Retirement Measures

The Board of Regents of the University of Wisconsin System

Retirement Measures in the Health and Retirement Study Author(s): Alan L. Gustman, Olivia S. Mitchell, Thomas L. Steinmeier Source: The Journal of Human Resources, Vol. 30, Special Issue on the Health and Retirement Study: Data Quality and Early Results (1995), pp. S57-S83 Published by: University of Wisconsin Press Stable URL: http://www.jstor.org/stable/146278 . Accessed: 24/10/2011 09:03
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

University of Wisconsin Press and The Board of Regents of the University of Wisconsin System are collaborating with JSTOR to digitize, preserve and extend access to The Journal of Human Resources.

http://www.jstor.org

Retirement Measures in the Health and Retirement Study

Alan L. Gustman Olivia S. Mitchell Thomas L. Steinmeier
ABSTRACT

This paper highlights unanswered research questions in the economics of retirement and shows how these questions can be addressed using the Health and Retirement Study (HRS). Unique features of the survey are described, including administrative records on earnings and Social Security benefits and employer-provided data on pensions and health insurance. Also collected by the survey are indicators of retirement plans, health status, family structure, income, wealth, and employer policies affecting job opportunities and constraints. Data from the first wave of the HRS are used to analyze retirement outcomes and constraints shaping retirement behavior.

I. Introduction
Researchersand policymakershave become increasinglyawareof the critical need for a new longitudinalsurvey to analyze work, retirement,and health patternsof older Americans. One reason is that existing data sets are out of date, and hence less useful for currentpolicy purposes. The economic, health, and social opportunitiesfacing older people are different now than in past decades. In addition, people reachingretirementage today may have differentex-

Alan L. Gustman is the Loren M. Berry Professor of Economics at Dartmouth College and a research associate of the National Bureau of Economic Research; Olivia S. Mitchell is International Foundation of Employee Benefit Plans Professor of Insurance and Risk Management and Executive Director of the Pension Research Council at the University of Pennsylvania's Wharton School and a research associate of the National Bureau of Economic Research; Thomas L. Steinmeier is a professor of economics at Texas Tech University. The authors acknowledge research support from the National Institute on Aging, the University of Michigan, and the Wharton School. Conclusions and interpretations are those of the authors. The data used in this article are from the alpha release of the HRS. THE JOURNAL OF HUMAN RESOURCES * XXX * Supplement 1995

S58

The Journalof HumanResources pectations about health and retirementthan in previous years.1While both the RetirementHistory Study (RHS) and the National Longitudinal Survey of Older Men (NLS-OM) were invaluablefor analysis of priorgenerations,they focused on people who are now in their late seventies and eighties. This cohort is now well past the period when most people are makingretirementdecisions.2 The Health and RetirementStudy (HRS) is a new longitudinal survey designed to fill this gap. It includes a comprehensiveset of questions which will permit econometric studies of labor force and health outcomes and their determinants. The first wave of the HRS focuses on a representativesample of older people aged 51-61 in 1992 as well as their spouses. These people will be followed with a longitudinalsurvey formatfor years to come, with the second wave fielded in 1994. The HRS collects a rich arrayof informationon income and time constraints affectingthis cohort's retirementexpectations, attitudes,and opportunities.The on study includes information earningsprofiles,currentand anticipatedprivately benefits such as pensions and health insurance,currentand anticipated provided governmentpayments such as Social Security, disability,and otherbenefits, and a variety of other data on income, debt, and assets. Finally the survey contains numerous measures deemed importantby sociologists, psychologists, and the medical profession, including employer and fellow-employee attitudes toward older workers, family demands, and an assessment of workers' and spouses' psychologicaland physical states. Here we identify how key researchand policy questionsabout retirementcan be addressed with the new HRS. After a brief overview of the most important questions that the retirementliteratureis confronting,we turn to an analysis of the new informationavailablein the first wave of the HRS on older Americans' labormarketoutcomes, along with data on the opportunities constraintsthey and face, includingincome and assets, health status, family structure,and transfers. We conclude with a discussionof what the findingsportendfor researchersexamining retirementand health using the HRS in the future.
1. Previous questionnaires gathered little information on women's work and retirement, mainly because of the low fraction of all women working, in the paid labor market. (Michael 1985 provides historical information on labor force participation by women.) The National Longitudinal Survey of Older Men excluded women entirely, while the Retirement History Study provided inadequate information on the labor market activities and opportunities of women, especially married women. Minority groups were also underrepresented in prior retirement surveys, making it difficult to use these to study ethnic differences in retirement patterns. Another problem with previous surveys was that they had inadequate pension data. Two recent reviews examine strengths and weaknesses of the existing literature on pensions and retirement: Gustman and Mitchell (1992) and Gustman, Mitchell, and Steinmeier (1994). 2. The RHS consisted of a biannual survey of people born between 1906 and 1911 who survived to enter the sample frame in 1969. The last wave of the RHS was completed in 1979. (Sample members who survived until 1995 would be 84-89 years of age.) Numerous studies using that data set for retirement analyses are listed in U.S. Department of Health and Human Services (1987). Respondents to the NLS-OM were aged 45-59 in 1966, and the last regularly scheduled labor market survey of this group occurred in 1983; an additional special survey wave was fielded in 1990 for determining circumstances after retirement. Studies using the NLS for retirement analysis appear in the Center for Human Resources Research (1988). Respondents to the National Longitudinal Study of Mature Women Survey (NLS-MW) were 30-44 years of age in 1967; that survey is only now becoming available for use in retirement research.

Gustman,Mitchell, and Steinmeier S59

II. How the HRS Can Help Address Important Retirement Questions
of To place in context the anticipatedcontributions the Healthand RetirementSurvey, it is useful to review briefly some lessons from the last two decades of retirementresearch. In general, most of those who have examined retirementpatternsfrom an economic perspectivewould agree with the following points:3
* Older people decide when to retire by taking into account not only current work and leisure opportunities, but future opportunities as well. Hence

retirementbehaviormust be modeledusingan intertemporal utilitymaximization frameworkratherthan with a single-periodmodel of cross-sectional labor force status.
* Jobs and pensions sometimes make it costly to continue working at older

ages. This can happen when a defined benefit pension subsidizes early retirement,or when a nationalretirementincome system penalizesdeferred retirement.Additionally,there may be problemsfinding jobs with reduced from emhours of work, and some older workers face age discrimination in ployers. All of these factors generate nonlinearities older workers' budget constraintsnear retirement.
* Retirement patterns vary according to family structure and marital status.

The presence, or absence, of spouses, dependent children, and elderly parents has substantial effects on retirementpatterns for both men and
women. * Retirement is influenced by health as well as economic factors.

Most retirement researchers would also agree that several importantunanswered questions need answers, in order to better understandwhy people retire when they do, and what effect health and retirementpolicy have on behavior. Six questions are worthy of special note here, though we recognize that the list could be expanded considerably:
1. What explains the long trend toward earlier retirement among men, and why did the downward trend level out in the last decade?

Men's labor force participationrates fell in the United States from the 1950s to the mid-1980s,and then leveled off (Anderson,Gustman,and Steinmeier1993). There remainsconsiderablecontroversyabout what explainsthese patterns,and
answers must be refined for better retirement income policy.4 The HRS will be
3. Campbell and Campbell (1976) review older retirement studies; more recent reviews include Mitchell and Fields (1982), Quinn, Burkhauser, and Myers (1990), and Sammartino (1987). See also Gustman and Steinmeier (1984, 1986), Hurd (1990), Lumsdaine, Stock, and Wise (1994), and Rust (1990). 4. A common theory is that the long downward participation trend has been due to trends in the wage structure, perhaps due to changes in the occupational and industrial mix of jobs disfavoring unskilled and older workers; however, the evidence does not appear to support this theory (Anderson, Gustman, and Steinmeier 1993). Worsening health cannot explain the trend either; recent evidence suggests that

S60

The Journalof HumanResources useful in providingbetter estimates of the incentives created by pensions and Social Security on retirement,and by facilitatingthe analysis of how changes in these incentives affect retirementoutcomes.
2. How does the family context affect decisions about work and retirement?

how retirement decisions are deterResearchershave only begunto understand minedwithinthe family. Risingrates of marketwork amongwomenand changing family structuresare likely to affect patternsof labormarketparticipation among women and men at older ages. Economic security has become more elusive for some older persons due to increasingdivorce rates and longevity, exposing increasingnumbersof people to a greaterriskof povertyandan increasedlikelihood of need for long-termcare. Linkedto this is the questionof how well today's aging generationis insured against possible drops in consumptionthroughpensions, life insurance, and disabilityinsurance.The HRS offers unique opportunitiesto examine how family structureaffects work and decisions about retirement.
3. What are useful ways to measure and model the impact of health status on retirement?

