Micro Determinants of Labor Force Status Among Older Americans
Transcript of Micro Determinants of Labor Force Status Among Older Americans
Micro Determinantsof LaborForceStatusAmongOlderAmericans†
HugoBenıtez-Silva‡
SUNY at Stony Brook
First version:May 1998Thisversion:September30,2000
Abstract
Thispaperusesthefirst threewavesof theHealthandRetirementSurvey (HRS)to investigatethedeter-minantsof laborforcestatusamongolderAmericans.Usingtransitionsat two-yearintervalswefind thatafterbeingretiredor unemployed,thosewhoareactively searchingfor a job haveahigherprobabilityofreturningto work. We alsofind thatbeingin goodphysicalandmentalhealth—measuredby objectiveandsubjective variables—increasesthe chancesof becomingemployed,asdoeshaving worked in thelast twelve months.Thosewho arereceiving disability paymentsarelesslikely to make this transition.If we focuson thosewho aremarried,we find a preferencefor joint leisurethroughthe influenceofthe labor force status,healthandageof the respondent’s partneron the transitiondecisions. We in-vestigatetransitionsin andout of employmentandself-employment,andfor subsamplesof malesandfemales.Usingmonthlyemploymentdummiesfor theperiod1989-97,we analyzemonthly, quarterly,semi-annualandannualtransitionsandfind thatmostof our conclusionsareindependentof theperiod-icity but thattheeffectsof thevariablesvaryacrossspecifications.
Keywords: LaborSupply, LaborForceTransitions,RetirementDecision,HealthandRetirementSurvey.
JEL classification: J14,J2
† I am grateful to JohnRust,MosheBuchinsky, Hiu Man ChanandSofiaCheidvasserfor their help, encouragementand
comments,andalsofor allowing meto usethedatasetwe areworkingwith. I would like to thankDonaldW.K. Andrews,Michael
Boozer, Agar Brugiavini, Kevin Foster, Eric French,Ann Huff-Stevens,Jenny Hunt, JohnKnowles, David Margolis, Massimo
Massa,Paul Mishkin, Olivia S. Mitchell, Giorgio Pauletto,T. Paul Schultz,Chris Sims, Anna Vellve-Torras, the participants
of the Yale SummerMicroeconomicsSeminar, the Joint GraduateStudentSeminarat the University of Pennsylvania, the Yale
MicroeconomicsWorkshopon LaborandPopulation,theNorth AmericanMeetingsof theEconometricSocietyin Madison,the
NBERAging SummerInstitute,theLatin AmericanMeetingsof theEconometricSocietyin Cancun,andtheESEM/EEAmeetings
in Santiagofor insightful commentsand discussions.I am also grateful for the financial supportof the “la Caixa Fellowship
Program,” theCowlesFoundationfor Researchin Economicsthrougha Carl Arvid AndersonDissertationFellowship,andto the
JohnPerryMiller Fundfor additionalfinancialsupport.All remainingerrorsaremy own.‡ EconomicsDepartment,StateUniversityof New York at Stony Brook. Stony Brook,New York 11794–4384,phone:(631)
632-7551,fax: (631)632-7516,e-mail:[email protected]
1 Intr oduction
This paperpresentsan empiricalanalysisof the determinantsof labor force statusin the HealthandRe-
tirementStudy(HRS).The HRS is a large, longitudinalsurvey that follows a cohort,ages51 to 61 asof
1992/93,andtheir spouses,from pre-retirementinto retirement.It is speciallydesignedto track the labor
market behavior of olderAmericansandtheirhealthstatus.
Laborforcetransitionshave beentheobjectof aconsiderablenumberof studies,andthey will continue
to beasbetterdatasetsbecomeavailableandinterestincreasesin theevolutionof thelabormarketbehavior
of olderworkers,in largepartdueto its policy implicationsin thecurrentdebateonthefutureandthereform
of theAmericanPensionandSocialSecuritySystems.
A significantnumberof studieshave usedthe RetirementandHistory Survey (RHS) asthe basisfor
their empiricalwork. TheRHSis a longitudinalsurvey of menandsinglewomenages58 to 63 asof 1969
thatfollowedrespondentsupto 1979.Rust(1990),Blau(1994)andRustandPhelan(1997)focusonmales,
Blau (1997and1998)studiesmarriedcouplesandPozzebonandMitchell (1989)restricttheir attentionto
marriedwomen.PeracchiandWelch(1994)usetheCurrentPopulationSurvey (CPS)to studylabor force
transitionsof malesandfemalesfrom 1968to 1991,andHurd (1990a)usesthe New BeneficiarySurvey
(NBS) to analyzethejoint retirementdecisionof husbandsandwives.
Theavailability of theHRShasfostereda new wave of empiricalwork in this area,respondingto the
needto updatethesourcesof datain a time of changein policy andattitudetowardsthe importanceof the
labor market behavior of older Americans.Also, the fact thata majority of HRS respondentsarefemales
hasallowed someresearchersto incorporatethe analysisof labor market behavior of older womeninto
their research.Hurd andMcGarry (1993),Gustman,Mitchell, andSteinmeier(1995),Blau andGilleskie
(1997),Blau,Gilleskie,andSlusher(1997),Friedberg (1997),Karoly andRogowski (1997),Rogowski and
Karoly (1997),Quinn(1998),Coile (1999),Costa(1999),andCoileandGruber(1999)areexamplesof this
research.Quinn (1998)is the only studythat usesthe threeavailablewavesof the HRS aswe do in this
paper, but concentratesonly on thoseemployedasof thefirst wave of interviews. BlauandRiphahn(1998)
usetheGermanSocioeconomicPanelto studymarriedcouplesin Germany, andit is theonly otherstudy
alongwith thisonethatmakesuseof monthlyemploymentindicators.1
Thispapercontributesto theliteraturein severalways:
� This is oneof thefirst papersto useall threewavesof theHRSto analyzethelatestevidenceontrends
1 Hurd (1990b),Lumsdaine(1995),LumsdaineandMitchell (1998),andCurrieandMadrian(1998)areexcellentsurveys oftheliteratureon retirementbehavior andrelatedissues.
1
in retirementandlaborsupplyamongtheAmericanelderly.
� This is thefirst panelstudy, usingAmericandata,to trackemploymentdown to themonthlylevel.
� Theuseof awiderangeof health,demographicandsocio-economicvariablesprovidesamoredetailed
andupdatedpictureof themicro determinantsof laborforcetransitions.
� This papertakesa closerlook at thosemakinga reversetransitiondecision,to exploremorein depth
thanprevious literaturetheir characteristicsandlabor forcedecisions.TheEconomicsof Aging has
focusedmuchmoreontheretirementdecisionbut changesin demographicandsocio-economictrends
(higherlongevity, improvementsin healthoutcomes,technologicalprogressthatcanchangethenature
of work, etc.) justify that we take a closerlook at the importantvariablesbehindthe decisionsof
comingbackto work aftera job displacementspellor retirement.
� This is oneof the first papersthat analyzesmarriedpeopleseparatelyusingthe HRS, allowing for
someof thespouse’s variablesto have aneffecton therespondents’decisions.
Someaggregatevariablesseemto show that in the last two or threeyears,as the economyhasgrown
steadilyandunemploymentrateshave gonedown without triggeringinflation, olderAmericansout of the
labormarket arefinding it easierto returnif they sochoose.2
Figures1 and2 show theevolution of the labor forceparticipationrateamongmalesandfemalesages
55 to 64, from January1988to August2000basedon monthly datafrom the Bureauof Labor Statistics
(BLS). For bothgroups,wecanseeanincreasein this ratein thelast3 or 4 yearsthatis sharperfor women,
confirmingahistoricaltrend,but it is alsopresentfor men,implying areversalin atrendthatstarteddecades
ago.This hasto beconsideredalongwith decreasingunemploymentratesamongthesamepopulation.We
canconcludethat peoplein theseagerangewho want to work, have a positionalmostassured,given the
currenteconomicconditions.3 For those65andover theparticipationratesaremuchlowerandmorestable
in thelast10years,but ashasbeennoticedrecentlyin themedia,thereis someevidencebasedonBLS data
of a changein thetrendin thelastyearsfor thisagegroupaswell.4
2 Theliteraturehasalreadyreferredto thepossibilityof this trend.Hurd (1990b,p.631)considerstheoverall rationalitybehindapossibleincreasein laborforceparticipationof theelderly: “I canwell imaginethatasthelaborpoolof youngerworkersshrinks,employment opportunitiesfor older workerswill expand,and that many will increasetheir work lives in responseto the betterjobs.” Schulz(1995,p.282)arguesthat: “Without dramaticchangesin the ability of countriesto moderatebusinesscyclesandkeepunemployment low over the long run, theremay be little changein currentprovisionsthat encourageretirementandearlyretirement.” Ruhm(1996,p.100)arguesthat: “...changesin theSocialSecuritySystemwill encouragelongerwork livesasbenefitsbecomelessgenerousfor futureretirees,theageof “normal” retirementis raised,andthepenaltyfor earlyretirementis increased.”
3 Ruhm(1996)mentionsthe possibility that official unemployment statisticsunderstatethe labor market problemsof olderworkers,whoaremorelikely to bediscouragedanddropoutof theworkforce.
4 “The New Facesof Retirement”.TheNew York Times,January3, 1999.
2
How canwe explain this changein trend?Onepossibleexplanationis that theSocialSecurityamend-
mentsof 1983andothermorerecentefforts to make work moreappealingat theendof the life cycle are
having animpacton behavior amongtheelderly. This canalsohappenasanindirectresultof thechangein
attitudetowardswork asAmericansreachtheretirementagein betterhealthandwith longerlife expectan-
cies,andasfirms seetheir workforcesagingasthebabyboomersgrow older. An alternative would be to
arguethatthis is atemporaryphenomenondueto thetight labormarketandtheeconomicexpansionthathas
forcedfirms to resortto lesstraditionalsourcesof workers,includingtheelderly.5 Wecancomplementthis
lastexplanationwith thefact thatthecurrenttechnologicalrevolution canhelpmake work moreaccessible
to older workers. Self-employment, part-timework usingthe internet,and telecommutingare just some
examplesof the radicalchangesin theconceptof employmentthatwe will seein the future,changesthat
couldfavor theincreasein employmentamongolderAmericans.
Thispapercannotfully addresstheseissues,but with themin mind,weareableto shedlight onwhether
themicro determinantsof labor forcestatusamongtheAmericanelderlyhave changedin the lastdecade.
The changein trendobserved in Figure1 justifiesour interestin learningmoreaboutwhat affects labor
market decisionsnow thatwe have accessto very rich dataon this population.Thereducedform natureof
ouranalysismakesit anexploratoryeffort contributing to theunderstandingof laborforcetransitionsof this
populationwith a richerandupdateddataset.
Our empirical resultsshow that amongthe non-employed, thosewho have healthinsuranceare less
likely to returnto work, revealinga “spike” at age62 anda 65+ effect, especiallyamongmen.6 Among
the employed, thosewith accessto retireehealthinsurancearemore likely to leave the workforce,with
theageeffectspeakingat age63-64.7 Theseresultsareconsistentwith Blau andGilleskie(1997)who use
thesamedatasetbut restrictattentionto malerespondentsin thefirst two wavesof thesurvey, andusea
somewhat controversial characterizationof the transitionperiods;andwith Karoly andRogowski (1997)
thatconcentrateon menandonly on theretirementdecision.Theresultscanalsobereconciledwith those
of RustandPhelan(1997), limited by the fact that they build a structuralmodel incorporatingthe Social
Securityincentives,anapproachbeyondthescopeof this study.
Our resultson theeffectsof healthstatusarefairly standard;thatis, thosewith healthproblemsareless
5 TheNationalCouncilon Aging launchedduring1998the“100,000jobsCampaign”thataimsat placingthatmany workersage50 or olderduringthenext five years.Many majorAmericanemployersareparticipantsin thecampaign.This canhelpmakethis increasein demanda lesstemporaryone.
6 We have groupedherethoserespondentsage65 with thoseover 65 becauseof the small numberof respondentsin eithergroup. We caninterpretthis 65+ effect asreflectingthe fact that individualsstartreceiving S.S.benefitsandbecomeeligible forpublichealthinsurancethroughMedicareafterreachingage65.
7 Theseeffectsprobablyincludetheinfluenceof theCOBRA legislationthatrequiresfirms of a certainsizeto provide contin-uationof healthbenefitsfor amaximumperiodof 18monthsonceemployeesleave thefirm.
3
likely to returnto work andmorelikely to leave work. We usea wide varietyof objective andsubjectives
measures,measuresof activities of daily living (ADLs), instrumentalactivities of daily living (IADLs), and
cognitionvariables.Theeffect of marriageis alsoconsistentwith the resultsof Blau (1994)andPeracchi
andWelch(1994),showing significantdifferencesbetweenmalesandfemales.Marriedmenaremorelikely
to returnto work andlesslikely to leave theworkforce,while theoppositeis truefor marriedwomen.We
find thateducationvariableshave a smalleffect andarerarelysignificantin the transitiondecisions.This
differsfrom theresultsof PeracchiandWelch(1994),but is consistentwith thesmalleffectsfoundby Blau
(1994). This is not surprisingsincewe arecontrolling for a largerarrayof variablesin which educationis
left with almostno explanatorypower.
An importantandoriginal result is the relevanceof the job searchbehavior of the respondentin the
re-entrydecision.Thosewho weresearchingfor a job in themonthprior to theinterview weremuchmore
likely to besuccessful.This resultcontrastswith Rust(1990),who did not find a significanteffect of the
searchingdecisionontheprobabilityof findinga job afterwards.Wewill show thatthis resultis notmerely
spuriousandtesttheexogeneityof this variablein oursampleof non-employedrespondents.
We alsoanalyzemarriedpeopleseparatelyandfind that the influenceof thespouse’s laborstatus,age
andhealthontherespondents’decisionscanbeinterpretedeitherasacomplementarityof timeeffect,or for
thosenon-employed,asa preferencefor joint leisure.This resultis consistentwith PozzebonandMitchell
(1989),Hurd (1990a),Blau (1998),andBlauandRiphahn(1998).
An importantcontribution of thispaperis theestimationof shorttransitionsusingmonthlyemployment
dummies. We calculatemonthly, quarterly, semi-annualand annualtransitionsin addition to the more
commonsurvey to survey transitions.Thisgivesusasenseof someof thedynamiceffectsof theexplanatory
variablesandcanbe interpretedasa testof our resultsusingthe traditionalbiennial transitions.Most of
thesignificantvariableshave strongereffectsfor the shortertransitions,implying that theestimatesusing
longertimeperiodscouldrepresenta lower boundfor thetrueeffectof somevariables.
In thenext sectionwe presentthedatausedin this studyanddiscussits limitationsandpotentialprob-
lems,aswell asits strengthsfor thepurposesof our research.In sectionthreewe presentandanalyzethe
empiricalresults.Sectionfour concludesandofferssomesuggestionsfor furtherresearch.
2 Data and Summary Statistics
Thedatausedin this paperis from theHealthandRetirementStudy(HRS).TheHRSis anationallyrepre-
sentative longitudinalsurvey of 7,700householdsheadedby anindividual ages51 to 61 asof wave 1. The
4
primarypurposeof theHRSis to studythelaborforcetransitionsbetweenwork andretirementwith partic-
ularemphasisonsourcesof retirementincomeandhealthcareneeds.It is asurvey conductedby theSurvey
ResearchCenter(SRC)at theUniversityof Michiganandfundedby theNationalInstituteonAging.8 Up to
now dataof thefirst four wavesof thesurvey areavailable,andin thispaperwewill considerthefirst three.
Wework with 12,652respondentsfrom wave 1 of thedatacollectedin 1992and1993,11,596respondents
from wave2 collectedmainly in 1994,and10,970respondentsfrom wave3, collectedduring1996andJan-
uaryof 1997.Thelasttwo waveswereconductedby phoneusingthecomputerassistedtechnology(CATI)
whichallows for muchbettercontrolof theskip patternsandreducesrecallerrors.
