RETIREMENT RESEARCH June 2017, Number 17-12 DOES PUBLIC HEALTH INSURANCE AFFECT HOW MUCH PEOPLE WORK? By Gal Wettstein* Introduction Most Americans get their health insurance through tively modest portion of the total population of older their employer, so they may be reluctant to leave a job Americans, this result does suggest that having the if such a change affects their coverage. This situa- option of public health insurance can reduce a barrier tion is known as “job lock,” which may be a particular to labor force transitions. concern for those with health problems.1 As a result, expansions of public health insurance, which are not tied to a job, could reduce job lock and result in some Medicare Part D workers scaling back from full- to part-time work or leaving the labor force entirely. Medicare has provided universal health insurance to One way to estimate the effect of public health all Americans age 65 and over since 1966. However, insurance on job lock is to look at policy changes that in its first 40 years, the program did not generally pro- offer a “natural experiment." This brief, based on a re- vide insurance for prescription drugs.3 This omission cent paper, uses the introduction of Medicare Part D grew more significant over time as the effectiveness in 2006 to assess the extent to which the availability of of drugs, as well as their costs, increased. To address drug coverage not tied to an employer induces older this large uninsured risk, Medicare was expanded in individuals to work less.2 January 2006 to cover prescription drugs through the The discussion proceeds as follows. The first sec- Part D program. tion provides brief background on Medicare Part D. The introduction of Part D was particularly im- The second section describes the data and sets up the portant for people considering retirement. Virtually methodology. The third section shows trends in labor all employer health insurance plans cover prescrip- force activity among the elderly before and after 2006. tion drugs for their current employees.4 Before Part The fourth section summarizes the main results. The D, drug insurance options for retirees were limited final section concludes that, prior to the introduction if they did not have employer-provided retiree health of Part D, “job lock” was a significant concern for indi- insurance (RHI). Thus, they would be vulnerable to viduals who would otherwise have lost their employer high drug costs if they left their employer plan. drug insurance at age 65. While this group was a rela- * Gal Wettstein is a research economist with the Center for Retirement Research at Boston College. 2 Center for Retirement Research In short, before 2006, access to drug coverage for 2006, such individuals who retired at or after 65 those 65 and older was limited mostly to individuals would lose their drug coverage when they transitioned with employer plans. After 2006, they could get drug from their employer plan to Medicare. The only way coverage through Medicare. The question is whether to keep their drug coverage was to keep working. the introduction of such coverage affected the labor After 2006, they could keep their coverage past age 65 force decisions of older workers. through Medicare regardless of when they retired. Data and Design Figure 1. Composition of HRS Sample Ages 55-68 The data used in the analysis are from the Health and by Employer Health Insurance Status Retirement Study, a large panel of Americans over age RHI after 65 unknown, 50 and their spouses. It was started in 1992, and fol- 5% lows up with its subjects every two years. RHI for life To isolate the effect of the introduction of Medi- (control), 12% care Part D on retirement, the study limits the sample No ESI, to individuals around age 65 (ages 55-68)5 and around RHI only to 65 41% the year 2006 (years 2000-2010). This restriction (treatment), provides a group of individuals ages 55-64 who saw 14% no change in their drug insurance availability, and a group of individuals ages 65-68 who had no access to Part D coverage in 2000-2004 and acquired it in 2006-2010. This approach allows for an estimation ESI, no RHI, 28% of the effect of subsidized drug insurance on labor outcomes for individuals ages 65-68. Note: Numbers reflect person-year observations. Before 2006, the change in work rates at age 65 Source: Author’s estimates from the Health and Retirement captures the baseline pattern of work over the life- Study (HRS) (2000-2010). cycle (e. g., the natural decline in work through aging, and institutional factors like access to Social Secu- rity). After 2006, the change in work rates at age 65 The second group, which functions as a control includes the same life-cycle trends, but also the effect group, is those who have RHI for life (in shaded red). of the new subsidized drug insurance. The difference They form a good control group, as they are quite between the change in work at age 65 after 2006 and similar to the treatment group. Both groups