Retiree Health Insurance for Public School Employees: Does it Affect Retirement?

Abstract

Despite the widespread provision of retiree health insurance for public sector workers, little attention has been paid to its effects on employee retirement. This is in contrast to the large literature on health-insurance-induced “job-lock” in the private sector. I use the introduction of retiree health insurance for public school employees in combination with administrative data on their retirement to identify the effects of retiree health insurance. As expected, the availability of retiree health insurance for older workers allows employees to retire earlier. These behavioral changes have budgetary implications, likely making the programs self-financing rather than costly to taxpayers.

I. Introduction

In the U.S., as in many other OECD countries, public sector pension funds are severely underfunded.¹ At the same time, the funds set aside to pay for the health insurance promised to retired state and local government employees, like teachers, represent an even smaller fraction of the estimated future health care liabilities than pension funds do pension liabilities (Clark and Morrill 2010). While the promised pension benefits to public employees are considered constitutionally protected in many states, retiree health insurance is not. Therefore, facing considerable deficits and a poor economic climate, state and local governments may decide to discontinue, significantly scale back, or otherwise alter retiree health insurance programs for their employees.

Economic theory would predict that the offer of retirement-contingent health insurance to public sector employees would decrease public sector employment of older workers for two reasons. First, there is an income effect of subsidized health insurance that discourages work. Second, health insurance that is tied to retirement from the public employer does not preclude employment in the private sector; the reduction of “job-lock” may lead employees to retire from the public sector but continue to work elsewhere. This option may be particularly attractive to older workers if private sector jobs offer more flexible hours than public employers.

However, there has been relatively little research into the magnitude of any effects of retiree health insurance on public sector employment. Research on similar programs, e.g. Medicare, COBRA and the Veterans Affairs insurance expansion, provides some indication (Gustman and Steinmeier, 1994; Karoly and Rogowski, 1994; Madrian, 1994; Gruber and Madrian, 1995; Lumstaine, Stock and Wise, 1996; Rust and Phelan, 1997; Blau and Gilleskie, 2001, 2006, 2008; Boyle and Lahey, 2010; Robinson and Clark, 2010; Strumpf, 2010; French and Jones, 2011; Coe et al., 2013; Martin and Woodbury, 2013; Nyce et al., 2013). This literature has generally shown that health insurance availability for retirement-aged individuals induces retirement, though the magnitudes of the effects vary across settings.² Yet, this work has focused on private sector employees. Differences between both public employee retiree health insurance plans and the other types of insurance studied and public and private employees suggest the effects of retiree health insurance may be different across the public and private sectors. For example, research has shown that public sector employees are particularly responsive to the nonlinearities in their pension benefits (Costrell and Podursky 2009; Brown 2010; Koedel et al. 2013; Grissom et al. 2013a, 2013b) and are more knowledgeable about their retirement benefits (DeArmond and Goldhaber 2010) than their counterparts in the private sector.

More recently, two studies have focused specifically on the effects of retiree health insurance on the labor supply of public sector retirees. Leiserson (2013) uses administrative data on public bureaucrats in Pennsylvania to investigate how employee exit responds to retiree health insurance eligibility. He leverages both the inherent variation caused by standing eligibility requirements for pensions and the retiree health insurance program and a natural experiment caused by an increase in the service requirement for retiree health insurance eligibility (but not pension eligibility) from 15 to 20 years. Shoven and Slavov (this issue) use data on all federal, state and local government employees from the Health and Retirement Study coupled with data on pension and retiree health insurance availability and generosity to determine the effects of retiree health insurance on the labor supply of older workers between ages 55 and 64. Both studies find that the availability of public employer provided retiree health insurance increases the likelihood that employees will be either out of the labor force or at least no longer working at their public employer.

The current study contributes to this emerging literature because, as described more below, I use administrative data on the single largest group of public sector workers, namely, teachers and other public school employees.³ I also leverage a natural experiment different in nature than those used by Leiserson and Shoven and Slavov. Its counterfactual is a world without retiree health insurance in the public sector. Also, similar to the other studies, the nature of my data allows me to pay careful attention to other endogenous factors that may be driving retirement, e.g. pension eligibility and generosity. Moreover, the weakness of my study my use of data and identifying variation from the early 1980s when labor supply patterns of older workers were likely somewhat different than they are today does not plague the other studies. As such, the findings of all three studies can be combined to more fully understand the relationship between retiree health insurance and public sector employee labor supply.

To be more specific, in this paper, I provide direct evidence about how public sector retiree health insurance availability affects the labor supply of public employees by examining the introduction of retiree health insurance for public school employees in Illinois. Today, former employees of Illinois Public Schools (IPS) who receive retirement benefits from the Illinois Teacher Retirement System (TRS) can participate in a health insurance plan called the Teachers Retirement Insurance Program. The state legislators introduced this retiree health insurance program for teachers and other public school employees, which I call TRHIP, in January of 1980 and permitted the first enrollments on July 1, 1980.⁴ At the time, premiums for enrollees were 50 percent subsidized. In order to enroll, former IPS employees needed to be receiving retirement benefits from the TRS and have at least 8 years of creditable service with the TRS.

