How Does Health Insurance Affect the Retirement Behavior of Women?

Kanika Kapur Jeannette Rogowski How Does Health Insurance Affect the Retirement Behavior of Women? The availability of health insurance is a crucial factor in the retirement decision. Women are substantially less likely to have health insurance from their own employment. Using the Health and Retirement Study, we examine the role of employerprovided retiree health insurance in the retirement decisions of single women, and women in single-earner and dual-earner couples. We compare the effect of health insurance on female and male retirement. Our results show that retiree health insurance increases retirement for all groups except single men. We find suggestive evidence that the role of health insurance for women hinges on their husbands labor force status. Women constitute a large and increasing share of people retiring. Women comprised 51% of the retiring cohort in 2006, compared to 44% in 1994. 1 The aging of the baby boom generation will fuel an increase in female retirement in the upcoming decades. Most of our existing knowledge on retirement comes from research on married men. A few studies have examined married couples; however, comparatively little attention has been paid to the retirement behavior of women. Yet, married women constitute twothirds of female workers over age 55. Women in dual-earner couples comprise almost half of female workers over 55, and single women account for more than one-third of the over- 55 female workforce. 2 These statistics highlight the importance of understanding female retirement behavior, since women in various demographic groups are a large and growing share of the retiring workforce. The availability of health insurance is a crucial factor in the decision to retire before Medicare eligibility. Because the prevalence of poor health rises with age, the near elderly have higher expected medical expenses than the young. For instance, average annual total health care expenditures for people ages 45 to 64 are $5,474 compared to $2,129 for people ages 18 to 44 (AHRQ 2007). 3 Few routes to public insurance exist: Unless blind or disabled, or belonging to an eligible low-income group, people under age 65 cannot qualify for Medicare or Medicaid. Individually purchased private health insurance can be very expensive. The challenge of finding affordable health insurance options may be particularly acute for women considering retirement because women face substantially higher health insurance costs than men in the market for individual health insurance (Codispoti et al. 2008). Workers who leave an Kanika Kapur, Ph.D., is a senior lecturer at the School of Economics and a senior research fellow at the Geary Institute, University College Dublin, Ireland. Jeannette Rogowski, Ph.D., is a university professor in the Department of Health Systems and Policy, School of Public Health, University of Medicine and Dentistry of New Jersey. Support for this research came from the Department of Labor, Employee Benefits Security Administration (contract no. J-9-P-7-0045). Address correspondence to Dr. Kapur at School of Economics, University College Dublin, Belfield, Dublin 4, Ireland. Email: kanika.kapur@ucd.ie Inquiry 48: 51 67 (Spring 2011). 2011 Excellus Health Plan, Inc. ISSN 0046-9580 10.5034/inquiryjrnl_48.01.04 www.inquiryjournal.org 51

Inquiry/Volume 48, Spring 2011 insured job have the option to continue group coverage known as COBRA for up to 18 months by paying 102% of the premium. Only a small fraction of those eligible to purchase COBRA coverage do so, however. 4 Thus, prior to the age of Medicare eligibility (65), employment-based opportunities for health insurance remain a valuable resource for the near elderly. In fact, among people ages 54 to 64 in 2006, 71% were covered by employment-based sources of health insurance and 58% of early retirees in this age cohort had such coverage. 5 Despite the importance of retiree health insurance, employer coverage for retirees has been declining over time. While 68% of retirees were covered by employer-provided health insurance in 1992, only 34% of retirees had such coverage in 2006. 6 Women have been particularly disadvantaged in their access to retiree health insurance. In 2006, while 42% of male workers had access to retiree benefits, only 33% of female workers did. Studying women s retirement behavior separately from that of men s is necessary to forecast future retirement patterns. The determinants of retirement for men and women are likely to differ. Moreover, women in the labor force are heterogeneous and retirement patterns for the growing cohorts of single-earner women in couples and single women may differ from dual-earner married women. As noted in the media, single-earner women with husbands who stay home is a growing group (Nyhan 2008; Dunleavey 2006). Another demographic trend shows that the percentage of single mothers in the labor force increased from 62% in 1995 to 73% in 2000, but went down to 70% in 2003. 7 Single women are not representative of women in general, and tend to be more risk averse when facing retirement than married women or single men (Bernasek and Schwiff 2001), suggesting that retirement patterns for single women may differ from those of married women. Among married women, retirement behavior may also change over time. There are an increasing number of dual-earner households. As women begin to take on the role of the primary earners in these households, retirement timing decisions may also change. In this paper, we examine the role of employer-provided health insurance in the retirement decisions of single women, women in single-earner couples, and women in dualearner couples. We also conduct a parallel analysis for men so that we are able to compare the results for women to those for men. Our analysis aims to answer the following questions: Does health insurance have a similar effect on the retirement decisions of men and women? Does the availability of coverage through a spouse allow dual-earner couples flexibility in retirement compared to a single worker or singleearner couples? Does the effect of health insurance on the retirement decision for married workers vary depending on the labor force and health status of the spouse? We examine the effect of health insurance on retirement behavior using data from the Health and Retirement Study (HRS) from 1992 to 2006. The HRS is particularly well suited to our study since it contains detailed longitudinal information on labor force participation, retirement, health, and health insurance for near-elderly and elderly couples. Background If health insurance were perfectly portable in the transition from employment to retirement, the prospect of losing health insurance would not deter an individual from retiring. However, in the U.S., employer-provided health insurance is not portable. Once retired, individuals who used to have employerprovided coverage but do not have employer-provided retiree coverage may incur disruptions in their relationships with providers and changes in policy quality as a result of changing insurance plans. 8 Employers, particularly large employers, can capitalize on economies of scale, lower administrative costs, and higher bargaining power to gain lower health insurance costs compared to individuals. 9 In addition, employers do not have complete flexibility to offer health insurance to some employees and not others, and to vary wages in accordance with each worker s insurance costs. Therefore, older workers with high health costs may be paying far less than the true costs of their insurance under employer-provided group insurance. As a result, health insurance can reduce the 52