have the change before 2006 isolates the effect of Part D on RHI; they only differ in whether that insurance is individuals’ labor force decisions. limited to age 65 (treatment) or not (control).6 The Before 2006, not everyone faced an incentive to control group is also observed at the same ages as the keep working in order to maintain insurance cover- treatment group in the same years, so if something age. Workers at firms that did not offer employer- unobservable happens to change the labor outcomes sponsored insurance (ESI) certainly would not be af- of 65-68 year olds after 2006, they would experience fected. Likewise, individuals who have retiree health that same shock and could be used to control for it.7 insurance (RHI) for life from their employer should Robustness checks show that using those with no ESI not be affected: both before and after Part D they can (in solid gray) as an alternative control group yields retire freely without losing their drug coverage. similar results.8 Figure 1 shows the breakdown of the HRS sample by type of employer insurance coverage. To focus on a relevant population, the study restricts attention Retirement Patterns Before to individuals who have RHI, and divides them into and After 2006 two groups. The first is a treatment group made up of those who have RHI only until age 65 (the solid An initial way to assess the potential impact of Part D red portion of Figure 1). This arrangement is fairly and evaluate the experimental design is to simply look common, applying to about half of those with RHI, as at the patterns in full-time work rates by age before everyone gains access to Medicare at age 65. Before and after 2006. Issue in Brief 3 Figure 2 shows the pattern for the treatment Figure 3. Percentage of Individuals Who Work group alone, where the gray line indicates those Full Time, Treatment and Control Groups, After with RHI only to age 65 in the period before 2006, 2006 while the red line shows the same insurance-status 90% group after 2006 (these two lines represent different birth-year cohorts).9 Those observed at ages younger than 65 move in lockstep both before and after 2006, suggesting that in the absence of Part D they would 60% continue to move in parallel throughout their lives. However, at age 65 a much sharper decline is evident after 2006 for the treatment group relative to the ear- lier period, suggesting that Part D may have caused a 30% large decline in the rate of full-time work. Treatment Control 0% Figure 2. Percentage of Individuals with RHI 55 56 57 58 59 60 61 62 63 64 65 66 67 68 Only Until Age 65 Who Work Full Time, Pre- and Post-Medicare Part D, by Age Source: Author’s estimates from the 2000-2010 HRS. 90% So far, Figures 2 and 3 suggest that Part D sub- stantially reduced job lock for the treatment group; 60% however, to test whether these effects are statistically significant and control for other factors, a regression analysis is needed. The basic estimation equation is: 30% Probability of working = ƒ (over65, post2006, treatment, personal characteristics), Pre-Part D with interactions of: Post-Part D (over65)(post2006), (over65)(treat), (post2006)(treat), 0% and (over65)(post2006)(treat) 55 56 57 58 59 60 61 62 63 64 65 66 67 68 Note: For clarity, pre-2006 observations are shifted up so the The same equation is estimated with several dif- mean of those under 65 is the same before and after 2006. ferent dependent variables that all measure aspects Source: Author’s estimates from the 2000-2010 HRS. of an individual’s labor market activity. The two measures highlighted in the results below are the Figure 3 displays the evolution of full-time work probability of working full time and the probability of for both the treatment and control groups, after 2006. working part time.10 The independent variables iden- As with the treatment group in isolation, the paral- tify whether the individual was over 65; was observed lel movement between the two groups before age 65 post 2006; and was part of the treatment group. The indicates that the groups had similar employment control variables include demographic variables, patterns before Medicare eligibility. health variables, and age, year, and individual fixed As expected, no marked change occurs in the de- effects. The main coefficient of interest is on the in- cline of employment at age 65 for the control group; teraction of (over65)(post2006)(treat), which compares the pattern for this group looks similar before and the change in the dependent variable at age 65, before after this age. For the treatment group, however, the and after 2006, for those whose RHI was limited to decline exactly at 65 after 2006 is dramatically larger under 65 (treatment group) to those who had RHI for than for the control group. This result is consistent life (control group). with the notion that Part D reduced job lock, affecting those who might have been constrained but having no effect on those for whom no job lock was possible. 