Using administrative data from IPS, I use a differences-in-differences framework to compare the labor supply of teachers old enough and with enough accumulated experience to be eligible for TRHIP to those who were ineligible (because they were too young or had too little experience to be eligible for retirement benefits and THRIP) just before and after the TRHIP was introduced. I control for age and experience fixed effects, thereby capturing any systematic variation in labor supply across the lifecycle and career. The identifying assumption is that, conditional on employee characteristics, there were no other concurrent policies or environmental factors that disproportionately affected the labor supply of teachers eligible for THRIP. Importantly, using historical TRS documents, I can confirm there were no concurrent policy changes by the Illinois TRS related to either pension benefit size or eligibility. To further support the identification assumption, I examine pre-treatment trends in differences in labor supply of the ineligible and eligible public school employees and find no differences.

The outcome of interest is the retirement from the IPS system, which is synonymous with leaving one’s career job. I find that eligibility for retirement-contingent health insurance induces a clear shift in the age profile of retirement for public school employees. Before retiree health insurance is introduced, the exit rate of employees from IPS is highest at age 65, when eligibility for Medicare begins. After TRHIP is introduced, the exit rate of employees who continue to be employed at age 65 decreases 40 percent, from 0.51 to 0.29. At the same time, exit rates when employees first become eligible for retirement benefits at age 55 were just 0.054 before the TRHIP was introduced. Afterwards, the exit rate at this age, which now determines not just eligibility for retirement benefits but also retiree health insurance, jumps to 0.098, an increase of 81 percent. As we would have expected, some employees move forward their timing of retirement when retiree health insurance becomes available.

In Section II, I provide an overview of the Illinois teacher pension and retiree health insurance programs as they exist today, and their historical genesis. I describe the data in Section III and the empirical methodology in Section IV. I present results in Section V before discussing implications, including how these changes in labor supply affect cost calculations, in Section VI and concluding with a brief discussion of the applicability of the results in today’s economy in Section VII.

II. Illinois Teacher Retiree Health Insurance Program

Today, former employees of IPS who receive retirement benefits from the Illinois TRS can participate in a health insurance plan officially called the Teachers Retirement Insurance Program. For retired employees enrolled in the managed care plan, the monthly premium is $203 if the retiree is under age 65 and $80.23 if the retiree is age 65 or older and enrolled in Medicare as the primary insurer. Premiums are higher in the alternative Teachers’ Choice Health Plan. Notably, unlike Medicare, the TRHIP program covers dependents, though premiums for dependents are more expensive than premiums for retired employees. Currently, there is no annual or lifetime limit on the coverage in any plan offered by the Illinois TRHIP, making it more generous than Medicare. In the managed care program, there is no annual deductible, though there are copayments for certain services, e.g. $20 for a physician’s visit. The out-of-pocket maximum is $3,000. The Teachers’ Choice Health Plan has a $500 deductible, but only a $1,200 out-of-pocket maximum for in-network services.

As mentioned in the introduction, in order to enroll in the TRHIP, former IPS employees must be receiving retirement benefits from the TRS and have at least 8 years of creditable service with the TRS. Eligibility for retirement benefits from TRS is determined by an employee’s age and years of creditable service in IPS.⁵ An employee is eligible to receive a full retirement benefit when she is age 55 with at least 35 years of experience, age 60 with at least 10 years of experience, or age 62 with at least five years of experience. During the period studied, employees who had accumulated up to one-half a year of sick leave could count this as creditable service. Additionally, throughout this period, employees could retire at age 55 with just 20 years of experience and receive an actuarially discounted annuity.

The state legislators introduced the health insurance program for retired teachers in January of 1980. The first enrollments were on July 1, 1980. At the time, premiums for enrollees were 50 percent subsidized such that the annuitant’s monthly premium cost was $81.70 for people under age 65 and $29.21 for people age 65 and older.⁶ The annual deductible was $847 and $7,056 for people on Medicare and those under age 65, respectively. After this deductible was met, all allowable expenses were covered. Initially, and until 1995, the retiree health insurance program was part of the TRS, so the subsidy was financed by funds in the pension system, which comprise state appropriations, employee contributions and investment returns.

In the absence of retiree health insurance, the single largest insurance provider for public school retirees was Medicare. In 1980, Medicare consisted of hospital insurance and medical insurance.⁷ For people ages 65 and older who have worked at least 10 years (40 quarters) in jobs that were subject to Medicare FICA taxes, Medicare hospital insurance is free, and was at the time of TRHIP introduction. The same is true if one’s spouse worked at least 10 years in a job covered by Medicare taxes. IPS teachers did not pay Medicare taxes on their earnings until after 1986, and even then Medicare taxes were levied only on teachers newly hired by a district. Therefore, even once employees were 65, unless they were eligible for the free hospital insurance through their spouse’s work credit, many IPS employees paid a hospital insurance premium.