Retirement Behavior of Women incentive to retire for older workers who have employer-provided health insurance, but cannot keep it once they retire. We would expect retirement rates to be higher among those who have access to employer-provided retiree health insurance since this benefit equips individuals with portable health insurance. We would also expect retirement rates to be higher for individuals who have access to health insurance from a spouse s employerprovided plan. Once individuals reach age 65 and are eligible for Medicare, losing employer coverage should be less of a concern for those workers previously covered by employerprovided health insurance. 10 A large body of literature has demonstrated that retiree health insurance facilitates early retirement among men (Gruber and Madrian 2004). There is relatively little work examining the effect of retiree health insurance on female retirement. Quinn (1996) included women in his analysis of the impact of retiree health insurance, but the paper did not estimate separate effects of retiree health insurance by sex or by family work status (dual-earner couple, couple with a single earner, or single worker). Rogowski and Karoly (2001) used data from the 1992 HRS and found that retiree health insurance benefits increased female retirement by five percentage points. Strumpf (2007) analyzed the effect of health insurance on retirement, health, and health care for men and women separately, and found quite similar effects of retiree health insurance for men and women; however, this analysis pooled single and married individuals. Several other papers have studied female retirement behavior; however, they did not focus on the role of retiree health insurance benefits (Honig 1985; Pozzenbon and Mitchell 1986; Dano, Ejarnaes, and Husted 2004). The recent literature has recognized the importance of modeling couples retirement decisions jointly. This work has demonstrated that husbands and wives tend to retire at the same time. 11 Furthermore, spousal health and Social Security and pension incentives are often found to affect retirement and labor force participation (Coile 2004; Johnson and Favreault 2001; Hurd 1990; O Rand, Henretta, and Krecker 1992; Pienta 1997). 12 Two recent studies focused on the effect of postretirement health benefits on joint retirement decisions among couples, and found that health benefits have a relatively modest effect on the overall labor force behavior of married couples (Blau and Gilleskie 2006; Kapur and Rogowski 2007). As already noted, we add to the existing literature with this paper by examining the role of retiree health benefits separately for three groups: dual-earner couples, couples with a single earner, and single workers. We examine retirement behavior for women and compare our results to those for men. While previous work has looked at some of these groups, our work aims to provide a comprehensive analysis using the same data set and analytic framework to compare and contrast the effect of retiree insurance on retirement among these key demographic groups. We also explicitly account for the role of spousal health, health insurance, and work status. Our results highlight the importance of analyzing these groups separately and show that the effect of retiree health insurance differs substantially depending on the demographic group. Data Our analysis uses the Health and Retirement Study, a nationally representative biennial panel survey of individuals born from 1931 to 1941. We used the first eight waves of the HRS (1992 2006). The HRS data have been linked to Social Security earnings histories and to employer pension plan data for the majority of the sample. Data were collected for sampled individuals and their spouses. The data are restricted to individuals who were working full time at the baseline year, were under age 65, and had non-missing data for our key analysis variables health insurance and labor force participation. We have restricted our data to the under-65 age group since employer-provided retiree benefits are likely to be most valuable before Medicare eligibility. The analysis data set has a panel structure with separate observations for each individual wave. Once an individual retires, the individual exits the data set. 13 Our final sample had 6,348 dual-earner couple-wave observations (6,348 men and 6,348 women), 53

Inquiry/Volume 48, Spring 2011 1,983 single male observations, 5,021 single female observations, 2,699 men belonging to single-earner couples, and 2,085 women belonging to single-earner couples. We measured retirement as a transition in the labor force status from full-time work at the baseline year to self-reported part or full retirement at the next survey date. 14 We classified health insurance for each individual using the following mutually exclusive variables: a) holds own employer-provided health insurance, but no retiree health insurance, b) holds own employer-provided health insurance and retiree health insurance, and c) does not hold own employer-provided health insurance and does not hold retiree health insurance. Employer rules commonly require that workers hold employer-provided health insurance to be eligible to participate in retiree health insurance. As a result, in the HRS data, the question on retiree health insurance is only asked of individuals who report having employer-provided health insurance. So, category b) defined previously, combines employer-provided health insurance and retiree health insurance. In 1992 and 1994, HRS measured retiree health insurance using a question that asked whether the employer-provided plan reported by the surveyed individual covered retirees. In 1996, HRS changed the wording of the question to ask whether surveyed individuals could continue coverage to age 65 if they left their jobs now. The focus of our study is on measuring retiree health insurance before age 65, and 99% of employers that offer retiree health benefits provide coverage for retirees younger than age 65; therefore, it is unlikely that this change in the HRS had much impact on our results (Kaiser/Hewitt 2006). In our analysis, we included year indicators to absorb the effect of changes in the survey instrument over time. 15 We used linked, restricted access data from the Social Security Administration (SSA) based on the Social Security Earnings and Benefits File to construct measures of Social Security wealth. Linked data were available for 88% of our sample. In cases where linked data were not available, we used measures of Social Security wealth imputed as part of the RAND HRS-SSA project (StClair et al. 2002). The Social Security variables included in the RAND HRS-SSA data are household Social Security wealth in 1992, projected household Social Security wealth at age 62, and projected household Social Security wealth at age 65. We used these variables to construct measures of household Social Security wealth at the baseline year and the gain in Social Security wealth obtained by delaying retirement past the follow-up survey year. 16 We constructed pension measures using self-reported pension data and employer reports of pension benefits using the supplemental pension data provided by the HRS. Validation research suggests that self-reported pension data are preferable to employer information for defined contribution (DC) plans. However, employer reports are likely to be a better measure of pension wealth for defined benefit (DB) plans. Therefore, the pension measures in our study were based on self-reported DC pension balances as the measure of DC pension wealth. 17 The employer-reported pension variables included in the HRS data are employer pension wealth in 1992 from current and past jobs, projected pension wealth at age 62, and projected pension wealth at age 65. We used these variables to construct measures of pension wealth at the baseline year and the gain in pension wealth obtained by delaying retirement past the follow-up survey year using a method similar to the construction of Social Security wealth. 18 We also included indicators for type of pension plan DB plan, DC plan, or no plan that were constructed using self-reports of pension benefits. Table 1 contains weighted means for the analysis sample. The table shows that over 40% of men have retiree health insurance from their own jobs. Women are less likely to have retiree health insurance from their jobs; however, this gap is the smallest for single women. Interestingly, single women are more likely than single men to hold employerprovided health insurance that is not accompanied by a retiree health plan. On whole, single men and single women are equally likely to have employer-provided health insurance. The table also shows that women have lower wages, are less likely to have a college degree, have lower tenure, and are less 54