4 Center for Retirement Research Regression Results Figure 5. Estimated Effect of Part D on Labor Outcomes of the Treatment Group by Health Figure 4 shows the key regression results for the ef- 15% fects of Part D on full- and part-time work.11 Part D Sick 9.9% led to a statistically significant decline of 8.4 percent- 10% Healthy age points in full-time work among individuals who were dependent on their employer insurance for 5% drug coverage. This estimate is large; the full-time 0.5% 0% work rate at the baseline was 35 percent, so Part D -1.1% led to a 24-percent reduction from that average rate. -5% Of course, this result does not mean that all of the affected individuals moved into retirement. Instead, -10% they may have shifted to part-time work.12 Indeed, -12.2% part-time work did increase in the treatment group by -15% Full-time Part-time 5.9 percentage points out of the 8.4-percentage-point overall effect (see Figure 4). Thus, the reduction in Note: Solid bars are statistically significant. full-time work can be decomposed into 70 percent Source: Author’s estimates from the 2000-2010 HRS. switching into part-time work and only 30 percent go- ing into full retirement.13 Conclusion Figure 4. Estimated Effect of Part D on Labor Overall, decoupling labor force decisions from Outcomes for Treatment and Control Groups insurance decisions can affect labor supply among 10% those near retirement. This study finds that, before Treatment the availability of Medicare Part D, many individu- Control 5.9% als worked past age 65 to maintain access to their 5% 2.0% employer-sponsored drug insurance. While this bar- 0.2% rier to retirement is only relevant for those who have 0% employer-sponsored health insurance, it seems to provide a large incentive to delay retirement for this -5% group. Knowing the pervasiveness of job lock is impor- -10% -8.4% tant for assessment of public policies that weaken the link between employment and insurance.14 If policies -15% remove an inefficient constraint on retirement, they Full-time Part-time could be beneficial. On the other hand, they may be Note: Solid bars are statistically significant. costly if they reduce employment and, corresponding- Source: Author’s estimates from the 2000-2010 HRS. ly, tax revenue. The large estimated labor responses imply a high valuation by near retirees on the health insurance subsidies in Part D. However, they also The results shown in Figure 4 are driven almost indicate that the fiscal cost of these subsidies is larger entirely by less healthy individuals. Sick individuals than their cost on paper when taking into account the (those with chronic health conditions such as diabetes reduced taxable earnings that result.15 or heart disease) see a decline in full-time work of 12.2 percentage points and an increase in part-time work of 9.9 percentage points (see Figure 5). In contrast, healthy individuals display no statistically significant response to Part D in their labor outcomes. Issue in Brief 5 Endnotes 1 For a comprehensive review of the early literature 9 Due to secular increases in labor supply among on this subject, see Gruber and Madrian (2004). For the elderly over time, the mean full-time work rate more recent papers see Boyle and Lahey (2010), who is higher after 2006 relative to before 2006. In this look at veterans, and Garthwaite, Gross, and Noto- figure, the pre-2006 observations are shifted up so the widigdo (2014), who consider the Medicaid-eligible mean of those below age 65 is the same before and af- population. These papers find similar magnitude ter 2006. This shifting is done for clarity. The graph effects on labor force participation as the study sum- with the raw means can be seen in Wettstein (2016). marized in this brief. 10 Other labor outcomes considered are weekly 2 Wettstein (2016). hours worked and earnings; all of the labor outcome results tell a consistent story. 3 Medicare did cover some drugs, such as those provided in hospitals, through Medicare Part A. 11 For complete results, see Appendix Table A1. Furthermore, Medigap and HMO plans covering drugs existed, but provided limited insurance for high 12 Few employers offer health benefits to part-time premiums, and were chosen only by a small minority workers (Kaiser Family Foundation, 2014), so the of those eligible. introduction of Part D could have made such a shift attractive to those ages 65 or over. 4 Kaiser Family Foundation (2014). 13 All results are robust to having no control group, 5 Construction of the treatment and control groups or using a different control group of individuals with relies on questions regarding retiree coverage which no employer insurance; excluding the Great Reces- are not asked of everyone. As a result, the oldest in- sion years; restricting the sample to ages 62-68; and to dividuals who can be included are those up to age 68. using other sets of control variables. For details, see the Data Appendix in Wettstein (2016). 