Coverage or Costs: The Role of Health Insurance in Labor Market Reentry Among Early Retirees

Abstract

Objectives.

This study evaluated the impact of insurance coverage on the odds of returning to work after early retirement and the change in insurance coverage after returning to work.

Method.

The Health and Retirement Study was used to estimate hierarchical linear models of transitions to full-time work and part-time work relative to remaining retired. A chi-square test was also used to assess change in insurance coverage after returning to work.

Results.

Insurance coverage was unrelated to the odds of transitioning to full-time work. However, relative to employer-provided insurance, private nongroup insurance increased the odds of transitioning to part-time work, whereas public insurance reduced the odds of making this transition. Additionally, after returning to work, insurance coverage increased among those who were without employer-provided insurance in retirement.

Discussion.

Results indicated that source of coverage may be more useful in explaining returns to part-time work than simply whether people have coverage at all. In other words, the mechanism underlying the positive relationship between insurance and returning to work appeared to be limited to those who return to work because of the cost of private nongroup insurance. Among these people, however, there was some evidence that they are able to secure new coverage once they return to work.

HEALTH insurance plays an important role in the lives of older Americans. Secure coverage limits the risks of experiencing poor health (Hadley & Waidmann, 2006) as well as increases access to and use of preventative medicine (Hoffman & Paradise, 2008). Insurance benefits also affect individuals’ retirement plans and decisions (Blau & Gilleskie, 2008; Maestas, 2010). In short, health insurance exerts extensive influence on the health, economic well-being, and retirement patterns of older Americans.

Despite a growing understanding of how insurance coverage influences the timing of retirement (Blau & Gilleskie, 2008; Johnson, Penner, & Toohey, 2008), little is known about how coverage affects other later life labor market decisions. Specifically, few studies consider how access to health insurance affects returning to work after a period of retirement. If people work longer to maintain insurance coverage, they may also return to work to (a) secure new coverage or (b) offset the costs associated with coverage. There are several reasons that uninsurance or underinsurance may push retirees back into the workforce: fewer companies offer retirement health benefits, the high cost of private nongroup insurance, and the expense of health care services (Weller, Wenger, & Gould, 2006).

This research used the Health and Retirement Study (HRS) (2008) to test three questions among those who retire prior to Medicare eligibility. First, was insurance coverage an important determinant of returning to work? Second, if insurance was indeed an important factor in reverse retirement, was this relationship conditional on the source of coverage? Third, among those that returned to work, did insurance coverage increase once they returned? I conclude with a discussion of how results suggest the Health Care and Education Reconciliation Act of 2010 may alter the impact of insurance on returning to work.

Background

Postretirement Employment of Older Workers.—

Recent scholarship has focused on the increased variation in pathways to full retirement (Flippen & Tienda, 2000; Han & Moen, 1999; Warner, Hayward, & Hardy, 2010). One emergent pattern is for workers to enter retirement then subsequently reverse their retirement decision (Ruhm, 1990). Estimates of the size of the population that reenter vary considerably: from as low as 2% (Hinterlong, 2008) to as high as 53% (Maestas, 2010) depending on the data source, the age range of respondents, and the operationalization of labor force reentry (Pleau, 2010). However, reentry is important to study because population trends suggest these numbers will grow. Many Baby Boomers are financially ill-prepared for retirement and may have to engage in postretirement employment to supplement their economic portfolios (Munnell & Sass, 2008).

Prior Research on Labor Market Reentry.—

The labor market patterns of older Americans are shaped, in part, by demographic characteristics. Gender, for example, influences the labor market experiences of most Americans (Padavic & Reskin, 2002). There is mixed evidence, however, as to how gender influences labor market reentry after retirement. For instance, Hardy (1991) found women were more likely than men to remain unemployed. More recently, however, Han and Moen (1999) found no gender differences in postretirement employment. It is likely that these differences are related to women’s increased labor force participation.