Retirement Behavior of Women Table 1. Means in the Health and Retirement Study (1992 2006) Dual-earner couples Singles Single-earner couples Men Women Men Women Men Women Own variables Retire in current wave.15.12.17.17.21.25 REPHI and EPHI.41.24.43.35.46.30 EPHI, no REPHI.25.23.25.31.29.27 No EPHI and no REPHI.34.53.32.33.25.43 Wage ($ per hour) 21.38 14.51 19.99 13.81 22.71 15.56 Age 55.76 52.49 56.05 56.66 56.91 57.30 Nonwhite.09.09.17.22.09.12 High school.55.67.56.62.57.66 College.32.25.27.21.25.17 Health conditions.93.87.94 1.22 1.00 1.20 Tenure (years) 15.72 10.71 14.30 11.72 16.85 13.39 Wealth ($) 319,972.50 309,644.90 179,548.90 138,743.30 368,060.90 280,080.10 IRA wealth ($) 39,580.00 41,011.02 19,509.97 15,254.14 42,865.45 46,804.02 DC pension plan.37.32.33.32.39.32 DB pension plan.40.29.31.29.42.33 DB wealth ($) 46,118.38 17,903.60 32,311.57 23,079.66 59,115.95 25,157.89 DB gain ($) 112,956.80 90,698.33 73,180.14 63,767.11 94,035.83 65,924.90 DC wealth ($) 39,334.20 18,347.83 31,007.00 19,339.26 42,764.56 22,541.87 Social Security wealth ($) 158,728.50 158,728.50 83,595.66 75,822.03 158,262.90 173,004.80 Social Security gain ($) 35,065.33 35,065.33 14,199.73 20,940.83 30,641.90 30,618.58 Spouse variables REPHI and EPHI.06.25 EPHI, no REPHI.04.09 No EPHI and no REPHI.91.65 High school.62.57 College.14.17 Spouse out of labor force.61.02 Spouse retired.25.83 Spouse unemployed.06.06 Spouse disabled.08.10 Number of observations 6,348 6,348 1,983 5,021 2,699 2,085 Notes: EPHI5 employer-provided health insurance; REPHI5 retiree employer-provided health insurance; DC5 defined contribution; DB5 defined benefit. 55

Inquiry/Volume 48, Spring 2011 likely to have pension plans than men. The gap between men and women is the smallest for single workers along several dimensions, such as pension plan holding, pension wealth, and health insurance. Most men belonging to single-earner couples have spouses who are out of the labor force. However, 83% of women belonging to single-earner couples have husbands who are retired. Twenty-five percent of single-earner women have access to retiree health insurance through a spouse s plan providing a valuable alternative source of health insurance. Single-earner men have higher retirement rates than men in dual-earner couples; they also have higher wealth, and are more likely to have access to retiree health insurance. Empirical Model Our analysis focuses on the role of health insurance in retirement decisions. We estimate reduced-form multivariate models of the decision to retire for single women, single men, and dual-earner couples. The advantage of the reduced-form approach compared to a structural approach is in the transparency of the modeling assumptions and the flexibility in incorporating changes in status from dual earner to single earner. We model the decision to retire for singles using a discrete time probit model. For dual-earner couples, we use a discrete time bivariate probit model to account for joint decision-making in the household. Our primary explanatory variable of interest is whether the worker holds: a) employer-provided health insurance (EPHI) that includes retiree health insurance, b) employer-provided health insurance without retiree health insurance, or c) no employerprovided health insurance and no retiree health insurance. These three variables are defined to be mutually exclusive. As Table 1 shows, for dual-earner women, 24% have EPHI and retiree health insurance, 23% have EPHI with no retiree health insurance, and 53% have no EPHI. There is a substantial group of women who have EPHI, but do not have retiree health insurance. We are interested in identifying the effect of retiree health insurance separately from EPHI (with no retiree health insurance) since workers lose EPHI after retiring; however, if workers have retiree health insurance in addition to EPHI, workers can continue to be covered by employer health insurance after retiring. We also include a number of demographic, health, and job control variables in our models. There are several advantages of the discrete time approach. This approach enables us to use data from individuals who do not retire during the year. We are also able to update health insurance information and other explanatory variables to the appropriate year, rather than rely on baseline characteristics measured in the first year of the survey. Furthermore, our measurement of retirement between survey dates implies that the analysis is less affected by seam bias in retirement reporting or errors in the retirement dates. The bivariate probit allows us to jointly model the retirement behavior of husbands and wives. When one member of the couple retires, the couple exits the dual-earner couple model and is treated as a single-earner couple in the descriptive analysis described in the next section. We assume that retirement behavior for dual-earner couples is determined by the following model: R hft ~azb 1 HI hft zb 2 HI wft zb 3 X hft zb 4 X wft zb 5 Z ft ze hft R wft ~azb 1 HI hft zb 2 HI wft zb 3 X hft zb 4 X wft zb 5 Z ft ze wft where R * hft and R * wft are the unobserved latent variables that measure the propensity for the husband (h) and wife (w) belonging to the family (f) to retire at time t. Since these are latent variables, we observe only R hft 5 1ifR * hft. 0 and R hft 5 0 otherwise. And similarly, for the wife, we observe only R wft 5 1ifR * wft. 0 and R wft 5 0 otherwise. We assume that the random error terms, e hft and e wft, have the following properties: E(e hft )~E(e wft )~0 Var(e hft )~Var(e wft )~1 Cov(e hft, e wft )~r This bivariate probit model is estimated using maximum likelihood in STATA. The 56