14 For example, the Congressional Budget Office 6 The treatment and control groups are also similar (2014) estimated a reduction of 1.5-2 percent in hours on observable characteristics such as demographics, worked due to the Affordable Care Act, partially due occupation, and industry. For details, see Wettstein to relaxation of retirement lock. This estimate is (2016). based on Gruber and Madrian (1995). 7 In practice, no such shock is found and this control 15 See Wettstein (2016) for more detailed analysis. group merely serves to reinforce the validity of the estimates found in the treatment group alone. 8 Observations in shaded gray are not considered in the study as they cannot be definitively allocated between the treatment and control groups. Those in black are not included because they are not very similar to the treatment group. 6 Center for Retirement Research References Boyle, Melissa A. and Joanna N. Lahey. 2010. “Health Insurance and the Labor Supply Decisions of Older Workers: Evidence from a U.S. Department of Veterans Affairs Expansion.” Journal of Public Economics 94(7-8): 467-478. Congressional Budget Office. 2014. “Budget and Eco- nomic Outlook.” Washington, DC. Garthwaite, Craig, Tal Gross, and Matthew J. Notowi- digdo. 2014. “Public Health Insurance, Labor Sup- ply, and Employment Lock.” The Quarterly Journal of Economics 129(2): 653-696. Gruber, Jonathan and Brigitte C. Madrian. 1995. “Health Insurance Availability and the Retirement Decision.” The American Economic Review 85(4): 938-948. Gruber, Jonathan and Brigitte C. Madrian. 2004. “Health Insurance, Labor Supply, and Job Mobil- ity,” In Health Policy and the Uninsured, edited by Catherine McLaughlin, 97-178. Washington, DC: Urban Institute Press. Kaiser Family Foundation. 2014. Employer Health Ben- efits, 2014 Annual Survey. Menlo Park, CA. University of Michigan. Health and Retirement Study, 2000-2010. Ann Arbor, MI. Wettstein, Gal. 2016. “Retirement Lock and Prescrip- tion Drug Insurance: Evidence from Medicare Part D.” Doctoral thesis. Cambridge, MA: Harvard University. Available at: http://scholar.harvard. edu/files/gwettst/files/jmp_0.pdf APPENDIX 8 Center for Retirement Research Table A1. Estimated Effect of Part D Eligibility on Labor Outcomes, Overall and by Health Full-time work Part-time work Variables Full sample Sick Healthy Full sample Sick Healthy Treatment effect -0.0836*** -0.122*** -0.00536 -0.0596* -0.099*** -0.0113 on treatment (0.0313) (0.0374) (0.0652) (0.0305) (0.0380) (0.0586) group Treatment effect 0.0199 0.0416 -0.0269 -0.00521 -0.0221 0.0353 on control group (0.0217) (0.0256) (0.0446) (0.0216) (0.0266) (0.0428) Health controls Yes No No Yes No No Observations 15,382 10,733 4,649 15,382 10,733 4,649 R-squared 0.219 0.21 0.217 0.013 0.014 0.035 Notes: Results are statistically significant at the 1-percent (***) or 10-percent (*) level. All equations include age and year interacted with treatment dummies, individual fixed effects, and demographic controls. Source: Author’s estimates from the 2000-2010 HRS. RETIREMENT RESEARCH About the Center Affiliated Institutions The mission of the Center for Retirement Research The Brookings Institution at Boston College is to produce first-class research Syracuse University and educational tools and forge a strong link between Urban Institute the academic community and decision-makers in the public and private sectors around an issue of criti- cal importance to the nation’s future. To achieve Contact Information Center for Retirement Research this mission, the Center sponsors a wide variety of Boston College research projects, transmits new findings to a broad Hovey House audience, trains new scholars, and broadens access to 140 Commonwealth Avenue valuable data sources. Since its inception in 1998, the Chestnut Hill, MA 02467-3808 Center has established a reputation as an authorita- Phone: (617) 552-1762 tive source of information on all major aspects of the Fax: (617) 552-0191 retirement income debate. E-mail: crr@bc.edu Website: http://crr.bc.edu © 2017, by Trustees of Boston College, Center for Retirement Research. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that the author are identified and full credit, including copyright notice, is given to Trustees of Boston College, Center for Retirement Research. The research reported herein was performed pursuant, in part, to a dissertation fellowship award from the U.S. Social Secu- rity Administration (SSA) funded as part of the Retirement Research Consortium. The opinions and conclusions expressed are solely those of the author and do not represent the opinions or policy of SSA, any agency of the federal government, Bos- ton College, or the Center for Retirement Research. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of the contents of this report. Reference herein to any specific commercial product, process or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply endorsement, recom- mendation or favoring by the United States Government or any agency thereof.