A debate continues on how best to measure health status in the context of retirementstudies. Existing data sets cannot resolve this debate since they do not provide health measures of sufficientlyhigh quality to determinewhether health measurescan be treatedas exogenous determinants retireof self-reported ment (Sammartino1987). The HRS offers researchersbetter informationon respondents'health than any previousretirementdata set, with detailedreportson chronic and acute health conditions, medical care insurance, and medical care utilization. Moreover, the appended questions on provisions of retirementand disability programs,together with these health measures, allow HRS users to determinehow health status interactswith benefitplans and health care plans to and retirementbehavior. shape labor force participation
4. How do retirement patterns respond to pecuniary and nonwage attributes of jobs?

Since wages are the most importantelement in compensation,a great deal of attentionmustbe devoted to theiraccuratecollection. Benefitsincluding pensions and retiree health insurance also affect the rewardsfor continuedwork among older individuals,as do nonwagejob attributesand job relationships,including
longer-lived recent generations are more able to work as compared to their older counterparts (Manton, Corder, and Stallard 1993). Increasing pension coverage and pension wealth coupled with improvements in Social Security benefits may be part of the explanation (Ippolito 1990). Nevertheless, Social Security incentives and wealth effects from unexpected benefit increases appear to have a relatively small effect on retirement incentives and outcomes (Burtless 1986), and the effects of unexpected wealth changes, including those from the early years of a growing Social Security System, should eventually be fully reversed. Pension incentives in defined benefit plans may also contribute to earlier retirement ages, although these plans do reduce retirement in the years before eligibility for early retirement age is attained (Stock and Wise 1990a, 1990b). Defined contribution and 401(k) plans embody little or no retirement incentives beyond wealth effects (Gustman and Steinmeier 1992; Ippolito, forthcoming). Early retirement window offerings and defined benefit plans do offer increased incentives to leave early (Brown 1995; Luzadis and Mitchell 1991). Trends in pensions raise questions not only about their direct effects, which do work towards encouraging earlier retirement, but also raise questions on a higher level about why defined benefit pensions continue to be changed to encourage earlier retirement.

Gustman,Mitchell, and Steinmeier S61 implicitthreatsof dismissal, companyunwillingnessto adaptto employee disabilon ity, job stress, or fellow-workerpressure to leave. Information these aspects of work is collect in the HRS. Workersand retirees are asked what they could and do earn as of the survey date as well as on their previousjob. These data are supplementedwith individual-specific earningsrecordssuppliedby the Social Pension and healthinsuranceexpectationsare collected SecurityAdministration. from the older person directly, supplementedwith outside informationcollected from the employer. Data on job demands, working conditions, and other job attributesare also included in the HRS.
5. How do retirement decisions interact with savings and consumption, as well as wealth accumulation and bequests?

A model which satisfactorilyintegratesboth savings and retirementdecisions has yet to be estimated empirically.5Retirementmodels which ignore savings may be misspecified,and conversely savings models which ignoreretirementare incomplete. We need to know more about asset accumulationas workers apabout those who accumulatevirtuallyno perproachretirement,and particularly sonal assets (Venti and Wise, forthcoming).Perhapspeople do not save because of high time preference or low after-taxinterest rates, or perhapsbecause pensions and Social Security more than meet their projected retirementneeds, or they greatlyunderestimateresource needs in retirement.Additionalexplanations include uncertaintyabout future health, combined with the availabilityof programslike Medicaidthat insurelosses only after assets are depleted. Information for examiningthese alternatehypotheses is containedin the HRS.
6. Whatfactors determine people's expectations about their future opportunities and constraints in retirement, and how accurate are these expectations?

Existing data sets do not permit a full explorationof how well older people understandwhat they will receive from Social Security and pensions when they retire, how much savings they will have had to accumulateto sustain their consumptionin retirement,and the way that theirpension and Social Securitybenefit The HRS offers a uniqueopportupaymentsare affected by additionalearnings.6 nity to compare company-providedinformationon actual health and pension plans providedto covered workers, with respondents'expectations of pensions, insurance,and Social Security benefits. The HRS also asks questions about people's anticipatedlife expectancy, healthoutlook, spousalretirement expectations, reportedplanninghorizon, etc. Eventually Medicareand mortalityinformation will be matched with HRS survey files, and these too can be compared with correspondinginformationon the questionnaire.Each of these links enables re-

5. Empirical retirement models tend to focus on labor market behavior alone, assuming implicitly that savings and consumption can be left in the background (and on rare occasions when savings and consumption are addressed in retirement models, it is generally assumed that capital markets are perfect). Conversely, empirical analyses of life-cycle consumption and savings tend to assume retirement is exogenous. Relaxing these assumptions requires gathering data on work, savings, wealth accumulation, and bequests, much of which is being undertaken in the HRS. 6. Pension and Social Security rules may not be well understood (Bernheim 1988; Mitchell 1988; Gustman and Steinmeier 1989).

S62

The Journalof HumanResources searchers more accurately to match people's stated expectations with realizations, and determinehow in turn these affect retirementoutcomes.

III. Retirement in the Health and Retirement Survey
This section outlines theoreticaland practicalaspects of the HRS which researchersshould be aware of when examiningretirementbehaviorwith this survey. In additionwe offer some initialevidence on the extent of retirement behavior observed in the first wave, and some evidence of patterns that will emerge as subsequentwaves are collected.
A. Conceptualizing Retirement

The term "retirement"has many meaningsand can be empiricallyrepresented from using a variety of labor marketmeasures.These includeworkerwithdrawal the laborforce, or the point when he or she leaves a careerjob or stops working full-time,or when the workerfiles for pension (or Social Security)benefits,among
others.7

Because many differenttypes of labor force transitionsoccur towardthe end of the work life, the HRS incorporatesmany detailed questions on labor market activity and the opportunities facing older individuals.In the firstwave, the baseline questionnairecollected in 1992,the HRS inquiresaboutrespondents'current employment status, pay and benefits, and working conditions (includinghours flexibilityand employerattitudes).Those who have pensionsandhealthinsurance coverage on their current jobs are also asked the nameof theiremployer.Pension and health plan descriptionsoffered by these firmsare being collected from the employer and the U.S. Departmentof Labor. For persons not employed at the time of the survey, questions are asked about prioremployment. Much of this labor market informationis collected for all respondents. For example, all are queriedregardingrecentjobs that lasted for five years or more, and on other jobs offering pension coverage. Moreover, all respondents who signed a release will have their Social Security earningshistory attachedto their file, permittingthe reconstructionof employmentand earningshistory in all covered employment(it will, however, not be possible to separatehoursand wages). In designingthe survey, an effortwas also madeto obtaininformation opportuon nities not taken. Specifically,respondentsare asked about recentjob search and unemployment;in additionbrief informationwas obtainedon past layoffs. The second wave, fielded in 1994, and subsequentwaves, can then be used to gauge the accuracy of retirementexpectations and whether retirementbehavior is af7. There is an additional complication that many older people pass through a partial retirement transition phase between full-time work and complete retirement; this has been variously defined as working part-time, having a low-wage job, or being employed in an occupation which is relatively undemanding and/or flexible in terms of hours requirements; others focus on changes in hours or wages, changes in occupation and industry, and working after acceptance of Social Security or pension benefits (Gustman and Steinmeier [1984] discuss many variants).