Race also shapes reentry into the labor market. Older African Americans and Hispanics are more likely to be poor, to have fewer assets, to be in poor health, and to have Medicare as their sole source of insurance (Oliver & Shapiro, 2006; Taylor, 2008). This suggests they may return to work for financial reasons and be limited by poor health. After accounting for economic well-being and health, however, research is mixed regarding how race affects reemployment. Peracchi and Welch (1994) found African Americans are more likely than Whites to return to part-time work after retirement. However, Hardy (1991) and Choi (2000) both found the probabilities of African Americans and Hispanics returning to work were no different than those of Whites.

Education also influences labor force attachment. Han and Moen (1999) found education was positively associated with having a postretirement job. They argued this is due to the greater opportunities for work that are available to the more educated. Similarly, Hardy (1991) found a college degree increased postretirement employment for those who were interested in it. However, she also found a college degree was associated with a desire to remain retired.

Economic well-being also influences the need to return to work after retirement. For example, in a study of older Canadian workers, Schellenberg and colleagues (2005) found economic concerns were the most common explanation for postretirement employment. Similarly, Han and Moen (1999) found those who were more financially prepared for retirement were less likely to return to work, and Lin (2005) found pension wealth reduced the risk of reemployment.

In sum, questions remain about the sources of variation in labor market reentry. The evidence is equivocal as to how race and gender affect labor market reentry. Most studies find that education increases the likelihood of securing postretirement employment, although this may be conditional on wanting to be reemployed. Finally, most studies also find that increased financial well-being lowers the risk of reentry.

Insurance and Later Life Labor Force Behavior.—

There is essentially no research on the influence of insurance on returning to work after retirement. However, clues can be drawn from research on the impact of insurance on the timing of retirement. This work shows that people will work longer to maintain their employer-provided benefits (Johnson et al., 2008). For example, Blau and Gilleskie (2008) found that access to health insurance influenced the labor market behavior of older men, although this was conditional on the source of the insurance. They argued that “budget constraint [and] aversion to risk” (p. 505) are two of the primary mechanisms through which insurance affects employment decisions at older ages.

Insurance coverage is important for the health and well-being of former workers, but retirees secure coverage through a variety sources. Prior to Medicare eligibility, most adults secure coverage through employment or through a spouse’s employment (Blau & Gilleskie, 2006; Rogowski & Karoly, 2000). For those whose coverage does not extend into retirement, employer provided insurance “locks” people into their jobs—making job changes and retirement difficult—and this is particularly true for sick people (those who need the benefit the most; Rashad & Sarpong, 2006). Additionally, coverage through a spouse is often lost when that spouse retires (Madrian, 2006). Thus, these people are more likely to postpone retirement until becoming eligible for Medicare.

In contrast, those with retiree health insurance (RHI) are more likely to retire early (French & Jones, 2007; Rogowski & Karoly, 2000). Although it was relatively common for employers to offer RHI through the 1980s, by 2000 only 36% of employers offered RHI (Quadagno, 2005). However, roughly half of those that retire prior to Medicare eligibility are covered with RHI (Johnson, 2007). Because RHI is not contingent upon continued work, those with RHI may be more likely to remain retired (Rogowski & Karoly, 2000).

For those without RHI or insurance through a spouse, there are few alternatives for securing insurance before Medicare. First, some early retirees purchase nongroup insurance on the private market. However, this coverage is expensive, benefits are limited, and coverage may be denied (Johnson, 2003). It is also unclear how private nongroup insurance may influence postretirement behavior. People with nongroup insurance may reenter to secure employer-provided insurance or to reduce the costs associated with private insurance. Second, some retirees are covered through Medicare’s disability benefit that provides insurance to those who are blind, disabled, suffer from permanent kidney failure, or who qualify for Social Security Disability Insurance (Buchmueller, Johnson, & Lo Sasso, 2006; Whittaker, 2005). However, because this type of insurance is often lost upon returning to work (Social Security Administration, 2011), having this type of private coverage may disincentivize retirees from returning to work—at least until they become age eligible for Medicare.