Retirement Behavior of Women key variables of interest are the husband s and wife s employer-provided health insurance, denoted by the vector HI hft for the husband s health insurance and HI wft for the wife s health insurance. HI hft includes an indicator for the husband having retiree health insurance offered through his own employer and an indicator for the husband having no employer-provided health insurance through his own employer. The omitted category is the husband having employerprovided health insurance, but no retiree health insurance from his own employer. HI wft for the wife s employer-provided health insurance is defined analogously. The control variables in the model include a full set of demographic controls for husband and wife characteristics (X hft and X wft ). These include categories for the husband s age and the wife s age. Since the gap between the husband s and wife s age is likely to be important in determining retirement, we have also included interactions between husband and wife age categories. Other control variables include husband and wife education (less than high school, high school, or college) and husband and wife health parameterized using the number of medical conditions. 19 Poor health is likely to reduce productivity in the workforce and the ability to work, and therefore, to increase the propensity to retire. Having a spouse in poor health may encourage an individual to retire sooner if caregiving is important. On the other hand, medical care for a health condition is costly; therefore individuals may have an incentive to keep working to pay for a spouse s health care. We would expect health insurance to play a role in determining whether couples with adverse health retire sooner. We test this hypothesis by including interactions of health insurance and health in the model. We have also included control variables for the husband s and wife s job characteristics. These include wages, pension benefits, and job tenure. Since the opportunity cost of retirement is higher for high wage workers, we would expect high wage workers to be less likely to retire. 20 On the other hand, DC employer-provided pension benefits should increase workers ability to retire by providing resources for retirement. The effect of DBs on retirement depends on the expected gain from waiting to retire a higher expected gain should reduce retirement in the current time period. When pension wealth data are missing, we include an indicator for missing data in the model and estimate the model on the full sample. 21 The vector Z ft includes family level variables such as race, household wealth, Social Security wealth, length of time between interviews, and year indicators. 22 Models are estimated using STATA s svy commands that take account of the HRS sample design by specifying sampling strata, household identifiers, and sampling weights. For single workers, the retirement decision is modeled by a single probit equation, where W denotes the standard normal cumulative density function: Pr (R it ~1)~W(azb 1 HI it zb 2 X it ) The variable R it determines whether the individual i retires at time t. The variables HI it and X it measure the individual s (i) health insurance, and demographic and job characteristics. In the single-worker model, spousal variables do not exist. Results The results from the bivariate probit models for dual-earner couples and from the probit models for singles are reported in Tables 2 and 3 for women and in Tables 4 and 5 for men. We report marginal effects and the p- values for the effects being statistically significantly different from zero. We find that r is.3 and statistically significant in the bivariate probit model for dual earners. This result supports the use of a bivariate probit model of joint decision-making in retirement for dual-earner couples. Effect of Health Insurance on Retirement Table 2 shows that retiree health insurance from a woman s own employer increases the probability of retirement for women in dualearner couples and for single women relative to the reference category of having one s own EPHI, but no retiree health insurance. The 57

Inquiry/Volume 48, Spring 2011 Table 2. Effect of health insurance on the probability of female retirement Women in dual-earner couples Single women Marginal effect p-value Marginal effect p-value Own EPHI, no REPHI (reference category) Own REPHI and EPHI.030.03.048.01 Own no EPHI, no REPHI.049.00.048.01 Spouse EPHI, no REPHI (reference category) Spouse REPHI and EPHI 2.008.47 Spouse no EPHI, no REPHI 2.009.40 Own wage 2.018.02 2.030.01 Spouse wage.015.10 Nonwhite 2.017.22.006.75 Own high school 2.059.00 2.059.00 Own college 2.063.00 2.074.00 Spouse high school 2.020.13 Spouse college 2.013.42 Own conditions.020.00.038.00 Spouse conditions.000.57 Own tenure.002.00.001.05 Spouse tenure.002.56 Log wealth.001.55.003.09 Log IRA wealth.000.69.002.28 DC pension plan 2.046.15 2.064.16 DB pension plan.045.15.077.12 DB wealth (log).001.34.003.05 DB gain (log) 2.002.36 2.004.36 DC wealth (log).006.07.002.63 SS wealth.003.71.003.43 SS gain (log) 2.001.85 2.004.38 Number of observations 6,348 5,021 Note: Age indicators, wave indicators, and time between interviews were also included in the model. EPHI5 employer-provided health insurance; REPHI5 retiree employer-provided health insurance. Omitted health insurance category is EPHI, no REPHI. DC5 defined contribution; DB5 defined benefit; SS5 Social Security. Table 3. Effect of health insurance on female retirement: predicted retirement probabilities based on model estimates Women in dual-earner couples (%) Single women (%) Predicted probability of retirement (sample average) 11 17 Own EPHI, no REPHI 8 13 Own EPHI and REPHI 11 18 Own no EPHI, no REPHI 13 18 Note: Predicted probabilities are obtained from the model estimates in Table 2. All predictions are based on own EPHI and own REPHI. EPHI5 employer-provided health insurance; REPHI5 retiree employer-provided health insurance. effect of retiree health insurance is quite similar for women in dual-earner couples and single women (three percentage points and 4.8 percentage points, respectively). Single women and dual-earner women who do not have their own employer health insurance are almost five percentage points more likely to retire compared to women who have employer health insurance, but no retiree health insurance (the reference group). For dual-earner women, the difference between having retiree EPHI and having no EPHI is statistically significant (p5.08). Table 3 presents predicted probabilities of retirement based on the model estimates obtained in Table 2. The first row of Table 3 simply presents the predicted probability of retirement for the estimation sample. As expected, the predicted probability is very 58

Retirement Behavior of Women Table 4. Effect of health insurance on the probability of male retirement Men in dual-earner couples Single men Marginal effect p-value Marginal effect p-value Own EPHI, no REPHI (reference category) Own REPHI and EPHI.047.00.039.14 Own no EPHI, no REPHI.025.10.053.03 Spouse EPHI, no REPHI (reference category) Spouse REPHI and EPHI 2.010.38 Spouse no EPHI, no REPHI 2.013.27 Own wage 2.010.19 2.015.34 Spouse wage.012.11 Nonwhite.009.59.028.20 Own high school 2.017.19.012.59 Own college 2.017.26.000.99 Spouse high school.003.83 Spouse college 2.027.15 Own conditions.026.00.024.00 Spouse conditions.002.76 Own tenure.001.21.002.01 Spouse tenure.002.01 Log wealth.008.01.003.34 Log IRA wealth 2.001.55.001.72 DC pension plan 2.068.05 2.069.20 DB pension plan.051.14 2.019.76 DB wealth (log).002.10.006.03 DB gain (log) 2.001.81.004.37 DC wealth (log).006.07.002.68 SS wealth.013.07.016.00 SS gain (log) 2.019.03 2.020.01 Number of observations 6,348 1,983 Notes: Age indicators, wave indicators, and time between interviews were also included in the model. Omitted health insurance category is EPHI, no REPHI. EPHI5 employer-provided health insurance; REPHI5 retiree employer-provided health insurance; DC5 defined contribution; DB5 defined benefit; SS5 Social Security. Table 5. Effect of health insurance on male retirement: predicted retirement probabilities based on model estimates Men in dualearner couples (%) Single men (%) Predicted probability of retirement (sample average) 14 17 Own EPHI, no REPHI 12 13 Own EPHI and REPHI 16 Own no EPHI, no REPHI 14 19 Note: Predicted probabilities are obtained from the model estimates in Table 4. All predictions are based on own EPHI and own REPHI. The marginal effect of own EPHI and REPHI for single men is not statistically different from own EPHI, no REPHI. Hence the prediction is replaced with a dash ( ). EPHI5 employer-provided health insurance; REPHI5 retiree employer-provided health insurance. similar to the raw sample average in Table 1. The predicted probabilities in the next three rows are obtained by successively switching on and off the indicators for the health insurance variables used in the model (EPHI, no retiree EPHI; EPHI and retiree EPHI; and no EPHI and no retiree EPHI). Eleven percent of dual-earner women with retiree EPHI retire in a wave compared to only 8% who have EPHI, but no retiree EPHI. Dualearner women who have no EPHI are more likely to retire than either of the insured groups. Single women with retiree EPHI and women with no EPHI are more likely to retire (18%) compared to women who have only EPHI, but no retiree EPHI. The effect of a husband s retiree health insurance is relatively small and statistically insignificant for dual-earner women in Table 2. 59