Deathandattrition have reducedthenumberof participantsin thesurvey. This cancreatesomeselec-
tivity issuesthatwe assumeaway in ourestimations.Additional individuals,who have beenincluded,have
enteredthesurvey lateronmainlyasspousesof previousrespondents.Wewill restrictourattentionto those
age50 andover giventheobjectivesof this paper. Thevariablesusedin this studyaretheresultof several
programsthatcalculatethehoursof work, earnings,employmenthistory, wealth,income,healthstatusand
severalotherrelevantvariablesfor therespondentof thethreewaves.Thehardesttaskhasbeento construct
theemploymenthistoriesof therespondents.Fortunately, thesurvey wasdesignedwith this in mind,but the
complexity of theskip patternandthemultiplicity of differentemploymenthistoriesover the threewaves
hasmadeour work a time consumingone. In theemploymenthistoriesof therespondentswe definethree
states:
� Employed: works for someoneelseandreceivesa salaryor wage. In general,this category hasthe
highestproportionof respondents.
� Self-employed: worksfor him/herselfreceiving asalary, profitsor evennothing.Theseparateanalysis
of theserespondentsis fairly new in theliterature.We have decidedto differentiatethemfrom those
employed for a third partybecausethosewho areself-employed have fairly differentcharacteristics
from thoseworking for a company, as shown below, and we conjecturedthat this could result in
obscuringsomeof theestimationresults.9
� Non-employed: doesnot do any work for pay. We includein this category peoplewho areout of the
laborforce,in thesensethatthey arenotnecessarilylooking for a job.10
8 SeeJusterandSuzman(1995),alsoGustman,Mitchell andSteinmeier(1994and1995)or theHRSwebpage.9 Wewill seein thenext sectionthatsomevariableshavedifferenteffectsfor employedandself-employedindividuals,justifying
thesplittingof thesample.Also theempiricaltransitionprobabilitiesfrom oneemploymentstateto anothershow differentpatternsfor employeesandself-employedindividuals.
10 This labor forcestatusis comparableto theoneusedin the literature. However, PeracchiandWelch (1994)includein this
5
We have flaggedthe respondentsappropriatelyto know their trajectoriesup to wave 3 or their last survey
interview. Although we usethe threewaves of data,we only have two decisionmaking periods. Our
datawill assumethat the respondentsmake labor force decisionsat eachinterview, given their statusand
characteristicsat that time. Theresultsof thesedecisionscanthenbemeasuredin thenext interview. We
canidentify thisapproximatelytwo-yearperiodasaproblem,sincearespondent’s statusin thenext wave is
somehow aresultof hisdecisionsandcharacteristicstwo yearsago,but it alsodependsonmany otherevents
thathaveoccurredin themeantime.To avoid thisproblemwehavealsorunmonthly, quarterly, semi-annual
andannualtransitionsfrom thedateof theinterview. Wewill presenttheresultsandwe will comparethem
with thesurvey to survey transitions.
Thedatafor therespondentsweremergedfromwave3backwardsto waves2and1,andweconstructeda
setof consistentvariablesondifferentsourcesof income,wealth,healthandsocio-economiccharacteristics
that were assignedto eachdecisionmaker appropriately. We work with a sampleof more than 20,000
person/periodobservations.We assumethateachindividual makesat mostonedecisionin wave 1 andone
in wave 2. Therefore,eachrespondentcanhave a maximumof two decisions,dependingon whetherwe
have beenableto follow him or herthroughthesubsequentwaves.
For ourestimationpurposeswe have brokendown this sampleinto sub-samples.For example,thesub-
sampleof thosenon-employed in any of the two waveshas7,117observations,that of the self-employed
has2,393observationsandthatof theemployed10,259observations.We dividedour sub-samplesfurther,
constructingsmallerdatasetsfor malesandfemalesfor eachsub-sample.This allows us to analyzesepa-
ratelymalesandfemalesto uncover relevantdifferencesandsimilarities.It is interestingto emphasizethat
femalesoutnumberedmalesin thesamplesof non-employedandemployed,andmenweremorenumerous
amongthe self-employed. We alsocreatesub-samplesof marriedrespondentsto analyzethat population
separately. Table1, panelsA to C offer meansandstandarddeviationsof mostof thevariablesusedfor the
threemainsub-samples.
Thesampleof non-employedcomparedwith thewholesamplehasrespondentswith lower educational
level, lower family andpersonalincomelevelsin theyearbeforetheinterview, a higherproportionof them
receivesHealthInsurancecoveragefrom the government,anda lower proportionfrom pastemployersor
their’sspousesemployers.They arein worsehealthby any measure,andthey haveworkedmuchlessduring
theyearprior to the interview. If we concentrateon the reversetransitioners(not shown) we seethat they
areyounger, they have higher incomelevels, and fewer of themreceive disability paymentsor consider
categoryonly thoseoutof thelaborforce,andincludethoseunemployedin thefull-time andpart-timecategorydependingontheirsearchingdecision.Thisallows themto introducea variablethatcontrolsfor unemploymentstatus.
6
themselvesretired. A muchhigherproportionof thosethat returnto work wassearchingfor a job in the
monthbeforetheinterview, anda higherproportionleft their last job becausethey quit or werefired. They
arealsoin betterhealththantheaveragenon-employed.
The sub-sampleof self-employed is white, male,married,andmoreeducatedthanthe whole sample.
They have higherincomeandwealthlevels. They worked morein theyearbeforethe interview andthey
arein betterhealth. Among the sub-sampleof employed, it is worth concentratingon the transitionersto
non-employment(not shown) who tendto beolder, in worsehealth,with lower incomelevelsanda lower
proportionof themexpectto work afterreachingage62.
3 Estimation Results
In this sectionwe first presentempiricaltransitionprobabilitiesto comparelaborforcestatusof therespon-
dentsin differentperiods.Wethenreportthemultinomiallogit estimatesof thesurvey to survey transitions
out of the different labor force statusand also the binomial logit resultsof the shortertransitions. We
will presentonly our preferredspecificationsandonly summarytablesof theshortertransitions,but other
specificationsandthebinomiallogits areavailableuponrequest.
3.1 Empirical Transition Probabilities
The transitionprobability for movementfrom stateA to stateB is the numberof respondentsthatmoved
from thefirst stateto thesecondbetweenthetwo periodsof analysisdividedby thetotal numberin stateA
in thefirst period.Hereweareonly consideringthoserespondentsthatappearin bothwaves.In oursample
we canlook at transitionprobabilitiesfrom wave 1 to wave 2 andfrom wave 2 to wave 3. In Table2 we
show thesematricesfor transitionsbetweensurvey waves.
FromTable2 thefirst importantissueto consideris thatcomingbackto work is fairly commonamong
this populationthereis, regardlessof thesampleandperiod,morethan11%probability that if a personis
non-employed in thepreviouswave he/shewill have a job in thenext wave,with a muchhigherprobability
for transitionsinto employmentthaninto self-employment.
PanelsA andB show theempiricaltransitionprobabilitiesfor menandwomen,respectively, from wave
1 to wave 2. PanelsC andD show thesamefor transitionsbetweenwave 2 andwave 3. Thecomparison
acrossperiodsrevealsvery similar transitionsprobabilities,with a slightly higherprobabilityof becoming
employedafterthesecondwave for men.
ComparingpanelsA and B we seethat the probability for self-employed womenof becomingnon-
7
employed is clearly larger thanfor menwho have a muchhigherprobabilityof keepworking at their busi-
nesses.The situationreverseswhenit comesto leave employment: womenhave a higherprobability of
keepingtheir jobsanda lower probabilityof becomingself-employed. Again, this differencepatternjusti-
fieshaving splittedthesamplebetweenemployeesandself-employedrespondents.Theprobabilitiesoutof
non-employmentarequitesimilar, with slightly higherprobabilityfor womenof stayingoutof work.
ComparingpanelsC andD wecanseeagainthatfemaleshadamuchhigherprobabilityof leaving work
if they wereself-employed—the probability increases12 percentagepoints,a very significantdifference.
On theotherhand,in thecaseof transitionsout of employment,womenhave transitionprobabilitiesvery
similar to thoseof men. Thehigherpersistenceof non-employmentamongwomenis maintainedin these
panels.
Anotherinterestingresultthat is truebothfor menandwomenis that it is morelikely to becomeself-
employed if you werenon-employed in theprevious interview thanif you wereandemployee,suggesting
that being out of work might foster, other things equaland comparatively to being and employee, that
individualsdecideto setup their own business.
3.2 Transitions from non-employment to self-employment and employment
We presentherethe estimationresultsfor the non-employed, trying to uncover the importantvariables
affectingthedecisionto returnto work eitherasemployeeor asself-employed. This is thespecificationthat
we will considermorein detailasbeingtheonelessstudiedin theliterature.
Table3 presentstheregressioncoefficientsof theMultinomial logit modelusingMaximumLikelihood
for the transitionsfrom non-employment to self-employment andemployment, and their standarderrors
adjustingfor heterocedasticity(White’s SE).Thebasecaseis remainingnon-employed. We alsoincludeat
thebottomof eachcolumntheaveragelog likelihoodandthenumberof observationsfor thatspecification.
Thistablehassix setsof resultssincewereportthreedifferentspecifications,andthecoefficientsfor thetwo
transitionchoicesareincluded.Thefirst sectionof thetablerefersto thetransitionsfrom non-employment
to self-employment,andthesecondrefersto thetransitionsto employment.
We include many independentvariables,up to 34, obtaininga good fit. A high proportionof them
aresignificant,themajority at the5% level or better. We have includedagerangedummiesto capturethe
differentageeffects(includingdummiesfor ages62and65+to capturetheearlyretirement“spike” thathave
beenstudiedin theliteratureandtheS.S.andMedicareeffectfor those65yearsof ageor older),andtheones
presentedarewith respectto thedummyfor thoseunder55. Wehavealsoincludedinteractiontermsbetween
the agedummiesand a dummy that takes the value 1 if respondentshave health insurancefrom a past
8
employer or their spouse’s employer. Herewe canseethatbeing62 or 65 andinsuredclearlydecreasesthe
probabilityof makingareversetransition,especiallyinto employment.Wealsoincludeadummythattakes
the value1 if the respondentshave governmenthealthinsurancecoverage,this decreasesthe probability
of making a transition,especiallyinto self-employment, as a matterof fact it is not significant for the
transitionsto employment,supportingour ideathat thedeterminantsof the transitionsto self-employment
andemploymentcanbedifferent.
Therearerelatively few respondentsin the62 to 65 agerangeamongthereversetransitioners,around
20%of thespellsbackto work comefrom theserespondents,but consideringthataround16%of oursample
is in thatagerangeindicatessomeunconditionalpropensityamongthis group. Most of themhave access
to retireehealthinsuranceaswell. Basedon analyzingthe characteristicsof the transitioners(not on the
estimationresults)thosethat returnat 65 or over tendto be marriedandmalein a lower proportionwith
respectto thosethatstaynon-employed in thesameagerange,but in generalareolder thantheir spouses,
13 yearsolder versus11 for the non-transitioners,andreceive retireehealthinsurance(especiallyhealth
insurancepaidby theemployer) andgovernmenthealthinsurancein a lower proportion.Thosethat return
to work at age62 tendto be male,marriedandwhite in a higherproportionthannon-transitionersat the
sameage.They alsotendto beolderthantheir spousesby around2 years,comparewith around9 months
for thosethat stay non-employed. The biggestdifferencebetweentransitionersand non-transitionersof
age62 is that thosethatchangelabor forcestatusreceive healthinsurancefrom thegovernmentin a much
lower proportion.Only around5% of transitionersreceive this kind of insuranceversusabout30%of the
non-transitioners.
The analysisof the effectsof healthinsuranceon labor force decisionsof the elderly is an important
issue,andour resultsareconsistentwith thoseof Blau andGilleskie(1997)who alsousetheHRSbut only
thefirst twowavesof thesurvey andconcentrateonly onmalesage51to 61asof wave1. Ourresultsarealso
consistentwith RustandPhelan(1997)who identify a groupof “health insuranceconstrained”individuals
that needto wait until they have accessto governmenthealthinsuranceto retire. Thoseindividualsseem
alsolesslikely to returnto work.11
A dummyvariablethat is 1 if the decisionis madein wave 2 is negatively relatedto the decisionof
changinglaborforcestatus,meaningthatdecidingin thesecondwave,1994/95,impliesa lowerprobability
of returningto work, proxyingfor somebusinesscycle effect. However, this dummyis only significantfor
thetransitionsto self-employmentandonly for women.
11 For a review of the literatureon theeffectsof healthinsuranceon thelabormarket outcomesof olderpeopleseeCurrieandMadrian(1998)andthediscussionin Lumsdaine(1995).
9
On the other hand, the effect of marriageis consistentwith Blau (1994) and Peracchiand Welch
(1994).12 Married menaremorelikely to returnto work asemployeesbut marriedwomenhave a lower
probabilityof makingthistransition.Thismightbereflectingthetraditionalroleof womenin thehousehold
amongthispopulation.Thiseffect doesnotappearin thetransitionsto self-employment.
Anothervery importantvariablein this estimation,andin fact acrossall specifications,is Searchj, a
dummythattakesthevalue1 for thosewhoweresearchingfor a job duringthemonthbeforetheinterview.
Thecoefficient is very largeandsignificantfor thetransitionsto employment,meaningthat theprobability
of returningto work increasessharplyif the individual is actuallysearchingfor a job. In the transitions
to self-employment the variableis lessimportant,andonly significantfor males. This is plausibleif we
believe thatestablishingone’s own businessis a decisionthat dependslesson one’s willingnessto search
for a job. This factalsoreinforcesour decisionto split thesampleof working respondentsinto employees
andself-employed individuals.
This resultmight seemobvious, after all if a personis not searchingfor a job the chancesof finding
onearevery slim. However, we have to considerthat this variablecanbemakingthedistinctionbetween
unemployedandoutof thelaborforceindividuals,andalsofor thebusinesscycleeffect, in thesensethatif
apersonis searchingandtheeconomyis doingrelatively well, thechancesof actuallyfindinga job increase
sharply. Unemploymentratesareat recordlows amongthis population,which meansthat almostanyone
whowantsto work findsa job. Thiscontrastsin aninterestingwaywith thedevelopmentsin Europewhere
giventhecurrenteconomicconditions,weconjecturethatasearchvariableof thetypeusedherewouldhave
amuchsmallereffect.
As canbeseenfrom thetable,thisvariableis veryrobustacrosssamplesandspecificationsandtherefore
it is a very goodpredictorof the decisionsto return to work, even after controlling for a wide rangeof
variables. We interpretour finding of the importanceof the searchvariableamongthis populationas a
confirmationof the main result in Flinn andHeckman(1983), that is, that unemploymentandout of the
labor force arebehaviorally distinct labor force states. They obtain this result using a small dataset of
youngmen. We find that thedistinctionbetweenthesetwo statesis alsomeaningfulfor older individuals,
in the sensethat the searchdecision(the variablethat we chooseto ultimately distinguishbetweenthese
two states)hasanlargeeffect on theprobabilityof finding a job. This resultis somewhatdifferentfrom the
resultsreportedby Rust(1990)usingtheDynamicProgrammingmodel,wheretheintentionto work hada
verysmallex-postprobabilityof becomingtrue.TheseresultswereobtainedusingRHSdataonly for males
12 However, theseauthorsonly considertransitionsinto part-timeor full-time jobs, not making the distinctionbetweenself-employmentandemployment,this therefore,qualifiesourcomparisons.Also, Blau (1994)concentratesonly onmen.