Because they are ineligible for Medicare, IPS employees younger than age 65 would have had one of five options for obtaining health insurance. (These same options also exist for those over age 65.) The first would be to remain employed and receive health insurance through an employer. Since all school districts provide employee health insurance and tenure rules protect the most senior teachers from termination of employment, continuing insurance coverage in this manner would have been at the discretion of the employee. A second way for IPS employees to obtain health insurance was through their spouse. Given that during the period studied most of the employees in my sample were female and likely secondary breadwinners, this may have been a viable alternative to continued employment until Medicare eligibility.⁸ Third, it may have been the case that individual school districts allowed former employees to continue participation in the district’s employee health insurance plan after leaving employment with the district. Fourth, retiring employees could have purchased their own health insurance, either through COBRA or the private market. Finally, Medicare ineligible retirees could have gone without any health insurance.

III. Data

The main sources of data are IPS administrative files called the Teacher Service Record (TSR) for 1970–1971 to 1991–1992 school years.⁹ (For the rest of the paper, I will refer to school-years by their spring year.) Each observation in the data contains information about an employee’s service in a particular year including her experience, salary, position (teacher, principal, etc.) and main assignment (math, reading, special education, etc.) at a school. I omit the 3 percent of employees who work at vocational schools, schools in prisons, special education cooperatives, etc. Employees in Chicago Public Schools (CPS) participate in a separate pension and retiree health insurance program from employees in other public schools in Illinois. For this reason, and because employee records for CPS are only available beginning in 1979, I drop employees of CPS from the analysis.¹⁰ Therefore, my results may not be representative of employees of public schools in extremely large cities.

An employee is eligible for the retiree health insurance with TRHIP when she leaves employment in IPS and begins collecting retirement benefits. The focus here is the effect this eligibility for retiree health insurance has on employee labor supply, namely on an employees’ decision to stop working in IPS. Therefore, I use termination of employment with IPS as the dependent variable, equal to one if the employee permanently discontinues employment with IPS after year t and 0 otherwise. I refer to this exit behavior as retirement, though it may not be synonymous with complete exit of the labor force if public school employees become employed elsewhere, e.g. the private sector, after discontinuing employment with IPS.

Other covariates of interest include age, years of experience, educational attainment, position (or title) and main employee assignment.¹¹ Because I am interested in the labor supply decisions of older workers with respect to their pension and retiree health insurance eligibility, I further limit the sample to include only workers ages 50 and older who are vested in the pension system (i.e. have at least 6 years of service).¹² For an employee’s years of experience, I use only the years of service accrued in IPS, which does not include service in public schools that are outside of Illinois or reciprocal service earned in other Illinois public employee retirement systems or private schools. While the former does not count as creditable service in the TRS system, the latter does. Not observing this information means there is measurement error in my controls for pension and TRHIP eligibility. To the extent that accrued experience in IPS does not capture the full credited experience with TRS, I will be underestimating some employees’ creditable experience.¹³ As such, I will be classifying some employees as ineligible who are actually eligible. This will lead me to underestimate the effects of both pension and TRHIP eligibility.

Educational attainment falls into one of four mutually exclusive categories: Bachelors degree, Masters degree, Doctorate, or other (including RN degrees and special certificates). I group employees into nine distinct position categories: superintendent, principal, dean, administrative staff, teacher, special education teacher or specialist, health related staff, other staff, and final category for those employees whose position information is missing in the data. I group teaching assignment categories in the following way: English as a second language, foreign languages, sciences, reading, social sciences, math, teachers in self-contained classrooms, other assignments, and a final category for employees whose assignment is unknown or missing in the data. Finally, I also include a measure of the employee’s annual salary in IPS, which I convert to 2013 dollars.

In Table 1, I present summary statistic information for the 405,139 IPS employee-year observations in the data. Ten percent of these older employees leave IPS each year, on average. The average age of employees in the sample over the period from 1971 to 1992 was 56. The mean experience in IPS of the employees in my sample is 22 years and the average salary is $66,519. Forty percent of employees hold Bachelor’s degrees and 55 percent Master’s degrees. The remaining employees are about equally likely to either hold a doctorate or some other degree. The vast majority of the employees in the sample are teachers (74 percent) or special education teachers and specialists (5 percent). Nine percent are superintendents, principals or deans. Another 12 percent of employees are staff. The assignments of teachers are also detailed in the table. Self-contained classroom teachers are approximately one quarter of the sample, and math and reading teachers are another 13 percent. The 25 percent of the sample that has an unknown main assignment classification is largely made up of administrators and staff without instructional duties.

Table 1.

Descriptive Characteristics of IPS Employees

	Mean	Standard Deviation
Exit	0.096	0.295
Age	56	5
Experience	22	8
Salary (thousands $2013)	66.519	19.991
Other Degree	0.028	0.166
BA Degree	0.401	0.490
MA Degree	0.552	0.497
PhD Degree	0.019	0.137
Position Held by Employee
Superintendents	0.034	0.182
Principals	0.056	0.230
Deans	0.004	0.062
Administrative Staff	0.015	0.121
Teachers	0.736	0.441
Special Education Position	0.047	0.211
Health Staff	0.052	0.222
Other Staff	0.056	0.230
Main Assignment of Teachers
ESL	0.002	0.040
Foreign Language	0.016	0.127
Science	0.041	0.199
Reading	0.090	0.286
Social Science	0.022	0.147
Math	0.036	0.187
Self-Contained Classroom	0.281	0.449
Other	0.258	0.438
Unknown	0.254	0.435
Number of Observations	405,139

Note: Based on author’s calculations using TRS and TSR data on employees of IPS schools ages 50 and older who are vested in the pension system from 1971 to 1992.

