and Retirement Study. Table 1 Sample Size by Gender, Survey Participation, and Work Status HRS Core: Respondents Aged 51-61 in 1992 Men Women Total Particpated in wave 1 n 5,869 6,783 12,652 Worked since age 49 n 5,344 5,196 10,540 % of HRS Core 91% 77% 83% Had FTC job after age 49 n 4,280 3,082 7,362 % of HRS Core 73% 45% 58% On FTC in 1992 n 3,057 2,513 5,570 % of HRS Core 52% 37% 44% HRS War Babies: Respondents Aged 51-56 in 1998 Men Women Total Particpated in wave 4 n 1,200 1,329 2,529 Worked since age 49 n 1,122 1,159 2,281 % of HRS WB 94% 87% 90% Had FTC job after age 49 n 890 675 1,565 % of HRS WB 74% 51% 62% On FTC in 1998 n 843 664 1,507 % of HRS WB 70% 50% 60% Source: Authors’ calculations based on the Health and Retirement Study. Table 2 Current Employment Status in 2004, by Gender Individuals with a Full-Time Career Job in their Work History and a FTC Job Since Age 50 HRS Core: Respondents Aged 51-61 in 1992 Full Time Bridge Don't % with n Career Job Job Know Bridge1 Men, Working 1,210 14% 24% 2% Men, Nonworking, Last job was 1,736 30% 22% 6% Total 2,946 45% 46% 9% 60% Women, Working 1,105 20% 24% 2% Women, Nonworking, Last job was 1,288 29% 20% 5% Total 2,393 50% 44% 7% 60% HRS War Babies: Respondents Aged 51-56 in 1998 Full Time Bridge Don't % with n Career Job Job Know Bridge1 Men, Working 699 44% 29% 3% Men, Nonworking, Last job was 212 13% 7% 3% Total 911 57% 36% 6% 74% Women, Working 653 36% 33% 2% Women, Nonworking, Last job was 269 10% 16% 3% Total 922 45% 49% 6% 83% 1: calculated as the ratio of those who moved to a bridge job among those who have made a transition. Source: Authors’ calculations based on the Health and Retirement Study. Table 3a Employment Status of Men in 2004, by Demographic Characteristics Core Men War Babies Men Percentage Percentage Percentage Who Moved to a Percentage Who Moved to a Percentage Still on Full-Time Bridge Job Percentage Still on Full-Time Bridge Job 1 Determinants Working Career Job2 in First Transition3 Working1 Career Job2 in First Transition3 Overall 38% 16% 55% 76% 54% 67% 4 Age in 2004 < 62 70% 32% 67% 81% 59% 70% 62 - 64 54% 25% 63% 72% 48% 64% 65 - 69 40% 14% 52% 59% 34% 57% 70 + 25% 6% 52% Subjective Health Status excellent or very good 45% 18% 58% 84% 48% 72% good 33% 13% 54% 74% 46% 58% fair or poor 22% 12% 44% 50% 39% 53% College Degree 48% 19% 61% 82% 44% 79% Less than College Degree 35% 15% 53% 75% 47% 63% Married 39% 16% 56% 77% 47% 65% Not Married 33% 15% 50% 72% 43% 67% Dependent Children 49% 23% 58% 77% 47% 65% No Dependent Children 36% 14% 55% 75% 46% 66% Spouse Employed 42% 17% 57% 79% 48% 70% Spouse Not Employed 33% 15% 53% 72% 43% 56% 1: among those who have worked since age 49 2: among those on a FTC job in the first wave of data 3: among those on a FTC job in the first wave and who have moved off of that FTC job 4. Age categories for the War Babies cohort are <60, 61-62, and > 62. Source: Authors’ calculations based on the Health and Retirement Study. Table 3b Employment Status of Women in 2004, by Demographic Characteristics Core Women War Babies Women Percentage Percentage Percentage Who Moved to a Percentage Who Moved to a Percentage Still on