This literature suggests that in addition to demographic and economic factors, postretirement work is likely impacted by the availability and costs of health insurance. Because some individuals work longer to maintain stable health insurance and reduce out-of-pocket spending, it may be that the costs of private coverage drive people back to work or that securing coverage motivates people to return to work. Indeed, Lin (2005) found having insurance reduced the risk of returning to work among older displaced workers. Lin speculates that reemployment may be the primary mechanism of providing insurance for many displaced older workers. However, it is unclear how similar displaced workers are to retirees.

METHOD

Although the impact of demographic, economic, and health characteristics on labor force reentry is becoming better understood, little is known about how different forms of insurance coverage influence this decision. I explored this question using the first eight waves of the RAND Health and Retirement Study (HRS) Data File (Health and Retirement Study, 2008; RAND Center for the Study of Aging, 2008). In 1992, HRS investigators began interviewing 12,600 people who were between 51 and 61 years of age as well as their spouses (Juster & Suzman, 1995), with new cohorts of 51- to 56-year olds added in 1998 and 2004. These people were reinterviewed every 2 years, for a total of eight waves in 2006.

Several steps were used to identify the analytic sample. First, people who reported having never worked were excluded. Second, individuals were not included until they were fully retired—defined as averaging 0 hr of work per week over the previous year and self-identifying as being retired (Maestas, 2010). Third, once a person returned to work, they were no longer retired, and were therefore excluded from the risk set unless they retired a subsequent time—in which case they were again included as retirees. Finally, because this research was concerned with the impact of insurance coverage prior to age eligibility for Medicare, individuals were included in the risk set as long as they were 64 years or younger; once they turned 65 years, they were no longer included. The resulting sample was an unbalanced panel of 8,153 person-waves nested within 4,513 individuals.

Dependent Variable

The outcomes used in the following analyses were dichotomous measures of transitioning out of retirement. Individuals were coded as “full-time work” if they worked 35 or more hr per week and at least 36 weeks per year, “part-time work” if they worked fewer than 35 hr per week or fewer than 36 weeks per year (Maestas, 2010), and “remained retired” if they continued to meet the definition of full retirement.

Insurance Measures

The analyses included two sets of time-varying measures for insurance coverage. In some models, a single dichotomous measure of “insured” was used. This measure was coded “1” for individuals who reported receiving insurance from any of the following sources: a former employer (RHI insurance), a spouse, Medicare, Medicaid, or the private nongroup market. In other models, four mutually exclusive measures of insurance were used. First, respondents who received insurance coverage through former employment or their spouse’s employment were coded as having “employer insurance” (for the regression analyses, this measure was used as the reference because 64% of the sample had this form of coverage). Second, respondents without employer insurance were coded “1” for “public insurance” if they were covered by Medicare (because of disability) or Medicaid. Third, respondents were coded “1” for “direct insurance” if their only source of coverage came from nongroup insurance purchased directly on the private market. Finally, respondents who were not covered by any of these three forms of insurance were coded “1” as “uninsured.”

Control Measures

An extensive set of controls were used in these analyses. Time invariant controls included “men,” four dichotomous measures of race (White, African American, Hispanic, and other race), and years of “education.” In addition to “age” in years, there were several other time-varying control measures. Economic characteristic included annual household “pensions,” annual household “social security,” annual household “earnings,” and total household “assets.” Each of these measures was adjusted to 2006 dollars and converted to the natural log. Marital status was measured using three dichotomous variables: “spouse working,” “spouse not working,” and “unmarried.” Three measures of health status were also included. “Physical limitations” was a scale summing a dichotomous report of difficulties with walking one block, climbing a flight of stairs, lifting, or carrying 10 pounds, picking up a dime, and pushing or pulling a large object. “Chronic conditions” was a scale summing a dichotomous report of experiencing hypertension, heart disease, stroke, nonskin cancer, diabetes, and chronic lung disease (Brown & Warner, 2008). “Poor health” was measured using a 5-point Likert item of self-assessed health. For each of the three health measures, larger numbers indicated poorer health. One final measure included a Likert item of “retirement satisfaction” with larger numbers reflecting a greater degree of satisfaction with retirement.

