transitioning out of unemployment: analysis using the...

111AUSTRALIAN JOURNAL OF LABOUR ECONOMICS

Volume 17 • Number 2 • 2014 • pp 111 - 137

Transitioning Out of Unemployment: Analysis Using the ABS Longitudinal Labour Force Survey File*

Cristian Ionel Rotaru, AustralianBureauofStatistics

Abstract What affects the probability that an individual who has just entered unemployment finds employment within a given timeframe? Does the probability of exiting unemployment depend on the length of the individual’s unemployment spell? This paper reflects on these questions and analyses the transitions from unemployment of those aged 20-65 years, over the 2008-2010 period. The analysis makes use of the ABS Longitudinal Labour Force Survey (LLFS) file – a dataset that includes households that were followed for eight consecutive months during the said period. This paper is the first longitudinal analysis conducted on the file. Building on the job-search theoretical framework, the paper builds a model aimed at analysing the factors that influence transitions from unemployment. A range of methodological techniques are implemented, including the creation of time intervals and the subsequent discrete duration analysis; the adoption of the competing-risks framework, to account for the different forms of exits from unemployment; and the inclusion of random effects in the modelling of the observed as well as unobserved heterogeneity.

Keywords: Unemployment, Hazard Rate, Survival Analysis, Longitudinal LabourForceSurvey

JELClassification:J64,J60,C41,J24,J21

Address for correspondence: Cristian Rotaru, Senior Analyst, Australian Bureau of Statistics,ACT,2617,Australia.Email:[email protected]:TheauthorisgratefultoYovinaJoymungulPoorun,FranklinSoriano,RuelAbello, Dr Siu-Ming Tam, Professor James Brown and Dr Marcia Keegan for their valuablesuggestions and comments. Views expressed in this paper are those of the author and do notnecessarilyrepresentthoseoftheAustralianBureauofStatistics.Wherequoted,theyshouldbeattributedclearlytotheauthor.Remainingerrorsaretheauthor’s.*Anearlierversionofthispaperwaspresentedatthe2013/14ALMRWorkshopattheUniversityofMelbourneinFebruary2014.©CentreforLabourMarketResearch,2014.

112AUSTRALIAN JOURNAL OF LABOUR ECONOMICSVOLUME 17 • NUMBER 2 • 2014

1. Introduction Whataffectstheprobabilitythatanindividualwhohasjustbecomeunemployedfinds(full- or part-time) employmentwithin a given timeframe?Does theprobability ofexitingunemploymentdependonthelengthoftheindividual’sunemploymentspell?

This paper addresses these research questions and analyses the transitionsfromunemploymentinAustraliaforthoseaged20-65years,overathree-yearperiod,fromthebeginningof2008totheendof2010.Thestudymakesuseoftherecentlyconstructed ABS Longitudinal Labour Force Survey (LLFS) file and is the firstlongitudinalanalysisstudyconductedonthefile.

Relevance The researchquestionsbeingaddressedareofconsiderable relevanceonanumberof fronts. From an economic perspective, the questions are concernedwith one ofthemajorchallengesfacedbynationaleconomies;sustainedunemploymentimposingsubstantialeconomic,personal,andsocialcosts(Feldstein,1978).Its importanceiswidelyhighlighted in the literaturewithsomestudiesconsidering itas ‘thecentraldilemmaforsomeofthemostprosperouscountriesinEurope’(Sen,1997)and‘oneofthemainchallengesofthemodernerainboththedevelopedanddevelopingcountries’(TanselandTasci,2010).

FollowingHubbardet al.,(2014),thecostsofunemploymentcanbecategorisedasfollows:

Costs to the individual First,thereisthelossofincometotheunemployed.Althoughtheaffectedindividualsmaybeeligibleforunemploymentbenefitsandotherformsofgovernmentassistance,theyarestill likelytoexperienceserioushardships(e.g., inmeetingtheirmortgagepayments) and a substantial decline in their standards of living. In addition to theloss of current output, previous studies have shown that prolonged unemploymentis associated with other pecuniary and non-pecuniary effects, such as the loss ofself-esteem (Goldsmith et al., 1996); adverse effects on mental health (Jackson,1984),well-being, and life satisfaction (Arulampalam, et al., 2001;Carroll, 2007);skillsdeterioration(EdinandGustavsson,2008);higher likelihoodofexperiencingunemploymentincidentsinthefuture(Arulampalam,et al.,2000);andlowerwageswhenreturningtowork(Arulampalam,2001).

Costs to the society Besides imposingsubstantial costson the individual, sustainedunemploymentalsoaffects the individual’s family, society, and thewider community.Amongst others,unemploymenthasbeenfoundtobe linkedtosocialdeprivationandtobeafactorinschooldropouts,drugabuseandalcoholism,suicide,familybreak-ups,crimeandracialinequality(Wattset al.,2000;Sen,1997).

Costs to the economy A rising unemployment increases the Government fiscal costs, due to higherwelfare payments. These costs are in addition complemented by the potential

113CRISTIAN IONEL ROTARU

Transitioning Out of Unemployment: Analysis Using the ABS Longitudinal Labour Force Survey Fi le

losses of tax revenue and social security premiums that could havebeen collectedhad the unemployed individuals insteadworked.Apart from these, there are othermore indirectcosts, suchas thecostsassociatedwith retraining the individual, thedeteriorationofhumancapital,thedecreaseinspendingongoodsandservices(astheunemployedarelikelytoexperienceadeclineinspendingpower),andthesubsequentlossofpotentialnationaloutput. (See, forexample,Sen (1997)andHubbardet al.,(2014)formoredetails.)

Fromapolicyperspective,theresearchquestionshavesubstantialimplicationsforthedesignandtargetingoflabourmarketpoliciesandprograms.Understandingwhich (and theway inwhich) variables are associatedwith longer unemploymentandwhichgroupsof individualsaremore likely toexperience longunemploymentspellsand/orarelesslikelytoexitunemploymentintoemploymentcanprovideusefulinsightsfordesigningefficientassistanceprograms.