Full-Time Bridge Job Percentage Still on Full-Time Bridge Job 1 2 3 Determinants Working Career Job in First Transition Working1 Career Job2 in First Transition 3 Overall 39% 21% 57% 71% 55% 68% Age in 20044 < 62 67% 39% 70% 73% 58% 68% 62 - 64 47% 22% 62% 64% 39% 70% 65 - 69 34% 11% 52% 67% 53% 50% 70 + 20% 4% 41% Subjective Health Status excellent or very good 45% 24% 61% 78% 50% 77% good 37% 17% 52% 71% 48% 58% fair or poor 22% 12% 43% 46% 36% 44% College Degree 47% 24% 63% 74% 47% 78% Less than College Degree 38% 20% 55% 70% 48% 64% Married 41% 23% 58% 72% 50% 63% Not Married 35% 14% 55% 69% 41% 72% Dependent Children 55% 35% 71% 65% 39% 57% No Dependent Children 37% 18% 55% 72% 50% 70% Spouse Employed 44% 23% 59% 71% 53% 65% Spouse Not Employed 34% 17% 54% 71% 44% 57% 1: among those who have worked since age 49 2: among those on a FTC job in the first wave of data 3: among those on a FTC job in the first wave and who have moved off of that FTC job 4. Age categories for the War Babies cohort are <60, 61-62, and > 62. Source: Authors’ calculations based on the Health and Retirement Study. Table 3c Employment Status of Men in 2004, by FTC Job Characteristics Core Men War Babies Men Percentage Percentage Percentage Who Moved to a Percentage Who Moved to a Percentage Still on Full-Time Bridge Job Percentage Still on Full-Time Bridge Job Determinants Working1 Career Job2 in First Transition3 Working1 Career Job2 in First Transition3 Health Insurance Status Not covered on career job 39% 21% 73% 65% 45% 79% "Covered and would maintain " coverage 35% 14% 53% 73% 51% 65% "Covered and would lose" coverage 50% 23% 52% 82% 60% 61% Pension Status No Pension 42% 22% 66% 71% 51% 83% Defined - Contribution only 44% 19% 59% 83% 57% 60% Defined - Benefit only 31% 11% 46% 75% 52% 53% Defined - Contribution and Defined - Benefit 49% 16% 63% 75% 55% 47% Self-Employed 58% 28% 77% 86% 63% 88% Wage and Salary 34% 13% 51% 80% 52% 63% Wage Rate < $6/hour 46% 20% 72% 90% 64% 90% $6 - $10/hour 39% 17% 57% 83% 51% 83% $10 - $20/hour 41% 14% 49% 82% 53% 63% $20 - $50/hour 45% 14% 57% 81% 57% 62% > $50/hour 55% 30% 69% 90% 39% 86% Occupation Status White collar, highly skilled 49% 18% 61% 80% 52% 72% White collar, other 49% 17% 58% 74% 53% 62% Blue collar, highly skilled 38% 13% 49% 74% 57% 61% Blue collar, other 39% 21% 55% 69% 54% 62% 1: among those who have worked since age 49 2: among those on a FTC job in the first wave of data 3: among those on a FTC job in the first wave and who have moved off of that FTC job Source: Authors’ calculations based on the Health and Retirement Study. Table 3d Employment Status of Women in 2004, by FTC Job Characteristics Core Women War Babies Women Percentage Percentage Percentage Who Moved to a Percentage Who Moved to a Percentage Still on Full-Time Bridge Job Percentage Still on Full-Time Bridge Job Determinants Working1 Career Job2 in First Transition3 Working1 Career Job2 in First Transition3 Health Insurance Status Not covered on career job 39% 23% 76% 70% 41% 82% "Covered and would maintain " coverage 38% 19% 55% 65% 50% 66% "Covered and would lose" coverage 46% 25% 54% 83% 62% 68% Pension Status No Pension 37% 20% 67% 