Gustman,Mitchell, and Steinmeier S63 fected by poor informationabout work and pension opportunities.Lastly, the HRS asks about people's reasons for retiringand/or changingjobs, information that has been unavailablein earlierstudies.8
B. Work and Retirement Outcomes

Table 1 shows that a relatively large fractionof first-waveHRS respondentsare not working, even though the target populationis quite young-ages 51 to 61 years old.9 The top six rows of the table define retirementas zero or few hours per year of work. By this definition, slightly over one-fifth of the men in the sampleare retired,as are abouttwo-fifthsof the women. Focusingon men, whites and Hispanicsreportequal rates of nonemployment, about one-fifth,while a third of black males report they are retired by this definition.Among women, about 40 percent of both white and black females report no currentjob, while more than half of Hispanic females are not workingfor pay. Different definitions produce different tallies of employment and retirement status. Since the HRS target population was 51 to 61 years old in 1992, the cohort is not yet eligible for Social Security retirementbenefits. Some people nevertheless reportthemselves as "retired," though the figuresare significantly lower than the objective labor force status measuresdiscussed above. Row 7 of Table 1 shows that only 15 percent of the men and 28 percent of the women The same relativerelationby race appearsfor men. consider themselves retired.10 Evidence on partialretirementpatternsalso appearsin Table 1. Among men, 4 to 6 percent are partiallyretiredusing definitionsbased on hours of work per week, weeks per year, or hours per year up to 1,200 hours, with more being classifiedas partiallyretiredwhen the 1,500-hour cutoff is used. Partialretirement rates are about double amongwomen, at 6 to 10percent, with 15 percentworking fewer than 1,500 hours; the gap is much smallerwhen the employmentmeasure
8. This information must be used with cautionto the extent that it is often unclearwhethera workeror the employerinstigatesexit from employment.In a long-termcontractsetting, where the wage profile is tilted or the pensionaccrualis such that the wage exceeds productivity near the end of the contract, an olderemployeemighttend to wantto worklongerthanwas mutually agreedon at the outset (Lazear 1979). 9. All HRS data in this paperuse the alpharelease tape, whichcontains"approximately three-quarters of the eventualHRS Wave 1 sample,and has been given only very preliminary cleaningand consistency checking.The data tape providesweights, but they are based only on the majorelementsof selection The weightsare not adjusted nonresponse for bias nor for some minorelementsof selection probability. (Healthand RetirementSurvey 1993). probability" 10. Some people in the HRS have no extendedperiodof earlierlaborforce participation, measuring so retirement exhibitingzero attachment the laborforce overstatesthe extent of transitions of the as to out laborforce. For example, in answerto the self-reported retirement statusquestion(variable4901), 910 individualsin the alphatape indicatethat the questionis not relevant,and 1,328indicatethat they are withthe number peoplewho indicatethatthey areretired,two-thirds of fullyretired.Thus, in comparison as many people indicatethat the questionis not relevantto them because the individual doesn't work for pay or is a homemaker, hasn'tworkedfor pay for ten or moreyears. Forpurposesof comparability or between the objective and self-reported who reportthat the self-reported measures,individuals retirement status question is not relevantwere counted as retired.Excludingthese individualsreduces the that percentageself-reporting they were retiredby aboutone-and-a-half percentagepoints for men and by about 14 percentagepointsfor women.

S64

The Journalof HumanResources
Table 1

Employmentand Retirementin the HRS underAlternativeDefinitions RetirementDefinition "Full retirement"status (percent) No currentjob White Black Hispanic Working<200 hours per year Working<400 hours per year Self-reportedas retired White Black
Hispanic 4"Partialretirement"status (percent)

Men

(N)

Women

(N)

21 19 33 22 21 22 15 14 26 14 4 5 4 6 6 5 5 8
15

(3,405) (2,726) (518) (161) (3,405) (3,405) (3,179) (2,534) (494) (151) (3,432) (3,418) (3,405) (3,405) (2,726) (518) (161) (3,405)
(3,405)

40 39 40 52 41 42 28 28 29 35
9

(3,818) (2,942) (699) (177) (3,818) (3,818) (3,798) (2,931) (690) (177)
(3,851)

Working<25 hours per week Working<40 weeks per year Working< 1,000 hours per year Working< 1,200 hours per year White Black
Hispanic

Working< 1,500 hours per year Left 10+ year job after age 45 Left 20 + year job after age 45 Self-reportedpartiallyretired White Black Hispanic Chances out of 10 of: Workingat age 62 Workingat age 65

8 8 8 6 5 5
3

(3,405) (3,179) (2,534) (494) (151) (2,688)
(2,680)

6 7 10 10 11 10 15 8 2 5 5 7 2 4 2

(3,827) (3,818) (3,818) (2,942) (699) (177) (3,818) (3,818) (3,818) (3,798) (2,931) (690) (177) (2,277) (2,269)

HRS responNote: Table percentagesare calculatedusing survey weightsand includeage-eligible dents (aged 51-61 in 1992)fromthe HRS alphareleaseof May 1993.Numbersin parentheses indicate
unweighted sample size is used to compute the reported fraction.

is weeks per year. It is possible that one should differentiatepartial retirees between those who always worked part-time,and those who previously held full-timejobs. Fifteen percent of men and 8 percent of women reporthavingleft a long-termjob after age 45, whereby "long term" is meant a job held for 10+ years. If instead the cutoff is having left a job of 20 + years after age 45, the figuresare 8 percent of men and 2 percentof women. Using selfcorresponding reports,8 percentof men and 5 percentof womendescribethemselvesas partially

Gustman,Mitchell, and Steinmeier S65 retired. White men are more likely to be measuredas, and to reportthemselves as, partiallyretired than are black or Hispanic males, while black women are more likely to be partiallyretiredthan are white females; Hispanicwomen report a low probabilityof partialretirement. A final set of measures in Table 1 describe HRS respondents' expectations with regardto work at future ages. Men anticipatethat the odds of working at age 62 slightlyexceed one-half,while women reportabouta 40 percentprobability of working. Forecastingto age 65, men reportonly a 30 percentchance of working, and women less than a 25 percent chance. Figure 1 summarizes graphicallythe patterns for full and partial retirement measures by age in the HRS baseline survey. Definingfull retirementas having no currentjob, the fraction retired increases with age, but the graph is not a smooth curve. Among 61-year-oldHRS respondents, one-thirdof the men and almost half the women have no currentjob; these are higherthan the rates at age 51, where 12 percent of the men and 36 percent of the women do not hold a job. For men the sharpestchange in retirementlevels occurs between ages 58 and 59, with other large increases at 54-55 and 60-61. For women the largest increases are at 52-53, 60-61, and 55-56. To what extent these changes are due to provisions of retirement programs awaits further investigation. Partial retirementin Figure 1 is definedas workingless than 1,200 hours per year; this rises by about five percentagepoints for men between ages 51 and 61, and less thanone percentage point for women. Otherretirementdefinitionswill also be used by researchersstudyingthe HRS, but most of these require additionalinformationto evaluate changes over time in work patterns. It is also worth notingthat retirementmay not be an absorbing 1 state, so that people may flow back and forth between work and retirement. To of this behavior at baseline and thereafter,HRS respondents are permit study asked retrospectivequestions about their last job to determinelaborforce transitions near the end of the work life.'2 In addition, other informationis to be and gatheredfrom the Social Security Administration from the employer.