Fromatheoreticalperspective,theresearchquestionsarerelevantinsheddingsome light on the recent debate over the concept of hysteresis of unemployment,advocatedbyBlanchardandSummers(1986).Contrarytoconventionalwisdom,thetheory states that the natural rate of unemployment can be influenced by shifts inaggregatedemand,viathehysteresischannelsofactualunemployment.Thetheory,which has received little attention in the literature, has significant implications formacroeconomicandmonetarypolicy.Thepaperwillprovidesomevaluableinsightsin thisareabycontributing toone importantavenueof researchsuggestedbyBall(2011)namely,onexaminingthedynamicbehaviourofshort-termunemployment.

The contributions of the paper Due to the limitedavailabilityof longitudinal labour forcedata,a largeproportionof studies that examined the labour market dynamics in Australia are descriptivein nature. Amongst the longitudinal analyses, these were mainly conducted on arestrictednumberofdatasetsandusedretrospectivedurationdata.Mostnotably,thedatasetsinclude:• theABSSurveyofEmploymentandUnemploymentPatterns(SEUP),whichwasconductedinthe1990s(see,forexample,ChalmersandGuyonne,2001;Carino-Abello,et al.,2000);

• theLongitudinalSurveyofAustralianYouth,whichisrestrictedtoyoungpeople(see,forexample,HardinandKapuscinski,1997;ChapmanandSmith,1992);and

• theHouseholdIncomeandLabourDynamicsinAustralia(HILDA)Survey,whichbeganin2001(see,forexample,Watson,2008;Carroll,2007andCarroll,2006).

Thispapermakesthreecontributionstotheliterature.First,itusesanewandanimportant longitudinaldatasource,namely, theLLFSfile.Byhavingmorethan1.8millionrecordsandaround150,000householdsobservedonamonthlybasis,foraperiodofuptoeightmonths,thedatasetiswell-suitedforshort-termdynamicsofunemploymentanalyses.Thesamplealsocoversaperiodofconsiderableinterest,theGlobalFinancialCrisis(GFC).

The second is in its use of actual reported unemployment data (collectedmonthly) instead of retrospective data, which has been used by the other studies


thatanalysedthedurationofunemploymentinAustralia.Ashasbeenshownintheliterature,retrospectivedatacansufferfromrecallbias.

Thethirdcontributionisintheapproachtakenindealingwiththenewfeaturesof thedataset,whichincludethehighfrequencyof thedata, thelimitednumberofwaves,andthetypeofunemploymentinformation(reportedinsteadofretrospective).Mostnotably,theapproachincludestheuseofthejobsearchframeworktobuildthemodel,thecreationoftimeintervalsandthesubsequentdiscretedurationanalysis,theadoptionofthecompeting-risksframeworktoaccountforthedifferentformsofexitsfromunemployment,aswellastheinclusionofrandomeffectsinthemodellingoftheobservedandunobservedheterogeneity.

Modelling approach The paper adopts discrete duration models in a competing-risks framework. Thisapproachisappealinginthatitaccountsforthediscretenatureofthedurationdata–thedatabeingcollectedonamonthlybasis.Threetypesofexitsfromunemploymentareconsidered.Thefirstiswhentheunemployedindividualgetsemployedonafull-timebasis (denotedby‘FT’).Thesecond iswhen the individualgetsemployedonapart-timebasis(denotedby‘PT’).Thethirdiswhentheindividualleavesthelabourforce(denotedby‘OLF’).AsindicatedinFlinnandHeckman(1983)thealternativescouldbebehaviourallydifferentmarketstatesandassuchitisimportanttotreatthemseparately.

Theanalysisisdividedintotwoparts.Thefirstisfocusedonnon-parametrictechniques,andincludesrawhazardandsurvivalfunctions.Thesecondincorporatesobservedaswellasunobservedheterogeneityinmodellingthehazardfunction.Thisisdonebyusingdiscretedurationmodels,whereboththeordinarylogitaswellastherandomeffectslogitmodelsareexamined.

Theplanofthepaperisasfollows.Section2providesaconceptualbackground.Section3describes thedata.Themethodologyused in theanalysis isdescribed insection4.Section5presentstheresults.Section6concludes.

2. Conceptual background Thispapermakesuseofthejob-searchtheoreticalframework(see,Mortensen,1970;LippmanandMcCall,1976a;andLippmanandMcCall,1976b)toanalysethefactorsthataffectthedurationofunemploymentandtheprobabilityofexitingunemployment.Toseehowthemodelworks,consideranindividualwhohasjustbecomeunemployed.This could happen either because the individual has moved from employment tounemploymentorbecausehehastransitionedfrombeingoutofthelabourforcetoanactivestateofsearchingforwork.Assumingthattheindividualcontinuestosearchforwork, the aim is to explain the factors that determine the expecteddurationofremaininginthecurrentstateofunemployment.Thisexpecteddurationisassumedto be inversely proportional to the probability of moving from unemployment toemployment,which in turn is assumed to depend on two essential aspects: (1) theprobabilityofreceivingajoboffer,and(2)theprobabilityofacceptingthejobofferconditionalonhavingreceivedthejob.

Theprobabilityofreceivingajobofferisdetermined,amongstotherthings,by the demand for the individual’s labour in the current market (Holzer, 1986).



Amongstthethingsthatemployeeslookforareeducation,skills,previousoccupation,and experience – factors that are aimed tomake the individualmore attractive topotential employers.Other factors that shift the demand include the local demandconditions,suchas thebusinesscycleandthephaseof thelocaleconomy(e.g., therelativestrengthofthelocaleconomyandwhethertheeconomyisinrecession)(Foley,1997),aswellasthesearchintensityoftheindividual(Holzer,1986).

Theprobabilityofacceptingtheofferdependsontheindividual’sreservationwage.Thisisthelowestwageatwhichtheindividualwillacceptajoboffer.Thiswagedependsonsuchfactorsastheexpectedwageintheirparticularoccupation,maritalstatus,familycomposition,otherincomesinthehousehold,unemploymentbenefits,aswellastheprobabilityofreceivingfuturejoboffersandtheexpectedworkhorizon(Long,2009).Notethatthereservationwagecouldalsodependonsome(orall)thedeterminantsofthedemandforthelabourprovidedbytheindividual(Holzer,1986).