62% 46% 70% Defined - Contribution only 44% 24% 53% 82% 59% 63% Defined - Benefit only 39% 18% 50% 79% 56% 68% Defined - Contribution and Defined - Benefit 57% 39% 75% 75% 50% 64% Self-Employed 49% 21% 75% 79% 51% 87% Wage and Salary 38% 21% 55% 77% 55% 66% Wage Rate < $6/hour 42% 25% 66% 64% 41% 50% $6 - $10/hour 43% 21% 56% 76% 56% 61% $10 - $20/hour 46% 18% 55% 81% 53% 69% $20 - $50/hour 44% 25% 54% 83% 64% 76% > $50/hour 39% 0% 63% 83% 100% ------ Occupation Status White collar, highly skilled 50% 23% 59% 79% 59% 77% White collar, other 46% 22% 57% 68% 52% 68% Blue collar, highly skilled 35% 14% 47% 72% 59% 55% Blue collar, other 41% 19% 59% 58% 47% 50% 1: among those who have worked since age 49 2: among those on a FTC job in the first wave of data 3: among those on a FTC job in the first wave and who have moved off of that FTC job Source: Authors’ calculations based on the Health and Retirement Study. Table 4a Marginal Effects from Logistic Regression Dependent Variable: Working at time t (working = 1) Men Who Have Worked Since Age 49 marg. effects p-value Age in 2004 57 or younger ------ ------ 58-61 -0.0792 0.000 62-64 -0.2659 0.000 65-69 -0.3229 0.000 70 or older -0.3257 0.000 Respondent Health Excellent/very good 0.0472 0.000 Good ------ ------ Fair/poor -0.2082 0.000 Spouse Health Excellent/very good -0.0506 0.000 Good ------ ------ Fair/poor 0.0464 0.004 Education Less than high school 0.0037 0.780 High school graduate ------ ------ College graduate 0.0506 0.000 Married -0.0050 0.876 Dependent Child 0.0427 0.001 Health Insurance Status Portable -0.5163 0.000 Non-portable ------ ------ None -0.5297 0.000 Pension Status Defined-benefit -0.3078 0.000 Defined-contribution -0.0661 0.000 Both 0.0910 0.001 None ------ ------ Occupational Status White collar - high skilled 0.1038 0.000 White collar - other 0.1478 0.000 Blue collar - high skilled 0.0952 0.000 Blue collar - other ------ ------ Self Employed 0.2096 0.000 Wage 0.0049 0.000 Wage Squared 0.0000 0.000 Wealth -0.0010 0.000 Wealth Squared 0.0000 0.000 Own Home -0.0079 0.606 Constant 0.6275 0.000 Regressors (continued) coef p-value War Baby Indicator 0.1589 0.000 Year Indicators 1992 1.1682 0.000 1994 0.4175 0.000 1996 0.2484 0.000 1998 0.1955 0.000 2000 0.1195 0.000 2002 0.0661 0.000 2004 ------ ------ War Baby Interaction Terms War Baby * 1998 0.8244 0.000 War Baby * 2000 0.1682 0.002 War Baby * 2002 0.0017 0.970 War Baby * 2004 ------ ------ Source: Authors’ calculations based on the Health and Retirement Study. Table 4b Marginal Effects from Logistic Regression Dependent Variable: Working at time t (working = 1) Women Who Have Worked Since Age 49 marg. effects p-value Age in 2004 57 or younger ------ ------ 58-61 -0.1021 0.000 62-64 -0.2422 0.000 65-69 -0.3106 0.000 70 or older -0.3958 0.000 Respondent Health Excellent/very good 0.0353 0.004 Good ------ ------ Fair/poor -0.2202 0.000 Spouse Health Excellent/very good -0.0474 0.002 Good ------ ------ Fair/poor 0.0384 0.032 Education Less than high school -0.0093 0.541 High school graduate ------ ------ College graduate -0.0305 0.062 Married -0.0526 0.170 Dependent Child 0.0485 0.000 Health Insurance Status Portable -0.4896 0.000 Non-portable ------ ------ None -0.4643 0.000 Pension Status Defined-benefit -0.2532 0.000 Defined-contribution -0.1423 0.000 None ------ ------ Both 0.2051 0.000 Occupational Status White collar - high skilled 0.0909 0.000 White collar - other 0.1467 0.000 Blue collar - high skilled 0.1116 0.000 Blue collar - other ------ ------ Self Employed 0.1809 0.000 Wage 0.0107 0.000 Wage Squared 0.0000 0.000 Wealth -0.0010 0.000 Wealth Squared 0.0000 0.000 Own Home -0.0024 0.876 Constant 0.5910 0.000 Regressors (continued) coef p-value War Baby Indicator 0.1354 0.001 Year Indicators 1992 0.9709 0.000 1994 0.4776 0.000 1996 0.2637 0.000 1998 0.2514 0.000 2000 0.1780 0.000 2002 0.1126 0.000 2004 ------ ------ War Baby Interaction Terms War Baby * 1998 0.6714 0.000 War Baby * 2000 0.1490 0.013 War Baby * 2002 -0.0119 0.814 War Baby * 2004 ------ ------ Source: Authors’ calculations based on the Health and Retirement Study. Table 5a Marginal Effects from Multinomial Logistic Regression Dependent Variable: First Transition from Full-Time Career Job Core Men on a Full-Time Career Job in 1992 Full-Time Career Job Bridge Job coef p-value coef p-value Age in 1992 less than 57 ------ ------ ------ ------ 57 to 61 -0.1202 0.000 -0.0785 0.001 62 to 64 -0.1642 0.000 -0.0634 0.171 65 and greater -0.1549 0.004 -0.0828 0.316 Education Less than high school 0.0193 0.220 -0.0298 0.279 High school graduate ------ ------ ------ ------ College graduate 0.0052 0.735 0.0815 0.005 Respondent Health Excellent/very good 0.0002 0.989 0.0978 0.000 Good ------ ------ ------ ------ Fair/poor -0.0198 0.304 -0.0944 0.008 Spouse Health Excellent/very good -0.0196 0.164 0.0181 0.504 Good ------ ------ ------ ------ Fair/poor -0.0049 0.800 -0.0342 0.363 Married 0.0072 0.740 0.0921 0.020 Dependent Child 0.0287 0.050 0.0001 0.998 Occupational Status White collar - high skilled 0.0260 0.172 -0.0642 0.058 White collar - other 0.0453 0.028 -0.0474 0.213 Blue collar - high skilled 0.0091 0.608 -0.1012 0.001 Blue collar - other ------ ------ ------ ------ Health Insurance Status Portable -0.0096 0.490 -0.0167 0.523 Non-portable ------ ------ ------ ------ None -0.0634 0.018 0.1154 0.021 Pension Status Defined-benefit -0.0793 0.000 -0.2294 0.000 Defined-contribution 0.0152 0.279 -0.0540 0.056 None ------ ------ ------ ------ Both -0.0300 0.424 0.1004 0.098 Self-Employed 0.0893 0.000 0.0443 0.219 Spouse Employed -0.0132 0.321 0.0542 0.027 Wage 0.0019 0.009 -0.0057 0.000 Wage Squared -0.0001 0.318 0.0001 0.007 Wealth 0.0001 0.848 -0.0003 0.691 Wealth Squared 0.0000 0.789 0.0000 0.721 Own Home 0.0694 0.001 0.0035 0.913 Constant -0.1801 0.000 0.1058 0.001 Source: Authors’ calculations based on the Health and Retirement Study. Table 5b Marginal Effects from Multinomial Logistic Regression Dependent Variable: First Transition from Full-Time Career Job War Baby Men on a Full-Time Career Job in 1998 Full-Time Career Job Bridge Job coef p-value coef p-value Age in 1998 less than 57 ------ ------ ------ ------ 57 and greater -0.0359 0.666 0.0197 0.776 Education Less than high school 0.1113 0.119 -0.0804 0.234 High school graduate ------ ------ ------ ------ College graduate -0.1188 0.025 0.1711 0.000 Respondent Health Excellent/very good -0.0512 0.249 0.0966 0.021 Good ------ ------ ------ ------ Fair/poor -0.1076 0.132 0.1106 0.083 Spouse Health Excellent/very good -0.1070 0.034 0.1480 0.002 Good ------ ------ ------ ------ Fair/poor 0.0458 0.539 -0.0300 0.686 Married 0.0500 0.497 -0.0786 0.249 Dependent Child 0.0135 0.746 -0.0051 0.897 Occupational Status White collar - high skilled 0.0634 0.323 -0.0960 0.107 White collar - other 0.0173 0.796 -0.0623 0.321 Blue collar - high skilled 0.0694 0.235 -0.0923 0.088 Blue collar - other ------ ------ ------ ------ Health Insurance Status Portable -0.1078 0.009 0.0192 0.626 Non-portable ------ ------ ------ ------ None -0.1881 0.046 0.0834 0.301 Pension Status Defined-benefit 0.1033 0.038 -0.1566 0.001 Defined-contribution 0.1122 0.020 -0.0628 0.170 None ------ ------ ------ ------ Both -0.1157 0.191 0.0579 0.517 Self-Employed 0.2235 0.001 -0.0577 0.343 Spouse Employed 0.0283 0.568 -0.0010 0.984 Wage 0.0064 0.013 -0.0063 0.005 Wage Squared 0.0000 0.212 0.0000 0.042 Wealth -0.0001 0.964 -0.0017 0.152 Wealth Squared 0.0000 0.361 0.0000 0.289 Own Home -0.0008 0.990 -0.0380 0.481 Constant 0.1069 0.196 0.0720 0.328 Source: Authors’ calculations based on the Health and Retirement Study. Table 5c Marginal Effects from Multinomial Logistic Regression Dependent Variable: First Transition from Full-Time Career Job Core Women on a Full-Time Career Job in 1992 Full-Time Career Job Bridge Job coef p-value coef p-value Age in 1992 less than 57 ------ ------ ------ ------ 57 and greater -0.2170 0.000 -0.0024 0.932 Education Less than high school -0.0185 0.450 -0.0959 0.004 High school graduate ------ ------ ------ ------ College graduate -0.0026 0.904 0.0867 0.009 Respondent Health Excellent/very good 0.0139 0.425 0.0833 0.001 Good ------ ------ ------ ------ Fair/poor -0.0441 0.127 -0.0246 0.516 Spouse Health Excellent/very good -0.0346 0.094 0.0154 0.627 Good ------ ------ ------ ------ Fair/poor 0.0260 0.296 -0.0507 0.214 Married 0.0158 0.519 0.0027 0.941 Dependent Child -0.0173 0.293 0.0302 0.220 Occupational Status White collar - high skilled 0.0580 0.034 -0.1003 0.007 White collar - other 0.0999 0.000 -0.0883 0.007 Blue collar - high skilled 0.1172 0.000 -0.0474 0.321 Blue collar - other ------ ------ ------ ------ Health Insurance Status Portable -0.0459 0.004 0.0777 0.003 Non-portable ------ ------ ------ ------ None -0.0744 0.045 0.2349 0.000 Pension Status Defined-benefit -0.0499 0.010 -0.2452 0.000 Defined-contribution 0.0250 0.165 -0.1866 0.000 None ------ ------ ------ ------ Both 0.0354 0.367 0.1974 0.004 Self-Employed 0.0420 0.141 -0.0875 0.052 Spouse Employed -0.0399 0.040 0.0620 0.040 Wage 0.0058 0.000 -0.0024 0.340 Wage Squared 0.0000 0.258 0.0001 0.321 Wealth 0.0006 0.206 -0.0020 0.011 Wealth Squared 0.0000 0.167 0.0000 0.012 Own Home 0.0285 0.245 0.0401 0.213 Constant -0.1749 0.000 0.1662 0.000 Source: Authors’ calculations based on the Health and Retirement Study. Table 5d Marginal Effects from Multinomial Logistic Regression Dependent Variable: First Transition from Full-Time Career Job War Baby Women on a Full-Time Career Job in 1998 Full-Time Career