IV. Difference-In-Differences Strategy

In order to identify the effects of availability of retiree health insurance on the labor supply of older public employees, I make use of the fact that the eligibility for TRHIP was determined by an employee’s age and experience. Note that this is true for two reasons. First, in order to collect the retiree health insurance an employee must be collecting a pension benefit from TRS and pension-benefit eligibility is determined only by age and years of service. Second, even among the pension eligible, those with 8 years of service or more are explicitly eligible for retiree health insurance, while those with less than 8 years of experience are not. Therefore, I use a difference-in-differences strategy that compares exit behavior of pension eligible employees with at least 8 years of service before and after they were also eligible for TRHIP to that of observationally equivalent employees who are either pension ineligible or have less than 8 years of service (or both) before and after TRHIP introduction.

The value of TRHIP is determined by the availability of alternative forms of insurance, most notably Medicare, and eligibility for Medicare is determined in part by age.¹⁴ In the pre-TRHIP period, if an employee did not have access to insurance through her spouse or to retiree insurance through her school district of employment, waiting until age 65 to retire allowed employees to avoid forgoing insurance or paying the high costs of privately insuring themselves. After TRHIP introduction, there was no longer a need to delay retirement until age 65 solely for the sake of insurance. Instead, employees were eligible for retiree health insurance any time after they reached age 55 and met the requirements for pension eligibility and TRHIP. Therefore, I examine how the effects of TRHIP introduction on retirement behavior vary by the age of employees.

To determine the effects of TRHIP on public employee retirement, I estimate the following equation, equation (1):

$$Y_{\mathit{idt}}=\alpha +{\sum }_{\mathit{age}}{\pi }^{\mathit{age}}{\mathit{Eligible}}_{\mathit{idt}}+{\sum }_{\mathit{age}}{\beta }^{\mathit{age}}{\mathit{EligibleXPost}}_{\mathit{idt}}+\gamma X_{\mathit{idt}}+{\delta }^{\mathit{exp}}+{\phi }^{\mathit{age}}+{\theta }_{dt}+{\epsilon }_{\mathit{idt}}.$$

The outcome measure, Y_idt, is a dichotomous variable with a value of one if the teacher i in district d exits IPS after school-year t and zero otherwise.¹⁵ The vector X includes measures of teacher and district characteristics that vary over time, including measures of employee’s educational attainment, position in the school (e.g. principal, teacher, staff, etc.), subject matter of specialization, and salary as described in the previous section. There is a full set of district-by-year fixed effects, θ_dt, which capture any district-year-specific shocks to employee labor supply across districts, such as those resulting from changing macroeconomic conditions. Standard errors are clustered at the experience-level to account for any serial correlation in the error terms within employees of similar experience.¹⁶

The variable Eligible_idt is a dummy variable indicating that teacher i employed at district d is eligible for full retirement benefits and has at least 8 years of service in year t. The variable EligibleXPast_idt is a dummy variable indicating that teacher i employed at district d is eligible for full retirement benefits has at least 8 years of service in year t and that the observation occurs after TRHIP was introduced, i.e. post-1979. Both of these variables measuring eligibility are interacted with age-specific fixed effects allowing me to obtain coefficient estimates of the relationship between full-retirement-benefit eligibility and THRIP eligibility and retirement behavior, π^age and β^age, respectively, for employees who remain employed until each age. The model also includes experience fixed effects. Therefore, the relationships between pension eligibility and the retirement of employees by age, π^age, are identified by the differences in retirement behavior across individuals of the different ages who have the same amounts of accrued service, e.g. the 59 year old with 30 years of experience who is ineligible as compared to the 60 year old with 30 years of experience who is eligible.¹⁷ Recall that I have also included age fixed effects, thereby controlling for average differences in retirement propensity throughout the lifecycle.

The effects of retiree health insurance eligibility, β^age, are identified off of differences between the retirement behavior of eligible employees before and after July 1980, the point when TRHIP was introduced. One limitation of much of the existing literature is the lack of data or identifying variation that allows for distinction between eligibility for retiree health insurance and eligibility for pension benefits. During the time period I study, the eligibility for and generosity of pension benefits did not change. Comparing the retirement behavior of employees who are pension eligible to the behavior of employees with of the same age and with the same accumulated experience who are both pension and TRHIP eligible allows me to disentangle the effects of pension benefit eligibility from TRHIP eligibility. The underlying assumption in using this interpretation of the β’s is that there were no other concurrent policy or environmental changes that would have affected pension-eligible employees with at least 8 years of service differently than other ineligible employees of similar age and experience. In Section V, I show results of event style models with distributed lags and leads that show no evidence of differential trends.

Before turning to the estimates, it is worth noting that the fact that the TRHIP was a newly introduced program means the counterfacutal in this setting is a world with pensions, but no insurance. Therefore, the effects of TRHIP I estimate may not be the same as introducing TRHIP for people who are not pension eligible. These results may, however, speak to the question of what changes in retirement patterns we would see if retiree health insurance eligibility ages are increased without concurrent increases in the age of eligibility for pensions.