Analysis Plan

The analyses proceeded in three steps. First, descriptive statistics were used to highlight the prevalence of returning to work from full retirement and show how insurance coverage varied across people, contingent upon their transitioned to labor force status. Second, because the data were composed of repeated observations nested within individuals, hierarchical linear modeling (HLM) was used to assess whether (a) insurance coverage affected the odds of returning to work and (b) if the impact of coverage was contingent upon the type of insurance. HLM is a powerful tool that can estimate equations for panel data using the logit-link function (Raudenbush & Bryk, 2002). Such a function was appropriate here, because of the dichotomous outcome measures (i.e., returned to part-time work and returned to full-time work, relative to remained retired). For these results, measures were led by one wave, so that covariates in wave_t predicted returning to work in wave_t+1. Finally, chi-square tests were used to assess whether insurance coverage changed after labor force reentry.

RESULTS

Descriptive Results

Table 1 provides descriptive statistics at the person level. It shows that 19.7% of the sample returned to work at some point during the period of observation: 15.07% to part-time work and 4.63% to full-time work. Consistent with prior research on early retirees (Rogowski & Karoly, 2000), most of this sample (64%) had employer insurance, whereas 14% was uninsured, 13% was covered through public insurance, and the remaining 8% had direct purchase insurance.

Table 1.

Person-Level Descriptive Statistics

Variable	Mean	SD	Min	Max
Work statusa
Part time	15.07%	0.21	0.00	1.00
Full time	4.63%	0.36	0.00	1.00
Remain retired	80.48%	0.40	0.00	1.00
Insurance coverageb
Insured	85.95%	0.35	0.00	1.00
Employer insurance	64.35%	0.48	0.00	1.00
Public insurance	13.23%	0.34	0.00	1.00
Direct insurance	8.38%	0.28	0.00	1.00
Uninsured	14.05%	0.35	0.00	1.00
Time invariant controls
Men	48.46%	0.50	0.00	1.00
White	75.09%	0.43	0.000	1.000
African American	16.88%	0.37	0.00	1.00
Hispanic	6.29%	0.24	0.00	1.00
Other race	1.73%	0.13	0.00	1.00
Education	12.35	2.88	0.00	17.00
Time-varying controlsb
Age	59.35	3.57	51.00	64.00
Pensions	4.56	5.75	0.00	22.11
Social security	3.58	5.38	0.00	19.79
Earnings	7.96	7.41	0.00	24.43
Assets	10.04	3.96	0.00	17.65
Spouse working	34.23%	0.47	0.00	1.00
Spouse not working	40.88%	0.49	0.00	1.00
Unmarried	24.88%	0.43	0.00	1.00
Physical limitations	1.14	1.76	0.00	10.00
Chronic conditions	1.12	1.07	0.00	6.00
Poor health	2.91	1.22	1.00	5.00
Retirement satisfaction	0.71	0.45	0.00	1.00

Notes. N = 4,513.

^aThese measures are the percentages for whoever makes the transition during the period of observation.

^bBecause these measures are time varying, the values shown above are what individuals report during the first wave in which they are observed.

Table 2 shows descriptive statistics by labor force transition. Relative to those who remained retired (a) smaller percentages of those who transitioned to part-time work had public insurance, whereas larger percentages had employer provided insurance or were covered with direct insurance and (b) smaller percentages of those who transitioned to full-time work had any kind of coverage or public insurance, whereas larger percentages had direct purchase insurance or were uninsured.

Table 2.