From an econometrics perspective it is important to consider both aspectswhenanalysing thedeterminantsofunemploymentduration.Failure to includeoneaspectmightresultinmissinganessentialcomponentofthemodelwhichinturncouldimpactontheresults.Notealsothatalthoughthejob-searchmodelsetsthetheoreticalframeworkforanalysingtheunemploymentduration,empiricalinterventionisoftenneededtodeterminetheeffects(ortheneteffects)ofthefactorsinthemodel.

Asanexampleoftheapplicationofthemodel,considertheeffectsofthelengthof unemployment spell on the probability of exiting unemployment.On one hand,a longerunemployment spell couldhavenegativeconsequenceson the individual’sprospectsoffindingwork(TanselandTasci,2010).Onereasonforthisisthelackofinvestmentinhumancapitalduetothelossofvaluableworkexperienceduringtheunemploymentspell.Anotherreasonisthepotentialchangeinattitude,astherepeatedfailuretosecureajobmightdiscouragetheindividualfromfully-exercisinghisskillsinfindingwork(Foley,1997).Finally,employersmaybemorereluctanttoofferjobofferstothosewithlongunemploymentspells.Thisisbecausetheymayperceivethelongspellofunemploymentasasignalof lowproductivity(Kroftet al.,2013).Allthesereasonsareassociatedwithalowerprobabilityofreceivingajoboffer.

On the other hand, the individual might decrease his reservation wage ashegets closer to theendofhisfinite timehorizon, soas to increasehisprospectsofsecuringa job (see,LippmanandMcCall,1976a;LippmanandMcCall,1976b).This in turnwill increase the conditional probabilityof acceptinganoffer.As thetwo effectsmove in opposite directions, it is not clear from theorywhich of themdominates.Empiricalapplicationwouldbeusefultosettlethisuncertainty.

3. Data and definitions Inthisstudy,unemploymentisdefinedas:Personsaged15yearsandoverwhowerenotemployedduringthereferenceweek,and• hadactivelylookedforfull-timeorpart-timeworkatanytimeinthefourweeksuptotheendofthereferenceweekandwereavailableforworkinthereferenceweek;or

• werewaiting tostartanew jobwithin fourweeks from theendof the referenceweekandcouldhavestartedinthereferenceweekifthejobhadbeenavailablethen.(AustralianBureauofStatistics,2013)


ThestudyusesdatafromtherecentlyconstructedABSLLFSfilethatcollectsmonthly information over a period of three years, from2008 to 2010. TheLLFSis compiled from 56 separate household surveys and records information aboutthe labourmarket participation and employment transitions for all individuals in ahousehold,overaperiodofuptoeightconsecutivemonths.Thoseinthescopeofthesurveyare15yearsofageorolder.

Onemainadvantageofusing theLLFSoverotherdatasets is itswealthofinformation – the file includes more than 150,000 households and more than 1.8millionrecords.Byrecordingmonthlydataforsucha largenumberofhouseholds,overaperiodofup toeightconsecutivemonths, thefile iswell-suited for in-depthanalysisofshort-termlabourmarketdynamics.

Forthepurposesofthisstudy,anumberofrestrictionswereimposed.First,thesamplewasrestrictedtoindividualswhowerebetween20to65yearsofageatthetimeofthefirstinterview.Thisrestrictionavoidstheinclusionofteenagers,whoselabourmarket behaviour is likely to differ substantially from that of theother agegroups,andtheinclusionofthoseolderthan65years,whoaremorelikelytoretire.Second,onlyprivatedwellingswereincluded.Boththeserestrictionswereimposeddue to the potential different labour market behaviour of the individuals in thesegroups.Third,duetotheaimofthestudyofanalysingthedurationofunemployment,the sample was further restricted to those who experienced unemployment atleast once during the interview period.Also note that similar to Foley (1997), forthe personswho experiencedmore than one spell of unemployment, only the firstspellofunemploymentwasconsidered.This approachavoids the serial correlationthat could result otherwise. As an extension, one could use multiple spells in theanalysisbytreatingthemasseparaterecords.Onewouldthenneedtodealwiththedependenceacrossspells1.Notealsothattheanalysisisrestrictedtotheindividualswhobecameunemployedduring the interviewperiod, i.e., restricted to the inflowsin unemployment. This avoids the complexity of dealing with left truncation andpotentiallyleft-censoring2.

Table 3.1 below shows the distribution of the different types of exits fromunemployment.Around47percentoftheunemployed,inscopeofthisanalysis,endinemployment,ofwhicharoundhalfendinfull-timeemploymentandhalfinpart-timeemployment.Around37percentoftheunemployedexitbyleavingthelabourforce,aproportionwhichissimilartowhatotherstudieshavefound(see,forexample,MorisonandBerezovsky,2001).Thebalanceof16.5percentremainsinunemployment.

Comparedtofemales,ahigherproportionofmalesendinemploymentandasubstantiallyhigherpercentageendinfull-timeemployment.Females,ontheotherhand,aremorelikelytoexitthelabourforce.Intermsofmaritalstatus,theresultsarenottoodifferentacrossthetwogroups.

1ThemethodsincludedinRotaru(2013)couldbeimplementedtodealwiththistypeofdependence.2Lancaster(1990)includessometechniquestodealwithleft-censoringandleft-truncation.Apartfrombeingconsiderablymorecomplex,themethodsrelyonretrospectivedata(whichmightsufferfromrecallbias),inthecaseofleft-truncation,andonadditionalassumptions,inthecaseofleft-censoring.



Table 3.1 - Percentage distribution of the exit states from unemployment spells

Exiting unemployment via: All Male Female Married Not marriedFull-orpart-timeemployment 46.9 51.3 42.7 43.3 41.9 Full-timeemployment 23.3 31.8 15.1 14.2 16.2 Part-timeemployment 23.6 19.5 27.6 29.1 25.8LeavingtheLabourForce(OLF) 36.6 31.2 41.8 43.4 39.9Remaininginunemployment 16.5 17.5 15.5 13.3 18.2

4. Methodology Thispapermodels the transitions from thefirstunemployment spellof individualsobservedduringtheeightconsecutive-monthsperioddescribedinsection3.Theaimsarefirst,tomodeltheprobabilityofexitingunemploymentandsecond,toaccountforthepotentialtimedependenceinthemodelling.