10
andlessdirectsearchvariables,whichmaybesubjectto highermeasurementerror.13
However, we have to acknowledgethe possibleendogeneityof Searchj. The respondent’s searching
decisioncanbecorrelatedwith unobservablesthatmakethatpersonmorelikely to find ajob, resultingin an
upwardbiasof ourestimatedueto thisspuriouscorrelation.Wearguethatthisbiasis nottoolargegiventhat
wearealreadycontrollingfor acomprehensive arrayof characteristicsof therespondents.Furthermore,we
testtheexogeneityof thesearchingdecision,following Heckman(1978),by usinga bivariateprobit model
of the decisionto changelabor force statusand the decisionto searchfor a job. In this set up, testing
for exogeneityreducesto testingthe independenceof the probit equations.The LagrangeMultiplier test
statisticis 0.01699,andsinceit follows a χ2 distribution with one degreeof freedom,we cannotreject
the null hypothesisof independenceof the probit equationsor equivalently the exogeneityof our search
decision.14 This importantresultdeservesfurtherattentionanda morepreciselook at thedeterminantsof
thesearchingactivity. This is partof ourongoingresearch.15
Receiving disability paymentsdecreasesthe probability of makinga reversetransition. This result is
only significantfor femalesandreflectsthefact that thosereceiving disability have majorhealthproblems
andalsothefactthatthey canonly work duringa trial period,otherwisethey losetheir benefits.
Personalincomeis positively and significantly relatedwith the decisionof returningto work as an
employee.Thiscouldseemapuzzle,but it mightmeanthatthosepeoplewhowereearningmorefrom their
previous jobshave morepressureto find a job, sincetheir familieshave grown accustomedto their living
standards.Rust(1989)considersincome(andwealth)asaproxy for unobservedjob skills whichmaymake
the worker moreemployable. This could alsomeanthat individualsworking continuouslybeforeexiting
work, andthereforeearningmorethanthosedisplacedfor longerperiods,have higherchancesof finding a
job if they needone. That is why we introducethevariableMonthp, that reflectstheproportionof months
worked in the twelve monthsprior to the interview. This variableis significantfor both the transitions
into self-employmentandthetransitionsinto employment,andthecoefficient is higherfor thosebecoming
employees. This is consistentacrossall specificationsfor the respondentsbecomingemployed, andalso
consistentwith theliteratureon thepersistenteffectsof job displacement.
Receiving pensionsisnegatively relatedwith becomingself-employedoremployed,somethingwecould
expect. Net wealth is positively relatedto the decisionof becomingself-employed, but the coefficient
arenot significant. However, it is negatively relatedandsignificantfor the decisionof employment,and
13 Rustusestwo variables:theself-reportedplannedhoursof work in theyearfollowing thesurvey andtheactualhoursworkedin theyearfollowing thesurvey.
14 AppendixA explainsin detail theconstructionof thetest,showing alsotheprobit results.15 SeeBenıtez-Silva (2000).
11
robustacrossmostspecifications.16 This is consistentwith Blau (1994)who foundthathigherassetswere
associatedwith adecreasedre-entryrate.Also consistentwith Blau(1994)wefind thatwhitesarelesslikely
to make reversetransitions.
Healthvariablesarealsoquiteimportantandin generalnegatively relatedwith thedecisionto returnto
work. This agreeswith theliterature,Rust(1990),Blau (1994),RustandPhelan(1997),Blau et al. (1997)
andQuinn(1998)find comparableeffects.17 Themostimportanthealthvariablesfor thedecisionto become
self-employedarethedummiesfor having ahealthlimitation thatpreventswork,having highbloodpressure
andpsychologicalproblems.18 For thedecisionto becomeemployed,having difficulty readinga maphasa
negative effect, andhaving a healthlimitation thatpreventswork doesaswell. We alsoincludea cognition
variable,constructedfrom the resultsof a memorytest. This variableis a numberbetween0 and20 and
reflectsthenumberof wordsthatrespondentscouldrememberout of a list readto them20 minutesearlier.
Thisvariableis positive andsignificantfor thetransitionto employment.
Anotherinterestingspecificationis shown in Table4, which usesa sub-sampleof marriedrespondents,
includinga dummyvariablethat is 1 if therespondent’s spousewasnon-employedat thetime of theinter-
view. We alsoincludethe differencebetweenthe spouses’ages,a variablethat wasfound to be relevant
amongmarriedwomenin PozzebonandMitchell (1989),andthehealthstatusof thespouse,avariableused
in theliterature.19
Thesevariablescouldpotentiallycapturesomeissuesof preferencefor joint leisureor thecomplemen-
tarity of time spentin andout of the labor market. The fact that the respondent’s spouseor partnerwas
non-employedsignificantlydecreases(moreclearlyfor returningasanemployee)theprobabilityof return-
ing to work. This agreeswith BlauandRiphahn(1998)andBlau (1998).
If a respondent’s spousehasa healthlimitation that limits work, he or sheis morelikely to returnto
work asself-employed, which is only significantfor womenandit is not significantfor the transitioninto
16 This is aninterestingresultgiventhatthis revealsaninterestingeffect of animportantvariablelike currentnetwealthon thetypeof employmentanindividual takes.Giving usa bettersensewhy is it worthsplitting thewaywe have thesampleof worker.
17 Theeffect of healthvariableson labormarket decisionshasbeen,andstill is, at thecenterof acontroversialdebateontheuseof self-reportedandobjectivevariablesin behavioral models.Thepossibleendogeneityof theregressorscoupledwith measurementerrorcould leadto substantialbiasesin thecoefficientsof interest.Most of thedebatehasconcentratedon theeffect of healthontheretirementdecision,but theextensionof theargumentsto thedecisionof returningto work is straightforward. We have testedtheexogeneityof theself-reportedhealthmeasureswith respectto thedecisionof returningto work with thesamemodelwe testedtheexogeneityhypothesisfor thesearchdecisionandagainfoundnoevidenceof endogeneityof any of thesemeasures.Theeffectof healthvariablesis importantbut doesnotseemto beoverstateddueto asupposetendency of therespondentsto rationalizetheirproblemsin thelabormarket usingtheirhealthcondition.
18 Thepsychologicalproblemdummyhasthecorrectsign in mostspecificationsin this andin theotherestimationsusingtheothertwo sub-samples.Theseresultssupportprevious literaturemorefocusedon thespecificeffectsof thesekindsof problems.SeeCurrieandMadrian(1998,p.11)for a survey.
19 Thedifferencebetweenthespouses’agesis calculateddifferentlyhere.PozzebonandMitchell subtractthewife’s agefromthehusband’s. Wesubtractthespouse’s or partner’s agefrom therespondent’s, regardlessof thesex.
12
employment.This is in generalagreementwith Blau andRiphahn(1998),consideringthatthey controlfor
theinteractionbetweenthehealthandlaborstatusof thespouse.
The differenceof agesis positively and significantly relatedwith the transitionto employment, and
giventhedifferentcharacterizationof thisvariable,it is consistentwith theresultsin PozzebonandMitchell
(1989). They find, usinga sub-sampleof marriedwomen,thatwhenthewife is youngerthanthehusband
shetendsto retireearlier, againmaybeproxyingfor somesortof joint laborforceparticipationdecision.In
our case,high agedifferencereflectsthattherespondentis older, andin this caseis morelikely to returnto
work.
Theseresultsareimportantandconsistentwith Hurd (1990a)andprobablyproxy for thefactthatolder
peopleseemto make joint labor forceparticipationdecisions,with a biastowardspreferringdecisionsthat
allow them to spendtime together, or that take into accountthe spouse’s characteristicsin the decision
makingprocess.We estimatethis very samemodelwith a sub-sampleof youngerrespondentstrying to
seeif this signcould reverseamongthem. We find that thevariablebecameinsignificantandthat for that
populationtheothercovariatesdid a goodenoughjob explainingthedecisions.A dynamicgamein amore
theoreticallychargedmodelthattriesto take into accountthedifferentvalueof onemoreperiodof work for
bothpartnerswouldbeneededto analyzethisquestionin full andis definitelyhigh in our researchagenda.
Thedirectcomparisonbetweenmalesandfemalesrevealsthat the importantvariablesremainsimilar,
with theclearimportanceof agedummieswheninteractedwith insurancestatus.Thesearchvariableand
having ahealthlimitation have thesamesignsmentionedabove. Theproportionof workingmonthsis more
importantfor womenandreceiving pensionshasa strongernegative effect for womenin the decisionto
returnasself-employed.
It is worth mentioningthe little significanceacrosstheboardof educationvariables.Oncewe control
for a largenumberof characteristicsthey areleft with relatively little explanatorypower. Thiscontrastswith
PeracchiandWelch(1994)who find thateducationis a goodpredictorof transitionsbut is moreagreement
with Blau (1994)whofindsvery smalleffectsof educationvariables.
3.3 Transitions from non-employment to employment: shorter transitions
Oneof the importantcontributionsof this paperis the useof monthly employment indicatorsto estimate
shortertransitions. Most of the literatureusesthe survey to survey transitions,but several authorshave
emphasizedthattheidealsituationwouldbeto have finer time periods.Blau (1994)is ableto usequarterly
data,and recentlyBlau and Gilleskie (1997) and Blau et al. (1997) useone year transitions. And as
mentionedbefore,Blau andRiphahn(1998)usemonthlyGermandata.
13
We run binomial logit modelsfor non-employed respondentswith the dependentvariablebeing the
employmentdummyvariableat the different time intervals. We run monthly, quarterly, semi-annualand
annualtransitions,andalsothe usualbiennial transition. In this casewe cannotdistinguishbetweenem-
ploymentandself-employmentandwe pool theminto onecategory of employed if they hada onein the
correspondingdummyvariable.We run all theseestimationsfor thesub-sampleof non-employedandalso
for malesandfemales.We presenta summaryof resultsin Table5, only reportinga subsetof variablesbut
thespecificationsareavailablefrom theauthoruponrequest.
Table5 presentsthreepanels.PanelA usesthefull sub-sampleof non-employed respondents50 years
or older. We report in eachpanelthe averageprobability, andthe marginal effectsasa proportionof the
averageprobabilityfor eachestimation.20
The first importantthing to notice is that the marginal effects vary dependingon the time periodof
our estimation. The effect of having a limitation that preventswork and the effect of beingmarriedare
the morestableones. The importanceof theSearchj variableon the probability of becomingemployed
seemsto be strongerfor shortertransitions,sinceit reflectssearchbehavior in a shortperiodbeforethe
interview. However, it hasa very large positive effect for the transitionto employment regardlessof the
periodicity. This is consistentwith our previousresults,andimpliesthatour coefficient usingthesurvey to
survey transitionscouldbea lower boundif we believe that therelevant transitionshappenin shortertime
periods.
Theeffectof thepercentageof monthsworkedin theyearprior to theinterview followsasimilarpattern
andis alsoconsistent,sincethepastattachmentto thelabor forceshouldhave a decreasinginfluenceonce
thepersonis outof thelabormarket for longerperiodsof time. Theeffectsof otherimportantvariables,such
asreceiving disabilitypaymentsor being62 andhealthinsured,arealsostrongerfor theshortertransitions.
PanelsB andC reporttheresultsfor malesandfemales.Theeffectsarevery similar, but it is important
to notice the strongereffect for womenof the searchvariableand the labor force attachmentvariables,
especiallyfor the shortertransitions. And it is amongthem that we seelarger differencesbetweenthe
effectsof the variablesusing the shortertransitionsand the two year results,implying that if we do not
considerthesefiner time intervalswecouldbeunderestimatingtheeffectsof thesevariablesfor women.
The conclusionfrom theseshorterestimationsis that the variableswe identified as importantin the
survey to survey transitionsremainthe relevantones.However, theeffectsvary significantly. We mustbe
20 For continuousvariablesthemarginal effectsarecomputedastheaveragederivative of theestimatedchoiceprobabilitywithrespectto the variablein question.For dummyvariablesthey arethe averageof the differencebetweenthe probability with thebinaryvariablesetto 1 andtheprobabilitywith thebinaryvariablesetto 0.
14
awareof this if we think that the relevant transitionsare the shorterones. However, it is not clearwhat
theappropriatetime interval to look at is if we weretrying to reacha sharperconclusionon theeffectsof
differentvariables. Onemight think that monthly transitionsarepotentiallyvery noisy, sincewe do not
considerthatmaybeanindividual hasaerraticlaborforcetrajectory. But again,thesamecanbesaidabout
thesurvey to survey transitionssincerespondentsareinterviewedat anarbitrarypoint in time.
It is importantto emphasizethatour resultsareconsistentregardlessof thetime periodchosen,but we
acknowledgetheneedfor adynamicperspective to analyzethelaborforcetrajectoriesof olderAmericans.
Thanksto the monthly employment indicators,we will be ableto characterizedifferentpatternsof labor
force trajectoriesover the1989-1997period,flaggingevery respondentdependingon his or herpatternof
attachmentto thelaborforce.Thiswill beanimportantcontribution to thestudyof olderworkersandtheir
retirementandlabormarket decisions.21
3.4 Transitions from self-employment to non-employment and employment
In Table6 we presentthe resultsfor the transitionsfrom self-employment to the other two states.These
areresultsrarelyreportedin theliterature,exceptasasubsetof thoseemployed.However, aswe have seen
above it seemsto bejustifiedto split thesamplein this waybecausetheeffectsof theexplanatoryvariables
canbedifferentfor thetwo typesof workers.Giventhatwe have fewer observations,thefit is not asgood
andtherearenotasmany significantvariablesacrossspecifications.
Theresultsfor thefull sampleshow thatfor transitionsfrom self-employmentto employmentonly afew
variablesaresignificant.Here,having highernetwealthhasa negative effect on switchingto employment,
which probablymeansthat,otherthingsequal,if a person’s businessis goingwell they have no incentive
to change.The effect of this variableis alsonegative for transitionsto non-employment. Having higher
wealthprobablymeansthat the businessis doing betterand thereforethe person’s chancesof becoming
non-employeddecreaseconsiderably. Wefind few significanthealthvariables.Having difficulty stoopingis
positively correlatedwith thetransitionto employmentandmight beproxyingfor activities morecommon
if thepersonrunshis or herown business.Having ahealthlimitation thatinterfereswith theperson’s work
hasthe correctsignsbut it is only significantfor the transitionsto non-employment. Cognitionvariables
alsohave thecorrectsigns:beingin bettermentalhealthmeasuredby memoryandcognitive testsdecreases
theprobabilityof becomingnon-employed.
21 Initial resultsshow that a high proportionof respondentsexhibit “normal” patterns,eithercontinuouswork or continuousnon-employmentalongwith differentpatternsof transitionsinto retirement.We alsoidentify animportantnumberof respondentswith lessconventionallaborforceparticipationpatternsandtheir analysisis partour researchagenda.
15
A very importantvariablefor the decisionto becomean employee is the proportionof monthsthat
respondents’workedduring thepreviousyear. Themorethey worked, the lesslikely they wereto switch.
This is againintuitive, if we believe that working a lot in their own businessesmeansthat they aredoing
well andwill probablynot join anothercompany. Total hoursworked in the last year is alsonegatively
relatedto exiting work, which is plausibleaswell. Having psychologicalproblemssuchasstressincreases
thechancesof leaving work, andsodoesbeingage62.
The higherthe self assessedprobabilityof working full time whenthey reach62 or 65 the lesslikely
they areto make a transition,proxying for the expectationof continuingwith their businesses.The male
indicatorhasa negative andsignificanteffect for transitionsto non-employment, as wasexpectedgiven
theempiricaltransitionprobabilities,wherewomenhada muchhighertransitionprobabilityof becoming
non-employed.
Comparingtheestimationresultsfor menandwomen,thedifferencebetweenthe two groupsin refer-
enceto the dummyfor marital statusis interesting. For women,beingmarriedis positively relatedwith
goingoutof work, andnegatively with joining acompany, while for menexactly theoppositeis true.Being
white is significantandnegative for malesexiting work, andpositive for becomingemployed, while for
womenraceis insignificant. Theexpectationsvariablecomesout significantfor menbut insignificantfor
women.