Inquiry/Volume 48, Spring 2011 Table 4 reports the results for men. Retiree health insurance increases retirement 4.7 percentage points for dual-earner men relative to having EPHI, but no retiree health insurance. However, the effect of retiree EPHI on single men is positive, but statistically insignificant. Men who have no EPHI are also somewhat more likely to retire than men who have EPHI, but no retiree health insurance. Dual-earner men with no EPHI are 2.5 percentage points more likely to retire and single men with no EPHI are 5.3 percentage points more likely to retire than men with EPHI, but no retiree EPHI. However, the difference between having retiree EPHI and having no EPHI is not statistically significant for either model. Among dual-earner couples, spousal retiree benefits have little effect on the propensity to retire for men. Table 5 shows the predicted probabilities of retirement obtained by switching on and off the health insurance variables in Table 4. For dual-earner men with retiree EPHI, the predicted probability of retirement is 16% compared to 12% for men with EPHI, but no retiree EPHI. For single men, the group with no EPHI has a relatively high retirement propensity of 19% compared to only 13% for men with EPHI, but no retiree EPHI. Effect of Other Factors on Retirement Women in dual-earner couples and single women with high wages are less likely to retire early, consistent with the notion that the opportunity cost of early retirement is higher for individuals with high wages. We also find that individuals with adverse health conditions are more likely to retire early across all demographic groups. An additional health condition increases the probability of retirement by between two and four percentage points depending on the demographic group of the worker. We find no effect of poor spousal health on the retirement propensity of the worker. The result that one s own poor health inhibits work is consistent with the results found in the literature. The effect of Social Security lacks statistical significance for the female models. However, the male models show some evidence that higher Social Security wealth is associated with an increased propensity to retire early. Furthermore, as expected, if the gain in Social Security wealth from postponing retirement is high, men are less likely to retire. DB pension wealth also significantly increases the probability of retirement for dual-earner men, single men, and single women, as does DC pension wealth for dual-earner men and women. Additional Models and Specification Checks Single-Earner Models The dual-earner models described earlier address the joint decision-making of husband and wife by using a bivariate probit framework. After one spouse retires, the retirement decision of the remaining working spouse is modeled using a discrete time probit model. We estimate multivariate models of the decision to retire for single-earner couples with a female head, and single-earner couples with a male head. A potential problem with these models is that they condition on the past labor force decision of the spouse who has already exited the labor force. These decisions may be endogenous. We have reestimated these models with and without the labor force decisions of the spouse, and found similar results for retiree health insurance; however, given these concerns we present these models for descriptive completeness and interpret the results with caution. For single-earner married female workers, the retirement decision is modeled by a single probit equation, where W denotes the standard normal cumulative density function: Pr (R wft ~1)~W azb 1 HI wft zb 2 HI hft zb 3 X wft zb 4 X hft zb 5 Z ft As in the bivariate probit model described previously, the variable R wft determines whether the wife (w) belonging to family f retires at time t. The variables HI wft,hi hft, X wft,x hft, and Z ft are defined in the same way as earlier. Since the category of single-earner married women may include individuals with spouses who have retired from the labor force and may have access to retiree health insurance from a previous job, we include indicators for the spouse s health insurance in 60

Retirement Behavior of Women the model. We also include variables describing the spouse s labor force designation (retired, disabled, unemployed, or out of the labor force). The probit model for singleearner married male workers is identical, except the dependent variable measures the husband s transition to retirement. Table 6 shows that the effect of retiree health insurance (relative to having EPHI, but no retiree EPHI) more than doubles for women in single-earner couples (10.9 percentage points) compared to women in dualearner couples. Women appear not to avail themselves of their retiree health benefits until their spouses leave the labor force, and then they seem to use their retiree health benefits to retire. 23 The predicted probability of retiring for women with retiree EPHI is 26%, compared to 16% for women with EPHI, but no retiree EPHI. For women with no EPHI, the predicted probability of retiring is 28%. 24 It is possible that the pattern we observe may stem from couples who value joint leisure and do not have retiree EPHI from the husband s job. After the husband retires, the wife chooses to join him in retirement if she has retiree EPHI to cover both of them. If she does not have retiree EPHI, she remains employed. We disaggregate women in single-earner couples into two groups short-term single earners and long-term single earners. Longterm single earners are women whose husbands did not work at any time in the five years before the first wave of the HRS. The labor force status of the spouse is less likely to be endogenous for long-term single earners. Short-term single earners refer to all other single-earner women. Not surprisingly, most women are short-term single earners, and retiree health insurance has a substantial effect on retirement for this group. The effect of retiree health insurance for long-term single earners is somewhat imprecise, given the small sample size. Retiree health insurance increases retirement by six percentage points for singleearner men relative to having EPHI, but no retiree health insurance. The predicted probability of retiring for men with retiree EPHI is 23%, compared to 17% for men with EPHI, but no retiree EPHI. For men with no EPHI, the predicted probability of retiring is also 17%. Unlike for dual earners, among singleearner men and women, spousal retiree health insurance has a relatively large effect on retirement. For single-earner women, spouse retiree EPHI increases retirement by almost 10 percentage points relative to women who have spouse EPHI, but no spouse retiree EPHI. For single-earner men, the effect is even larger (16 percentage points), suggesting that spousal retiree health insurance is an important benefit that is available to married workers but not to single workers. Single-earner men with wives who have already retired are 7.7 percentage points more likely to retire than single-earner men whose wives have not been in the labor force, suggesting that single-earner men have a preference for timing their retirement close to their wives retirement. It is unclear why single-earner men who have retired wives hasten their own retirement compared to single-earner men whose wives are not in the labor force. One possibility is that it may be relatively more important for men who have always been sole earners to stay in the labor force to maintain income. Another possibility is that couples in which the man has always been the sole earner are not as reliant on the joint consumption of leisure. Women whose husbands are unemployed are 19.2 percentage points less likely to retire, suggesting that women stay in the labor force to maintain family income when a spouse is unemployed. Among single-earner couples, women are less likely to retire early if their spouses have poor health. This suggests that women continue to work when their spouses have to leave the labor force. There is some evidence that DC pension wealth increases the probability of retirement for single-earner men and women. This suggests that single earners may be waiting for adequate pensions before they retire. Endogeneity of Retiree Health Insurance Our analysis assumes that health insurance is determined exogenously from the employment decision. We assume that individuals do not choose health insurance based on their expected retirement choices. The exogeneity 61

