BRIGITTE C. MADRIAN Harvard University The Effect of Health Insurance on Retirement FOR DECADES health insurancein the United States has been provided to most nonelderlyAmericans through their own or a family member's employment. This system of employment-basedhealth insurance has evolved largelybecause of the substantialcost advantagesthat employ- ers enjoy in supplyinghealth insurance. By pooling largenumbers of in- dividuals, employers face significantlylower administrativeexpenses thando individuals.In addition,employer expenditures on healthinsur- ance are tax deductible,but individualexpenditures are generallynot. I Despite these cost advantages, there is widespread dissatisfaction with this system of employment-basedhealth insurance. Many people are excluded because not all employersprovide health insurance and not all individualslive in households in which someone is employed. An es- timated 36 million Americans were uninsured in 1990.2 Even among those fortunate enough to have employer-providedhealth insurance, there is mountingconcern that it discourages individuals with preex- isting conditionsfrom changingjobs and, for those who do changejobs, it often means findinga new doctor because insuranceplans vary from firmto firm. I wish to thankGary Burtless, David Cutler, Jon Gruber,Jerry Hausman, Jim Poterba, andAndrew Samwick for theircomments and suggestions. 1. Undercurrent tax law, there are two circumstancesin which individualsmay de- duct theirexpenditures on healthinsurance: (1) those who are self-employedand who do not have access to employer-providedhealth insurance through anotherjob or their spouse's employmentmay deduct 25 percent of their expenditureson health insurance, and(2) those who itemizemay deducttheir medical expenses, includingthe cost of health insurance,to the extent thatthese expendituresexceed 7.5 percentof income. 2. Piacentiniand Foley (1992,p. 185). 181 182 Brookings Papers on Economic Activity, 1:1994 This dissatisfaction with the present system of employer-provided health insuranceis one of the factors drivingthe currentmove toward healthcare reform.Most proposalsspecifically address the problemsof access and continuityof coverage outlined above. For example, under PresidentClinton's plan, employers must help pay for the insuranceof theiremployees but need not provideit directly. Employees at firmsthat opt not to provide insurance would be covered by a plan of the em- ployee's choosing provided through a regional health alliance, which would be able to achieve the same cost savings as employers. Although individualswho are unemployedor who choose not to work would have to pay the full cost of their insurance, government subsidies would be available to help defray the costs for those with low incomes and for early retirees. And all plans, whether provided by an employer or a healthalliance, would prohibitinsurers from excludingpreexisting con- ditions. While other proposals differ in their approachto reform, most, like the Clintonplan, will fundamentallychange the way that healthcare is either financedor delivered. Given the currentrelationship between healthinsurance and employ- ment, any health care reformwill affect the structureof the labor mar- ket. This paperexplores one piece of the employmentrelationship that is likely to change:the decision to retire.Just as employer-providedhealth insurance is thought to deter job mobility, it is also likely to deter re- tirement. Variousaspects of the proposalsraise three importantissues regard- ing the effects of reformon retirement.The first is the extent to which the provision of health insurance is tied to employment. Severing the link between employment and the receipt of health insurance will in- crease the incentives to retire early among those who fear losing health insurancecoverage if they do so. But mandatingthat all employerspro- vide health insurancemay also postpone the retirementof some older workerswhose employersdid not previously provide health insurance. The second issue is whether the plan establishes purchasingpools that makenonemployment-based health insurance available to individualsat the lower rates usuallyachieved by firms.These pools would reduce the cost of health insurancein the private marketand would therefore en- courageearly retirement.The thirdissue is whetherthe plan subsidizes the purchaseof individualhealth insuranceby early retirees. Such sub- sidies furtherreduce the cost of purchasingprivate health insurance and would thereforeprovide an additionalincentive for early retirement. Brigitte C. Madrian 183 To evaluate the effect that differentreform proposals might have on retirement,one must know the extent to which health insurance influ- ences the retirementdecision. This paper attempts to quantifythis ef- fect. I begin by examiningwhether health insuranceaffects other types of labormarket behavior, such as job mobilityand laborforce participa- tion. To the extent that health insuranceaffects these other types of be- havior, it is likely to influenceretirement. I then examine the factors that make health insurancean importantconsideration in deciding whether to retire. These include the need for health insurance, which rises with age, and the availabilityof health insuranceafter retirement.Among in- dividualsaged 55-64, those who are not employed are much less likely to be covered by employer-providedhealth insuranceand much more likely to be eitheruninsured or covered by an individualnongroup policy or government-sponsoredhealth insurance. To the extent that the latter types of insuranceare inferiorto employer-providedhealth insurance, they may defer retirement. Given these reasons why health insuranceshould affect retirement,I then brieflyreview the existing literatureon healthinsurance and retire- ment. Previouswork in this areais limited, mainlybecause there are few data sets that describeboth healthinsurance and labormarket behavior. My analysis uses three individual-leveldata sets: the 1987 National Medical ExpenditureSurvey and two subsets of the Survey of Income and ProgramParticipation. All three include questions on the age at re- tirement and the availability of employer-provided postretirement healthinsurance. The next section of the paperdescribes the availabilityand structure of postretirementhealth insurancebenefits. Although the provision of these benefitshas increasedover time, less than half of the workforce is currentlyemployed by firms offering health insurance to retirees. As mightbe expected, workers in large firms and those employed in high- wage industriesare more likely to receive such benefits. Surprisingly, eligibilityto receive retiree health benefits is tied to the receipt of other retirementbenefits, such as a pension, for only one-quarterof workers. Using the data described above, I then examine the effect of retiree health insuranceon age at retirement.I find that those with employer- provided postretirementhealth insurance retire 5-16 months earlier than those without such insurance. This result is robust using various definitionsof retirementand regardlessof whetherindividuals receive a pension. It is less clear, however, whether this is the effect of actually 184 Brookings Papers on Economic Activity, 1:1994 having retiree health insuranceor whether a firm's provision of retiree healthinsurance is correlatedwith overalljob quality, the generosity of pension benefits, and any incentives for early retirementthat are partof the pension plans in these firms. Finally, I try to answer three questions about health insurance and retirement.The firstis what impacthealth care reformmight have on the labor force participationof older individuals. I estimate that universal coverage would reduce the laborforce participationof men aged 60-64 by 4.3 percentagepoints. To the extent that reformis financedthrough payroll taxes ratherthan throughindividual contributions, such a labor supplyresponse would reduce the revenue that the governmentcollects from these taxes. Second, I ask to what extent the increased availability of retiree health insuranceexplains the aggregatetrend toward early retirement. The labor force participationof men aged 55-64 has fallen steadily over the past several decades. Much of this decline is attributableto the in- creased generosity of both social security and private pensions. How- ever, these two factors together can explain only one-thirdto one-half of the decline in aggregatelabor force participation.Over this same time period, the availabilityof retiree healthinsurance has increasedas well. My estimates suggest that the increased availabilityof postretirement health insurancecan explain between 10 and 20 percent of the overall decline in the laborforce participationof older men. The third question is whether eligibility for Medicare explains the "excess" spike in the probabilityof retirement(the retirementhazard) at age 65. The social security system and many private pensions provide incentives to retireat age 65; consequently, some spike in the retirement hazard at this age is not surprising.However, the financialincentives associated with social security and pensions do not fully account for the spike at 65.3One potentialexplanation is that individualswait until they are eligible for Medicarebefore they retire in orderto avoid being unin- sured or having to purchase insurance in the private market. If this is true, the excess spike should exist only for those with retiree health in- surance. I find little evidence of this. Althoughhealth insuranceclearly