If werestrictourattentionto marriedrespondents,Table7,andintroduceadummythatis 1 if therespon-
dent’s spouseor partneris self-employedat thetime of theinterview, theprobabilityof makinga transition
outof self-employmentdecreases,aresultthatcanbeinterpretedassupportingthecomplementarityof time
hypothesis.This variable,however, is only significantfor women. We alsointroducea dummythat is 1
if theperson’s spousehasa healthlimitation that limits work. This variableis positive andsignificantfor
womenfor bothtransitions,but negative andinsignificantfor men.
3.5 Transitions from employment to non-employment and self-employment
In Table8 we presentthe resultsfor the transitionsfrom employment to the other two states.Given that
this is themostnumeroussub-sample,thenumberof significantvariablesis quitehighacrossspecifications,
especiallyfor thetransitionsto non-employment.
The effect of having accessto retireehealthinsuranceis positive, but only significantfor males,on
the probability of exiting work, a resultconsistentwith Karoly andRogowski (1997). However, by also
consideringin the analysisrespondentsolder than 61 we are able to captureinterestingeffects for this
agegroup,somethingnot presentin mostof the literature. The effectsof agepeakat 63-64,something
16
consistentwith the fact that oncerespondentsreachthat agethey will not losetheir healthcoverage(for
example)if they chooseto leave the work force given the currentCOBRA laws, andthe fact that by the
next interview they will probablyreachthenormalretirementageandthereforehave accessto Medicareor
Medicaid.22 Themaindifferencesbetweenthoseage62 at thetime of the interview thatmake a transition
and thosethat stayemployed, are that the transitionersreceive healthinsurancefrom their employersor
pastemployersin a lowerproportion,especiallyinsurancepaidby theemployer, andin ahigherproportion
receive governmenthealthinsuranceanddisability payments.In a sense,thesecharacteristicsreflect less
attachmentto their employersandhealthconditionsthatresultin changesin laborforcestatus.If we focus
on those65 and over, the transitionersare males,married,and white in a higher proportion. A higher
proportionof themreceivesgovernmenthealthinsuranceor disability payments,andin a lower proportion
have accessto paid retireehealthinsurance.They alsohave, comparatively, a lower agedifferencewith
their spousesor partners.Theserespondentscanagainbe identifiedasthe “health insuranceconstrained”
describedby RustandPhelan(1997).
Also, receiving pensions,having a healthlimitation, psychologicalproblemsor difficultiesperforming
activities of daily living have theexpectedpositive impacton thetransitionoutof work. On theotherhand,
having workedmorein thelastyearseemsto benegatively relatedwith this decisionandsodoesreporting
beingin excellentor goodphysicalhealthor mentalhealth. Theseresultsarein generalagreementwith
previousliterature.Theself-assessedprobabilityof workingfull timeby age65or overhasanegativeeffect
on this transition,proxyingfor generalexpectationsaboutthelaborforceattachmentof therespondent.
Thetransitionto self-employmentpresentslesssignificantvariablesbut somevery clearresults.Being
white,male,wealthyandwith ahealthlimitation significantlyincreasesthelikelihoodof startingabusiness,
andsodoesexpectingto beworking by age65. However, having difficulty walking multiple blocks,self-
reportingfair health,or having workeda lot in thelastyeararenegatively relatedto thisdecision.
Looking separatelyat menandwomen,the importantvariablesarethesameones.Someof theresults
sharpenwhen looking at thesesub-samples,like the strongeffect of being62 or 63-64andwith retiree
healthinsuranceon thetransitionout of work, or thenegative effect of theexpectationsto work after65 on
thesametransition.
The specificationconsideringonly marriedemployees,Table9, is againof interest. We introducea
dummythat is 1 if the respondent’s spousewasanemployeeat the time of the interview. In this case,the
respondentis lesslikely to exit the employmentstatus,consistentwith our previous resultson the com-
22 In alternative specificationswe interactedtheagedummieswith thehealthinsurancevariableandobtaineda similar patternof coefficientsthatleadto theverysameinterpretation.
17
plementarityof spouses’time. Anotherdummythat takesthe value1 if the person’s spousehasa health
limitation that limits work is positive andsignificantfor womenexiting work, but insignificantfor men.
This resultdiffersfrom thefindingsin PozzebonandMitchell (1989)andBlau (1998).They reportthatfor
women,thespouse’s healthconditionhasa negative effect on theexit rate,meaningthat respondentstend
to delayretirement.Both groupsof researchersusetheRHSwherewomenareunderrepresented,but more
importantly, they do not control for healthinsurancestatus,and in fact justify the resultasmeaningthat
couplesdependon thewife’s insuranceif thehusbandis in poorhealth,forcing thewife to stayemployed.
Oncethis is controlledfor, thesignschange.
Theexpectationsvariableis alsoveryimportantfor thissub-sample,decreasingtheprobabilityof exiting
work and increasingthe transitionto self-employment, probablycharacterizingpeoplethat feel they can
continueworking for a long timeafterthey reachage65.
3.6 Transitions from employment to non-employment: shorter transitions
Here we turn againto analyzethe shortertransitionsusing binomial logits for the decisionto leave the
employment status. In Table10 we presentthe summaryresultsalong threepanels. We report in each
tablethe averageprobability, andthe marginal effectsasa proportionof the averageprobability for each
estimation.PanelA usesthe full sub-sampleof employed respondents50 yearsor older. Our resultsare
consistentwith thesurvey to survey transitionestimates,but now someof thecoefficientsarenotsignificant
soweseesomeswingsin themarginal effectsof theimportantvariables.
We canagainseethat mostvariableshave a strongereffect for the shortertransitions,except for the
dummythat interactsbeing63-64andhaving accessto retireehealthinsurance,this effect is higherfor the
annualandtwo yeartransitions,resultdrivenby men. This couldreflectthefact thatoncetherespondents
reachage63-64(62 for our femalesub-sample)they retireat differenttime intervals,but by age65,a high
proportionof thosethatwantedto retirewill have doneso.Wemustemphasizethestabilityof themarginal
effectof theexpectationsvariable:theself-assessedprobabilityof workingfull timeby age65 is negatively
andsignificantly relatedto becomingnon-employed, and its effect is very similar regardlessof the time
periodof the transitionwe pick. This is expectedif we believe this variableis picking up unobservable
characteristicsof therespondentsthataremoreor lessstableover time.
18
4 Conclusions
In this paperwe usethe first threewavesof the HRS to addsomeinsight to the literatureon labor force
transitions,usingamorecurrentandcompletesourceof data.This is oneof thefirst attemptsto analyzethe
trendsin retirementandlaborsupplyamongthispopulationusingall threewavesof thesurvey. This is also
thefirst panelstudyto trackemploymentto themonthlylevel with Americandata
We have paidspecialattentionto reversetransitions,somethinglessstudiedthantransitionsinto retire-
ment. Oneof our initial concernswas to study thesereversetransitionsin depthto verify whetherthey
have increaseddueto the healthystateof the economy, and to learnmoreaboutthesedeterminantsin a
time that moreof this kind of transitionscanbe expectedfor individuals55 andolder. Several important
variablesin our specificationcanbe interpretedassupportingthis trend,andthe evidencefrom the labor
forceparticipationratesusingBLS dataseemto point in thatdirection. We emphasize,however, that this
studyconcentrateson themicrodeterminantsof laborforcestatusanddoesnotaddressthemacroeconomic
issuesthatarealsoimportantin explainingthis trend.
Wealsoestimatethelaborforcedecisionsof marriedrespondents,includingdummiesfor thelaborforce
statusof the spouse,their agesandtheir health,andthe resultsshow a preferencefor joint leisureand/or
complementarityof time betweenspouses.Theseresultsareconsistentwith previous literatureusingdata
from the1970’s andearly80’s. Theanalysisof joint retirementbehavior deservesspecialattentionin future
research.
It is alsoworth emphasizingthatwe usea slightly differentcharacterizationof labor forcestatusfrom
the oneusein the literature,sincewe distinguishbetweenbeingself-employed andbeingemployed for
someoneelse.This makessomeof our resultslesscomparablewith previousresearch,but at thesametime
we investigatethis issuein muchmoredepththanpreviousstudiesdo, andwe find that thanksto splitting
thesamplein thisfashionwecaptureddifferenteffectsof theexplanatoryvariablesonthevarioustransitions
probabilitiesandavoid possibleoffsettingeffectsof somevariables.
We usea multinomial anda binomial logit model for estimatingthe decisionsto changelabor force
statusat eachinterview dateasanapproximationto a moreinvolvedmodelof thetransitionbehavior. The
reducedform natureof our analysisuncoversinterestingrelationshipsthat,however, cannotbeinterpreted
in a causalway. We run separateestimationsfor the threesub-samplesof non-employed, employed and
self-employedrespondents,andalsofor morenarrowly definedsub-samplessuchasmalesandfemales.
In the decisionsto returnto work the interactionof agedummieswith healthinsurancestatusreveals
a “spike” at 62 anda 65+ effect, anda negative effect on the probability of returningto work. Among
19
the employed, we find that accessto retireehealth insurancehasa positive effect on the probability of
becomingnon-employed (especiallyfor men). Theseresultsareconsistentwith the importanteffectsof
healthinsurancein labormarket outcomesthathave beenfoundin theliterature.
Theeffect of marriageis quiteinterestingandconsistentwith previousresearch.Marriedmenaremore
likely to returnto work asemployeesandlesslikely to leave theworkforce.Theoppositeis truefor married
women.Healthandcognitionvariablesarealsosignificantandhavetheexpectedsignsacrossspecifications,
wherebeingin worsehealthstatus,by a wide varietyof measures,hasa negative effect on theprobability
of returningto work andapositive effecton becomingnon-employed.
Oneof themostimportantandsignificantvariablespositively affecting the decisionto returnto work
(especiallyasa paidemployee)is therespondent’s job searchdecision.We interpretthis asa variablethat
allows us to distinguishbetweenthoseout of the labor forceandthosestill active in the labormarket, and
for somesortof implied businesscycle effect, sincethewillingnessto find a job hassucha strongpositive
effecton theprobabilityof finding it, whichmeansthatthelabormarket is healthyenoughto accommodate
mostof thereversetransitioners.Theseresultscontrastwith previousresultsin theliteraturethatdid notfind
asignificanteffectof thesearchingdecisionontheprobabilityof findinga job afterwards.Weacknowledge
thepossibleendogeneityof this variable,andtesttheexogeneityof thesearchingdecisionusingabivariate
probitmodel.Wedonot rejectits exogeneityatany standardlevel of significance.
Incomein theyearbeforetheinterview is positively relatedto returningto work. This couldmeanthat
thosethatwereearningmorearemorelikely to re-entry. This possibility is confirmedwhenwe introduce
a variablethat shows the proportionsof monthsworked during the yearprior to the interview. This is a
significantvariablefor the transitionsto employment, consistentwith the literatureon the persistenceof
job displacement.The longertheperiodthat thepersonspendsnon-employed the lesslikely he or sheis
to return to work. Also, morework in the yearprior to the decisionreducesthe likelihoodof makinga
transitionfrom employmentor from self-employment.
Finally, an importantcontribution of this paperis the estimationof shortertransitionsusingmonthly
employmentindicators.We run monthly, quarterly, semi-annualandannualtransitionsfor thedecisionsto
exit theemploymentstatusor to returnto it. Our resultsareconsistentwith thefindingsusingthesurvey
to survey intervals, but most of the importantvariableshave strongereffects for the shortertransitions,
implying that the estimatesusinglongertime periodscould be a lower boundfor the true effect of some
variables.Wearguethata dynamicstudyof laborforcetrajectoriesusingthesemonthlydummieswill bea
relevantcontribution to theliterature,andit is partof ourundergoingresearch.
We considerthat theseresultsmake a contribution to the literature.However, many issuesarestill not
20
addressed,for example,therole of smallercompaniesandnon-profitorganizationsin hiring olderworkers
sincethey could be cheaper, becausetheseworkers might needlower investmentsin pensionandhealth
benefits,or training costs. Thosecompaniescan usethe savings and the experiencedeliveredby older
workers. Also, we might think that theservicesectoris theonepreferredby theelderlybecauseit allows
for moreflexibility andconvenience.
Anotherimportantissueis the incidenceof part-timework in thesetransitions,which onemight think
would be higher amongthe elderly. This is somethingthat could be includedin the model as another
employmentstate.
Many of thesestudiesfocusonolderworkers,but a lot of insightcouldalsobeobtainedby studyingim-
migrants,temporaryworkers,low incomeindividualsor minorities.23 An interestingareaof futureresearch
is to integratetheseanalysesto comparehow public policy andthebusinesscycle is affectingthedecisions
of thesepopulations.
It is a fact thatthelabormarket is changing,andthelaborsupplyanddemandof andfor olderworkers
is not an exception. We interpretsomeof our resultsassupportingthe increasingtrendof longercareers
andsecondcareersamongolder individuals,in a healthyandgrowing labormarket. This hassomepolicy
implicationsbecauseit cangive usa differentperspective on the issueof thefuturesolvency of theSocial
SecurityandRetirementsystem.In ahealthyeconomymaybeit will notmatterthatmuchthattheproportion
of older peopleincreaseswith respectto previous decades,becausetheir working liveswill last longeras
life expectancy increasesanda majority of jobs requiremoreknowledgeandlessphysicalstrength.This
alsomeansthattheelderlyin thenearfuturewill have to keepupwith thetechnologicalchanges,otherwise
they will find themselveswith thejobsthatothersdo notwantor will decideto stayoutof thelabormarket
increasingtheinflationarypressuresin astrongeconomicenvironment.24
23 Evencombinationsof thosecharacteristicsareworthstudyingasthelaborsupplydecisionsof particulargroupscanbestronglyinfluencingtheentiremarket. SeeFlippenandTienda(1996)for anstudyonHispanicolderworkersin theUS laborforce.
24 SeeFriedberg (1999a)for a discussionof theimpactof technologicalchangeon olderworkers,andBenıtez-Silva (2000)fora modelthattriesto shedlight onhumancapitalinvestmentdecisionsat theendof thelife cycle.