IV. Elements of the Opportunity Set in the HRS
In designing the HRS, it was deemed essential to map carefully the constraints and opportunitiesolder people face, includingmoney and time constraints.This section describes findingsfrom the firstwave of the HRS about each of these factors.13
11. Reverse flows are discussed by Quinn,Burkhauser, Myers(1990)and Rust (1990). and 12. WhileHRS questionsaboutpastjobs are less completethanaboutemployees'current jobs, survey lengthprecludedthe inclusionof an entirejob history. 13. Becausethe HRS focuses on individuals families,it is not nationally and of representative employer to practices. As a result, the survey can make only a modest contribution answeringthe question of and why companiesoffer the particular compensation employment policiesthey do. Analystswho model retirement behaviorfrom the supply side should neverthelessbe awarethat workers'preferencesmay be correlatedwith companycharacteristics, the extent that employersdesigncompensation to packages to attractand keep employees with specific attributes.In particular, may be controversial HRS it for

S66

The Journal of Human Resources

55.0 -

50.0 -

45.0 -

40.0 '

35.0

t 30.0
CL|

? 20.0
15.0

10.0

5.0

an. V.V

-

I 1
50

I r- r 51- 52 53

54

I 55 56 57 Age

I I I *U 58 59 60

61 62

_ B _ _4

Females(Retired) Males (Retired)

*
-A*

*

Females(Partially Retired)

*A Males (Partially Retired)

Figure 1 Retirement and Partial Retirement By Age* *Retirement is defined as no current job. Partial Retirement is defined as less than 1200 hours per year.

Gustman,Mitchell, and Steinmeier S67
A. Labor Market Earnings and Job Opportunities

In developing retirementmodels, analysts must predict a range of wage offers for all HRS respondents, correctingobserved pay measures for selectivity bias and estimatingpotentialwages for work on a full-time"'main" on a postretirejob, ment but full-timejob, and/or on a part-time job.14Here we reportonly evidence on job earningsand opportunitiesfor those employed in 1992. Several differentpay measuresappearin Table2. One approachclassifies HRS respondentsas full-timeworkers if their usual annualhours total 1,200 or more. The first three lines of Table 2 show that full-time workers earn higher median pay than do part-timeworkers, but the differencesbetween the rates depend on the time period over which earnings are measured. On an hourly basis, men counemployedfull-timehave hourlywages 45 percenthigherthantheirpart-time terparts ($14 versus almost $10 per hour); however, on an annual basis male full-time workers earn more than three times as much ($32,000versus $10,000 per year). An even more pronounceddifferentialapplied to women: full-timers earn 32 percent more on an hourly basis, and more than three times as much on an annualbasis. We also note that pay rates for part-timeself-employedmen are relatively high.15 Table 3 shows that many workers in the HRS age-range face hours constraints.16 Among full-time male workers, 12 percent report they would like to
users to assumethat pay and benefitsare exogenousdeterminants retirement of outcomes(for a discussion of this issue in the pensionliterature,see Gustman Mitchell[1992]and Gustman, and Mitchell,and Steinmeier[1994].Studies of labordemandappearin Hamermesh [1993]). 14. Some of the often-citedinvertedU-shapedage-earnings profilesare due to changeof employersand hoursreductionsamongolder workers,accordingto a studyof the Retirement HistoryStudy(Gustman and Steinmeier1985). Earningsappearto decline less with age amongworkerswho remainwith the same employeras they grow older. 15. Table 2 covers currentlyemployedworkerswith valid annualhours; earningsfiguresare median amountsamong all individualswhose usual hours place them in one or the other of these categories. Medianfull-timehourly and weekly wages for men are based on 2,111 observations,while the correspondingpart-timewages are based on only 148 observations.Earningsin 1991 are based on 1,939 observationsfor full-timeearnings,and 123observationsfor part-time earnings.In the case of women, there are 1,699 full-timeand 331 part-timewage observationsfor hourlyand weekly wages, and 1,563 and 282 observationsfor full-timeand part-time earningsin 1991. It shouldalso be noted that differencesbetween means are smallerthan the differencesbetween the medians.Men employedfull-timeaverage81 percentmorethanpart-timers an annualbasis ($36,769 on versus $20,288),and the mean of the usual hourlywage variableis actuallyless for full-timers thanfor ($18.23 versus $27.06 per hour). The latter findingis not due solely to outliers, since the part-timers thirdquartilevalue for a part-timer's usual wage exceeds the thirdquartilevalue for the full-timewage versus $19.95for full-timers).It shouldbe noted that the part-time ($21.64for part-timers information is based on a sampleof only 148 observations,of which 49 are self-employedindividuals whose mean usualhourlywage is $30.04. Amongwomen HRS respondents,meanvalues are closer for full-timeand part-timeworkers, but once again the hourly wage for part-timeself-employedexceeds that for the full-timeself-employed($15.33per hourfor part-timers versus $11.19for full-timers). both cases it is In work. Also, there may be some possible that selectivity bias favors individualswho choose part-time individualswho reportunusuallyor temporarily levels of hours worked,raisingcalculatedhourly low wages. 16. These numberscombinethe responsesfroma questionregarding whetherthe individual reduce can or increasehours of work, with anotheron whetherthe workerwould like to changehours given that he or she cannot. We recognize that constraintson work hours are not necessarilyinefficient.One

S68

The Journal of Human Resources Table 2 Earnings in the HRS
Men Full-Time Workers (?1,200 hours per year) 14.00 615 32,000 14.50 10.68 10.00 14.10 13.46 14.00 17.23 Part-Time Workers (<1,200 hours per year) 9.62 210 10,000 10.50 6.58 9.00 9.42 11.54 9.60 * Women Full-Time Workers (?1,200 hours per year) 9.26 370 18,800 9.39 8.79 7.30 9.38 7.00 9.05 10.75 Part-Time Workers (<1,200 hours per year) 7.00 122 6,000 7.00 6.49 5.25 6.73 7.70 7.00 8.60

Median Earnings Measure Usual hourly wage ($) Usual weekly pay ($) Earnings last year ($) Usual hourly wage ($) White Black Hispanic Employees Self-employed Private sector Public administration

Note: * denotes fewer than five observations. Figures given are medians and calculated using survey weights. The sample includes only age-eligible HRS employed respondents (aged 51-61 in 1992) from the HRS alpha release of May 1993.

Table 3 Hours and Other Job Constraints in the HRS Men Full-Time Workers Part-Time Workers Women Full-Time Workers Part-Time Workers

Constraint Hours constraints Would like to work fewer hours but cannot (percent) Would like to work more hours but cannot Laid off > age 45 from 10 + year job and: Currently employed at new firm Currently self-employed

12 (1,957)

5 (104)

15 (1,662)

3 (295)

15

18

15

21

6 (2,459) 2

8 (187) 4

5 (1,862) 1

3 (375) 1

Note: Table percentages calculated using workers reporting valid annual hours and survey weights; numbers in parentheses indicate the sample size used to compute the reported fraction, including only age-eligible HRS respondents (aged 51-61 in 1992) from the HRS alpha release of May 1993. Full-time is defined as 21,200 hours/year; part-time is defined as <1,200 hours/year.

Gustman,Mitchell, and Steinmeier S69 work fewer hours than permitted to on their currentjobs. Almost 15 percent would like to increasetheir hoursof work. Increasinghoursis a goal of 15 percent of the full-time working women, while slightly fewer than 15 percent wish to reduce work hours. Part-timeemployees appearmuch less constrainedin terms of wishingto providefewer hours (5 percentof the men, 3 percentof the women), but many more would like to increase their hours of work (18 percentof the men and 21 percent of the women). Table 3 also shows that a reasonablylarge segmentof the HRS full-timework force is continuingto work after having been laid off from a job held for more thanten years. About 8 percentof full-timemen and 6 percentof full-timewomen are still working after having been laid off from a long-timejob, most of them workingfor a new employer ratherthan being self-employed.
B. Social Security Benefits and Taxes