Inordertomeettheseaimsandtoadequatelydealwiththeparticularsofthedurationdata–thedataforeachhouseholdinthesurveybeingcollectedonamonthlybasis,foraperiodofuptoeightwaves–afewchallengesneedtobeaddressed.Thefirstchallengeisdealingwithleft-censored/truncateddurationdata,assomeindividualswerealreadyunemployedatthetimeofthefirst interview.Asaforementioned,thiswasaddressedbyfocusingontheindividualswhoenteredunemploymentduringtheinterviewperiod.

Thesecondchallengeisdealingwiththespellsofindividualswhohavenotyetexitedunemploymentatthetimeofthelastinterviewandwiththediscretenatureofthedata,thedataforeachindividualbeingcollectedonamonthlybasisforaperiodofuptoeightmonths.Toaddresstheseaspectsofthedata,thepaperconstructstimeintervalsandimplementsdiscretedurationmodellingtechniques.Asthesetechniquesrequireamorethoroughexposition,theyareelaboratedmorefullybelow.

The third challenge is controlling for the effects of covariates that are notavailableinthedataset,suchasmotivationandability.Thisisaddressedbyincludingrandomeffectsinthemodelling.

The fourth challenge is accounting for the different ways of exitingunemployment.Toaddressthis,thepaperadoptsthecompeting-risksframework.

Finally, the fifth challenge is dealing with the longitudinal aspect of theconstructed person-period dataset, in which each individual has multiple records,one for each period. For this challenge, it is important to note that by using themaximumlikelihoodestimation,theresultinglikelihoodfunctionbecomesaproductof independent Bernoulli likelihood functions. This means that relatively simpletechniquesareneededtoestimatetheparametersandthatthewell-knowninferentialstatisticscanbeappliedinthiscase.(See,SingerandWillett,2003;SingerandWillett,1993;andMuthénandMasyn,2005.)

Setting the Framework InageneralsettingconsiderarandomsampleofNunemployedindividualsthatareobservedoveraperiodoftime,whichinthecontextofthisstudyisuptoeightmonths


long.Thisperiodisallowedtovaryacrossindividualsandisrecordedonadiscretescale.Theaimistotrackindividualsfromthetimetheyenteredintounemploymentuntil theyfirst exit that state (i.e., the focus is on single spells) and to analyse thecharacteristics thatcontribute to thedifferences in thedurationexperiencedby theunitsinthesample.

Theobservedcharacteristicsarecapturedinthevector(X81,X82)8 ,whereinthecontextofthejob-searchmodelpresentedinSection2,thefirstsetofcharacteristics,assembledinvectorX1,determinetheprobabilityofbeingofferedajob,whereasthesecondset,assembledinvectorX2,influencetheprobabilityofacceptingthejoboffer.For example,X1 could include previous occupation, education,English proficiency,andpreviousworkexperience,whereas,X2couldincludefamilycomposition,sex,andmaritalstatus.Notethatthetwosetsneednotbemutuallyexclusive.Asemphasisedinsection2,itisimportanttocontrolforbothsetsofcharacteristicsinthemodel.Tofurthersimplifythenotation,allfactorsarecollapsedintovectorX=(X81,X82)8 .

At the end of the spell, each unit i in the sample is assumed to end up inone of four states: exits unemployment and becomes employed full-time (zi = 1);exitsunemploymentandbecomesemployedpart-time(zi=2);remainsunemployed(zi=3)–casewhentheobservationiscensored;orexitsthelabourforcealtogether(zi=4).Wherehere,aswellasintherestofthesection,subscriptidenotesthevaluesforindividuali.Tosimplifytheexpositionconsiderthecasewherethereisonlyoneexitstate,i.e.,onlyonedestination,casewhenthevaluesofzarecollapsedintoabinaryvariabley,whereyi=1ifindividualiexitsunemployment,i.e.,whenziP{1,2,4},andyi=0iftheindividualremainsintheinitialstateofunemployment,i.e.,whenzi=3.Notethatwithmorethanoneexitstateonecanusethecompeting-risksframeworkpresentedinSingerandWillett(2003)andAllison(2010).

Let T *i be a random variable capturing the duration of unemployment forindividuali,i.e.,thedurationuntilyi=1,andlet(0, tni

]=(0,t0]< (t0, tni]betheinterval

overwhichtheindividualisobserved.Astheanalysisisrestrictedtotheinflowsintounemployment,s0:=(0,t0]istheintervalinwhichtheindividualbecomesunemployed.Further, tni

captures the last time the individual’s responses are recorded (for thecensoredcases)orthefirsttimetheindividualisknowntohaveexitedunemployment.niisthenumberofwavesuntilindividualiexitsunemploymentoruntilcensoringafterbecomingunemployed.ItfollowsthatforcensoredcasesT *iisnotobservedandallthatisknownisthatT *i>tni

.Further,withthecurrentdataset,evenwhenyi=1,onedoesnotnecessarilyknowtheexactT *i,andratheronlyknowsthatT *i P(tni-1

,tni].This

isbecausethedurationofunemploymentisrecordedonamonthlybasis.Althoughtheapproachtakenbymostempiricalstudiesistomodeltheexact

timingofeventoccurrenceandtreatdurationasacontinuousrandomvariable,thispaper takes a different path and instead models the probability that T * falls intodiscrete time-intervals.Toseehowthisworks,consideragaintheinterval(0, tni

]=(0,t0]< (t0, tni

]overwhich individual i isobserved.Themain idea is to transformthecontinuoustimehorizonintoasequenceofdiscreteintervals.Now,partitionthe



intervalwheretheindividualis‘atrisk’ofleavingunemploymentinto niadjacentandmutuallyexclusive intervals, calledperiods, andwhich in thecontextof this studycorrespondtothetimeperiodbetweentwoconsecutivewavesofthesurvey,suchthat(t0, tni

]=(t0,t1]< (t1, t2]<...< (tni-1,tni

].Asahypotheticalexample,consideranindividualwhoisinterviewedforall

eightwavesofthesurveyandwhoseresponsesaregiveninfigure4.1.Inparticular,atthetimeofthefirstinterviewheisemployedfull-time,inthesecondheisemployedonapart-timebasis,thenhebecomesunemployedandheremainssountilbeforehissixthinterview,atwhichtimeheindicatesthatheisemployedpart-time.Duringthetimeofthefinaltwointerviewsheisoutofthelabourforce.