Descriptive Statistics by Transition Status (person level, initial wave)

	Remained retired		Full-time		Part time
Variable	Mean	SD	Mean	SD	Mean	SD
Insurance coverage
Insured	86.54%	0.34	77.03%*	0.42	85.59%	0.35
Employer insurance	63.96%	0.48	58.85%	0.49	68.24%*	0.47
Public insurance	14.95%	0.36	6.22%*	0.24	6.03%*	0.24
Direct insurance	7.63%	0.27	11.96%*	0.33	11.32%*	0.32
Uninsured	13.46%	0.34	22.97%*	0.42	14.41%	0.35
Controls
Men	48.40%	0.50	56.94%*	0.50	46.18%	0.50
White	74.83%	0.43	74.16%	0.44	76.76%	0.42
African American	16.69%	0.37	20.57%	0.41	16.76%	0.37
Hispanic	6.61%	0.25	4.31%	0.20	5.15%	0.22
Other race	1.87%	0.14	0.96%	0.10	1.32%	0.11
Education	12.24	2.92	12.87*	2.70	12.84*	2.65
Age	59.55	3.53	57.73*	3.80	58.74*	3.54
Pensions	4.67	5.83	3.63	5.25	4.25*	5.41
Social security	3.81	5.51	2.43*	4.45	2.67*	4.78
Earnings	7.38	7.30	10.24*	7.55	10.37*	7.33
Assets	9.95	4.03	10.13	3.85	10.50*	3.54
Spouse working	32.85%	0.47	44.02%*	0.50	38.82%*	0.49
Spouse not working	41.93%	0.49	32.54%*	0.47	37.79%*	0.49
Unmarried	25.22%	0.43	23.44%*	0.42	23.38%*	0.42
Physical limitations	1.26	1.85	0.53*	1.09	0.66*	1.29
Chronic conditions	1.18	1.08	0.84*	1.00	0.93*	0.96
Poor health	3.01	1.23	2.44*	1.11	2.51*	1.09
Retirement satisfaction	0.75	0.43	0.51*	0.50	0.60*	0.49
	N = 3,632		N = 209		N = 680

Notes. Tests for differences between those who transition to full-time work and those who work part-time (relative to remaining retired) were conducted using two-tailed t-tests for continuous measures and chi-square tests for dichotomous measures.

*p < .05.

HLM Regression Results

Table 3 shows the results of HLM regression models using the logit-link function for dichotomous data. The coefficients represent odd ratios of transitioning to full-time and part-time work, relative to remaining fully retired in wave_t+1. Models 1 and 2 provided a baseline composed of control variables. In these models, men had greater odds than women of transitioning back to full-time work but lower odds of transitioning to part-time work. Consistent with earlier research (Choi, 2000; Hardy, 1991), these results did not indicate racial differences in the odds of returning to work. Increases in poor health, physical limitations, and retirement satisfaction were associated with reduced odds of returning to either form of work. In contrast, increases in household earnings were associated with larger odds of returning to both part-time and full-time work. Because earnings are generally similar within households (Schwartz, 2010), it may be that this represented workers with the greatest human capital and who were most qualified to secure new employment. This may also explain why, in contrast to earlier studies, there was no impact of education on returning to work; that is, variation based on the ability to secure new employment was being captured in the earnings measure rather than the education measure. Results from these models also indicated that increases in age and assets were associated with lower odds of transitioning to full-time work but had no significant impact on part-time work, whereas having a working spouse was associated with a reduction in odds of transitioning to part-time work but had no significant impact on returning to full-time work.

Table 3.