The HRS cohort is still too young to be eligible for Social Security based on its own earnings,and most samplemembersare not now eligiblefor paymentsbased on spouse status. In future survey waves, Social Security benefits will be the focus of much attentioninasmuchas they constitutethe majorsource of income for large segments of the older population.It is anticipatedthat Social Security benefits will be computedthree differentways and the results compared.A first approachwill use first-wavequestions, where respondentsare asked what they expect to receive in Social Security payments at the point they retire. Second, when earningshistories are attachedto the files of respondentswho have granted permissionto the researchorganizationto obtaindata, benefitcomputationalgorithms can be used to predict respondents'benefits at various future retirement dates. Analogouscalculationswill also indicatethe retirementincentives created by Social Security regulationsincludingthe earningstest, benefit recomputation rules, and the delayedretirementcredit(whichmay differacross samplemembers dependingon year of birth).Futuresurvey waves will also reportbenefitsactually received by retirees. Available informationregardingSocial Security pertains mainly to expected coverage and benefits. As seen in Table 4, fewer than 1 percent currentlyreceive Social Security benefits from disability or other programs.The table includes only currently employed workers, but almost all of the employed workers (92 percent)expect to receive Social Security benefits. Benefit expectancy does not vary much by maritalstatus, though there is a gap across ethnic groups:whites are 6 percent more likely and blacks are about 3 percentmore likely thanHispanics to anticipatereceiving benefits. Other differences observed are less notable by firm size, union status, and pension status, though manufacturing employees are five percentagepoints more likely to expect benefits.
reason is that they may reflect the terms of an implicit contract which supplies backloaded compensation despite productivity which flattens or even falls with age. In this event workers will want to supply too much labor late in life, and a mechanism must be found for terminating the contract (Lazear 1979). Other reasons that hours may be inflexible are fixed costs of employment and requirements for coordination in team production.

S70

The Journalof HumanResources A related question is what happens when people misunderstand Social the Security benefit structure. This is a concern promptedby prior surveys which concludedthat older people tend to stop workingwhen theirpay rises to the point that their Social Security benefits are subjectto an earningstest; the anomalyis that otherfeaturesof the benefitformulaoffset the earningstest, makingit worthIf while for most people to continue in the labor market.17 future HRS waves reveal a similar spike in the frequency distributionof earnings at the income for disregard the earningstest, this will suggestthat analysts shouldrethinkolder people's understandingof the benefit computationprocess. A related issue is future Social how to model workers' evaluationsof the uncertaintysurrounding benefits and taxes. These future streams should have attachedto them Security people's valuations of their riskiness, and the HRS can help make headway in measuringhow these streams may vary in the future. The middle columns of Table 4 display the HRS respondents' expectations with regardto future Social Security benefit changes. Few believe that benefits are likely to rise on average, with odds of only 2.5 out of 10, while people offer much higher odds than benefits will be cut, about 6 out of 10. Indeed HRS membersare as pessimistic about the prospects of a majordepressionand high inflationas they are about cuts in Social Securitybenefits. While the means are fairly similaracross most groupsshown in Table 4, both Hispanicsand blacks are more optimistic than others about their benefit prospects under Social Security.
C. Employer-ProvidedPensions and Health Insurance

To date, nationally representativeretirementsurveys have not supplied highThis is an quality data on company-providedpensions and health insurance.18 importantomission inasmuch as benefits are believed to influence retirement patterns profoundly, because they comprise a majorportion of older workers' wealth, and because the benefit rules impartlarge discontinuitiesto older workers' budget constraints. The HRS seeks to remedy this data deficit by linkingemployers'descriptions of their pension and health care plans to each individual'ssurvey record. Survey respondentswere asked to identify their employers, and the Institutefor Social Research(ISR) is collecting benefitplan reportsfrom various sources for subseformat. Additionally,a computer softquent conversion to computer-readable ware programis being written at the University of Michiganto compute participants' pension eligibility ages and expected benefits, which will streamlinethe
17. Studies on this problem are reviewed by Hurd (1990). On the other hand, B3ernheim(1988) concluded about the RHS that ". . . people seem to be reasonably competent at forming relatively accurate expectations conditional on the information that they do choose to use. In addition, it is somewhat comforting to note that few individuals exhibit the kind of extreme optimism that might be responsible for catastrophic error in financial planning; indeed, there is a general bias toward conservatism" (p. 314). 18. The Survey of Consumer Finances (SCF) and the National Longitudinal Study of Mature Women (NLS-MW) are the only nationally available surveys which provide matched employer pension data. The number of retirees in the SCF is relatively small. Firm-side pension plan details on the NLS-MW were just coded in mid-1993 and retirement analysis with the data set has not yet utilized the employerprovided plan descriptions.

Gustman,Mitchell, and Steinmeier S71
Table 4 Social Security and Other Expectations in the HRS
Anticipated Odds Out of 10 Over the Next Decade Expect to Receive Social Security Benefits (percent) 92 93 92 90 92 90 86 91 92 92 92 96 91 93 91 Now Gets Social Security Benefits (percent) 1 0 1 3 1 1 0 1 1 1 1 0 1 0 1 Social Security Benefits Will Increase 2 2 2 2 2 4 3 2 2 2 3 2 3 2 3 Social Security Benefits Will Decrease 6 6 6 6 6 5 5 6 6 6 6 6 6 6 6

Respondent Group All By sex Men Married women Unmarried women By race Whites Blacks Hispanics By union status Union Nonunion By firm size Large firm Small firm By industry Manufacturing Nonmanufacturing By pension status Has pension No pension

Major Depression 5 5 6 6 5 6 5 6 5 5 5 5 5 5 6

Inflation >10 Percent 6 6 7 6 6 6 6 6 6 6 6 6 6 6 6

Note: Large firms have 100 employees or more. Table results are calculated for currently employed workers using survey weights including only age-eligible HRS respondents (aged 51-61 in 1992) from the HRS alpha release of May 1993.

process of estimatingpension wealth. As of this writingthe employer-sidelink is not yet available, so the discussion here describes only what HRS respondents state they expect to receive, ratherthan what their employersplan on providing after retirement. HRS respondentsare asked whetherthey are covered by a privatepension and if so what type of plan they have. Employed respondents'responses appearin Table 5 (excludingthe self-employed).Two-thirdsreport having pension coverage. Consistent with earlier surveys, the data show that women and nonwhites are less likely to have a pension than are men and whites. People most likely to have a pension are union members,employees of large firms, and manufacturing employees.19
19. The percent unionized in the HRS is about 26 percent, a finding virtually identical to the Current Population Survey figure for 45-64-year-olds (Curme, Hirsch, and Macpherson 1990).

Table 5 Pension Coverage and Pension Plan Type in the HRS

RespondentGroup All employees By sex Men Marriedwomen women Unmarried By race
Whites Blacks Hispanics By union status Union Nonunion

Workers with Pension (percent) 67 72 62 59
68 62 50 89 58 81 45

Covered Workerswith Pension Plan Type( Only DB 42 41 45 42
41 55 44 59 33 43 41

Only DC16 14 17 18
16 14 20 9 19 12 26

Only 401(k) 12 11 12 13
12 11 12 6 15 11 17

Both DB & DC Only 12 14 9 12
12 11 7 14 11 14 6

Both DB & 401(k) Only 11 12 10 9
12 5 11 10 12 12 3

B DB & O

By firm size
Large firm Small firm

By industry Manufacturing
Nonmanufacturing

77
63

37
44

11
17

14
11

12
12

15
10

Note: DB = Definedbenefitpension;DC = Definedcontribution pension. Largefirmshave 100employeesor more. Table fig HRS respondents(aged 51-61 in 1992)from the HRS alph workersusing survey weightsfor age-eligible but not self-employed