V. Effects of Retiree Health Insurance Availability on Retirement Behavior Difference-in-Difference Estimates

The difference-in-difference estimates show that exit of public school employees is responsive to the incentives of both the pension and retiree health insurance systems. This can be seen in Table 2, where I report the difference-in-difference estimates of retirement rates for pension eligible (first column) and TRHIP eligible (second column) employees of different ages. In the table, I report results from a specification that includes district-by-year, age, experience, age-group-by-year and experience-group-by-year fixed effects. The omitted group is therefore the set of employees who are ineligible for either pension or retiree health insurance. Importantly, all of the coefficients are estimates of differential retirement probabilities for the group of employees who remain employed at a given age. As I detail next, this may be a differentially selected group of employees at each age depending on whether employees are eligible for retiree health insurance or not.

Table 2.

Difference-in-Difference Estimates of the Effects of Pension and Retiree Health Insurance Eligibility on IPS Employee Retirement, by Age of Employee

	Eligible	Treat
Age 55	−0.018 (0.036)	0.052** (0.022)
Age 56	−0.010 (0.019)	0.047*** (0.012)
Age 57	−0.010 (0.022)	0.007 (0.018)
Age 58	0.015 (0.020)	−0.018 (0.018)
Age 59	0.023 (0.016)	−0.014 (0.013)
Age 60	0.096*** (0.010)	0.004 (0.011)
Age 61	0.068*** (0.012)	−0.002 (0.010)
Age 62	0.045*** (0.015)	0.007 (0.011)
Age 63	0.094*** (0.018)	0.008 (0.010)
Age 64	0.094*** (0.008)	0.008 (0.013)
Age 65	0.243*** (0.016)	−0.135*** (0.015)
Age 66	0.400*** (0.021)	−0.266*** (0.019)
Age 67	0.187*** (0.036)	−0.085*** (0.018)
Age 68	0.206*** (0.040)	−0.072** (0.032)
Age 69	0.181*** (0.040)	−0.097*** (0.029)
Age 70	0.269*** (0.043)	−0.087** (0.035)
Age 71	0.195*** (0.048)	−0.100*** (0.035)
Age 72	0.185*** (0.046)	−0.108** (0.046)
Age 73	0.171*** (0.060)	−0.143** (0.064)
Age 74	0.359*** (0.073)	−0.260*** (0.068)
Age 75	0.168*** (0.053)	−0.155*** (0.047)
Observations	405,139

Note: Based on author’s calculations using TRS and TSR data on employees of IPS schools from 1971 to 1992. Only employees ages 50 and older and those vested in the pension system are included in the analysis. Specification includes year-by-district, age, experience and age-group-by-year and experience-group-by-year fixed effects.

The pension eligible 55 to 59 year olds are no more likely than their pension ineligible counterparts to retire. However, retiree health insurance eligibility leads those who are employed at ages 55 and 56 to be 5 percentage points more likely to retire. At the same time, those who are pension eligible and still employed at ages 60 to 64 are between 5 and 10 percentage points more likely to retire than the ineligible who remain employed to ages 60 to 64. The pension eligible who are not also eligible for TRHIP and remain employed until at least age 65 are between 18 and 40 percentage points more likely to retire than their similarly aged counterparts who are ineligible for either pensions or retiree health insurance. However, those eligible for both retiree health insurance and a pension who have not yet retired by age 65 are more likely to leave than the pension ineligible that remain at the same age, but not as likely to leave as the pension eligible that remain employed to at least age 65.

In Table 2, the retiree health insurance eligibility induced increases in exit rates at ages 55 and 56 are much smaller than the decreases in exit rates for employees ages 65 and older. However, the number of employees working at age 55 is substantially larger than the number of employees who remain employed until age 65. In order to understand how retiree health insurance affects the composition of teachers, it is useful to consider how it would affect a particular cohort. Predicted conditional density functions (CDFs) and probability density functions (PDFs) of exit by age are presented in Figure 1, Panels A and B, respectively.¹⁸ In each panel, the solid line traces the CDF or PDF of exit for a cohort of employees before the retiree health insurance is introduced, while the dashed line represents the CDF or PDF for a cohort of employees after TRHIP is introduced.

Figure 1.

Predicted Density Functions of Retirement by Age for a Hypothetical Cohort, by Eligibility for Retiree Health Insurance

Note: Based on author’s calculations using TRS and TSR data on employees of IPS schools from 1971 to 1992. Only employees ages 50 and older and those vested in the pension system are included in the analysis. The solid black line presents the predicted CDF or PDF for pension-eligible workers at each age indicated on the horizontal axis from a specification that includes age-by-year, experience-by-year and district fixed effects, as well as other available information on employee characteristics. The solid grey line presents predicted CDF or PDF for pension- and retiree health insurance-eligible workers from the same specification. The plotted PDF of exit rates are relative to those of ineligible employees. The asterisks next to each age in Panel B indicate whether the coefficient estimate for the effect of TRHIP eligibility is statistically different than that of pension eligibility (i.e. the 95 percent confidence intervals of the two lines do not overlap).