Binomial Growth Models of Returning to Work After Full Retirement

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
	Full time	Part time	Full time	Part time	Full time	Part time
Intercept	283.880** (1.428)	0.581 (0.899)	408.126** (1.454)	0.630 (0.900)	189.116** (1.462)	0.643 (0.903)
Insurance
Insured	—	—	0.438** (0.202)	0.856 (0.131)	—	—
Public insurancea	—	—	—	—	0.751 (0.334)	0.668* (0.177)
Direct insurancea	—	—	—	—	1.858* (0.257)	1.578** (0.141)
Uninsureda	—	—	—	—	2.357** (0.221)	1.133 (0.139)
Time invariant controls
Menb	1.460* (0.163)	0.793* (0.091)	1.430* (0.164)	0.790* (0.091)	1.453* (0.166)	0.805* (0.092)
African Americanc	1.323 (0.197)	1.100 (0.120)	1.320 (0.196)	1.099 (0.121)	1.354 (0.200)	1.129 (0.121)
Hispanicc	0.770 (0.367)	1.016 (0.191)	0.718 (0.362)	1.008 (0.190)	0.728 (0.363)	1.031 (0.191)
Other racec	0.564 (0.742)	0.752 (0.362)	0.538 (0.737)	0.748 (0.363)	0.543 (0.738)	0.746 (0.366)
Education	1.015 (0.031)	1.008 (0.018)	1.024 (0.031)	1.009 (0.018)	1.021 (0.031)	1.006 (0.018)
Time varying
Age	0.877** (0.026)	0.989 (0.014)	0.876** (0.026)	0.989 (0.014)	0.875** (0.026)	0.988 (0.014)
Pensions	0.984 (0.014)	0.992 (0.008)	0.993 (0.014)	0.993 (0.008)	0.996 (0.014)	0.994 (0.008)
Social security	1.028 (0.018)	0.983 (0.010)	1.024 (0.018)	0.982 (0.010)	1.023 (0.018)	0.981* (0.010)
Earnings	1.052** (0.013)	1.073** (0.007)	1.052** (0.013)	1.073** (0.007)	1.055** (0.013)	1.073** (0.007)
Assets	0.955* (0.022)	0.993 (0.014)	0.966 (0.023)	0.995 (0.014)	0.960 (0.023)	0.984 (0.015)
Spouse workingd	0.885 (0.228)	0.612** (0.137)	0.985 (0.233)	0.619** (0.138)	0.979 (0.237)	0.619** (0.138)
Spouse not workingd	0.830 (0.214)	0.849 (0.122)	0.884 (0.219)	0.854 (0.122)	0.875 (0.223)	0.848 (0.122)
Physical limitations	0.761** (0.076)	0.808** (0.041)	0.773** (0.076)	0.809** (0.041)	0.785** (0.079)	0.823** (0.041)
Chronic conditions	0.873 (0.085)	0.980 (0.049)	0.887 (0.085)	0.982 (0.049)	0.898 (0.085)	0.993 (0.049)
Poor health	0.775** (0.093)	0.820** (0.051)	0.772** (0.091)	0.820** (0.051)	0.783** (0.091)	0.830** (0.051)
Retirement satisfaction	0.319** (0.182)	0.411** (0.096)	0.339** (0.178)	0.415** (0.097)	0.346** (0.181)	0.418** (0.097)
Model statistics
Deviance	16704.179	19330.814	16686.198	19329.396	16678.588	19312.038
Parameters	18	18	19	19	21	21
Akaike Information Criterion	16740.18	19366.81	16724.20	19367.40	16720.59	19354.04

Notes. n = 4,513; N = 8,153. Coefficients are reported as odds ratios; standard errors in parentheses. Outcome measures are led so that covariates in wave_t predict work status in wave_t+1 relative to remaining retired.

^aEmployer insurance

^bMen

^cWhite

^dUnmarried.

*Significant at p < .05, **significant at p < .01.

In models 3 and 4, a dichotomous measure of insurance coverage was added to the baseline models. In these results, having insurance coverage was associated with 56% lower odds of leaving retirement for full-time work but had no impact on returning to part-time work. However, it may be that the impact of coverage was conditional on the source of insurance. As such, in models 5 and 6, three types of coverage were compared with having employer-provided insurance. The results showed that individuals with private nongroup insurance had significantly larger odds of returning to both part-time and full-time work. Similarly, the uninsured had 1.36% higher odds of returning to full-time work but not significantly different odds of returning to part-time work. Finally, those with public insurance had lower odds of transitioning to part-time work but not to full-time work.