Gustman,Mitchell, and Steinmeier S73 Policymakersare currentlyquite interestedin the types of pensions that workers have among those who have a plan, and the HRS offers informationon this matter. Approximately42 percent of the HRS pension-coveredsample reports havinga definedbenefitpension alone, while another25 percentindicateshaving a defined benefit plan paired with another type of plan. Defined contribution plans, particularly401(k) plans, have grown quickly over the last decade. This trend is reflected in the HRS with more than one-quarterof all covered respondents having a 401(k) plan (either alone or in combinationwith other plans). Fewer than 3 percent of covered workers cannot classify their plan type, a far smaller proportionthan in previous surveys. In the 1983 Survey of Consumer Finances, for example, 19 percent of respondentscould not identify their plan type (Mitchell 1988). Retirementincentives in pensions depend on a numberof plan characteristics includingthe "normal" retirementage, or the age at which retirees are eligible to receive "full" or unreducedbenefits. This age has been decliningand is now quite low, as is evident from Table 6. HRS pension-coveredworkersreportthat their pension plans allow retirementwith unreducedbenefits at a normalretirement age averaging61, or a median age of 62. Most pension plans also permit early retirement,though usually with reduced benefits. In the HRS, workerswith definedbenefitpensions face a mean (and a median) early retirementage of 58, with a range from 57 to 60 for various subgroups. Respondentswith defined benefit plans who know how much their early retirement benefits are reduced report that their pension reductionfactor is about 5 percent per year. A reductionfactor of this magnitudeusually implies that early retirees receive subsidized benefits.20 Eventuallythe HRS will permita comparisonof these datawith plan characteristics and pension accrualprofilescomputedfrom employer-supplied descriptions of the pension. Withthe computersoftwareit is possibleto calculateeach covered worker'sexpected "pension wealth profile,"takingas inputthe worker'spension plan description and assumptions regardingexpected future earnings, inflation rates, anticipatedSocial Securitybenefits, and longevity information.In addition to providinga straightcalculationof the expected benefitat alternativeretirement dates, the programcan also be used to help answer "what if" questions (such as how the pension might change if the Social Security offset changed, or if the pension contributionor benefit formulachanged).21 health insurance benefits should also be included among Employer-provided the factors influencingworker mobility and retirementbehavior. While costs of
20. Benefit accrual varies among plans with formulas of different types and depends on such factors as adjustments in the benefit formula in future years, the extent of postretirement adjustments in benefits, and other factors that vary among plans. For a discussion see Mitchell (1992) and Gustman and Steinmeier (1989). 21. Calculating pension wealth, and the changes in pension wealth if retirement is deferred, requires that the analyst know each worker's expected retirement age and how benefits are likely to change if retirement is deferred. In addition, spousal benefits must be taken into account, as must temporary early retirement windows, postretirement cost-of-living benefit adjustments, and potential disability pensions. The HRS asks each pension-covered person for such plan details, which can be compared with information available in the employer-supplied pension Summary Plan Description.

S74

The Journalof HumanResources
Table 6 Pension Plan Features in the HRS

RespondentGroup Employees covered by definedbenefit pension By sex Men Marriedwomen Unmarriedwomen By race Whites Blacks Hispanics By union status Union Nonunion By firm size Large firm Small firm By industry Manufacturing Nonmanufacturing

Normal Retirement Age (mean) 61

Early Retirement Age (mean) 58

Early Retirement Reduction (percent/year) 5.3

61 61 62 61 61 62 61 61 61 61 61 61

58 59 59 58 58 59 58 58 58 58 58 58

5.2 5.0 6.6 5.0 7.2 7.0 5.4 5.2 5.3 4.7 5.9 5.0

Note: Largefirmshave 100employeesor more. Tablefigurescalculatedusing surveyweightsincluding only age-eligible(aged 51-61 in 1992)HRS workerswith definedbenefitpensionsin the HRS medianfor the normalretirealpharelease of May 1993.Figuresreportedare means;the (unreported) ment age is 62 for all groups;the medianearly retirement variesbetween57 and 60 amongthe age groups.

health insuranceare not availablein the HRS, there are data employer-provided on currentcoverage as well as health care benefits anticipatedafter retirement. Table 7 shows that health insurancecoverage is widespreadin the HRS working cohort: 86 percent have some coverage, and 80 percent of those enjoy coverage healthcoverageis higherfor men throughtheirown employer.Company-supplied and unmarriedwomen, though many marriedwomen receive insurancethrough their spouses. This genderdifferenceprobablyexplainswhy coveragerates from own employmentare higherfor blacks and Hispanicsthan for whites overall. In general, employees are more likely to be covered if they are unionizedand have pensions (coverage rates are 96 percent or higher), and work in large firms or firms(92 percent). manufacturing

Gustman,Mitchell, and Steinmeier S75
Table 7

Current and Expected Retiree Health Insurance Coverage for Employed Persons in the HRS
Anticipate Retiree Health Insurance Current Health Insurance Among Covered Workers, Fraction with Own Coverage 80 90 56 96 79 82 85 88 76 84 72 89 77 85 63 Fraction among Those Currently Covered by Health Insurance from Own Employment Don't Know 16 14 19 20 15 20 35 12 19 14 23 17 16 14 27

Respondent Characteristics

Fraction of Employed with Coverage 86

Yes 69 73 62 62 69 66 55 80 63 73 52 70 68 74 50

No 15 13 19 18 15 13 10 8 19 12 25 13 16 13 23

Fraction among Those Currently Covered Via Spouse's Job 68 58 72 100 69 70 32 75 66 69 68 60 69 72 63

By sex Men Married women Unmarried women By race Whites Blacks Hispanics By union status Union Nonunion By firm size Large firm Small firm By industry Manufacturing Nonmanufacturing By pension status With pension Without pension

89 87 79 88 80 76 96 83 92 80 92 83 97 66

Note: Largefirmshave 100employees or more. Figuresare percentagesof relevantsamplecalculated using survey weights includingage-eligibleHRS respondents(aged51-61 in 1992)fromthe HRS alpha release of May 1993.In the last columnemployeeswho do not know if they are coveredby healthinsurancewhile workingare includedin the base and accountfor an averageof 30 percentof the cell as (ranging high as 60 percent).

Of those who have health insurancefrom their employers while actively employed, more than 69 percentexpect continuedretireehealthinsurancecoverage, with rates even higherfor union employees (80 percent)and employees of large companies (73 percent). Only half of Hispanics and workersin small firmshope to receive retiree health care coverage. Interestingly,only 15 percent of those with currenthealth coverage do not expect retiree coverage, but a largergroup,

S76

The Journalof HumanResources 16percent, does not know what retireecoverageis offered. Some may be entitled to continuedcoveragethroughtheirspouse's continuedemployment,as is evident in the last column of Table 7. While a comparisonof the HRS with other data bases is beyond the scope of this study, it is useful to ask whether the HRS pension and health insurance coverage data appear broadly consistent with coverage informationfrom other the surveys. Two data sets lend themselves to a naturalcomparison: 1991Survey of Income and ProgramParticipation(SIPP), and the 1988 CurrentPopulation Survey (CPS). These two surveys include people in broaderage rangesand pose the coverage somewhat differentlythan in the HRS, so the figureswould not be expected to be identical. Nevertheless, all three appear to tell a similar story. Unpublishedtabulations from the SIPP indicate that 75 percent of men aged 50-59 who are wage and salary workers are covered by a pension, as are 64 figuresfor the slightlyolder HRS sample are 72 percent of women. Comparable and 62 percent. CPS full-time men and women employees aged 50-59 percent report health coverage rates from their own employmentof 78 percent and 62 percent, respectively, in 1988,versus comparableHRS coveragerates of 80 percent and 58 percent for men and women, respectively.
D. Nonwage Aspects of the Job

The HRS asks respondentsmany questions about nonwageaspects of theirjobs, includingphysical and mentaljob requirements,workerattitudestowardthe job and its constraints, and future prospects for continued employmentas well as alternativeprospects. Answers to these questions should permitresearchersto derive variablesuseful to measuringHRS participants' preferencestowardwork andleisure. Wherephysicaldemandsof jobs are involved, it will also be natural to interactthese with individuals'health status in the models explainingretirement. Table 8 describesjob attributesfor HRS membersworkingfull-timeat the time of the survey. A majorityof respondentsreport that their jobs require skill in dealing with others much of the time (57 percent of men and 70 percent of women). Around90 percent of HRS men and women reporttheir work environof ments to be friendly most or all of the time. Three-quarters men and women freedomto decide how they do their work much of the time. Most reporthaving men (81 percent) and women (73 percent) believe they are paid fairly, though, surprisingly,fewer than half of the men (43 percent) and women (35 percent) believe that their pay depends on their job performance.Fewer than a fifth of men and women believe that younger people are given preference over older people. Amongboth men and women, 80-90 percentreject the idea that employers or fellow workersexert pressureto retire. Finally, abouta thirdof older men and women work on jobs where they believe they can partiallyretire, namely where they would be allowed to move to a less demanding with less pay.22 job In general, most of these employees seem to like work: less than a third would
22. For further discussions of this issue see Hurd and McGarry (1995).