Inquiry/Volume 48, Spring 2011 Table 6. Effect of health insurance on single-earner couples All single earners Marginal effect p-value Women in single-earner couples Men in single-earner couples Short-term single earners Marginal effect p-value Long-term single earners All single earners Marginal effect p-value Marginal effect p-value Short-term single earners Marginal effect p-value Long-term single earners Marginal effect p-value Own EPHI, no REPHI (reference category) Own REPHI and EPHI.109.00.128.00.251.19.064.00.051.10.071.01 Own no EPHI, no REPHI.128.00.148.00.388.04 2.006.82 2.007.84 2.011.79 Spouse EPHI, no REPHI (reference category) Spouse REPHI and EPHI.096.02.102.02.163.01.163.04 Spouse no EPHI, no REPHI.059.12.050.19.064.23.100.11 Own wage 2.001.93 2.002.94.015.87 2.035.00 2.052.00 2.010.65 Nonwhite.000.99 2.021.47.251.22.001.97 2.020.62.023.49 Own high school 2.035.18 2.060.09.101.64 2.028.19 2.035.31 2.034.32 Own college.009.82 2.048.36.415.14 2.025.44 2.004.93 2.055.22 Spouse high school.050.06.050.08.220.24 2.006.84 2.017.59.021.53 Spouse college.065.10.091.03.081.77 2.014.70 2.017.72 2.007.89 Own conditions.025.01.025.02.172.01.027.00.029.03.024.03 Spouse conditions 2.013.09 2.018.05 2.024.68.000.99.001.91.000.96 Own tenure.000.76 2.002.15.018.02.003.00.004.00.003.01 Log wealth 2.001.93.002.66 2.034.46.006.21.005.51.005.31 Log IRA wealth.002.34 2.001.81.030.05.004.05.004.16.003.20 DC pension plan 2.136.04 2.154.06 2.714.20 2.087.13 2.094.19 2.125.19 DB pension plan.115.16.097.15.573.26.000.99 2.133.07.112.06 DB wealth (log).002.57.005.11 2.037.08.005.02.006.12.003.41 DB gain (log) 2.007.19 2.006.27 2.037.35 2.001.65.009.13 2.011.01 DC wealth (log).013.04.015.08.070.22.000 1.00 2.002.80.007.48 SS wealth.007.49.007.48.192.04.024.04.036.07.012.26 SS gain (log) 2.002.86 2.001.96 2.155.06 2.035.01 2.037.09 2.037.00 Spouse out of labor force (reference category) Spouse retired 2.123.18 2.057.53.077.00.111.00 Spouse unemployed 2.192.08 2.155.17 2.007.86 2.014.78 Spouse disabled 2.127.20 2.079.44.015.67.046.35 Number of observations 2,085 1,561 524 2,699 1,436 1,263 Note: Age indicators, wave indicators, and time between interviews were also included in the model. Omitted health insurance category is EPHI, no REPHI. EPHI5 employer-provided health insurance; REPHI5 retiree employer-provided health insurance; DC5 defined contribution; DB5 defined benefit; SS5 Social Security. 62

Retirement Behavior of Women assumption also implies that retiree health insurance is uncorrelated with unmeasured factors that affect retirement. We conduct two specification checks in this section to validate this assumption. First, we drop workers with short tenures (fewer than five years), since it is most likely that these individuals may be shopping for retiree health benefits. The effect of retiree health insurance is very similar for this subsample of workers (results available on request). Second, we add additional variables to the models that include measures of the physical demands of the job, job stress, weeks worked, hours worked, and detailed industry and occupation controls. Our original models already included pension and Social Security wealth information, wages, and tenure. Despite the addition of these comprehensive job controls, we find that the estimate of retiree health insurance continues to be very similar. Given that we have included a host of potentially important observable job attributes and found little change in the effect of retiree health insurance, it is unlikely that an unobservable attribute could be responsible for our finding. Third, we conduct propensity score matching to ensure that the characteristics of our sample with retiree health insurance is as similar as possible to the sample with no retiree health insurance according to the observable variables in our data. We estimate a probit model for the probability of having retiree health insurance with all the relevant explanatory variables in our analysis data set. Next, we weight our models with the propensity score constructed from this model. These models yield very similar results to our original estimates. 25 Other than these empirical checks, there are several reasons that we believe that our assumption of exogeneity is valid. Employees typically must satisfy tenure requirements before qualifying for retiree health benefits. The Kaiser/Hewitt employer survey (2004) found that 89% of employers had age and tenure requirements for retiree health benefit eligibility. The most frequent tenure requirement was 10 years of service (49%), followed by 15 years of service (14%). In our sample, average job tenure was 19 years for male retirees and 13 years for female retirees, showing that individuals choose jobs well before retirement. It is usually impossible for individuals to change jobs close to their retirement dates in order to obtain retiree health benefits, since in most jobs they would not be eligible for benefits. In our data, only 2.2% of men and 1.9% of women switched from jobs that did not offer retiree health benefits to jobs that offered these benefits; this suggests that there are very few job transitions that are consistent with the notion of shopping for retiree health benefits. Furthermore, given the large drop in the employer offering of retiree health insurance plans, individuals who wanted to shop for retiree benefits would have no guarantee that the plan would be around when they decided to retire. Therefore, like most of the existing literature (see for example Kapur and Rogowski 2007 and Blau and Gilleskie 2006), we assume that health insurance choice is determined exogenously. Alternative Model Specifications We conducted a number of specification checks that are not reported in the tables. First, we reran the models using measures of health insurance coverage from the current job only. These measures are available from 1996 onward. The health insurance measure used in the main models does not specify whether the health insurance is from a current or previous job. We found that the results for the health insurance measures were very similar to the results from the main model. We do not report these as our main results since we prefer a health insurance measure that is consistently defined over time. Second, we included interactions between the health insurance variables and health measures available in the HRS, including number of medical conditions for the husband and wife, number of functional limitations, and self-reported health. We did not find any statistically significant interactions in our models. Third, we re-estimated the models without restricting the sample to individuals under age 65. We would expect employer health insurance to have a smaller effect on the Medicare-eligible; however, we found very similar results probably due to the fact that most of our sample is under age 65. 63