Job Bridge Job coef p-value coef p-value Age in 1998 less than 57 ------ ------ ------ ------ 57 and greater -0.0857 0.369 0.0938 0.242 Education Less than high school 0.2367 0.033 -0.2397 0.024 High school graduate ------ ------ ------ ------ College graduate -0.0556 0.363 0.1024 0.063 Respondent Health Excellent/very good -0.0296 0.576 0.0766 0.126 Good ------ ------ ------ ------ Fair/poor -0.0739 0.369 -0.0612 0.418 Spouse Health Excellent/very good 0.0157 0.794 0.0394 0.490 Good ------ ------ ------ ------ Fair/poor 0.0804 0.373 0.0116 0.892 Married 0.0176 0.845 -0.1419 0.098 Dependent Child 0.1079 0.027 -0.0993 0.030 Occupational Status White collar - high skilled 0.1879 0.012 -0.1098 0.101 White collar - other 0.1975 0.003 -0.1381 0.024 Blue collar - high skilled 0.2894 0.002 -0.1318 0.128 Blue collar - other ------ ------ ------ ------ Health Insurance Status Portable -0.0681 0.157 0.0687 0.129 Non-portable ------ ------ ------ ------ None -0.2067 0.090 0.1832 0.069 Pension Status Defined-benefit -0.0153 0.783 -0.0440 0.412 Defined-contribution 0.0609 0.246 -0.0965 0.054 None ------ ------ ------ ------ Both 0.0384 0.720 -0.0084 0.938 Self-Employed 0.1856 0.052 -0.0215 0.794 Spouse Employed -0.0383 0.611 0.0665 0.366 Wage 0.0079 0.206 -0.0026 0.644 Wage Squared 0.0001 0.683 -0.0001 0.301 Wealth 0.0000 0.996 -0.0004 0.759 Wealth Squared 0.0000 0.895 0.0000 0.834 Own Home 0.0305 0.721 -0.1205 0.093 Constant -0.0532 0.647 0.2312 0.022 Source: Authors’ calculations based on the Health and Retirement Study. Table 6a Coefficients from OLS Regression Dependent Variable: Hours Worked per Year Men with a FTC Job Since Age 49 Prior to Transition from FTC Job coef. p-value Age in 2004 57 or younger ------ ------ 58 - 61 27.1 0.002 62 - 64 -72.3 0.000 65 - 69 -133.2 0.000 70 or older -207.3 0.000 Respondent Health Excellent/very good 24.7 0.002 Good ------ ------ Fair/poor -17.7 0.143 Spouse Health Excellent/very good 12.1 0.172 Good ------ ------ Fair/poor 0.6 0.961 Education Less than high school -8.7 0.386 High school graduate ------ ------ College graduate -31.0 0.004 Race White ------ ------ Black -33.4 0.003 Other -29.6 0.145 Married 47.1 0.049 Dependent Children 11.6 0.258 Health Insurance Status Portable -24.6 0.003 Non-portable ------ ------ None 1.8 0.927 Pension Status Defined-benefit 14.1 0.094 Defined-contribution 72.3 0.000 None ------ ------ Both -24.2 0.100 Occupational Status White collar - high skilled 57.0 0.000 White collar - other 54.8 0.000 Blue collar - high skilled -2.7 0.778 Blue collar - other ------ ------ Self-employed 1.9 0.896 Wage -1.7 0.000 Wage Squared 0.0 0.000 Wealth ($1,000) 0.1 0.169 Wealth Squared ($1,000) 0.0 0.290 Own Home 18.9 0.082 Constant 2230.8 0.000 Regressors (continued) coef p-value War Baby Indicator -36.4 0.179 Year Indicators 1992 -110.3 0.000 1994 -72.2 0.002 1996 -147.2 0.000 1998 -75.5 0.002 2000 -76.3 0.002 2002 -113.2 0.000 2004 ------ ------ War Baby Interaction Terms War Baby * 1998 18.1 0.574 War Baby * 2000 4.6 0.892 War Baby * 2002 75.7 0.034 War Baby * 2004 ------ ------ Source: Authors’ calculations based on the Health and Retirement Study. Table 6b Coefficients from OLS