In the CDFs in Panel 1A, note that the proportion of a cohort that exits at age 55 is much larger after retiree health insurance is introduced, as indicated by the estimates presented in Table 2. Furthermore, at every age between 55 and 65, a larger fraction of the cohort is predicted to have left employment at IPS if the cohort was eligible for retiree health insurance. It is only beginning at age 65 that slightly more of the cohort ineligible for retiree health insurance has exited than the cohort eligible for the insurance. The differences in exit rates at each age are detailed in Panel B, where the asterisks next to each age label on the horizontal axis indicate whether the coefficient estimate for the effect of TRHIP eligibility is statistically different than that of pension eligibility (i.e. the 95 percent confidence intervals of the two lines do not overlap). Most strikingly, as we saw in Table 2, there is a large statistically significant difference between the retirement rates of employees over age 65 who are eligible for either pensions and TRHIP or pensions alone. There is also a smaller, statistically significant, increase in the rate of retirement at ages 55 and 56 because of the TRHIP eligibility. From this evidence, we can conclude that employees are shifting their retirement forward in their careers based on their eligibility for health insurance that is not contingent on their employment.

Graphical Evidence Using the Raw Data

To be sure the results presented in Figure 1 are not artifacts of the specification choice, it is useful to examine the raw data for evidence of an effect. In Panel A of Figure 2, the dashed line presents information on the retirement rates of public school employees by age before retiree health insurance was provided by the state. There are local peaks at the places where pension eligibility changes discretely, ages 55 and 60, and a rather large global peak in retirement rates at age 65 and 66, when an employee becomes eligible for Medicare.

Figure 2.

Retirement Rates of IPS Employees by Age and Existence of TRHIP

Note: Based on author’s calculations using TRS and TSR data on employees of IPS schools from 1971 to 1992. The ‘before TRHIP’ period covers 1971 to 1979 and the ‘after THRIP’ period covers 1980 to 1992. Only employees ages 40 and older are included in the calculation of exit rates. The exit rate plotted on the vertical axis represents the rate at which employees of each age leave the sample between school year t and t+1.

The solid line in the figure traces retirement rates for public school employees by age in the period after TRHIP is introduced. Notice that, relative to pre-TRHIP levels, the retirement rate increases at the age when employees first become eligible for either retiree health insurance or a pension, i.e. age 55, but the rate of retirement at age 65 decreases significantly. Employees whose labor supply is elastic with respect to the availability of retiree health insurance retire when it becomes available to them, e.g. at age 55 if they have accrued enough experience. The retirement rates at older ages are no longer as sensitive to the Medicare eligibility because only those employees who have a strong attachment to the labor force (and whose labor supply is therefore less responsive to health insurance availability) remain employed until reaching these older ages.

Because TRHIP was introduced at a point in time, not for a particular cohort, many public school employees may not have been eligible for TRHIP at the same time they became eligible for their pension. This is because many employees may have already passed the age and years-of-service combination required for pension eligibility pre-1980. (For example, consider the 60 year old person in 1979. The introduction of retiree health insurance in 1980 cannot shift her retirement to be at any age earlier than 61.) For these constrained employees, the shifting forward of retirement rates with retiree health insurance eligibility in Figure 1 will be an underestimate of the true shift in retirement rates had they always had access to retiree health insurance.

Therefore, in Panel B of Figure 2, I plot exit rates of three groups of employees by their age in 1980. First is the solid line, indicating the retirement rates of employees by age before the TRHIP is introduced. (This is the same as the dashed line in Panel A.) Second, I have split the retirement rates after the program into two groups: those who were ages 54 and younger before TRHIP was introduced (and could therefore more fully adjust their retirement decision to its availability) and those who were 55 or older when TRHIP was introduced (and may therefore have been limited in terms of their behavioral adjustment). The line with smaller dashes traces exit rates of the former, and the line with longer dashes traces those for the latter. As would be expected, the retirement of employees who are younger when the program is introduced has more time to adjust and there is a larger response of their labor supply across ages.

Pensions and retiree health insurance might not only make retirement at certain ages more attractive due to their income effects, but the nonlinearities in the eligibility rules mean there are large returns to some employees of remaining employed until they become eligible for pension benefits. Since the introduction of TRHIP made retirement-benefit collection more attractive, in addition to making eligible employees more likely to retire, it may have also made those approaching eligibility less likely to retire. Looking at Figure 2, there is little evidence of this happening. Individuals ages 40 to 54 have very similar exit rates regardless of the availability of retiree health insurance.¹⁹

Event Study-Style Estimates of the Effects of Retiree Health Insurance

The difference-in-difference estimates show that the retirement of public school employees is sensitive to their retiree health insurance eligibility. The underlying assumption with the identification strategy used is that there were no other changes to the pension system or other relevant factors that would have affected the retirement behavior of eligible employees differently than that of ineligible employees. To find support for this assumption, I estimate equation (1) using the measure of pension eligibility interacted with dummy variables for each year relative to TRHIP introduction. These eligibility related distributed leads and lags are then interacted with age-specific fixed effects to determine if there is any time-pattern to the retirement of pension-eligible employees of each age that predates the introduction of TRHIP.