Figure4.2showshowtheintervals/periodswereconstructed.Notefirstthattheanalysisisfocusedontheperiodthatstartswiththeentranceintounemployment(i.e.,theperiodbetween0andt0)andendswiththetimeatwhichtheindividualexitsunemployment(i.e.,theperiodbetweent2andt3).Notealsothatastheinformationabout his labour force status is collected at the timeof the interview, it is unclearwhereexactlythetransitionbetweenthedifferentstatesoflabourforceoccurred.Asanexample,considerthefirstintervals0,wheretheindividualentersunemployment.Asitisonlyknownthathewasemployedpart-timeattime0andthatbythetimeofthenext interviewhehasenteredunemployment, it isunclearwhereexactly in theintervals0heenteredunemployment.Ratherthanpinpointingtotheexacttime,theanalysis insteadfocuseson intervalsorperiods.Using thesameapproach, thenexttwoperiods(s1ands2)areconstructedduringwhichheisstillunemployed.Finally,thefinalintervals3indicatestheperiodwhentheindividualexitsunemploymentintopart-timeemployment.

Bydisaggregating theduration intodiscrete timeperiods,onecanproceedwiththeanalysisbyconsideringthediscreterandomvariableTitakingvaluesfrom{1,2,…,ni},valueswhichcorrespondtotheniintervals,suchthatTi=j(i)wheneverT *i P(tj(i)-1, tj(i)]:=sj(i)forj(i)=1,2,…,ni.Notethathereafter,inordertosimplifythenotation,j(i)willbereplacedbyj.

ThisstrategyhasthereforeshiftedthefocusofanalysisfromthecontinuousrandomvariableT *tothediscrete-randomvariableT.Thisisappealinginthisstudybecause(1)therearealimitednumberofobservationsforeachindividualand(2)theexact timingof events isunknown.Although the timingof theunemployment andthetransitionfromthisstatemightbecapturedbyintervals,theirexacttimingisnotcovered in thedata.By treating the timeofdurationusingfinite intervals,discretesurvivalmodelsmakeadjustmentsforthislimitation(SingerandWillett,2003).


Figure 4.1 - Example of a response

Figure 4.2 - Example of a response – constructing the intervals

Modelling TheresearchquestionthenanalysesthedurationT,giventheobservedcharacteristicsX,theunobservedcharacteristicsW,andthedependenceovertime,aftercontrollingfortheeffectofcensoring.

Since T is by assumption intrinsically conditional (as it is assumed thatindividualsexperiencingthetargeteventhavenotexperienceditbefore),interestliesinderivingitsconditionalprobabilityfunction.Thehazardfunction,whichiswell-suitedandiscentralatanalysingdurationdata,canbeusedforthisscope.

In a discrete-time context, the hazard hij, is defined as the conditionalprobability that individual i exits the state of unemployment in period j, whichcorrespondstointerval(tj-1, tj],giventhattheeventhasnotoccurredpriortoperiodj.Mathematicallythisisgivenby

hij :=hij(xij ,wij ) :=P(Ti =j |Ti ≥j ,Xij=xij ,Wij =wij)(4.1)

whereXijandWijarethevectorofobservedandrespectivelyunobservedcovariatesfor individual i andwhere xij andwij denote some particular values ofXij andWij,respectively. Note that the hazard given in (4.1) is very flexible in that it includescovariatesthatareallowedtovaryovertime,asindicatedbythesubscriptj.

Animportantattractivefeatureofthehazarddescribedin(4.1)isthatsinceTiisdiscrete,thehazardissimplyapropensityandthus,onecanusediscretechoice

1 3 5 72 4 86

FT U U OLFPT U OLFPT

Wave

Response

0 t0 t1 t2 t3

s0 s1 s2 s3

1 3 5 72 4 86

FT U U OLFPT U OLFPT

Wave

Response



models to model the duration of unemployment. In particular the common logit(usedinthispaper),theprobit,andthecomplementarylog-logmodelscanbeused.Specifically,theconditionalprobabilityofexitcanbemodelledas

P(Ti =j |Ti ≥j ,Xij=xij ,fi ) =F(gm +x8ij b + fi )(4.2)

where gm is a polynomial that needs to be specified by the researcher and whichcapturesthedurationdependenceacrosstheperiodsinthedataset,F(.)isthecdf,andfi isarandomcomponentwithknowndistributionwhichcontrolsfortheeffectsoftheunobservedcovariates.Forexample,ifthelogisticdistributionisimposed,aftersomesimplemathematics,(4.1)and(4.2)leadtotheconditionallog-odds:

log[hij/1–hij ]=gm +x8i b + fi.(4.3)

Given the discrete-time framework and the limited period of analysis, this paperassumesthatgm =a1D1+...+amDm,whichisacompletegeneralspecificationfortime.Herea=(a1,...,am)8isavectorofcoefficientstobeestimated,Di areperiodindicators(durationdummies)withavalueof1forperiodiand0otherwise(i=1,...,m),andm isthenumberofriskperiodsinthedataset.Notealsothatbydroppingfifromequation(4.3)onegetsbacktothestandardlogitmodelappliedtodiscrete-durationdata.

5. Model application One of the main aims of this analysis is to investigate the factors that affect theprobabilityofexitingunemployment.As therearedifferentexitstates, theanalysisusesthecompeting-risksframeworktoexaminethedurationofunemployment.Underthisframework,oneimportantassumptionisthatafterconditioningontheregressorsincludedinthemodel,theoccurrenceofeachofthethreeeventsisnon-informativeforalltheotherstates.Thisassumptionallowsforrelativelysimpleparallelanalyseswheretheanalysisforeachexitstateisconductedonthesameperson-perioddatasetandwhere adjustments are onlymade to the censoring variable – i.e., treating thecompetingeventsascensored.