Regression Dependent Variable: Hours Worked per Year Men with a FTC Job Since Age 49 After Transition from FTC Job coef. p-value Age in 2004 57 or younger ------ ------ 58 - 61 -4.7 0.862 62 - 64 -253.6 0.000 65 - 69 -397.2 0.000 70 or older -602.1 0.000 Respondent Health Excellent/very good 24.5 0.227 Good ------ ------ Fair/poor -77.1 0.006 Spouse Health Excellent/very good -14.6 0.511 Good ------ ------ Fair/poor 60.2 0.034 Education Less than high school 85.0 0.001 High school graduate ------ ------ College graduate -107.5 0.000 Race White ------ ------ Black -73.5 0.010 Other 56.0 0.312 Married 109.4 0.056 Dependent Children 90.0 0.000 Health Insurance Status Portable -234.6 0.000 Non-portable ------ ------ None -171.9 0.000 Pension Status Defined-benefit 175.7 0.000 Defined-contribution 301.7 0.000 None ------ ------ Both -174.0 0.006 Occupational Status White collar - high skilled 61.3 0.037 White collar - other 68.6 0.018 Blue collar - high skilled 38.1 0.120 Blue collar - other ------ ------ Self-employed -207.3 0.000 Wage -2.5 0.000 Wage Squared 0.0 0.040 Wealth ($1,000) -0.5 0.007 Wealth Squared ($1,000) 0.0 0.000 Own Home -124.8 0.000 Constant 1695.8 0.000 Regressors (continued) coef p-value War Baby Indicator 8.1 0.905 Year Indicators 1992 91.2 0.056 1994 129.1 0.005 1996 95.8 0.025 1998 103.7 0.010 2000 123.8 0.002 2002 175.6 0.000 2004 ------ ------ War Baby Interaction Terms War Baby * 1998 274.7 0.001 War Baby * 2000 181.7 0.045 War Baby * 2002 1.2 0.989 War Baby * 2004 ------ ------ Source: Authors’ calculations based on the Health and Retirement Study. Table 6c Coefficients from OLS Regression Dependent Variable: Hours Worked per Year Women with a FTC Job Since Age 49 Prior to Transition from FTC Job coef. p-value Age in 2004 57 or younger ------ ------ 58 - 61 -18.2 0.039 62 - 64 -15.7 0.318 65 - 69 -0.4 0.989 70 or older -59.2 0.365 Respondent Health Excellent/very good 10.1 0.204 Good ------ ------ Fair/poor -17.0 0.136 Spouse Health Excellent/very good -4.3 0.666 Good ------ ------ Fair/poor 18.8 0.124 Education Less than high school 9.6 0.441 High school graduate ------ ------ College graduate 25.4 0.027 Race White ------ ------ Black -40.3 0.000 Other 4.7 0.789 Married -77.7 0.007 Dependent Children -22.9 0.002 Health Insurance Status Portable -12.8 0.095 Non-portable ------ ------ None -47.9 0.026 Pension Status Defined-benefit 38.9 0.000 Defined-contribution 79.3 0.000 None ------ ------ Both -33.5 0.055 Occupational Status White collar - high skilled 50.2 0.000 White collar - other -4.0 0.721 Blue collar - high skilled 15.2 0.265 Blue collar - other ------ ------ Self-employed 26.5 0.196 Wage -4.1 0.000 Wage Squared 0.0 0.029 Wealth ($1,000) 0.4 0.003 Wealth Squared ($1,000) 0.0 0.028 Own Home 8.2 0.457 Constant 2124.9 0.000 Regressors (continued) coef p-value War Baby Indicator 62.6 0.010 Year Indicators 1992 -53.5 0.004 1994 -1.7 0.924 1996 -80.7 0.000 1998 -1.6 0.933 2000 2.9 0.881 2002 -18.6 0.390 2004 ------ ------ War Baby Interaction Terms War Baby * 1998 -53.6 0.078 War Baby * 2000 -53.1 0.086 War Baby * 2002 -28.1 0.406 War Baby * 2004 ------ ------ Source: Authors’ calculations based on the Health