21
Table1. PanelA: Meansand Standard Deviations
Variable Full Non-Employed Self-Employed EmployedSample Sub-Sample Sub-Sample Sub-Sample
N 19,769 7,117 2,393 10,259
Age 57.63 59.21 57.48 56.58(0.06) (0.15) (0.12) (0.05)
White 73.54 68.92 82.76 74.69(0.30) (0.53) (0.74) (0.43)
Male 45.28 38.68 61.54 50.05(0.34) (0.56) (0.95) (0.49)
Married 80.76 79.17 87.29 78.19(0.27) (0.47) (0.65) (0.41)
Divorced 16.44 17.65 10.71 18.77(0.25) (0.44) (0.61) (0.39)
BachelorDegree 20.74 12.96 26.60 24.17(0.28) (0.39) (0.87) (0.42)
ProfessionalDegree 7.31 3.78 9.87 9.09(0.18) (0.22) (0.58) (0.28)
Family Income 60,350 49,119 84,086 61,591(1,618) (2,976) (3,886) (2,416)
Resp.Income 22,229 5,357 46,187 28,533(652) (841) (4,717) (371)
Incomefrom Pensions 4,712 6,415 2,616 4,372(785) (987) (256) (1,484)
Net Worth 236,249 215,813 528,982 182,027(3,372) (4,850) (18,146) (3,401)
HousingWealth 73,054 69,548 102,565 69,674(899) (1,217) (4316) (1,208)
HoursWorked 1,441 327 2,061 1,937(7) (9) (22) (7)
% MonthsWorkedlastyear 0.66 0.10 0.97 0.98(0.00) (0.00) (0.00) (0.00)
GovernmentHealthIn. 0.18 0.34 0.10 0.09(0.00) (0.01) (0.01) (0.00)
EmployerHealth-In. 0.71 0.54 0.55 0.85(0.00) (0.01) (0.01) (0.00)
PrivateHealth-In. 0.15 0.15 0.27 0.12(0.00) (0.00) (0.01) (0.00)
Receiving Disability 0.06 0.18 0.00 0.00(0.00) (0.00) (0.00) (0.00)
Retired 0.16 0.38 0.09 0.03(0.00) (0.01) (0.01) (0.00)
22
Table1. PanelB: Meansand Standard Deviations
Variable Full Non-Employed Self-Employed EmployedSample Sub-Sample Sub-Sample Sub-Sample
N 19,769 7,117 2,393 10,259
Non-elegibleSSI 0.92 0.86 0.94 0.95(0.00) (0.00) (0.00) (0.00)
Pr.Working Full Time62+ 44.28 22.55 55.86 45.28(0.33) (0.86) (0.84) (0.41)
Pr.Working Full Time65+ 28.35 23.15 43.77 25.70(0.31) (1.03) (0.81) (0.36)
Pr.Living to 75 64.44 60.40 69.41 66.11(0.21) (0.38) (0.57) (0.28)
Pr.Living to 85 44.10 42.29 47.33 44.37(0.22) (0.40) (0.64) (0.32)
HealthLimitation 22.20 45.48 13.90 8.80for work (0.28) (0.57) (0.68) (0.28)
HealthLimitation 10.56 29.79 0.42 0.18preventswork (0.21) (0.52) (0.13) (0.04)
AgeSpouse 55.24 56.98 54.30 54.96(0.08) (0.16) (0.13) (0.12)
AgeDifference 1.29 1.44 2.11 1.67(0.05) (0.09) (0.14) (0.06)
SpouseN-Employed 0.37 0.47 0.30 0.31(0.00) (0.01) (0.01) (0.01)
SpouseEmployed 0.46 0.37 0.38 0.53(0.00) (0.01) (0.01) (0.01)
SpouseS-Employed 0.11 0.09 0.26 0.09(0.00) (0.01) (0.01) (0.01)
SpousehasHlimwk. 20.66 24.88 7.73 18.75(0.22) (0.44) (0.81) (0.44)
SpousehasHlimpw. 9.24 12.71 5.71 7.87(0.22) (0.44) (0.49) (0.31)
23
Table1 PanelC: Meansand Standard Deviations
Variable Full Non-Employed Self-Employed EmployedSample Sub-Sample Sub-Sample Sub-Sample
N 19,769 7,117 2,393 10,259Objective HealthMeasures:
HospitalStays 0.23 0.42 0.12 0.14(0.01) (0.02) (0.01) (0.01)
DoctorVisits 5.26 7.44 3.63 4.17(0.06) (0.13) (0.11) (0.06)
High Blood pressure 22.18 25.19 18.58 21.66(0.28) (0.50) (0.76) (0.41)
Diabetes 6.76 9.72 4.72 5.37(0.17) (0.34) (0.42) (0.22)
Cancer 2.39 3.24 1.61 2.10(0.10) (0.20) (0.25) (0.14)
Stroke 1.03 2.45 0.19 0.30(0.07) (0.18) (0.09) (0.05)
PsychologicalProblems 6.83 10.67 4.49 4.52(0.17) (0.35) (0.41) (0.21)
BackProblems 32.12 39.87 27.95 27.98(0.32) (0.56) (0.88) (0.44)
FeetProblems 33.03 43.75 25.60 27.88(0.32) (0.57) (0.86) (0.44)
NursingHomeStays 0.57 1.38 0.15 0.14(0.05) (0.13) (0.08) (0.04)
MemoryTest1 7.72 7.27 7.92 7.88(0.02) (0.04) (0.06) (0.03)
MemoryTest2 5.78 5.34 6.00 5.91(0.02) (0.04) (0.06) (0.03)
Cognitive Test 6.20 5.72 6.61 6.35(0.02) (0.04) (0.06) (0.03)
Subjective HealthMeasuresandADL:ExcellentHealth 20.85 13.21 28.10 23.60
(0.28) (0.39) (0.88) (0.42)Fair Health 14.36 21.45 10.29 10.81
(0.24) (0.47) (0.60) (0.31)PoorHealth 7.05 16.73 2.00 1.93
(0.17) (0.43) (0.27) (0.14)WorseHealth 2.98 3.09 2.92 2.93
(0.00) (0.01) (0.01) (0.01)ExcellentCognition 17.12 13.91 2.84 17.77
(0.26) (0.40) (0.82) (0.38)PoorCognition 2.95 6.28 0.73 1.28
(0.11) (0.28) (0.17) (0.11)Diff. Wlk. M.Blocks 14.35 27.39 8.09 7.49
(0.24) (0.51) (0.54) (0.26)Diff. ClimbingStairs 9.48 19.63 3.49 4.23
(0.20) (0.46) (0.36) (0.20)Diff. Stooping 27.43 41.24 8.51 20.70
(0.30) (0.57) (0.76) (0.40)
24
Table2: Empirical Transition Probabilities
Panel A. Transitionsfrom wave1 to wave2. Males50 andover.EmployedWave 2 S-EmployedWave 2 N-EmployedWave 2 Observations
EmployedWave 1 0.8065 0.0367 0.1568 2,832S-EmployedWave 1 0.0782 0.8142 0.1076 818N-EmployedWave 1 0.0842 0.0454 0.8704 1,389
Observations 2,465 833 1,741 5,039
Panel B. Transitionsfrom wave1 to wave2. Females50andover.EmployedWave 2 S-EmployedWave 2 N-EmployedWave 2 Observations
EmployedWave 1 0.8258 0.0160 0.1582 2,686S-EmployedWave 1 0.0745 0.7404 0.1851 416N-EmployedWave 1 0.0822 0.0313 0.8866 2,045
Observations 2,417 415 2,315 5,147
Panel C. Transitionsfrom wave 2 to wave3. Males50 andover.EmployedWave 3 S-EmployedWave 3 N-EmployedWave 3 Observations
EmployedWave 2 0.7968 0.0326 0.1706 2,303S-EmployedWave 2 0.0899 0.7844 0.1257 756N-EmployedWave 2 0.0853 0.0420 0.8728 1,501
Observations 2,031 731 1,798 4,560
Panel D. Transitionsfrom wave2 to wave3. Females50 andover.EmployedWave 3 S-EmployedWave 3 N-EmployedWave 3 Observations
EmployedWave 2 0.7961 0.0197 0.1842 2,438S-EmployedWave 2 0.0819 0.6725 0.2457 403N-EmployedWave 2 0.0752 0.0289 0.8960 2,182
Observations 2,138 382 2,503 5,023
25
Tabl
e3:
Est
imat
ions
for
Not
-Em
ploy
edT
rans
ition
ers
toS
elf-
Em
ploy
men
tand
Em
ploy
men
t
Not
-Em
ploy
edto
Sel
f-E
mpl
oyed
Not
-Em
ploy
edto
Em
ploy
edN
o.V
aria
ble
Sub
sam
ple
Mal
esF
emal
esS
ubsa
mpl
eM
ales
Fem
ales
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
1C
onst
ant
-3.0
776
0.44
42-3
.094
90.
6417
-2.8
490
0.62
64-1
.604
70.
2864
-2.1
960
0.49
54-1
.492
00.
3689
2M
ale
0.60
820.
1798
——
——
0.02
130.
1303
——
——
3M
arrie
d0.
1311
0.23
940.
0895
0.33
300.
1751
0.36
15-0
.394
30.
1478
0.26
590.
2910
-0.6
860
0.18
814
Whi
te-0
.058
80.
1971
-0.3
563
0.26
630.
3413
0.31
83-0
.150
50.
1395
-0.2
542
0.21
97-0
.032
20.
1945
5N
oD
iplo
ma
0.27
380.
1752
0.33
080.
2571
0.27
070.
2497
0.03
750.
1235
-0.0
356
0.19
640.
0422
0.16
406
Voc
.Tra
inin
g0.
2265
0.17
740.
5292
0.24
68-0
.113
40.
2892
0.16
710.
1342
0.08
280.
1995
0.23
440.
1802
7W
ave2
-Ind
-0.3
451
0.20
170.
1093
0.30
24-0
.839
90.
2686
-0.1
349
0.15
850.
1913
0.26
93-0
.223
90.
2130
8A
ge55
-59
-0.0
232
0.25
90-0
.038
50.
3798
0.01
370.
3627
-0.3
385
0.17
92-0
.244
60.
3081
-0.3
169
0.22
149
Age
60-6
10.
1656
0.31
130.
2249
0.50
210.
2732
0.40
93-1
.122
80.
3139
-0.3
678
0.47
72-1
.587
00.
4512
10A
ge62
0.08
390.
4388
0.43
920.
5895
-0.2
748
0.77
97-0
.707
00.
3985
-0.0
683
0.56
44-1
.315
70.
6668
11A
ge63
-64
0.07
180.
4434
0.28
240.
5539
0.00
680.
7825
-0.8
538
0.41
74-0
.429
50.
5529
-1.4
086
0.66
4912
Age
65+
-0.1
925
0.45
47-0
.121
60.
5465
0.27
471.
0057
-0.5
029
0.31
65-0
.651
80.
4575
-0.0
516
0.52
7013
Insu
red5
5-59
-0.5
773
0.28
05-0
.468
30.
4029
-0.5
802
0.39
73-0
.255
40.
2011
0.01
200.
3142
-0.5
182
0.26
7514
Insu
red6
0-61
-1.3
216
0.40
87-1
.346
20.
6084
-1.4
914
0.59
740.
0364
0.36
12-0
.861
50.
5747
0.54
230.
4985
15In
sure
d62
-1.5
620
0.59
87-1
.957
20.
7558
-0.8
533
1.03
24-0
.964
70.
5028
-1.8
369
0.69
09-0
.024
60.
7867
16In
sure
d63-
64-1
.011
10.
5310
-1.1
951
0.63
69-0
.593
10.
9522
-0.0
716
0.47
40-0
.280
40.
5728
0.02
080.
8471
17In
sure
d65
+-0
.654
00.
5582
-0.6
357
0.60
99-0
.913
41.
3949
-0.4
719
0.36
87-0
.379
60.
4324
-0.6
281
0.72
6618
Gov
.In
sura
nce
-1.4
102
0.28
42-1
.474
00.
3703
-1.2
933
0.47
69-0
.213
50.
1633
-0.1
146
0.25
39-0
.313
00.
2182
19S
earc
hj0.
7242
0.20
761.
2172
0.28
050.
0246
0.39
671.
4702
0.14
221.
7267
0.22
461.
2348
0.19
8120
Inco
me(
$1,0
00)
0.01
100.
0047
0.01
720.
0052
-0.0
130
0.02
120.
0091
0.00
370.
0131
0.00
460.
0190
0.01
1321
Pen
tot.(
$1,0
00)
-0.0
184
0.01
270.
0051
0.01
42-0
.053
70.
0252
-0.0
049
0.00
800.
0060
0.01
26-0
.009
90.
0101
22N
wor
th($
105)
0.00
700.
0190
0.00
100.
0310
0.00
700.
0300
-0.0
800
0.03
70-0
.108
00.
0690
-0.0
630
0.04
7023
Mon
thP
0.83
370.
2493
0.54
570.
3689
1.45
240.
3544
1.49
940.
1711
1.34
380.
2649
1.55
590.
2454
24R
eceiv
ing
DI
0.13
810.
3832
0.57
780.
5224
-1.3
771
0.93
85-0
.296
20.
2791
0.06
260.
3933
-0.6
355
0.39
8125
Hbl
oodp
-0.6
726
0.23
14-0
.632
10.
3126
-0.7
619
0.36
050.
0634
0.15
48-0
.020
00.
2422
0.13
550.
2103
26P
shyc
-1.0
398
0.47
11-1
.321
60.
7510
-0.9
552
0.60
650.
1331
0.21
58-0
.025
50.
3746
0.12
840.
2757
27M
emor
yTe
st0.
0465
0.02
470.
0645
0.03
800.
0431
0.03
320.
0261
0.01
930.
0238
0.03
340.
0285
0.02
4328
Dif-
Wkm
b0.
1662
0.24
290.
1309
0.36
960.
3232
0.32
97-0
.376
10.
1868
-0.2
834
0.28
79-0
.465
60.
2485
29D
if-rm
ap-0
.103
40.
2315
0.14
050.
3852
-0.2
241
0.29
19-0
.202
60.
1640
-0.3
640
0.37
26-0
.155
40.
1892
30H
limpw
-1.0
615
0.35
54-0
.805
20.
5183
-1.6
031
0.57
95-1
.034
80.
2485
-1.2
193
0.38
92-0
.958
60.
3093
31E
x-H
ealth
0.33
540.
2160
0.53
290.
3124
0.09
480.
3090
0.19
150.
1630
-0.1
659
0.26
160.
4180
0.21
8632
Vg-
Hea
lth0.
0958
0.20
810.
4362
0.30
19-0
.236
40.
2933
0.02
850.
1438
-0.0
216
0.21
700.
0432
0.19
7233
Fair-
Hea
lth0.
0510
0.24
680.
0158
0.34
170.
0930
0.37
08-0
.112
60.
1889
-0.4
043
0.28
650.
1074
0.25
4434
Poo
r-Hea
lth-0
.322
60.
4278
-0.5
721
0.55
800.
3867
0.66
34-0
.377
00.
2987
-0.9
755
0.52
940.
0298
0.35
95A
vg.L
ogL/
Obs
.-0
.373
15,
125
-0.3
693
2,24
1-0
.363
02,
884
-0.3
731
5,12
5-0
.369
32,
241
-0.3
630
2,88
4
26
Tabl
e4:
Est
imat
ions
for
Not
-Em
ploy
edM
arrie
dT
rans
ition
ers
toS
elf-
Em
ploy
men
tand
Em
ploy
men
t
Not
-Em
ploy
edto
Sel
f-E
mpl
oyed
Not
-Em
ploy
edto
Em
ploy
edN
o.V
aria
ble
Sub
sam
ple
Mal
esF
emal
esS
ubsa
mpl
eM
ales
Fem
ales
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
1C
onst
ant
-3.5
172
1.03
43-2
.691
11.
3802
-2.4
522
1.07
68-0
.477
20.
6092
0.97
550.
9278
-0.5
497
0.65
842
Mal
e0.
9563
0.35
05—
——
—0.
1862
0.26
69—
——
—3
Whi
te0.
2713
0.32
92-0
.265
70.
4209
1.22
240.
6092
-0.6
799
0.22
24-1
.052
30.
3610
-0.4
486
0.31
254
BA
0.17
780.
3784
-0.0
940
0.51
100.
3957
0.52
63-0
.160
50.
2900
-0.5
414
0.48
150.
1027
0.36
635
Pro
f.D
egre
e0.
1115
0.58
210.
6676
0.68
15-0
.663
01.
1957
0.27
270.
4435
0.55
100.
7007
-0.4
911
0.65
956
Nss
i0.
7913
0.79
320.
2343
0.81
95—
—-0
.458
30.
4302
-0.6
772
0.73
11—
—7
Wav
e2-I
nd-0
.366
00.
5145
0.02
660.
5828
-0.4
291
0.86
54-0
.579
40.
3917
-0.7
232
0.63
54-0
.928
70.
4633
8A
ge55
-59
0.26
680.
4454
0.56
670.
6572
0.00
290.
4002
-0.6
843
0.36
680.
0274
0.56
89-0
.840
80.
2543
9A
ge60
-61
0.85
430.
5670
1.34
501.
1073
-0.8
297
0.60
57-0
.440
70.
5876
1.80
551.
0279
-1.8
650
0.50
2410
Age
620.
4987
0.82
310.
4976
1.29
770.
3506
0.71
86-0
.615
90.
9393
-0.2
388
1.58
53-1
.019
10.
5616
11A
ge63
-64
1.32
830.
5980
1.78
420.
7553
-0.3
782
0.97
15-2
.027
31.
1204
-0.0
973
0.99
29-1
.382
50.
6985
12A
ge65
+-0
.152
91.
1904
-1.2
428
0.67
06-1
.648
81.
3259
-1.0
562
0.92
86-1
.003
10.
5719
-2.0
014
0.78
2013
Insu
red5
5-59
-0.6
077
0.44
90-0
.940
60.