Thissectionhasthreeparts.Thefirstidentifiestheexplanatoryvariablesusedinmodelling the unemployment duration. The second presents the non-parametricresults.Includedarerawhazardratesforthewholesampleaswellasforsomekeycovariates and a life table. The third part focuses on themodelling results of thediscretehazardfunctionpresentedinsection4.Thispartincludesboththeordinarylogitaswellastherandomeffectslogitmodels.

Explanatory variables In choosing the explanatory variables to be included in the models, the paperconsideredtheinformationavailableinthedataset,therelevantliterature,andthejob-searchtheoreticalframework.Attentionhasbeenpaidtoincludeboth,variablesthatareexpectedtoaffecttheprobabilityofreceivingajobofferaswellasvariablesthatarelikelytoinfluencethereservationwage.


The following personal characteristics were included: age, gender, maritalstatus,familycomposition,educationalattainment,previousoccupation,whethertheindividualhaspreviousworkexperience,English-speakingability,andperiodsincearrivalinAustralia.

Fromthislist,educationalattainment,previousoccupation,English-speakingability,periodsincearrivalinAustralia,andpreviousworkexperienceareexpectedtoinfluencetheprobabilityofreceivingajoboffer.Thesevariablestendtobewidelyused in the literature asproxies for skill orhumancapital andare therefore likelyto affect the attractiveness for the individual’s supply of labour (Borland, 2000).Twoothervariables,namely,maritalstatusandfamilycompositionareexpectedtoinfluencethereservationwage(seeLong,2009;Carroll,2006).

Inadditiontothese,thepaperalsoincludesthegeographicallocationoftheindividual(stateandsectionofstate)andavariableindicatingthequarterwhentheindividualfirstbecameunemployed.ThislattervariableisexpectedtocapturesomeoftheeconomiccyclesandisparticularlyrelevantsincethedatasetincludestheGFC.Thevariablesareexpectedtoplayimportantrolesinaffectingboththeprobabilityofreceivingajobofferaswellastheconditionalprobabilityofacceptingit(see,Borland,2000;ProductivityCommission,2014;Long,2009).

Non-parametric results Beforeproceedingwiththeregressionanalysis,thissectionbeginswithsomesimplenonparametricplots–rawhazardandsurvivalfunctionplots–andalifetable.Theresultsarepresentedintable5.1,figure5.1,andfigure5.2.

Theresultsintable5.1indicatethattheproportionofindividualsmovingoutofunemploymentdecreasesthelongertheyareunemployed.Inparticular,thelargestproportionofexitsoccursduringthefirstperiodaftertheyhavebecomeunemployed(around60percent)anddecreasesthereafterforeachoftheconsecutiveperiods.

Figures5.2and5.3providetwographicaldisplaysofthisphenomenon.Figure5.2,whichdisplaystherawhazardratefunction,showsthatiftheheterogeneityacrossindividualsisignored,theriskofleavingunemploymentpeaksduringthefirsttimeintervalanddecreasesthereafter,prettysharplyatfirstbutquitesmoothlyafterthat.Figure5.3,whichshowsthesurvivalfunction,describesthesamephenomenon–thesurvivalfunctiondecliningmostsharplyduringthefirsttimeintervalanddecreasingatadecreasingratethereafter.TheresultsaresimilartothosefoundbyFoley(1997)andTanselandTasci(2010).

Tocomplement theseresults,appendixAincludesotherexploratoryresultsintheformofalifetableforthedifferenttypesofexitsandhazardfunctionsforsexandmarital status.The results indicate that after ignoring theheterogeneityacrossindividuals,thehazardratestendtobedecreasingwithtimeforallexittypesandthatthereseemtobedifferencesinthehazardfunctionsacrossbothsexandmaritalstatus.



Table 5.1- Life table describing the number of periods spent in unemployment

Number of those Proportion of those Unemployed at the beginning of Still Unemployed the period who unemployed at the left at the end at the end beginning of Censored at of the period of the period Time the period Who left the end of (Hazard (SurvivalPeriod Interval (Risk set) unemployment the period function) function)0 [0,1) 11,073 - - - 1.0001 [1,2) 11,073 6,719 854 0.607 0.3932 [2,3) 3,500 1,553 401 0.444 0.2193 [3,4) 1,546 588 216 0.380 0.1364 [4,5) 742 246 150 0.332 0.0915 [5,6) 346 99 85 0.286 0.0656 [6,8) 162 40 122 0.247 0.049

Figure 5.1- Hazard function for the duration of unemployment

Figure 5.2 - Survival function for the duration of unemployment

0.0

Time Interval

Haz

ard

Func

tion

0

0.4

0.8

2 4 5

0.6

0.2

1 3 6

0.0

Time Interval

Surv

ival

Fun

ctio

n

0

0.4

1.0

2 4 5

0.6

0.2

1 3 6

0.8


Modelling results Table5.2reportstheresultsoftheOrdinaryLogitmodel,whereastable5.3thoseoftheRELogitmodel.Inbothcasesthecoefficientsofthemodelsarereportedashazardratios.As theresultsof the twomodelsaresimilar,unlessotherwise indicated, thediscussionbelowisbasedontheOrdinaryLogitmodelresults.

Personal characteristics Theresultsdifferbyagegroupswiththeoddsofexitingintofull-timeemployment,aftercontrollingfortheeffectoftheothercovariates,generallydecreasingwithage.The odds of exiting into part-time employment or exiting the labour force, on theotherhand,aregenerallyincreasingwithage,althoughtheincreaseissmallacrossthesecond,third,andfourthagegroups.TheseresultsareconsistentwiththefindingsofBorland(2000).

Oneexceptiontothesepatternsistheyoungestgroup(aged20-24),whichhassignificantlyloweroddsofexitingintofull-timeemploymentthanthoseaged25-34andoneofthehighestoddsofexitingintopart-timeemployment.Thereishowever,nosignificantdifferencebetweentheiroddsofexitingthelabourforceandthoseoftheotheragegroups–betweentheagesof25and54.