The results of this event-study-style analysis are presented in Appendix Figure 1.²⁰ Though the estimates of the coefficients on the year-specific effects and the eligibility variable are noisy, the estimates in the figure suggest that there were no pretreatment trends in retirement rates for these employees. The evidence therefore supports the underlying assumption in my use of a differences-in-differences framework to estimate the effects of retiree health insurance availability on employee exit.

Appendix Figure 1.

Event Study Style Estimates of the Effects of Retiree Health Insurance Eligibility on Employee Retirement

Note: Based on author’s calculations using TRS and TSR data on employees of IPS schools or older from 1971 to 1992. The vertical line represents the last period before TRHIP introduction. Only employees ages 50 and older and those vested in the pension system are included in the calculation of exit rates. The exit rate plotted on the vertical axis represents the rate at which employees of each age leave the sample between school year t and t+1. Coefficients and confidence intervals plotted are from an event-style model where eligibility for pensions and retiree health insurance are interacted with age-by-year specific dummies. Also included in the specifications are year-by-district, age, experience and age-group-by-year and experience-group-by-year fixed effects, as well as controls for employee characteristics.

Mandatory Retirement

Although the estimates presented in Figure 4 show no evidence of pre-existing trends or discrete changes in retirement behavior at any point before retiree health insurance was introduced, over this period there were significant changes to federal laws regarding mandatory retirement. Specifically, in 1978, Congress increased the minimum age for mandatory retirement from 65 to 70 and, in 1986, mandatory retirement was abolished in most industries. Unfortunately, to my knowledge, no data exist on any mandatory retirement policies that applied to the employees in my sample. However, if mandatory retirement age policies had been constraining behavior of IPS workers by forcing them to retire before they would have chosen to leave the public school system, we would expect to see the age at retirement increase, rather than decrease as it does here after 1980. As such, the rolling back of mandatory retirement may lead the estimates presented to be downward biased.²¹

Effects for Different Types of Employees

The elasticity of labor force participation with respect to subsidized retiree health insurance may differ across different types of employees. The retirement of people with particularly strong labor force attachment, e.g. those more educated and those who have spent effort training to become school leaders or management, is likely to be less elastic than their counterparts who have stronger preferences for leisure. On the other hand, the highest quality employees might have the best options for employment outside of the public school system, in which case they would be more likely to respond given the outcome as its measured here, exit from the public school system. To investigate whether labor supply responses differ across employees systematically, in Appendix Table 1, I present results for different types of employees. Specifically, I estimate the effects of pension and retiree health insurance eligibility by age for groups of employees based on either their position at the school or their highest degree held.

Appendix Table 1.

Difference-in-Difference Estimates of the Effects of Pension and Retiree Health Insurance Eligibility on IPS Employee Retirement, by Age Group, Position and Highest Degree Held

	By Position			By Highest Degree Held
	Teachers	Staff	Leadership	Bachelors	Masters	PhD	Other
Pension Eligibility, Ages 55 to 59	0.007 (0.016)	−0.007 (0.028)	0.011 (0.016)	0.008 (0.017)	−0.002 (0.017)	0.010 (0.044)	0.050* (0.025)
Pension Eligibility, Ages 60 to 64	0.081*** (0.008)	0.060*** (0.009)	0.083*** (0.010)	0.082*** (0.008)	0.072*** (0.008)	0.066** (0.026)	0.098*** (0.010)
Pension Eligibility, Ages 65+	0.274*** (0.024)	0.234*** (0.026)	0.169*** (0.022)	0.270*** (0.025)	0.248*** (0.022)	0.092 (0.061)	0.273*** (0.024)
TRHIP Eligibility, Ages 55 to 59	−0.001 (0.011)	0.008 (0.022)	0.023 (0.021)	−0.005 (0.012)	0.015 (0.013)	−0.036 (0.050)	−0.044 (0.039)
TRHIP Eligibility, Ages 60 to 64	0.004 (0.009)	0.014 (0.012)	0.016 (0.012)	0.008 (0.010)	0.009 (0.010)	−0.003 (0.032)	0.007 (0.020)
TRHIP Eligibility, Ages 65+	−0.194*** (0.014)	−0.143*** (0.023)	−0.110*** (0.019)	−0.184*** (0.015)	−0.177*** (0.015)	−0.015 (0.065)	−0.105*** (0.026)

Note: Based on author’s calculations using TRS and TSR data on employees of IPS schools from 1971 to 1992. Only employees ages 50 and older and those vested in the pension system are included in the analysis. Estimates presented are from a differences-in-differences specification in which eligibility for pensions and retiree health insurance are interacted with age-group fixed effects as indicated by the row headers. The specification includes controls for employee characteristics and district fixed effects as well as age-by-year and experience-by-year fixed effects.

^{***, ** and *} represent statistical significance at the 1, 5 and 10 percent levels, respectively.

There is no difference in the responsiveness of the retirement of teachers and staff to either their pension eligibility or their eligibility for retiree health insurance. In contrast, the retirement of school leadership is much less elastic with respect to both pension eligibility and TRHIP eligibility at ages 65 and older than that of the teachers and staff at those ages. This suggests that the set of school leaders who remain employed at older ages has stronger labor force attachment than the set of teachers and staff who remain employed at those ages. This is as expected given the extra training and work required to become a school principal, dean or superintendent. The results by highest degree attained corroborate the evidence for school leaders, but because there are very few employees with PhDs or other degrees, the results are not statistically significant.