and Retirement Study. Table 6d Coefficients from OLS Regression Dependent Variable: Hours Worked per Year Women with a FTC Job Since Age 49 After Transition from FTC Job coef. p-value Age in 2004 57 or younger ------ ------ 58 - 61 -22.8 0.316 62 - 64 -223.2 0.000 65 - 69 -360.2 0.000 70 or older -532.7 0.000 Respondent Health Excellent/very good -6.2 0.762 Good ------ ------ Fair/poor -26.9 0.391 Spouse Health Excellent/very good -32.4 0.219 Good ------ ------ Fair/poor -4.0 0.903 Education Less than high school 16.5 0.567 High school graduate ------ ------ College graduate -130.5 0.000 Race White ------ ------ Black -148.1 0.000 Other -48.9 0.291 Married -113.4 0.059 Dependent Children -9.6 0.636 Health Insurance Status Portable -249.7 0.000 Non-portable ------ ------ None -190.1 0.000 Pension Status Defined-benefit 164.2 0.000 Defined-contribution 271.4 0.000 None ------ ------ Both -140.4 0.017 Occupational Status White collar - high skilled 19.6 0.541 White collar - other -26.0 0.309 Blue collar - high skilled 12.5 0.759 Blue collar - other ------ ------ Self-employed -55.9 0.082 Wage -4.5 0.000 Wage Squared 0.0 0.008 Wealth ($1,000) -1.7 0.000 Wealth Squared ($1,000) 0.0 0.000 Own Home -39.4 0.141 Constant 1842.5 0.000 Regressors (continued) coef p-value War Baby Indicator 50.0 0.536 Year Indicators 1992 35.4 0.489 1994 100.4 0.034 1996 227.4 0.000 1998 165.4 0.000 2000 144.7 0.001 2002 169.6 0.000 2004 ------ ------ War Baby Interaction Terms War Baby * 1998 240.2 0.014 War Baby * 2000 106.4 0.286 War Baby * 2002 -41.7 0.664 War Baby * 2004 ------ ------ Source: Authors’ calculations based on the Health and Retirement Study. RECENT WORKING PAPERS FROM THE CENTER FOR RETIREMENT RESEARCH AT BOSTON COLLEGE The Trajectory of Wealth in Retirement David A. Love, Michael G. Palumbo, and Paul A. Smith The Rising Age at Retirement in Industrial Countries Gary Burtless The Implications of Career Lengths for Social Security Melissa M. Favreault and C. Eugene Steuerle Do Out-of-Pocket Health Care Costs Delay Retirement? Richard W. Johnson, Rudolph G. Penner, and Desmond Toohey How the Income and Tax Treatment of Saving and Social Security Benefits May Affect Boomers’ Retirement Incomes Barbara A. Butrica, Karen E. Smith, and Eric J. Toder Saving and Wealth Accumulation in the PSID, 1984-2005 Barry P. Bosworth and Sarah Anders Older Women’s Income and Wealth Packages in Cross-National Perspective Timothy M. Smeeding, Janet C. Gornick, Eva Sierminska, and Maurice Leach How Many Struggle to Get By in Retirement? Barbara A. Butrica, Daniel Murphy, and Sheila R. Zedlewski, January 2008 The Impact of Late-Career Health and Employment Shocks on Social Security and Other Wealth Richard W. Johnson, Gordon B.T. Mermin, and Dan Murphy, December 2007 The Effect of Economic Conditions on the Employment of Workers Nearing Retirement Age Till von Wachter, December 2007 The Costs of Owning Employer Stocks: Lessons From Taiwan Yi-Tsung Lee, Yu-Jane Liu, and Ning Zhu, December 2007 All working papers are available on the Center for Retirement Research website (http://www.bc.edu/crr) and can be requested by e-mail (crr@bc.edu) or phone (617-552-1762).