Fortheolderworkers(aged55-65),theresultssuggestthatonceunemployedtheyareconsiderablylesslikelytoexitintofull-timeemploymentandmuchmorelikelytoexitoutofthelabourforcethanalltheotheragegroups.Toputitinperspective,whencomparedtotheoddsoftheyoungestgroup,theoldestgrouphavearound52percentloweroddsofexitingintofull-timeemploymentand73percenthigheroddsofexitingthelabourforce.Inthecontextofrecentdebatesonyouthunemploymentandmatureageworkersremaininginthelabourforcelonger,theseresultsareparticularlyrelevant.

Intermsofgenderandmaritalstatus,beingmaleincreasestheoddsofexitingintofull-timeemploymentby27percent,whereasbeingfemaleincreasestheoddsofexitingintopart-timeemploymentby40percent.Formales,beingmarriedincreasestheoddsofexitingintofull-timeemployment,butdecreasestheoddsofleavingthelabourforceandexiting intopart-timeemployment.Theresults indicate thatbeingmaleandmarriedalmostdoublestheoddsofexitingintofull-timeemployment.

Compared to those with only secondary school completed, having highereducation(BachelororTAFE)increasestheoddsofexitingintofull-timeemployment(byatleast23percent)anddecreasestheoddsofexitingthelabourforce(byatleast28percent).TheseresultssupportthefindingsofBorland(2000).

Overall,theindividualswithahigher-ormiddle-skilledlastoccupationaremorelikelytoexitintofull-timeemploymentandareless-likelytoexitintopart-timeemploymentthanthosewithalower-skilledlastoccupation–theseresultsbeinginlinewiththefindingsofProductivityCommission(2014)3.Therearetwoexceptionstothis.Thefirstrelatestotheresultsofthosewholastworkedasmachineoperators

3 The paper uses the categories defined in the Productivity Commission (2014) report, whereoccupations are classified as higher-skilled occupations–managers andprofessionals;middle-skilledoccupations–techniciansandtradeworkers;communityandpersonalservicesworkers;andclericalandadministrativeworkers;andlower-skilledoccupations–salesworkers;machineryoperatorsanddrivers;andlabourers.



or drivers and which are not significantly different from those who last workedin a higher-skilled occupation.A reason for thismight be due to the employmentgrowthinminingoperationsthatoccurredduringtheperiodcoveredbythefileandwhichislikelytohaveimpactedonthedemandforlabourfromthisoccupation.Thesecondiswithregardstothosewholastworkedascommunityorprofessionalserviceworkers andwhoareassociatedwithamuch lowerpropensityof exiting into full-timeemploymentandamuchhigherpropensityofexitingintopart-timeemployment.Thesefindingsreflecttheparticularnatureofworkinthisoccupation,withworkersbeingmorelikelytoworkpart-timeandwithwomenandolderworkersconstitutingamajorpartoftheworkforceABS(2011).

Oneparticularresultthatstandsoutistheconsiderablyloweroddsofexitingintoemploymentandtheconsiderablyhigheroddsofexitingthelabourforceforthosewhohavelastworkedmorethantwoyearsagoorwhoarelookingforworkforthefirsttime.

Overall,whencomparedtocoupleswithnochildrenandnootherdependents,coupleswith dependents (childrenor other dependents) have lower odds of exitingunemployment via full-time work and higher odds of exiting unemployment viapart-time work or into the OLF status. Based on the magnitude of the estimatedcoefficients,one-parentfamilieswithchildrenareassociatedwiththelowestoddsofexitingunemploymentviafull-timeemployment,followedbycoupleswithchildren.One-parent familieswithchildrenhavealso thesecondhighestoddsofexiting thelabourforce,surpassedonlybyone-parentfamilieswithnochildrenunder15years,butwithotherdependents.

Those from a non-English speaking background are associatedwith loweroddsof exitingunemploymentvia full-timeemployment and theodds are lower iftheindividualarrivedrecently(i.e.,after2001).TheseresultssupportthosefoundinCarroll(2006)andBorland(2000).Non-Englishspeakerswhoarrivedafter2001areassociatedwithhigheroddsofexitingunemploymentviapart-timeemploymentorintotheOLFstatus.

Geographical location The results differ by geographical location, with NT, ACT,WA, and Qld (in thatorder)beingassociatedwiththehighestoddsofexitingunemploymentviafull-timeemployment.SAandTasmania,ontheotherhand,havethelowestodds.ThesefindingsmirrorthedifferencesinunemploymentratesacrossstatesreportedinBorland(2000).TheyarealsoconsistentwiththefindingsofProductivityCommission(2014)–theminingstateshavingthehighestemploymentgrowthratesoverthelastdecade.Aninterestingfindingis thatfortheotherexit typesthedifferencesintheoddsacrossstates are very small. This is informative in that although other previous studies,summarised by Borland (2000), report similar differences in employment acrossstates,theydonotdistinguishbetweenfull-timeandpart-timeemployment.Thispaperprovidesevidencethatthesedifferencesaremainlydrivenbyfull-timeemployment.

In addition, the results indicate that, compared to the balance of the state/territory,thecapitalcitiesareassociatedwithsignificantlyhigheroddsofexitingintofull-timeemployment,significantlyloweroddsofexitingintopart-timeemployment,andsimilaroddsofexitingthelabourforce.


Initial unemployment quarter The results for the initial unemployment quarter variable suggest a potentialGFCeffect.Thisisindicatedbythebigchangeinthemagnitudeoftheoddsaroundthefirstquarterof2009.Inparticular,theindividualswhohaveenteredunemploymentduringthisquarterhavemuchloweroddsofexitingunemploymentviafull-timeorpart-timeemployment.Thoseindividualswhohaveenteredunemploymentduringthefirsttwoquarters of 2009 are also associatedwith the largest oddsof exiting into theOLFstatus.Thisisaninterestingresultthatwarrantsfurtherinvestigation.