VI. Implications

The numbers in Figure 1 imply that roughly 200 more people from each cohort (15 percent) retire either at age 55 or 56, while 15 percent fewer people wait until after age 65 to retire. The median older worker in IPS retires 2 years, or 8 percent, earlier than they otherwise would have because of the availability of retiree health insurance. This statistic does not tell us, however, whether this shift corresponds to a small share of workers who would have retired at 65 now retiring at much earlier ages or a larger share of workers retiring just a bit earlier. Bounding exercises such as those suggested by Djebbari and Smith (2008) suggest the latter is more likely, but I cannot rule out the former.²²

Elasticity of Labor Force Participation

These estimates can be used to calculate a measure of the elasticity of labor force participation with respect to income for older workers in IPS. (This requires assuming that these employees are not moving to other bridge jobs when they exit IPS.) Upon introduction, the annual premiums paid for TRHIP by employees younger than 65 were $2,400.²³ This represents 12 percent of the annual salary of 55 year olds in 1979. In Figure 1, we see the labor force participation of 55 year olds in about 9 percentage points lower when retiree health insurance is available, which is about a 10 percent decrease in the labor force participation of these 55 year olds. Therefore, a 10 percent increase in annual income would decrease participation of 55 year olds by 8 percent, which implies a labor force participation income elasticity of −0.8. This is in line with other estimates of the labor force participation of older workers (e.g. Bound 1989) and the response is comparable to that of other similarly aged private and public sector workers to retiree health insurance (Karoly and Rogowski 1994; Blau and Gilleskie 2001; Leisserson 2013).

Budgetary Implications

The sensitivity of the retirement behavior of older teachers to the provision of retiree health insurance shown here has many implications for government budgets. The most direct cost of retiree health insurance in Illinois, the employer’s share of the current annual premium, is $5,208 per retiree aged 55 to 64 (and less for older workers). This assumes the state continues its 50 percent subsidy of the premium. Although policymakers are most concerned with the current costs of retiree health insurance premiums, it is useful to think carefully about the implications of retiree health insurance over a longer time horizon and on other components of the budget.

A true calculation of the lifetime costs of the policy would take into account the full budgetary implications of the results shown here: workers with retiree health insurance are retiring earlier. This has implications for the calculation of annual and lifetime pension benefits, districts’ wage bills and the lifetime cost of the retiree health insurance program. For example, employees who retire earlier receive lower pension benefits but for a longer period of time than those who retire later. In this setting, the median retiree leaves at age 57 with 25 years of service rather than 59 with 27 years of service. Because this median retiree is an early retiree and faces an actuarial penalty for retiring early, the present discounted value of her pension is $76,731 lower than if she had not had retiree health insurance.²⁴ This represents a net savings to the pension system.

Additionally, the true cost to the government (or taxpayer) of the retiree health insurance includes its effects on the wage bill of public school employees. This includes a savings in the first two years (because 1^st and 2^nd year teachers have lower salaries than 26^th and 27^th year teachers) and extra expense in the 3^rd through 25^th year (since by the third year, the school district is employing a 3^rd year teacher rather than a 1^st year teacher, and so on). The present discounted value of this change in the wage bill is a net savings of $10,108 per teacher. Finally, because the median employee retires at 57, the lifetime present discounted value of the employer’s direct cost of the subsidized retiree health insurance for her is $70,248.²⁵

Combining these three components gives a total net lifetime savings of retiree health insurance $16,591 per worker. Based on this extremely simplified calculation, if governments revoked their retiree health insurance, rather than saving themselves (and taxpayers) the lifetime costs of the retiree health insurance, $70,248, they would actually pay about $16,591 more per public school employee. In other words, the cost of a teacher-year would increase by about $500 if retiree health insurance were discontinued. Of course, there are components of costs that I have not included in this calculation (e.g. I did not account for the growth in health care costs, factor in the costs of employee health insurance or include the employee contribution paid to the pension system while she is employed), but the implication remains: understanding the behavioral implications of retiree health insurance alters the cost-benefit calculus of such programs.

VII. Conclusion

The evidence presented here shows that the retirement behavior of public school employees in Illinois was responsive to the introduction of a retiree health insurance program. The retiree health insurance program served to move forward the timing of job separations for older workers. Although still a salient age for retirement, after retiree health insurance becomes available, many fewer employees wait until age 65 to retire. Instead, retiree health insurance leads more employees to retire 2 years, or 8 percent, earlier on average.

Because most retiree health insurance programs in the public sector are not constitutionally protected in the same way pensions are, the programs may become less generous, be dissolved, or be replaced by other forms of insurance. If this happens, the resulting retirements of older public school employees will have productivity effects that will depend on the quality of older versus younger employees.²⁶ The effects of changes in retiree health insurance will depend on the counterfactual environment, which, due to changes embedded in the Affordable Care Act, is likely to be different in the future than it is now or was in the early 1980s.²⁷ Regardless, there are likely to be some labor supply effects of changing public sector employee access to retiree health insurance that will in turn have effects on public sector budgets and liability funding.