Inquiry/Volume 48, Spring 2011 We also re-estimated the models including an interaction between the over age 65 and health insurance variables, but found no statistically significant effects. Fourth, we have re-estimated our models including controls for government coverage (Medicare, Medicaid, and other sources) and individually purchased health insurance. These variables often have statistically significant direct effects on retirement. For instance, we observe that government and privately purchased insurance are associated with an increase in the propensity to retire. However, our findings for the effect of retiree health insurance on retirement remain unchanged. We prefer not to include the alternative insurance variables in our main models since we believe that they are particularly likely to be endogenous. For instance, individuals who wish to retire early may seek to purchase individual health insurance. We also have re-estimated all models dropping individuals who have these alternative sources of coverage and found similar results. Conclusion This paper has analyzed the effect of health insurance on the retirement decision for women, parsing out the effect for single women, women in single-earner couples, and women in dual-earner couples. We also have compared the role of health insurance in female retirement to male retirement. Our findings from our empirical models suggest that men and women in most demographic groups respond similarly to their own retiree health insurance availability. However, married women become substantially more responsive to the availability of retiree health insurance from their own jobs once they become sole earners. This pattern is robust to various specification checks; however, it still should be interpreted with caution since our single-earner models do not jointly model the labor force status of nonworking spouses. Married women in dual-earner couples are less likely to avail themselves of their retiree health benefits to retire until their husbands have retired; however, once their husbands have retired, married women who have retiree benefits tend to leave the labor force. Single earners are also more likely to retire if their spouses can obtain retiree health coverage to cover themselves since this reduces the need for the worker to remain in the labor force to provide their nonworking spouse with health coverage. Relative to single workers, dualearner couples benefit from the option of spousal retiree health insurance. These results suggest that examining retirement patterns for each demographic group individually is important in forecasting changes to the workforce. Women are less likely to have retiree health insurance than men. For single women, who do not have the option of coverage through a spouse, the lack of retiree health benefits is a serious impediment to financial security in retirement. Women are also far more likely than men to have no source of employerprovided insurance. While these women have relatively high retirement rates, possibly because they do not need to stay in their jobs to maintain health benefits, they too potentially face high health costs with no source of labor income to pay for these expenses. Over the past decade, the provision of retiree health benefits has steadily declined. In addition, an increase in cost-sharing requirements and premiums appears inevitable (Kaiser/Hewitt 2006). These trends suggest workers will find it more difficult to retire early with health benefits than they did previously. For dual-working couples, having one s own retiree benefits has a significant effect on retirement. Furthermore, one s own and spousal retiree benefits have an even larger effect on retirement for single-earner couples who have a retired spouse. Retiree health benefits enable couples to time their retirement with their spouse; however, declining retiree health benefits are likely to restrict retirement options for singles and couples. The recently enacted Patient Protection and Affordable Care Act (PPACA) stipulates that individuals will be able to purchase health coverage from insurance exchanges. The availability of such insurance may turn out to be equivalent to universal access to retiree health insurance. If this is the case, we may expect early retirement rates to increase, particularly among married couples. Investigating the impact of these changes on retirement patterns is an important area for future research. 64

Retirement Behavior of Women Notes The authors are grateful to Paul Devereux, Alan Gustman, Lynn Karoly, Nicole Maestas, Constantijn Panis, and Susann Rohwedder for helpful comments. Sandy Chien provided invaluable programming support. The authors are able to make all data and programs available to researchers who obtain clearance to use the restricted version of the Health and Retirement Study data. 1 Authors calculations using the HRS sample of workers who reported retiring in 1994 and 2006. 2 Authors calculations using the 2006 HRS sample of individuals who reported working. 3 For the uninsured, privately insured, and publicly insured groups, annual expenses are $715, $2,321, and $3,789, respectively, for people ages 18 to 44 vs. $2,196, $5,645, and $9,193, respectively, for people ages 45 to 64. 4 The high cost of COBRA coverage ($7,000 to $8,000 for family coverage) may be a deterrent for many, especially for those who have just left a job (Scandlen 2001). 5 Authors calculations based on the Health and Retirement Study. 6 Authors calculations based on the Health and Retirement Study. 7 http://www.cbpp.org/cms/index.cfm?fa5view& id51986 8 HIPAA and COBRA aim to reduce some of these frictions, but ultimately individuals who retire cannot permanently keep their employer policy unless they are eligible for retiree coverage. 9 http://www.rwjf.org/pr/synthesis/reports_and_ briefs/pdf/no2_policybrief.pdf 10 Individuals still may prefer their employerprovided health insurance policy to Medicare if it allows them to access preferred providers. For such reasons, even after reaching Medicare eligibility, individuals may prefer to keep their employer policy; however, the incentive should be much lower than before age 65. 11 See for example, Hurd 1990, Gustman and Steinmeier 2000, Maestas 2001, and Johnson and Favreault 2001. 12 A few studies have found no significant link between spousal factors and continued work (Campione 1987; Maury 2001). 13 We do not model re-entry into the labor force after partial or full retirement. We model only the first observed transition to retirement. Twelve percent of the sample has multiple observed transitions to retirement. 14 Since employer-provided health insurance is usually offered to full-time workers and not to part-time workers, we believe that this is the appropriate definition. However, we have reestimated our model with an alternative definition of retirement that allows for transitions from full-time or part-time work to full retirement. We find somewhat smaller effects of health insurance; however, this effect remains significant. 15 We also have estimated secondary models that include an interaction between an indicator variable for survey years from 1996 on with retiree health insurance to determine whether the effect of retiree health insurance varies due to the change in survey instruments. The interaction was small and statistically insignificant, suggesting that our results are not sensitive to the change in survey instrument. 16 For couples who were under age 62, we assumed that baseline Social Security wealth was equal to the wealth in 1992. For couples who were between the ages of 62 and 65, we assumed that baseline Social Security wealth was equal to wealth at 62. The gain in Social Security wealth was calculated as the difference between baseline wealth and projected wealth at 65. 17 The HRS data imputes missing pension values. We re-estimated the model on the subsample with non-missing pension data and found a similar pattern of results; however, these results were less precise due to the smaller sample size. 18 Since DB pension wealth information was only available for 1992, this variable is potentially mismeasured for individuals who changed jobs. In supplementary analyses, we checked the sensitivity of our results to this measurement issue by estimating the DB pension wealth coefficients on the subsample that did not change jobs and by including an indicator for job change in our model. We found very similar results for the health insurance effects. 19 Excluding health resulted in very similar estimates for the effect of retiree health insurance on retirement. In addition, selfreported health status had a similar effect, in most cases, as the number of conditions variable, but when both were included in the model, they were imprecisely measured. 20 It can be argued that wages are endogenous to the retirement choice. We re-estimated the models in the paper without controls for wages and found very similar results for the health insurance variables of interest. Also, models 65