Although these results suggested that insurance coverage may be an important factor in explaining transitions from retirement back to both full-time and part-time work, assessments of change in Akaike Information Criterion (AIC; calculated as deviance plus 2 times the number of parameters) suggested a slightly different conclusion. Indeed, difference tests (McCoach & Black, 2008) provided evidence for including insurance measures in the part-time equation—suggesting that the inclusions of these measures fit the data better than the model that did not account for insurance coverage. [It should be noted that Bayesian Information Criterion (BIC) difference test did not support the inclusion of the more parameterized models. BIC tests, however, tend to favor more parsimonious models and this tendency increases along with sample size (McCoach & Black, 2008). Thus, the AIC difference test is appropriate here.] Among the full-time work models, however, there was not evidence that Model 5 or Model 3 was more appropriate than Model 1—leading to the conclusion that Model 1 was the best fit for full-time work. However, because most (76%) returns to work were to part-time work, these models generally supported the proposition that for many people who transitioned from retirement back to work, insurance coverage—or rather the costs of private nongroup insurance—played an important role in making this transition. Although I consider why this may be in the Discussion section subsequently, there was enough evidence to suspect that part of the reason people reversed their retirement decision was to secure new or less-expensive coverage or to secure additional funds to afford expensive private insurance.

Coverage After Reentry

The preceding analyses indicated that insurance coverage may have been an influential determinant of why people return to part-time work. As such, it is important to consider how insurance coverage changed for people once they returned to work. Table 4 shows the percentage change in coverage among individuals who returned to work prior to becoming age eligible for Medicare and had data on insurance coverage in both waves. Chi-square tests indicated that among those who were uninsured or relied on private nongroup insurance, employer-provided insurance increased by 23.29% and 42.59%, respectively, after returning to work. This suggests that some of those without employer-provided coverage in retirement were able to secure it through new employment.

Table 4.

Comparison of Change in Insurance Coverage Before and After Returning to Work^a

	Coverage after returning to work
Coverage prior to returning to work	Uninsured (n = 72)	Direct insurance (n = 51)	Public insurance (n = 29)	Employer insurance (n = 310)
Uninsured (n = 73)	−42.47%*	9.59%*	9.59%*	23.29%*
Direct insurance (n = 54)	9.26%*	−55.56%*	3.70%	42.59%*
Public insurance (n = 24)	16.67%*	12.50%	−29.17%*	0.00%
Employer insurance (n = 311)	6.75%*	5.47%*	0.96%	−13.18%*

Notes. n = 462. Chi-square test of change of insurance before and after returning to work.

^aLimited to those who transition prior to Medicare eligibility.

*p < .05.

In contrast, uninsurance increased by 16.67% among those who had public insurance prior to returning to work. This likely represents those who lost Medicare disability coverage because they were unable to continue coverage once they secured new employment (Social Security Administration, 2011). Additionally, uninsurance and direct insurance increased by 6.75% and 5.47%, respectively, among those who had employer-provided insurance prior to returning to work. It is likely these were people who lost employer-provided coverage due to the retirement of a spouse and returned to work to offset the new costs of expensive private nongroup coverage or to pay for uncovered premiums. Although the HLM models did control for financial well-being, it is quite possible that for many people, returning to work to offset the costs of private insurance was preferable to paying for insurance out of savings, regardless of how wealthy an individual may have been. This interpretation would be consistent with the contention of Blau and Gilleskie (2008) that aversion to risk is the primary way insurance coverage influences labor market behavior among older workers. Moreover, gaining coverage or offsetting the costs of retirement is only part of the picture. Existing coverage protects some of the sample from returning to work, and these people were excluded from Table 4.

DISCUSSION

In recent decades, the pathways to retirement have become increasingly varied. Moreover, retirement is no longer the absorbing state it once was. With nearly 20% of early retirees reversing their retirement decision, this paper sought to answer three questions regarding the role of insurance in returning to work after retiring before Medicare coverage. First, to the question of whether insurance coverage was an important determinant of returning to work, these results indicated that it depended on the type of work to which one was returning. Insurance coverage appeared to be of little importance in the transition back to full-time work but was important for transitioning back to part-time work—but only when considering different forms of coverage. This suggests that insurance was generally important in the process because more than three quarters of the transitions back to work were to part-time employment.