Gustman,Mitchell, and Steinmeier S77 retire if they lost their jobs, and more than two-thirdsstate that "Even if I did not need the money, I would probablykeep on working." Respondentsare also asked whether theirjobs requireparticular physical and mental requirements.Two-thirds of the HRS men and women say their jobs include requirephysicaleffort at least some of the time, wherethese requirements stooping, repetitive work, and keeping a fast pace on the job. More than half of the women but fewer men reportneedinggood eyesight and intense concentration almostall of the time. This may be because morewomen than men use computers at work (28 percent versus 14 percent). Almost half of the HRS workers agree that they could performbetter with more training;more than half state that their job is becoming more difficultover time; and two-thirdsreportsubstantialstress on theirjobs.
E. Wealth Measures

The HRS promisesgreatlyimprovedmeasurement financialstatusas compared of to priorsurveys. Results on economic status in the HRS are discussed elsewhere by Moon, Juster, and Abrams (1995), and wealth measures are discussed by Smith(1995).Eventuallyall the differentexplanationsfor asset accumulation and decumulationshould be integratedwith those for retirement,savings, consumption, and bequests.
F. Other Factors Relevant to RetirementAnalysis

The HRS incorporatesa richerand morereliableset of indicatorsof healthstatus, family structure, and disability plan participationthan have ever before been availablein previous surveys of retiremnent-age people. Knowledge of these will improve social scientists' ability to measure older workers' opportunityset and should facilitate estimation of key behavioralparametersin retirementmodels. Given the centralimportanceof health status in determining retirementbehavior, careful estimationof health and disabilitystatus will improveour understanding of their effects on retirementand should reduce bias in retirementmodels which mightotherwise result from imprecise measurementof the impactof poor health on retirementdecisions. The HRS permitsimprovedmodelingof retirement decisions in a familysetting. Earliersurveys did not provideinformation both spouses' healthand disability on status, a shortcomingwhich makes it difficultto determinehow poor health of the husbandinfluences the wife's work and retirementbehaviorand vice versa. Moreover, the HRS collects data on each spouse's earnings, pension, Social Security, and employment opportunitiesindependently,offering better quality data than heretoforeavailable.23 The HRS also offers good information other on
23. This paper does not summarize retirement patterns or elements of the opportunity set for HRS respondents falling outside the age range of 51-61, who generally appear in the data file because they are spouses of age-eligible sample members. Considered by themselves, these individuals are not representative of their age group in the population. However, data on these people will be of immense importance to analyses of family retirement behavior, since these individuals are representative of spouses of a population falling within the age range.

Table 8 Job Requirements, Job Characteristics, and Attitudes toward Work of Full-Time HRS Workers

Men Almost All or All the Time None or Almost None of the Time Almost All or All the Time

Job Requirements Physical demands Physical effort Heavy lifting Stooping Good eyesight Otherdemands Concentration Dealingwith people Computers Analyze information Keep up pace Repetitivework Learn new things Freedomto decide Friendlywork environment

Most of the Time

Some of the Time

Mo the

21
10 15 46 47 57

20 10 15 41 38 27 9 20 24 27 27 37 45

30 32 37 9 13 13 21 22 20 30 38 16 11

30 49 34 4 2 3 55 33 31 13 9 9 1

21 8 11 61 52 70 28 26 38 43 29 35 49

1

1 3

14 24
25

30
25

38 42

3 2 1 1 2 2 2 3 4

Job Characteristics and WorkerAttitudes Job characteristics Need training More difficult Need good memory Involves stress Attitudestowardwork Retire if lost job Don't work for money Wantjoint retirement Employee attitudes
Pay is fair

Strongly Agree
10

Agree 38 44 65 45 20 54 45 67 33 15 15 32

Disagree 40 38 6 32 50 24 37 16 48 67 68 55

Strongly Disagree 12 6 1 3 22 9 6 4
10

Strongly Agree 11 13 32 23 8 13 13 13 6 4 2 2

Ag

12 29
19

4

4

8 14 12 14
10

4

Workinfluencespay Boss likes youth Pressureto retire Can partiallyretire

4 3 2

13
15 10

Note: Figuresgiven are fractionsof relevantsample. First panel includesself-employed workers;second panel excludes them survey weightsincluding only age-eligibleHRS respondents(aged 51-61 in 1992)from the HRS alpharelease of May 1993.

S80

The Journalof HumanResources family members besides spouses, recognizingthat older workers' labor supply decisions respondto family needs and resourcesas a whole. This information on family status can be used to assess older peoples' reactionsto changes in household structureincludingdivorce, migration,and death of family members.

V. Data Matching and Estimation in the HRS
As of this writing(June 1995),the respondentpartof the HRS has just been released as a public use file, but none of the proposeddata matching efforts have yet been completed. The overall plan of the HRS calls for merging of the respondentdata with data from the Social Security Administration, from health pension plan descriptions,and from employer-supplied employer-supplied insurance information.Work is under way on all three of these matchingdata sets, and currentestimates are that they should be availableby late summeror fall of 1995. Duringthe 1992 HRS survey, interviewersattemptedto gain permissionfrom the respondents to access their Social Security files. Of the full set of 12,654 individuals who completed the survey, 9,498 signed the required permission forms. Of those who grantedpermission,9,089 gave valid Social Security numreceived 8,416 of the forms within bers, and the Social Security Administration the 60-day period requiredin order to have the data released. In the 1994wave of the survey, another attempt was made to secure permissionto access the Social Security records, both from individualswho refused permission in the initialsurvey and fromindividualswho gave permissionbut for whomthe attempt to gain the records failed. This resulted in another 1,157forms being signed;the percentage of these forms that will yield actual Social Security records is unknown at this time. Due to the sensitive natureof these records, Social Securitywould only agree to release summaryinformationon a public use data set. This informationincludes average earnings, annual benefit amounts, estimates of the discounted value of the benefits, earningsby 10-yearperiod, and the numberof years in each 10-yearperiod when earningswere above a trivialamount.The complete Social Securityrecords will also be available,but only as a restricteduse data set. Both the public use data set and the restricted use data set are to have only broad geographicidentifiers, at a level equivalent to Census divisions. For those researchers who require state-level geographicidentifiers,there will be a second restricteduse data set that has only the summarySocial Security information. For the 5,713 individualswho indicatedthey were covered by a pension, an attempt was made to gather summaryplan descriptionsfrom their employers. of The pension files maintainedby the U.S. Department Laborwere searchedas a backup. This strategy resulted in plan descriptionsbeing gatheredfor 3,834 individuals,or about 67 percent of those who indicatedthat they were covered. These plan descriptionshave been coded in a formatsimilarto that used for the Survey of ConsumerFinances and for the MatureWomencohort of the National to Survey. Effortsto develop a personalcomputerprogram evaluate Longitudinal these pensions under various sets of assumptionsare within a month or two of

Gustman,Mitchell, and Steinmeier S81 completion.Plans also call for potentialannualbenefitamountsand the projected total value of the pension to be placed on the mainpublic use file, similarto the values calculatedfor Social Security. The third matchingdata set contains health insuranceplan descriptionsgathered from telephone interviews with the insurance providers. Of the approximately 8,900 individualswho indicated that they were covered by health insurance in the 1992survey, the insuranceproviderwas interviewedin approximately 6,400 cases, or about 72 percent of those who were covered. A 10-15 minute questionnairecovered, among other things, the type of plan (fee for service, HMO, etc.), the contributionrates, the nature and limits of the coverage, and the availabilityof retirementcoverage. At present, the data are in the cleaning stage, again with a projectedrelease in the late summeror fall of 1995.