Inquiry/Volume 48, Spring 2011 with additional job controls that included industry indicators, occupation indicators, and tenure yielded similar results. 21 If we restrict the sample to observations where we have non-missing pension data, we find imprecise estimates of health insurance. This loss of precision is due solely to the change in the sample rather than the inclusion of pension controls since we continue to find imprecise health insurance effects in the restricted sample even without the inclusion of pension controls in the model. 22 We estimated an alternative set of models that dropped the controls for Social Security wealth. The estimates for the health insurance variables of interest were virtually identical. 23 This result is not simply an artifact of singleearner women being older, since we are controlling for the effect of age on retirement. 24 The difference between the effect of retiree EPHI and no EPHI is not statistically significant. 25 Strumpf (2007) conducted similar endogenity checks in her work on retiree health insurance. References Agency for Healthcare Research and Quality (AHRQ). 2007 Medical Expenditure Panel Survey: MEPSNet Query Tables. Rockville, Md.: AHRQ. http://www.meps.ahrq.gov/ mepsweb/data_stats/mepsnethc.jsp Bernasek, A., and S. Shwiff. 2001. Gender, Risk and Retirement. Journal of Economic Issues 35(2):345 356. Blau, D., and D. Gilleskie. 2006. Health Insurance and Retirement of Married Couples. Journal of Applied Econometrics 21(7):935 953. Campione, W. 1987. The Married Woman s Retirement Decision: A Methodological Comparison. Journal of Gerontology 42(4):381 386. Codispoti, L., B. Courtot, J. Swedish, et al. 2008. Nowhere to Turn: How the Individual Health Insurance Market Fails Women. Washington, D.C.: National Women s Law Center. http:// action.nwlc.org/site/docserver/nowheretoturn. pdf. Accessed November 1, 2008. Coile, C. 2004. Retirement Incentives and Couples Retirement Decisions. Topics in Economic Analysis and Policy 4(1):1 28. Dano, A., M. Ejrnaes, and L. Husted. 2004. Do Single Women Value Early Retirement More than Single Men? Copenhagen, Denmark: Centre for Applied Econometrics, University of Copenhagen. Dunleavey, M. 2006. The Secret Lives of Breadwinner Wives. MoneyCentral, MSN. Gruber, J., and B. Madrian. 2004. Health Insurance, Labor Supply, and Job Mobility: A Critical Review of the Literature. In Health Policy and the Uninsured, C. McLaughlin, ed. Washington, D.C.: The Urban Institute Press. Gustman, A., and T. Steinmeier. 2000. Retirement in a Family Context: A Structural Model for Husbands and Wives. Journal of Labor Economics 18:503 545. Honig, M. 1985. Partial Retirement among Women. Journal of Human Resources 20(4):613 621. Hurd, M. 1990. The Joint Retirement Decisions of Husbands and Wives. In Issues in the Economics of Aging, D. Wise, ed. Chicago: University of Chicago Press. Johnson, R., and M. Favreault. 2001. Retiring Together or Working Alone: The Impact of Spousal Employment and Disability on Retirement Decisions. CRR Working Paper 2001-01. Boston: Center of Retirement Research at Boston College. Kaiser Family Foundation and Hewitt Associates. 2004. Retiree Health Benefits Now and in the Future: Findings from the Kaiser/Hewitt 2003 Survey on Retiree Health Benefits. The Henry J. Kaiser Family Foundation, Menlo Park, Calif., and Hewitt Associates, Lincolnshire, Ill.. 2006. Retiree Health Benefits Examined. The Henry J. Kaiser Family Foundation, Menlo Park, Calif., and Hewitt Associates, Lincolnshire, Ill. Kapur, K., and J. Rogowski. 2007. The Role of Health Insurance in Joint Retirement among Married Couples. Industrial and Labor Relations Review 60(3):397 407. Maestas, N. 2001. Labor, Love and Leisure: Complementarity and the Timing of Retirement by Working Couples. Working Paper. Berkeley, Calif.: Department of Economics, University of California, Berkeley. Maury, D. 2001. Couples Retirement Decisions: The Impact of Health and Race Differences. Mimeograph. Paris, France: Institut d Etudes Politiques de Paris. Nyhan, P. 2008. Stay Home Fathers Figure Out their New Full-time Job. Seattle Post-Intelligencer May 9. http://www.seattlepi.com/local/ 362567_optin10.html O Rand, A., J. Henretta, and M. Krecker. 1992. Family Pathways to Retirement. In Families and Retirement, M. Szinovacz, D. Ekerdt, and B. Vinick, eds. Newbury Park, Calif.: Sage. Pienta, A. 1997. Older Couples: An Examination of Health and Retirement within the Context of the Family. Penn State University Population Research Institute Working Paper Series 97 03. University Park, Pa.: Penn State University. 66

Retirement Behavior of Women Pozzenbon, S., and O. Mitchell. 1986. Married Women s Retirement Behavior. NBER Working Paper No. W2104. Cambridge, Mass.: National Bureau of Economic Research (NBER). Quinn, J. 1996. The Role of Bridge Jobs in the Retirement Patterns of Older Americans. Boston College Working Papers in Economics 324. Boston: Boston College. Rogowski, J., and L. Karoly. 2001. Health Insurance and Labor Force Transitions of Older Women Workers. RAND DRU-2728- DOL. Santa Monica, Calif.: RAND. Scandlen, G. 2001. Helping Laid-Off Workers Keep Insurance. Brief Analysis No. 373. Washington, D.C.: National Center for Policy Analysis. StClair, P., D. Bugliari, P. Pantoja, S. Ilchuk, G. Lopez, S. Haider, M. Hurd, D. Loughran, C. Panis, M. Reti, and J. Zissimopoulos. 2002. SSA-HRS Data Documentation, PM-1172- SSA. Santa Monica, Calif.: RAND. Strumpf, E. 2007. Employer-Sponsored Health Insurance for Early Retirees: Impacts on Retirement, Health, and Health Care. Manuscript. Montreal: McGill University. 67