6572
——
0.10
000.
3780
-0.2
761
0.57
31—
—14
Insu
red6
0-61
-2.2
675
0.76
43-2
.106
41.
1868
——
-0.9
041
0.66
86-2
.446
61.
1388
——
15In
sure
dAge
62-1
.047
80.
9234
-1.0
515
1.31
72—
—-1
.399
21.
0762
-2.5
654
1.96
06—
—16
Insu
red6
3-64
-1.4
693
0.61
05-1
.658
60.
6874
——
1.05
701.
1737
-0.2
458
1.02
99—
—17
Insu
red6
5+
-1.4
166
1.21
05—
——
—-0
.514
40.
9001
–—
——
18S
earc
hj0.
5100
0.43
760.
8610
0.64
210.
3165
0.63
071.
6430
0.24
551.
3366
0.44
721.
5031
0.34
3219
Inco
me(
$1,0
00)
-0.0
150
0.02
01-0
.016
10.
0222
0.05
240.
0526
0.00
870.
0107
0.03
510.
0153
0.03
640.
0194
20P
ento
t.($1
,000
)-0
.001
80.
0231
0.00
310.
0301
0.05
580.
0207
-0.0
311
0.02
57-0
.163
30.
0959
-0.0
098
0.02
2721
Nw
orth
($10
5)
-0.0
760
0.07
400.
0030
0.05
00-2
.896
01.
7910
-0.1
010
0.11
50-0
.807
00.
3430
-0.0
760
0.07
8022
Mon
thP
1.35
950.
3525
1.06
790.
4996
2.15
140.
5726
1.59
560.
2630
1.38
960.
4040
1.63
330.
3599
23R
eceiv
ing
DI
-0.4
835
0.58
78-0
.066
20.
7868
——
-0.4
357
0.50
98-0
.265
00.
7101
——
24H
ealth
Insu
red
-0.5
732
0.32
92-0
.472
20.
2234
-1.2
473
0.37
39-0
.063
20.
234
-0.0
543
0.15
67-0
.047
90.
3048
25R
etire
d-0
.945
90.
3577
-0.8
907
0.55
31-0
.929
60.
6743
-0.2
181
0.25
65-1
.307
70.
4993
0.17
890.
3173
26S
pous
eNE
-0.4
355
0.31
250.
0901
0.47
23-1
.284
40.
5723
-0.3
767
0.21
32-0
.527
30.
3212
-0.4
199
0.29
3227
Age
Dif
f.0.
0189
0.02
40-0
.002
00.
0486
0.04
870.
0272
0.03
330.
0154
0.02
160.
0202
0.04
410.
0273
28H
ealth
Sp.
0.50
320.
3937
0.20
480.
5427
1.07
020.
6942
-0.0
712
0.29
98-0
.574
60.
5823
-0.0
146
0.34
8229
Hbl
oodp
-1.1
148
0.58
18-0
.517
40.
5891
-0.0
131
0.03
920.
3499
0.31
340.
3890
0.60
96—
—30
Psh
yc-1
.219
41.
0935
-0.8
018
1.30
370.
5846
0.42
72-0
.026
00.
4283
-1.2
198
0.72
26—
—31
Mem
ory
Test
0.00
130.
0339
0.04
890.
0558
-0.4
776
0.47
520.
0409
0.03
000.
0982
0.05
430.
0083
0.03
6032
Dif-
Wkm
b0.
0698
0.39
25-0
.583
70.
6027
——
-0.1
566
0.30
28-0
.741
60.
5768
0.23
180.
3457
33D
if-rm
ap0.
0038
0.32
110.
5839
0.42
97—
—-0
.137
80.
2558
-0.3
695
0.55
76-0
.105
50.
3023
34H
limpw
-0.7
023
0.52
09-0
.546
70.
6535
-1.0
482
0.69
54-1
.156
90.
4536
-2.0
418
0.65
24-0
.873
50.
5015
35E
x-H
ealth
-0.0
548
0.35
83-0
.007
00.
5064
0.00
140.
5414
-0.1
255
0.29
06-0
.875
90.
5795
0.23
290.
3483
36V
g-H
ealth
0.02
760.
3134
0.31
300.
4388
-0.2
016
0.44
83-0
.158
10.
2208
0.28
500.
3356
-0.4
773
0.31
5737
Fair-
Hea
lth0.
0606
0.37
85-0
.170
50.
5389
0.16
340.
5376
-0.5
820
0.31
09-0
.661
00.
4451
-0.6
073
0.42
6038
Poo
r-Hea
lth-0
.044
60.
5555
0.06
730.
6827
-1.0
129
1.13
71-0
.611
70.
5166
-1.4
473
0.97
62-0
.356
90.
5382
Avg
.Log
L/O
bs.
-0.2
72,
918
-0.3
011,
106
-0.3
141,
316
-0.2
72,
918
-0.3
011,
106
-0.3
141,
316
27
Table 5: AverageProbabilities and Mar ginal Effects for the Non-Employment toEmployment Transitions. Binomial Logits25
Panel A. Sub-Sampleof thoseage50 andover1 month 3 months 6 months 12 months 24 months
AverageProbability 0.033 0.040 0.053 0.065 0.126
HealthLimitation Preventswork -0.315 -0.209 -0.675 -0.628 -0.627% of Monthsworkedlastyear 1.791 1.705 1.277 1.191 1.309Searchedfor a job lastmonth 2.361 2.847 2.628 2.479 1.307Insuredand62 -0.678 -0.569 -0.491 -0.718 -0.608Receiving DI -0.675 -0.584 -0.317 -0.141 -0.226Married -0.341 -0.334 -0.309 -0.020 -0.1819
Panel B. Sub-Sampleof malesage50andover1 month 3 months 6 months 12 months 24 months
AverageProbability 0.039 0.045 0.059 0.071 0.131
HealthLimitation Preventswork -0.401 -0.460 -0.865 -0.861 -0.609% of Monthsworkedlastyear 1.353 1.551 0.919 0.780 0.945Searchedfor a job lastmonth 2.146 2.128 2.408 2.512 1.574Insuredand62 -0.489 -0.492 -0.052 -0.731 -0.817Receiving DI -0.479 -0.423 0.061 0.3425 0.083Married -0.080 -0.064 -0.353 0.065 0.149
Panel C. Sub-Sampleof Femalesage50 andover1 month 3 months 6 months 12 months 24 months
AverageProbability 0.0297 0.0369 0.0494 0.0595 0.1217
HealthLimitation Preventswork -0.508 -0.044 -0.5535 -0.480 -0.667% of Monthsworkedlastyear 2.221 1.672 1.193 1.363 1.628Searchedfor a job lastmonth 2.655 3.483 3.070 2.430 0.978Insuredand62 -0.870 -0.851 -0.890 -0.720 -0.179Receiving DI — -0.755 -0.726 -0.565 -0.564Married -0.501 -0.4794 -0.183 -0.025 -0.425
25SelectedVariables. Marginal Effectsare in timesthe AverageProbability. All regressionsalso includedummiesfor beingmale,white, andhaving madethedecisionin wave 2. Also educationlevels,agesinteractedwith healthinsurancestatus,income,wealth,self reportedhealthvariables,ADLs, objectivehealthmeasuresandcognitionvariables.All themarginaleffectscomefromsignificantcoefficientsat the5%or 10%level, exceptfor thosereportedin italics.
28
Tabl
e6:
Est
imat
ions
for
Sel
f-E
mpl
oyed
Tra
nsiti
oner
sto
Non
-Em
ploy
men
tand
Em
ploy
men
t
Sel
f-E
mpl
oyed
toN
on-E
mpl
oyed
Sel
f-E
mpl
oyed
toE
mpl
oyed
No.
Var
iabl
eS
ubsa
mpl
eM
ales
Fem
ales
Sub
sam
ple
Mal
esF
emal
esE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
or1
Con
stan
t0.
6688
1.28
131.
5568
1.59
78-2
.010
82.
2355
-0.1
723
1.19
970.
1880
1.60
54-1
.092
22.
0210
2M
ale
-0.5
108
0.25
28—
——
—0.
3414
0.31
69—
——
—3
Mar
ried
0.09
390.
3898
-0.0
338
0.62
660.
0959
0.51
89-0
.545
90.
3408
0.32
870.
5627
-1.1
526
0.49
974
Whi
te-0
.365
80.
3229
-0.5
279
0.40
45-0
.176
70.
5410
-0.1
289
0.38
85-0
.061
50.
4602
0.16
880.
7688
5B
A-0
.127
30.
3422
0.24
120.
4118
-0.0
724
0.40
66-0
.112
20.
3461
-0.2
847
0.34
500.
6950
0.60
846
Pro
f.D
egre
e0.
6151
0.44
160.
5119
0.41
18-0
.189
10.
4391
-0.3
586
0.47
940.
1058
0.41
02-0
.423
90.
7272
7W
ave2
-Ind
-0.6
526
0.27
59-0
.610
80.
3721
-0.9
226
0.47
32-0
.081
10.
3658
-0.8
458
0.50
520.
7179
0.70
598
Age
55-5
9-0
.374
30.
2818
-0.6
013
0.38
71-0
.242
80.
4148
0.43
800.
3040
0.25
160.
3690
0.80
180.
5932
9A
ge60
-61
0.74
280.
3067
0.33
340.
4492
1.01
420.
4571
-0.1
304
0.43
76-0
.238
20.
5639
-0.1
005
0.88
0010
Age
621.
0225
0.42
840.
7242
0.53
241.
4926
0.77
510.
2430
0.59
79-0
.455
20.
7945
1.64
281.
0187
11A
ge63
-64
-0.1
422
0.81
75-1
.340
21.
0287
——
0.67
670.
8428
0.47
120.
9944
——
12H
ealth
Insu
ranc
e0.
3586
0.39
710.
1119
0.55
410.
2921
0.54
29-0
.076
40.
3977
0.47
000.
5855
-1.0
836
0.66
2013
Inco
me(
$1,0
00)
-0.0
050
0.00
49-0
.000
00.
0037
-0.0
107
0.02
040.
0025
0.00
360.
0074
0.00
34-0
.003
20.
0155
14P
ento
t.($1
,000
)-0
.003
30.
0014
0.00
570.
0198
-0.0
067
0.02
60-0
.055
80.
0048
-0.2
979
0.09
510.
0354
0.03
0715
Nw
orth
($10
5)
-0.0
250
0.02
60-0
.089
00.
0440
0.01
100.
0310
-0.0
340
0.05
60-0
.207
00.
0900
0.03
100.
0280
16To
talH
wkd
.-0
.002
20.
0013
-0.0
252
0.02
03-0
.025
70.
0248
-0.0
157
0.01
34-0
.028
10.
0176
0.00
080.
0203
17M
onth
P-0
.706
31.
0713
-1.5
673
1.19
431.
1750
1.99
33-2
.110
80.
9696
-1.9
965
1.33
09-1
.748
31.
2111
18P
r.WF
T65
-0.3
113
0.34
33-0
.356
00.
4262
-0.1
213
0.64
54-0
.518
00.
3725
-0.1
723
0.44
52-1
.642
10.
7114
19P
shyc
0.80
890.
4837
0.02
061.
0592
——
0.71
750.
5085
1.49
840.
6051
——
20M
emor
yTe
st-0
.060
20.
0484
-0.0
112
0.07
20-0
.096
40.
0640
-0.0
239
0.04
78-0
.064
30.
0624
-0.0
176
0.09
5821
Cog
nitio
nTe
st-0
.027
70.
0438
-0.0
093
0.06
240.
0039
0.06
130.
0904
0.04
620.
1218
0.05
410.
0059
0.08
1422
Dif-
Sta
irs0.
9125
0.58
08-0
.636
60.
9653
2.14
330.
8414
0.85
900.
7831
0.00
801.
0877
2.01
860.
9885
23D
if-S
toop
0.30
890.
2910
0.02
700.
4910
0.71
730.
4087
0.53
190.
3085
0.61
360.
3631
0.60
010.
6623
24H
limw
k0.
4540
0.34
790.
6049
0.41
850.
3162
0.58
70-0
.232
40.
3734
-0.2
360
0.43
05-0
.508
90.
9482
25E
x-H
ealth
-0.5
358
0.42
42-1
.125
90.
5357
0.42
950.
7302
-0.1
125
0.58
69-0
.759
50.
6670
0.88
671.
1165
26V
g-H
ealth
-0.4
214
0.40
49-0
.882
70.
4928
0.51
560.
7041
0.16
020.
5303
-0.3
843
0.58
291.
1676
1.15
2227
Goo
d-H
ealth
-0.4
306
0.37
96-1
.009
40.
4773
0.41
290.
6897
0.62
340.
5234
0.50
040.
5446
0.84
251.
2493
Avg
.Log
L/O
bs.
-0.5
4495
8-0
.488
163
2-0
.573
332
6-0
.544
958
-0.4
881
632
-0.5
733
326
29
Tabl
e7:
Est
imat
ions
for
Mar
ried
Sel
f-E
mpl
oyed
Tra
nsiti
oner
sto
Non
-Em
ploy
men
tand
Em
ploy
men
t
Sel
f-E
mpl
oyed
toN
on-E
mpl
oyed
Sel
f-E
mpl
oyed
toE
mpl
oyed
No.
Var
iabl
eS
ubsa
mpl
eM
ales
Fem
ales
Sub
sam
ple
Mal
esF
emal
esE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
or1
Con
stan
t2.
0503
1.76
223.
0263
2.65
914.
0836
2.49
83-0
.592
11.
7962
0.25
472.
2461
-2.8
029
3.32
552
Mal
e-0
.428
80.
3094
——
——
0.00
640.
4037
——
——
3W
hite
0.39
390.
5239
0.63
500.
8037
0.21
740.
8616
-0.7
332
0.51
320.
1354
0.69
05-1
.952
30.
8894
4B
A0.
0992
0.42
93-0
.488
20.
6883
0.73
190.
6447
-0.1
853
0.51
39-0
.240
20.
5952
0.48
760.
8386
5P
rof.
Deg
ree
0.53
040.
5516
1.07
680.
7906
0.88
640.
8156
-0.6
931
0.92
11-0
.980
41.
1807
0.80
691.
5677
6W
ave2
-Ind
-0.8
879
0.47
71-0
.654
50.
6300
-1.0
483
0.98
41-0
.089
40.
5974
-0.4
612
0.69
551.
4924
1.01
737
Age
55-5
9-0
.507
90.
4223
-1.1
940
0.67
34-0
.633
00.
6677
0.36
750.
3942
0.37
470.
4464
-0.8
917
1.09
298
Age
60-6
10.
7842
0.44
970.
3760
0.60
101.
5698
0.68
130.
1966
0.60
57-0
.268
10.
7852
-0.2
726
0.84
859
Age
621.
4187
0.55
231.
1284
0.70
34—
—0.
6088
0.78
65-0
.115
61.
2219
——
10A
ge63
-64
1.79
971.
1320
——
——
3.19
961.
7948
——
——
11In
com
e($1
,000
)0.
0017
0.00
280.
0008
0.00
300.
0245
0.01
31-0
.002
70.
0048
0.00
060.
0035
0.02
670.
0202
12P
ento
t.($1
,000
)-0
.026
20.
0240
——
——
-0.6
748
0.81
39—
——
—13
Nw
orth
($10
5)
-0.0
220
0.02
80-0
.037
00.
0270
0.00
600.
0490
0.01
600.
0250
-0.0
290
0.03
900.
0890
0.03
7014
Tota
lHw
kd.
-0.0
293
0.01
77-0
.025
60.
0221
-0.0
463
0.02
61-0
.026
80.
0185
-0.0
297
0.02
12-0
.030
80.
0295
15M
onth
P-1
.613
01.
4919
-2.5
153
2.16
03-2
.416
51.
8522
-0.1
952
1.42
73-0
.555
32.
0452
2.27
702.
5026
16P
r.WF
T65
-0.3
072
0.44
40-0
.115
90.