VI. Conclusion
The Health and RetirementStudy offers researchersample scope to explore current practices and to answer outstandingquestions about retirement. The survey also affords new informationwith which to evaluate current programsand improve policy design for the future. This is because the survey contains better measures than have ever before been availableof older people's and behavopportunities constraints,as well as insightsinto healthandretirement ior for the generationon the verge of retirement.Criticallyimportantquestions can be addressed with the survey because of its richness of detail and linkages with Social Security records, companyhealth and pension data, and (eventually) Medicare and Vital Statistics records. Thus, for instance, the HRS will permit researchersto study how retirementresponds to changes in income supportprograms, the Social Security System, pension regulationsand trends, and requirements affectinghealth insurance, spouse equity, disabilitypolicy, and others.

References
Anderson,PatriciaM., Alan L. Gustman,and ThomasL. Steinmeier.1993. "The Trendto EarlierRetirementamong Males." Final Reportto the Social Security GrantNumber 10-P-98343-1-01. Administration. Bernheim,B. Douglas. 1988. "Social Security Benefits:An EmpiricalStudy of Expectationsand Realizations."In Issues in Contemporary Retirement,eds. R. RicardoCampbelland E. Lazear, 312-45. Stanford:Hoover Institution. Brown, Charles. 1995. "Early RetirementWindows:Windowsof Opportunity? Defenestrations?(and even refenestrations.. .)." Unpublished. Burtless, Gary. 1986. "Social Security, Unanticipated Benefit Increases, and the Timing of Retirement."Review of Economic Studies 53(5):781-805. Campbell,Colin D., and RosemaryG. Campbell.1976. "Conflicting Views on the Effects of Old Age SurvivorsInsuranceon Retirement."EconomicInquiry14(3):369-88. Centerfor HumanResource Research. 1988.NLS BibliographyUpdate. Columbus, Ohio: The Ohio State University.

S82

The Journal of Human Resources Curme,MichaelA., BarryT. Hirsch, and David A. Macpherson.1990."Union and ContractCoveragein the United States, 1983-1988."Industrialand Membership
Labor Relations Review 44(1):5-34.

Gustman,Alan L., and Olivia S. Mitchell. 1992. "Pensionsand the LaborMarket:
Behavior and Data Requirements." In Pensions and the U.S. Economy: The Need for

Good Data, eds. Zvi Bodie and Alicia Munnell,39-87. Philadelphia: PensionResearch Council. Gustman,Alan L., Olivia S. Mitchell, and ThomasL. Steinmeier.1994. "The Role of
Pensions in the Labor Market." Industrial and Labor Relations Review 47(3):417-38.

Gustman,Alan L., and Thomas L. Steinmeier.1984. "PartialRetirementand the
Analysis of Retirement Behavior." Industrial and Labor Relations Review

37(3):403-15. ---. 1985. "The Effects of PartialRetirementon Wage Profilesof OlderWorkers."
Industrial Relations 24(2):257-65.

---. ---. ---.

RetirementModel." Econometrica54(3):555-84. 1986. "A Structural 1989. "An Analysis of Pension Benefit Formulas,Pension Wealth,and 1992. "The StampedetowardsDefinedContribution PensionPlans: Fact or

Incentives from Pensions." Research in Labor Economics 10:33-106. Fiction?" Industrial Relations 31(2):361-69.

Hamermesh,Daniel. 1993.LaborDemand. Princeton:PrincetonUniversityPress. of Health and RetirementSurvey. 1993. "Early Release Tape: Memorandum about Conditionsof Use." May 28, 1993. Understanding Hurd, MichaelD. 1990. "Researchon the Elderly:Economic Status, Retirement,and
Consumption and Saving." Journal of Economic Literature 28(2):565-637.

between Job Hurd, Michael, and KathleenMcGarry.1995. "The Relationship
Characteristics and Retirement." Journal of Human Resources 30

(Supplement):S260-S267. Ippolito, RichardA. 1990. "TowardExplainingEarlierRetirementafter 1970."Industrial
and Labor Relations Review 43(5):556-69.

----

Pension . Forthcoming."TowardExplainingthe Growthof DefinedContribution

Plans." Industrial Relations. Lazear, Edward P. 1979. "Why Is There Mandatory Retirement?" Journal of Political

Economy87(6):1261-84. Lumsdaine,Robin, James Stock, and David A. Wise. 1994. "PensionPlan Provisions and Retirement:Men & Women, Medicare,and Models." In Studies in the Economics of Aging, ed. David A. Wise, 183-220. Chicago:Universityof ChicagoPress. Luzadis, Rebecca A., and Olivia S. Mitchell. 1991. "ExplainingPension Dynamics."
Journal of Human Resources 26(4):679-703.

Manton,KennethG., LarryS. Corder,and Eric Stallard.1993. "Estimatesof Changein ChronicDisabilityand InstitutionalIncidenceand PrevalenceRates in the U.S. Elderly Care Survey." Journal Populationfrom the 1982, 1984, and 1989NationalLong-Term
of Gerontology 48(4):S153-S166.

Michael, RobertT. 1985. "Consequencesof the Rise in Female LaborForce Rates: Questionsand Probes." Journalof LaborEconomics(3, part Participation 2):S117-S146. Mitchell, Olivia S. 1988. "WorkerKnowledgeof Pension Provisions."Journalof Labor
Economics 6(1):28-39.

. "Trendsin Pension Benefit Formulasand RetirementProvisions." In Trendsin D.C.: U.S. Pensions 1992, eds. J. Turnerand D. Beller, 177-216. Washington, of Department Labor, PWBA. Mitchell, Olivia S., and Gary S. Fields. 1982. "The Effects of Pensionsand Earningson
Retirement: A Review Essay." Research in Labor Economics 5:115-56.

Gustman, Mitchell, and Steinmeier Moon, Marilyn,and F. ThomasJuster. 1995. "Economic Status Variablesand the
HRS." Journal of Human Resources 30 (Supplement):S138-S157.

S83

Peracchi,Franco, and Finis Welch. 1992. "Labor Force Transitionsof OlderWorkers." New York University WorkingPaper92-36. Quinn,Joseph F., RichardV. Burkhauser,and Daniel A. Myers. 1990.Passing the
Torch: The Influence of Economic Incentives on Work and Retirement. Kalamazoo:

W. E. UpjohnInstitute. Rust, John. 1990. "Behaviorof Male Workersat the End of the Life Cycle: An
Empirical Analysis of States and Controls." In Issues in the Economics of Aging, ed.

David A. Wise, 317-79. Chicago:Universityof ChicagoPress. FrankJ. 1987. "The Effects of Health on Retirement."Social Security Sammartino,
Bulletin 50:31-47.

Smith, James P. 1995. "Racialand Ethnic Differencesin WealthUsing the HRS."
Journal of Human Resources 30 (Supplement):S158-S183.

Stock, James H., and David A. Wise. 1990a."The Pension Inducementto Retire:An
Option Value Analysis." In Issues in the Economics of Aging, ed. David A. Wise,

205-24. Chicago:Universityof ChicagoPress. . 1990b."Pensions, the OptionValue of Work, and Retirement."Econometrica
58(5): 1151-80.

U.S. Departmentof Health and HumanServices, Social SecurityAdministration, Office
of Policy, Office of Research and Statistics. 1987. Bibliography of Studies Based on the

RetirementHistory Study. Washington,D.C.: SSA Publicationno. 13-11759. Venti, Steven F., and David A. Wise. Forthcoming."The Wealthof Cohorts,the ChangingAssets of OlderAmericans,EmployerPensions, and PersonalRetirement
Saving." In Public Policy Toward Pensions, ed. S. Schieber and J. Shoven.

Washington,D.C.: TwentiethCenturyFund.