Second, to the question of whether the relationship between insurance and returning to work was conditional on the source of coverage, these results suggest that the impact of insurance coverage on returning to part-time work was complex. For instance, the relationship between coverage and part-time work was nonsignificant in the models of any coverage, but a more nuanced relationship was evident when comparing different patterns to those with employer insurance. Specifically, relative to those with insurance through a former employer or through their spouse’s employer, individuals who purchase private nongroup insurance were at greater risk of returning to work. Moreover, in these results, the uninsured were no more likely to reverse their retirement than were those with employer insurance. Taken together, these results suggest that the expense of nongroup coverage may have been a key factor driving people back to the labor market. So in contrast to Lin (2005)—who finds that “uninsurance” increased reemployment among displaced workers—these findings suggest that “underinsurance” increased the odds of part-time reemployment among retirees.

It is also noteworthy that public insurance reduced the odds that individuals will transition back to part-time work. However, unlike Medicare coverage from age eligibility, eligibility rules make it difficult to keep Medicare disability coverage upon securing new employment (Social Security Administration, 2011). As such, it is not surprising that people with this form of coverage were less likely to return to work nor that many lost their public insurance once they did return.

Third, to the question of whether insurance coverage increased once retirees returned to work, analyses of the change in coverage after returning to work suggest that, after returning to work, employer-provided insurance increased for those who were uninsured or insured through nongroup coverage during retirement. This not only highlights the importance of employer insurance for successful early retirement but also provides limited evidence that those without this form of insurance are seeking out new coverage as they return to work. Thus, although the costs of private nongroup insurance may be what is driving people back to work, some people who return to work are successfully securing better coverage than they had prior to reentry.

With the passage of the Health Care and Education Reconciliation Act of 2010, health insurance should be more affordable, public insurance more available, and Medicare increasingly comprehensive (Health Care and Education Reconciliation Act, 2010; White House, 2010). Although uninsurance should drop dramatically as a result of this legislation, individuals will still secure coverage through a variety of sources. The results presented here suggest that different forms of coverage will continue to exert distinct influences on individuals’ postretirement labor force behavior. Taken as a whole, however, these findings indicate that the Health Care and Education Reconciliation Act may impact labor force reentry in at least two ways. First, if the expansion of coverage increases individuals’ health and limits disability, people should be less likely to retire for health-related reasons, and more able to seek out postretirement employment should they so choose. Second, these results suggest returning to part-time work should increase if individuals meet the mandate of coverage through private nongroup insurance rather than from employer-provided insurance—unless, of course, such a mandate decreases early retirement as people become more locked to the insurance provided by their employers.

There are two notable limitations to the present study. First, there is a lack of consensus on how to measure postretirement labor force behavior. Some studies include any work after retirement—regardless of whether or not there is a period of nonwork between employment transitions (Pleau, 2010). Others, like the method used here, require a period of nonwork between employment and returning to work in order to constitute reentry. Empirically, this lack of definitional consensus leads to widely varying estimates of the size of the population that reenters. Theoretically, inasmuch as these different postretirement labor force transitions are distinct processes, it becomes difficult to disentangle the mechanisms that underlie the transitions between them.

Second, the discussion of returning to work used in this analysis has implied that returning is a risk to be avoided. However, for some people, work is preferable to retirement, and these people elect to reverse their retirement for reasons other than need. Although I attempted to control for this with a variety of measures (including education and retirement satisfaction), data limitations prohibited me from directly assessing whether individuals transition from full retirement out of need or because of choice.

These limitations aside, this work makes several contributions to the research on work and retirement. First, it highlights the importance of considering insurance coverage in reversing retirement. Second, this research suggests why insurance may not be more influential on returning to work. That is, insurance coverage may be driving people to return for reasons other than securing new coverage (e.g., to offset the costs of private coverage). Finally, this work points out that even though coverage may be more comprehensive and more widely available because of the Health Care and Education Reconciliation Act of 2010, people will still secure supplementary private coverage through a variety of ways. This research suggests that the form this private coverage takes will differentially shape the odds that people will transition back to work after early retirement.

Funding

National Institutes of Health (2T32AG000139-21); Claude and Mildred Pepper Dissertation Fellowship at Florida State University.