5388
-0.6
376
0.89
64-0
.411
90.
5396
-0.5
289
0.62
50-0
.015
50.
9705
17S
pous
eSE
-0.4
517
0.42
19-0
.018
40.
5009
-1.7
597
0.93
95-0
.207
10.
5039
-0.6
750
0.61
300.
0711
0.91
4218
Age
spou
se62
+0.
5979
0.48
16-0
.017
90.
0332
-0.0
230
0.06
37-1
.529
71.
0154
0.03
520.
0240
-0.0
163
0.03
1419
Hea
lthS
p.0.
2927
0.37
07-0
.388
10.
5181
1.01
680.
5966
0.05
840.
4776
-0.5
823
0.62
150.
9455
0.74
7220
No.
Hos
p.0.
2297
0.33
980.
1669
0.42
24—
—0.
1738
0.33
630.
1989
0.36
20—
—21
Psh
yc0.
7027
0.60
58-0
.606
30.
9775
——
0.85
940.
9088
0.75
440.
9705
——
22M
emor
yTe
st0.
0152
0.05
300.
0320
0.07
43-0
.046
40.
0873
-0.0
921
0.07
31-0
.005
00.
0836
-0.2
260
0.13
3723
Cog
nitio
nTe
st-0
.103
20.
0599
-0.1
452
0.08
170.
0107
0.08
870.
0782
0.07
060.
0683
0.08
32-0
.000
60.
1285
24D
if-S
tairs
0.11
370.
8429
—-
——
0.24
921.
0586
——
——
25D
if-S
toop
0.72
160.
3788
0.59
090.
5189
0.82
330.
5418
0.39
700.
3778
0.66
160.
4364
0.25
590.
8578
26H
limw
k0.
9723
0.42
351.
1450
0.54
810.
8568
0.59
83-0
.482
40.
5271
-0.5
493
0.53
12-0
.073
21.
2041
27W
orse
Hea
lth-0
.355
00.
2295
-0.4
313
0.40
41-0
.539
60.
3301
-0.1
435
0.23
96-0
.372
30.
2955
-0.2
518
0.49
0428
Ex-
Hea
lth-0
.051
90.
5384
-0.2
180
0.64
17-0
.511
00.
6050
-0.8
663
0.86
89-1
.902
21.
1069
0.35
991.
0028
29V
g-H
ealth
-0.2
889
0.51
96-0
.421
90.
6020
-0.5
560
0.53
650.
1635
0.76
73-0
.507
40.
8474
1.00
940.
8423
30G
ood-
Hea
lth-0
.137
20.
4816
-1.0
863
0.66
84—
—0.
9202
0.73
300.
6830
0.71
88—
—
Avg
.Log
L/O
bs.
-0.5
063
0-0
.462
413
-0.5
0921
5-0
.50
630
-0.4
6241
3-0
.509
215
30
Tabl
e8:
Est
imat
ions
for
Em
ploy
edT
rans
ition
ers
toN
on-E
mpl
oym
enta
ndS
elf-
Em
ploy
men
t
Em
ploy
edto
Non
-Em
ploy
edE
mpl
oyed
toS
elf-
Em
ploy
edN
o.V
aria
ble
Sub
sam
ple
Mal
esF
emal
esS
ubsa
mpl
eM
ales
Fem
ales
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
1C
onst
ant
-1.2
058
0.47
43-0
.632
90.
7943
-0.8
284
0.55
08-4
.946
61.
1268
-3.8
698
1.13
60-7
.086
01.
6684
2M
ale
-0.0
534
0.09
36—
——
–0.
7853
0.22
80—
——
—3
Mar
ried
-0.0
177
0.10
51-0
.312
20.
1789
0.12
90.
1272
-0.2
054
0.24
14-0
.079
30.
3571
-0.4
002
0.38
464
Whi
te0.
0015
0.10
100.
0269
0.15
370.
0167
0.13
620.
7269
0.30
350.
5497
0.36
511.
0762
0.58
215
No
Dip
lom
a0.
1423
0.09
660.
0266
0.14
250.
2203
0.13
130.
4042
0.23
890.
6309
0.32
590.
0611
0.37
876
Voc
.Tra
inin
g0.
2065
0.10
150.
1677
0.14
500.
2516
0.14
31-0
.109
80.
2675
-0.1
271
0.33
030.
0357
0.44
967
N-e
ligib
leD
I/SS
I0.
2480
0.22
44-0
.088
30.
2962
0.10
180.
1706
0.46
030.
6286
0.09
260.
5984
0.76
080.
5816
8W
ave2
-Ind
-0.0
972
0.10
35-0
.150
80.
1514
——
0.53
390.
2210
0.31
830.
2702
——
9A
ge55
-59
0.44
720.
1016
0.64
490.
1565
0.27
810.
1345
-0.2
701
0.21
760.
0367
0.26
92-0
.888
30.
4213
10A
ge60
-61
1.43
600.
1175
1.62
850.
1782
1.28
760.
1595
0.24
230.
2934
0.32
530.
3834
0.12
440.
4838
11A
ge62
1.88
180.
1696
2.29
500.
2328
1.37
770.
2684
0.18
450.
4575
0.03
210.
6483
0.60
590.
6316
12A
ge63
-64
2.03
510.
2756
2.37
880.
3189
——
1.18
980.
5433
1.55
230.
5780
——
13A
ge65
+1.
0910
0.44
971.
8672
0.51
19—
—0.
8101
0.77
271.
3628
0.79
61—
—14
Inco
me(
$1,0
00)
0.00
200.
0010
0.00
130.
0019
0.00
790.
0048
0.00
290.
0015
0.20
670.
3784
-0.0
171
0.01
9515
Pen
tot.(
$1,0
00)
0.00
730.
0060
0.00
440.
0106
0.00
930.
0086
-0.0
131
0.02
180.
0036
0.00
120.
0063
0.02
6116
Nw
orth
($10
5)
0.01
800.
0120
0.01
400.
0180
0.02
600.
0170
0.04
600.
0180
-0.0
211
0.03
000.
1010
0.03
5117
Tota
lHw
kd.
-0.0
093
0.00
78-0
.012
30.
0122
-0.0
075
0.01
06-0
.017
00.
0201
0.02
200.
0150
0.01
280.
0360
18M
onth
P-1
.274
50.
4224
-1.0
065
0.73
26-1
.353
60.
5014
-0.9
993
0.84
920.
6113
0.32
330.
7744
1.20
7319
Hea
lthIn
sura
nce
0.13
490.
1249
-0.0
483
0.17
570.
3057
0.18
31-0
.040
30.
2915
0.81
240.
3594
-0.4
938
0.48
3420
P-W
rk.F
T65
+-1
.213
10.
1597
-1.4
852
0.23
69-0
.930
70.
2185
0.64
720.
2637
-1.0
509
0.60
780.
7372
0.47
6921
Psh
yc0.
2047
0.18
260.
2716
0.27
65—
—-0
.042
40.
5410
0.92
850.
5975
——
22M
emor
yTe
st-0
.014
40.
0142
-0.0
439
0.02
360.
0047
0.01
790.
0855
0.02
98-0
.359
40.
3180
0.02
120.
0464
23D
if-W
kmb
0.33
890.
1518
0.45
210.
2442
0.20
240.
1956
-1.2
756
0.60
490.
1379
0.65
54-1
.532
41.
2071
24D
if-st
airs
0.02
930.
2005
0.27
510.
3599
-0.0
547
0.24
710.
7653
0.51
880.
1242
0.03
940.
5801
0.93
2125
Dif-
stoo
p0.
1772
0.10
530.
1142
0.16
250.
2363
0.13
90-0
.171
70.
2569
-0.0
249
0.02
650.
0860
0.43
8926
Hlim
wk
0.28
630.
1469
0.47
600.
2032
0.09
670.
2198
0.87
160.
2995
-1.4
602
1.02
820.
8325
0.54
8927
Ex-
Hea
lth-0
.235
50.
1174
-0.1
602
0.17
00-0
.866
60.
2080
-0.1
563
0.25
64-0
.129
40.
3303
1.20
151.
0053
28V
g-H
ealth
-0.1
775
0.10
20-0
.066
10.
1467
-0.8
416
0.19
01-0
.173
10.
2344
0.00
300.
2903
0.82
580.
9850
29G
d-H
ealth
——
——
-0.5
524
0.18
14—
——
—1.
4372
0.94
8130
Fair-
Hea
lth0.
3463
0.13
440.
1470
0.20
11—
—-0
.480
60.
4219
-0.1
838
0.48
35—
—31
Poo
r-Hea
lth0.
5627
0.27
780.
4983
0.35
15—
—0.
3294
0.74
980.
7052
0.81
29—
—A
vg.L
ogL/
Obs
.-0
.448
75,
881
-0.4
586
2,93
3-0
.429
52,
950
-0.4
487
5,88
1-0
.458
62,
933
-0.4
295
2,95
0
31
Tabl
e9:
Est
imat
ions
for
Mar
ried
Em
ploy
edT
rans
ition
ers
toN
on-E
mpl
oym
enta
ndS
elf-
Em
ploy
men
t
Em
ploy
edto
Non
-Em
ploy
edE
mpl
oyed
toS
elf-
Em
ploy
edN
o.V
aria
ble
Sub
sam
ple
Mal
esF
emal
esS
ubsa
mpl
eM
ales
Fem
ales
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
Est
.S
t.Err
orE
st.
St.E
rror
1C
onst
ant
-1.6
642
0.51
08-1
.524
90.
7748
-1.4
782
0.65
93-5
.114
81.
2383
-4.4
598
1.62
04-6
.510
01.
9904
2M
ale
-0.0
461
0.11
99—
——
—0.
9343
0.29
96—
——
—3
Whi
te0.
1631
0.13
930.
0676
0.18
740.
3797
0.20
430.
6741
0.36
310.
5014
0.41
681.
1599
0.76
304
BA
-0.0
253
0.14
780.
0005
0.20
200.
0543
0.20
650.
7565
0.25
241.
2310
0.29
59-0
.514
70.
6599
5P
rof.
Deg
ree
-0.0
051
0.22
32-0
.150
70.
3050
0.26
740.
3270
-0.8
377
0.41
27-1
.273
20.
4589
0.52
580.
9775
6N
-elig
ible
SS
I0.
2396
0.30
12-0
.002
90.
4111
0.38
080.
4232
-0.7
505
0.53
53-0
.853
30.
5697
-0.3
085
1.16
927
Wav
e2-I
nd-0
.227
50.
1773
-0.5
168
0.25
48-0
.473
20.
2453
0.03
460.
3755
-0.0
816
0.44
26-0
.554
60.
6627
8A
ge55
-59
0.30
280.
1203
0.40
940.
1765
0.11
110.
1578
-0.2
950
0.24
82-0
.330
80.
2867
-0.3
789
0.45
879
Age
60-6
11.
2360
0.15
281.
1879
0.21
031.
0636
0.21
710.
0725
0.37
10-0
.251
00.
4876
0.61
400.
5741
10A
ge62
1.60
230.
2206
1.90
480.
2787
0.70
520.
3932
0.21
010.
5132
0.12
740.
6110
0.77
480.
9677
11A
ge63
-64
1.81
980.
3926
1.65
300.
4081
——
0.95
650.
8144
0.89
470.
8296
——
12In
com
e($1
,000
)-0
.004
40.
0039
-0.0
066
0.00
53-0
.024
70.
0092
-0.0
047
0.00
80-0
.007
10.
0085
-0.0
215
0.03
5413
Pen
tot.(
$1,0
00)
0.01
620.
0107
0.01
740.
0168
0.01
330.
0139
-0.0
816
0.04
72-0
.257
50.
1294
0.01
800.
0271
14N
wor
th($
105)
0.02
400.
0160
0.04
400.
0230
0.03
300.
0200
0.01
500.
0340
0.03
300.
0360
-0.0
910
0.12
8015
Tota
lHw
kd.
-0.0
297
0.00
85-0
.025
70.
0126
——
-0.0
438
0.01
79-0
.041
80.
0211
——
16P
r.WF
T65
-0.9
491
0.20
47-1
.086
00.
2744
-0.7
624
0.30
400.
6420
0.31
320.
6807
0.37
001.
0647
0.60
1717
Spo
useE
.-0
.311
40.
1608
-0.5
478
0.23
34-0
.151
70.
2163
-0.1
979
0.29
400.
0764
0.37
23-0
.852
20.
5483
18S
pous
eAge
62+
0.09
640.
2197
0.63
920.
4504
0.24
570.
2552
0.13
780.
5224
1.36
230.
6821
-1.0
440
0.68
4919
Hea
lthS
p.0.
2782
0.14
840.
0804
0.22
770.
4625
0.18
920.
4959
0.30
260.
4043
0.39
560.
3955
0.47
9120
Hos
pita
lsta
ys—
——
——
——
——
——
—21
Psh
yc0.
1846
0.24
96-0
.200
00.
4639
-0.0
761
0.07
660.
0215
0.60
870.
3930
0.62
89-0
.926
10.
9580
22M
emor
yTe
st-0
.027
00.
0182
-0.0
915
0.02
970.
0182
0.02
200.
0478
0.03
790.
0648
0.04
890.
0063
0.05
7223
Dif-
Rea
ding
Map
——
——
0.11
400.
1783
——
——
-0.0
092
0.55
3224
Dif-
Wal
k.M
.B0.
1660
0.19
330.
4584
0.29
21—
—-1
.155
50.
5009
-0.6
392
0.49
90—
—25
Dif-
Sta
irs0.
2130
0.24
830.
4450
0.41
93—
—0.
1256
0.48
740.
7059
0.48
82—
—26
Pr.
Liv
to75
0.28
560.
2019
0.16
300.
2877
0.08
060.
2604
0.34
750.
4546
0.23
740.
5266
1.26
890.
7788
27H
limw
k0.
1771
0.18
840.
1976
0.26
610.
3426
0.23
540.
6799
0.35
810.
2582
0.45
281.
0872
0.63
2528
Wor
seH
ealth
0.09
090.
0875
0.27
630.
1367
-0.0
948
0.11
640.
3930
0.23
560.
3828
0.27
380.
4551
0.36
9829
Ex-
Cog
nitio
n-0
.167
20.
1975
0.06
810.
2661
-0.3
049
0.29
580.
4129
0.52
710.
7749
0.68
270.
1197
0.88
1730
Vg-
Cog
nitio
n-0
.272
90.
1716
-0.0
737
0.24
00-0
.446
00.
2362
0.01
720.
5262
0.38
910.
6776
-0.1
916
0.82
2531
Gd-
Cog
nitio
n-0
.125
90.
1711
0.08
250.
2430
-0.3
256
0.23
120.
2612
0.53
490.
4150
0.71
150.
4016
0.81
0632
Ex-
Hea
lth-0
.241
50.
1544
-0.0
717
0.21
42-0
.585
50.
2012
-0.2
028
0.31
42-0
.330
00.
3721
0.42
610.
5724
33V
g-H
ealth
-0.1
310
0.13
12-0
.022
30.
1840
-0.3
944
0.16
73-0
.140
60.
2871
-0.2
815
0.34
590.
5533
0.52
1334
Fair-
Hea
lth0.
2951
0.18
12-0
.105
50.
2601
——
-0.6
647
0.51
76-0
.434
00.
5393
——
35P
oor-H
ealth
1.13
260.
3974
1.00
210.
4957
——
-0.0
011
0.94
300.
3057
0.96
66—
—
Avg
.Log
L/O
bs.
-0.4
443,
830
-0.4
42,
119
-0.4
539
1,84
0-0
.444
5828
-0.4
42,
119
-0.4
539
1,84
0
32
Table10: AverageProbabilities and Mar ginal Effects for the Employment toNon-EmploymentTransitions. Binomial Logits26
Panel A. Sub-Sampleof thoseage50 andover1 month 3 months 6 months 12 months 24months
AverageProbability 0.0189 0.030 0.051 0.085 0.121
HealthLimitation for work 0.943 0.296 0.411 0.202 0.360% Monthsworkedlastyear -1.046 -1.331 -2.024 -1.128 -1.06755-57with RetireeH-I. 0.238 0.240 -0.121 0.327 0.22163-64with RetireeH-I. — 0.674 1.959 2.919 2.255No High SchoolDiploma 0.314 0.643 0.126 0.188 0.293Prob. Working FT by age65 -0.495 -0.293 -0.594 -0.671 -0.792
Panel B. Sub-Sampleof malesage50andover1 month 3 months 6 months 12 months 24months
AverageProbability 0.0145 0.0280 0.0543 0.0749 0.1086
HealthLimitation for work 2.252 0.384 0.297 0.070 0.505% Monthsworkedlastyear -1.193 -1.826 -2.175 -1.457 -1.18355-57with RetireeH-I. 0.306 0.108 0.017 0.261 0.24463-64with RetireeH-I. — — 1.661 1.983 2.871No High SchoolDiploma 0.526 0.377 0.007 0.138 0.234Prob. Working FT by age65 -1.149 -0.711 -0.729 -0.923 -1.014
Panel C. Sub-Sampleof Femalesage50 andover1 month 3 months 6 months 12 months 24months
AverageProbability 0.0233 0.0326 0.0478 0.0945 0.132
HealthLimitation for work 0.173 0.141 0.613 0.351 0.234% Monthsworkedlastyear -1.005 -0.978 -1.925 -0.873 -1.03055-57with RetireeH-I. 0.246 0.381 0.285 0.360 0.19962with RetireeH-I. 0.567 2.817 1.724 1.612 1.539No High SchoolDiploma 0.185 0.822 0.258 0.218 0.311Prob. Working FT by age65 -0.137 0.113 -0.470 -0.597 -0.804
26Marginal Effectsarein timestheAverageProbability. All regressionsalsoincludedummiesfor beingmale,married,white,non-eligiblefor disability insuranceandhaving madethe decisionin wave 2. Also educationlevels, agesinteractedwith healthinsurancestatus,income,wealth,self reportedhealthvariables,ADLs, objective healthmeasuresandcognitionvariables.All themarginaleffectscomefrom significantcoefficientsat the5% or 10%level, exceptfor thosereportedin italics.
33
Figure1: Labor Forceparticipation. Males 55 to 64
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 63
64
65
66
67
68
69
Labo
r For
ce P
artic
ipatio
n
Figure2: Labor Forceparticipation. Females55 to 64
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 42
43
44
45
46
47
48
49
50
51
52
53
Labo
r For
ce P
artic
ipatio
n
34
Appendix A
In this Appendixwe borrow from Heckman(1978),Kiefer (1982)andGreene(1993)to show how we testfor theexogeneityof thesearchingdecisionin ourestimations.27
Considera two equationsystem:
y�1i� X1iβ1
�diα1
�y�2iγ1
�U1i (1)
y�2i� X2iβ2
�diα2
�y�1iγ2
�U2i (2)
wherethedummyvariabledi is definedby: di� 1 iff y
�2i � 0, anddi
� 0 otherwise.It is importantto emphasizethat thesystemabove representstwo continuouslatentvariablesthatgen-
erateobservablediscretedummyvariables(y1i andy2i) whenthey reacha threshold.This modelis flexibleenough,asHeckman(1978)shows, to includea numberof importantspecifications.Thecasethatwe areinterestedin is onewhereequation(1) representsthestructuralequationof interest,in ourcasethedecisionto becomeemployed in thenext period(employmentfor a third partyandself-employmentarenot distin-guishfor thepurposeof this test)and(2) representsthe searchingdecisionthatmight be endogenous.Inequations(1) and(2), X1i andX2i, arerespectively, 1 � K1 and1 � K2 row vectorsof boundedexogenousvariables.Thejoint densityof continuousrandomvariablesU1i � U2i is g � U1i � U2i � which is assumedto beabivariatenormaldensity, with meannormalizedto be0 andthe2 � 2 covariancewith variancesnormalizedto 1, andcorrelationcoefficient ρ ��� 1 � 1� .
Our objective is to test exogeneityof the searchingdecisionwith respectto the structuraldecision.Without lostof generalitywecanconsiderasimplecharacterizationof thesystemunderthenull hypothesisof exogeneityof thesearchvariable,by settingα2
� 0 � γ1� 0 � and γ2
� 0. This modelthenreducesto astandardbivariateprobit model,wherethe test for independenceof the probit equationsis equivalent forus to a test of exogeneityof the searchingdecision. If we cannotreject that ρ � 0 then we can safelyassumeexogeneityin ourestimations.If werejectthenull hypothesiswewill have to estimatethestructuralparametersthroughabivariateprobitwith astructuralshift.
Therefore,we test the independenceof the probit equationsresultingfrom (1) and(2). The simplestmethodto constructthetestof thehypothesisthatρ � 0 follows Kiefer (1982)andGreene(1993).Thecon-structionof theLagrangeMultiplier (LM) testonly requirestheestimationof thetwo independentprobits:
LM � f 2
h(3)
where f andh arecalculatedasfollows:
f � ∑i
q1iq2iφ � w1i � φ � w2i �Φ � w1i � Φ � w2i � (4)
h � ∑i
φ � w1i � φ � w2i ��� 2
Φ � w1i � Φ ��� w1i � Φ � w2i � Φ ��� w2i � (5)
where,
q1i� 2y1i � 1 � and q2i
� 2y2i � 1 (6)
w1i� q1iβ �1X1i � and w2i
� q2iβ �2X2i (7)
27 Benıtez-Silva et al. (2000)alsousethis approachin their studyof theSocialSecurityDisability awardprocess.
35
TableA1 presentstheestimatesof theindependentprobitsfor theemploymentandsearchdecisions.TheLM teststatisticfor thisspecificationis 0.0169andfollowsaχ2
1, deliveringaP-valueof 0.896,sowecannotrejectthenull hypothesisof exogeneityof thesearchdecision.TheLM teststatisticis simplerto calculatethantheWaldstatisticandtheLikelihoodRatiothatrequiretheestimationof thebivariateprobit,andgiventhatwe have morethan4,000observationswe believe our resultsarenot sensitive to theteststatisticused.
TableA.1: Probit Estimatesof the Employment and Search decisions
Not-Employedto Employed SearchingDecision
No. Variable Estimate StandardError Estimate StandardError1 Constant -0.785 0.178 -0.285 0.1822 Male 0.1664 0.065 0.448 0.0793 Married -0.199 0.082 -0.428 0.0884 White -0.079 0.072 -0.429 0.0795 No Diploma 0.080 0.061 0.156 0.0756 VocationalTraining 0.024 0.066 0.016 0.0827 Wave2-Ind -0.144 0.096 -0.228 0.1228 Age55-59 -0.137 0.111 0.210 0.1259 Age60-61 -0.182 0.164 -0.094 0.17810 Age62 -0.269 0.204 -0.449 0.23211 Age63-64 -0.375 0.214 -0.073 0.23612 Age65 + -0.312 0.178 -0.138 0.22613 Insured55-59 -0.169 0.116 -0.463 0.12614 Insured60-61 -0.395 0.180 -0.484 0.19515 Insured62 -0.537 0.241 -0.180 0.26616 Insured63-64 -0.121 0.233 -0.793 0.27617 Insured65 + -0.190 0.188 -0.809 0.26918 GovernmentInsurance -0.288 0.087 -0.491 0.11619 OwnHealthInsurance 0.035 0.096 -0.440 0.10420 Income($1,000) 0.003 0.002 -0.000 0.00321 Pentot.($1,000) -0.005 0.004 -0.020 0.00622 Nworth($105) -0.026 0.016 -0.062 0.02423 MonthP 0.736 0.093 0.953 0.09624 Receiving DI -0.093 0.129 -0.264 0.15225 Hbloodp -0.066 0.085 -0.002 0.10326 Pshyc -0.040 0.124 0.273 0.12827 MemoryTest 0.014 0.009 0.014 0.01028 Dif-Wkmb -0.201 0.091 -0.038 0.11229 Dif-rmap -0.137 0.080 -0.058 0.09430 Hlimpw -0.454 0.110 -0.621 0.12131 Ex-Health 0.110 0.083 0.141 0.09732 Vg-Health 0.016 0.073 0.094 0.08833 Fair-Health -0.024 0.092 0.070 0.10534 Poor-Health -0.093 0.138 -0.270 0.17635 Searchj 0.816 0.084 — —
Avg. Log L/Obs. -0.2991 4,175 -0.2088 4,175
36
References
Benıtez-Silva, H., M. Buchinsky, H-M. Chan,J.RustandS.Sheidvasser:(2000),“How Largeis theBiasin Self-reportedDisability Status?”NBERWorkingPaperNo. 7526.
Benıtez-Silva, H. (2000): “A DynamicModel of JobSearchBehavior over the Life Cycle,” manuscript,YaleUniversity.
Blau, D., andR. T. Riphahn(1998): “Labor ForceTransitionsof Older Married Couplesin Germany.”CEPRDiscussionPaper1911.
Blau,D., D. B. Gilleskie(1997):“RetireeHealthInsuranceandtheLaborForceBehavior of OlderMeninthe1990’s,” manuscript,Universityof North Carolinaat ChapelHill.
Blau,D., D. B. GilleskieandC.Slusher(1997):“The Effectof HealthonEmploymentTransitionsof OlderMen,” manuscript,Universityof North CarolinaatChapelHill andSocialSecurityAdministration.
Blau, D. (1998): “Labor ForceDynamicsof OlderMarriedCouples,” Journal of Labor Economics, 16-3595–629.
Blau, D. (1997): “Social SecurityandtheLaborSupplyof OlderMarriedCouples,” Labor Economics, 4373–418.
Blau,D. (1994):“Labor ForceDynamicsof OlderMen,” Econometrica, 62 117–156.
Bound,J.,M. Schoenbaum.,T. Stinebrickner, andT. Waidmann(1998): “Modeling theEffectsof Healthon Retirement,” Draft presentedat theNBERAging & HealthCareSummerInstitute,July, 1998
Bound,J.,M. Schoenbaum.,T. Stinebrickner, andT. Waidmann(1997):“MeasuringtheEffectsof Healthon RetirementBehavior,” Draft presentedat the“Economicsof Aging: InternationalHealthandRe-tirementSurvey Conference,” Amsterdam,August7-8,1997.
Coile, C., andJ. Gruber(1999): “Social Security, PrivatePensions,andRetirement,” manuscript,Mas-sachusettsInstituteof Technology.
Coile, C. (1999): “Social Security, Pensions,and the RetirementDecisionsof Couples,” manu-script,NBER.
Costa,D. L. (1999): “ TheRiseof JointRetirement,1940-1998,” manuscript,MassachusettsInstituteofTechnology.
Currie,JandB.C. Madrian(1998):“Health,HealthInsuranceandtheLaborMarket,” forthcomingin TheHandbook of Labor Economics, Orley AshenfelterandDavid Card,eds.
Flinn, C. J., andJ. J. Heckman(1983): “Are UnemploymentandOut of the Labor ForceBehaviorallyDistinct LaborForceStates?”Journal of Labor Economics, 1-128–42.
Flippen,C.,andM. Tienda(1996):“Labor ForceBehavior of Hispanicelderly, insightsfrom HRS,” Work-ing Paper96-032.HRSwebpage.
Friedberg, L. (1999a):“The impactof TechnologicalChangeon OlderWorkers: Evidencefrom DataonComputers,” Universityof California,SanDiego,DiscussionPaperNo. 99-11.
37
Friedberg, L. (1999b):“The TrendTowardsPart-Time Work AmongOlderWorkers,” Universityof Cali-fornia,SanDiego.
Greene,W.H (1993):Econometric Analysis, SecondEdition,PrenticeHall, New Jersey.
Gustman,A. L., O. Mitchell, andT.L. Steinmeier(1994): “RetirementResearchUsing the HealthandRetirementSurvey,” NBER WorkingPaperNo. 4813.
Gustman,A. L., O. Mitchell, andT.L. Steinmeier(1995):“RetirementMeasuresin theHealthandRetire-mentStudy,” The Journal of Human Resources 30 (Supplement).
Heckman,J.J.(1978): “Dummy EndogenousVariablesin a SimultaneousEquationSystem,” Economet-rica, 46 931–959.
HRSwebpage:www.umich.edu/ hrswww/
Hurd,M. (1990a):“The joint RetirementDecisionof HusbandsandWives,” in David A. Wise(ed),Issuesin the Economics of Aging, Chicago,Universityof ChicagoPressfor theNationalBureauof EconomicResearch.
Hurd,M. (1990b):“ResearchontheElderly: EconomicStatus,Retirement,andConsumptionandSaving,”Journal of Economic Literature, 28565–637.
Hurd,M., andK. McGarry(1993):“TheRelationshipBetweenJobCharacteristicsandRetirement,” NBERWorkingPaperNo. 4558.
Juster, T.F., andR. Suzman(1995): “An Overview of theHealthandRetirementStudy,” The Journal ofHuman Resources, 30 (Supplement):S7–S56.
Karoly, L., andJ. Rogowski (1997): “Health InsuranceandLaborMarket Transitionsof OlderWorkers,”RAND WorkingPaper.
Kiefer, N.M (1982):“Testingfor Dependencein multivariateprobitmodels,” Biometrika, 69161-166.
Lumsdaine,R. (1995):“FactorsAffectingLaborSupplyDecisionsandRetirementIncome,” NBERWork-ing PaperNo. 5223.
Lumsdaine,R.,J.StockandD. A. Wise(1992):“ThreeModelsof Retirement:ComputationalComplexityvs. Predictive Validity,” In Topics in the Economics of Aging, D. A. Wise,ed.Chicago:UniversityofChicagoPress.
Lumsdaine,R. andO. Mitchell (1998): “New Developmentsin the EconomicAnalysisof Retirement,”forthcomingin The Handbook of Labor Economics, Orley AshenfelterandDavid Card,eds.
Peracchi,F., andF. Welch(1994):“Trendsin LaborForceTransitionsof OlderMenandWomen,” Journalof Labor Economics 12-2210–242.
Pozzebon,S. andO. S. Mitchell (1989); “Married women’s retirementbehavior,” Journal of PopulationEconomics 2, 39-53.
Quinn,J.,andR. Burkhauser(1994):“RetirementandLaborForceBehavior of theElderly,” In Demogra-phy of Aging, NationalResearchCouncil,NationalAcademyPress,WashingtonD.C.
Quinn,J. (1998):“RetirementPatternsandBridgeJobsin the1990’s.” manuscript,BostonCollege.
38
Rogowski, J. and L. Karoly (1997): “Health Insuranceand RetirementBehavior: Evidencefrom theHealthandRetirementSurvey,” RAND.
Ruhm,C.J,(1996):“Historical TrendsandtheFutureof OlderAmericans,” in The Handbook on Employ-ment and the Elderly, William H. Crown, ed.Greenwood.
Rust,J. (1989): “A DynamicProgrammingModel of RetirementBehavior,” in David A. Wise(ed),Eco-nomics of Aging, Chicago,University of ChicagoPressfor the NationalBureauof EconomicRe-search.
Rust,J. (1990): “Behavior of Male Workersat theendof theLife Cycle: An EmpiricalAnalysisof Statesand Controls,” in David A. Wise (ed), Issues in the Economics of Aging, Chicago,University ofChicagoPressfor theNationalBureauof EconomicResearch.
Rust,J.(1997)with C.Phelan:“How SocialSecurityandMedicareAffectRetirementBehavior in aWorldof IncompleteMarkets,” Econometrica, 65 781–831.
Schulz,J.H. (1995):The Economics of Aging, Auburn House,sixth edition.
Stevens,A. H. (1997a):“PersistentEffectsof JobDisplacement:Theimportanceof Multiple JobLosses”.Journal of Labor Economics, 15 165–188.
Stevens,A. H. (1997b):“Changesin EarningsInstability andJobLoss,” manuscript,YaleUniversity.
39