Baseline hazard function For the time interval, the results provide evidence that the probability of exitingunemployment into employment depends on the duration of the spell, which isconsistent with the findings of other studies, such as Caroll (2006), Borland andJohnston (2011), and Trivedi and Hui (1987). The results were maintained aftercontrolling for unobserved heterogeneity. In addition, the baseline logit hazardfunctionconfirmsthepreviousresults,asitisgenerallydecreasingovertimeforallexittypes.Thisisinterestingresultasitindicatesthat,ononehand,contrarytothediscouragedjobseekereffect,theconditionalprobabilitiesofexitingintoemploymentaswellasthatofexitingintotheOLFstatusarelowerwithalongerunemploymentspell.On theotherhand, thediscouraged job seeker effect cannotbe ruledout, asthe decline in the hazard rates, of exiting unemployment into employment, couldalsobe explainedby a decline in search intensity (the repeated failure to secure ajobdiscouragingtheunemployedindividualsfromfully-exercisingtheirskillswhensearching forwork).Besides, the employer’s perceptionmight alsobe a factor, theemployersbeingreluctanttoofferjobstothosewithlongunemploymentspells.

Ordinary vs RE model Table5.3showstheestimationresultsoftherandomeffects(RE)logitmodel.TheRElogitmodelwasappliedtoaccountforunobservedcovariates,likemotivation,ability,ortheintensityofjobsearch.

Overall, the RE logit model results are similar to those of the ordinaryprobitmodel,withthelikelihoodratiotestrejectingthenullhypothesisinfavourofthe random effectsmodel only in the case of the exits into full-time employment.However,eveninthiscase,therejectionisonlymarginalatthe5percentsignificancelevel.WhencomplementingtheseresultswiththereportedAICandBIC,theredoesnotseemtobesufficientevidencetosupportselectingonemodelfromamongthetwo(i.e.,ordinaryandrandomeffectsmodels).Otherstudieshavefoundsimilarresults,see,forexample,BorlandandJohnston(2011)andTanselandTasci(2010).

.



Table 5.2 - Effect of covariates on exit rates – Ordinary logit model – hazard ratios

Variables FT PT OLFAgeGroup(20-24 years) 25-34years 1.252*** 0.736*** 0.960 35-44years 1.043 0.769*** 0.997 45-54years 0.925 0.807*** 1.043 55-65years 0.482*** 1.038 1.730***Sex(Female) Male 1.273*** 0.712*** 0.856**MaritalStatus(Not married) Married 0.812** 1.250** 1.168* Male*married 2.309*** 0.668*** 0.703***Education(Secondary completed) Bachelor 1.242** 1.080 0.718*** TAFE 1.229*** 1.024 0.708*** Secondarynotcompleted 1.147 0.781*** 0.858** Missing 1.322*** 0.671*** 0.776***FamilyComposition(Couple, no children, no dependents) Couple,nochildren,otherdependents 0.691*** 1.265** 1.242** Couple,children,otherdependents 0.640*** 1.276*** 1.449*** Oneparent,children,otherdependents 0.486*** 1.024 1.498*** Oneparent,nochildren,otherdependents 0.753 1.061 1.621*** Oneparent,nochildren,nootherdependents 0.828 0.855 1.177 Loneperson 0.814* 0.971 1.023 Others 0.968 1.358*** 1.005LastOccupation(Professional) Manager 0.934 0.589*** 1.155 Technician 1.184* 0.676*** 1.207* Community 0.531*** 1.274*** 1.259** Clerical 1.017 0.736*** 1.105 Sales 0.728*** 0.874 1.140 Operator 0.955 0.621*** 1.069 Labourer 0.590*** 1.004 1.289*** Lastworkedmorethan2yearsago 0.046*** 0.154*** 3.165*** Firsttimelookingforwork 0.098*** 0.164*** 2.751*** Missing 0.142*** 0.127*** 18.897***State(NSW) Vic 0.897 1.138** 0.989 Qld 1.206*** 1.015 0.850** SA 0.799** 1.076 0.957 WA 1.344*** 1.082 1.096 Tas 0.765** 1.169 1.018 ACT 1.455*** 1.039 1.020 NT 2.036*** 0.946 1.043LanguageSpoken(English) Non-English 0.770*** 0.907 1.062YearofArrivalinAustralia(Before 2001) After2001 0.801** 1.223** 1.204**InitialUnemploymentQuarter(Quarter 1, 2008) Quarter2,2008 0.854 0.885 1.147 Quarter3,2008 0.853 0.923 0.968 Quarter4,2008 0.837** 0.874 0.932 Quarter1,2009 0.613*** 0.656*** 1.189* Quarter2,2009 0.763** 0.820* 1.218* Quarter3,2009 0.813* 0.829 1.065


Table 5.2 - Effect of covariates on exit rates – Ordinary logit model – hazard ratios (continued)

Variables FT PT OLF Quarter4,2009 0.745*** 0.934 0.999 Quarter1,2010 0.651*** 0.838* 0.945 Quarter2,2010 0.964 0.771** 1.057 Quarter3,2010 0.720** 0.961 0.970 Quarter4,2010 0.892 1.047 0.881TimeInterval 1 0.333*** 0.409*** 0.145*** 2 0.212*** 0.268*** 0.128*** 3 0.179*** 0.214*** 0.125*** 4 0.147*** 0.173*** 0.115*** 5 0.149*** 0.136*** 0.094*** 6 0.020*** 0.274*** 0.082***AreaofUsualResidence(Balance of state/territory) CapitalCity 1.203*** 0.810*** 0.994Loglikelihood -6307.3 -6676.4 -7895.2AIC 12724.6 13462.7 15900.4BIC 13151.5 13889.7 16327.4Observations(n) 17369 17369 17369

Note: ***p<0.01;**p<0.05;*p<0.10.Referencecategoryisinbrackets.Robuststandarderrorswerecomputed.

Table 5.3 - Effect of covariates on exit rates – Random effects logit model – hazard ratios

Variables FT PT OLFAgeGroup(20-24 years) 25-34years 1.297*** 0.705*** 0.958 35-44years 1.053 0.744*** 0.997 45-54years 0.918 0.784** 1.045 55-65years 0.432*** 1.044 1.775***Sex(Female) Male 1.326*** 0.680*** 0.849**MaritalStatus(Not married) Married 0.796* 1.279** 1.172*Male*married 2.622*** 0.639*** 0.696***Education(Secondary completed) Bachelor 1.260** 1.095 0.708*** TAFE 1.252** 1.036 0.697*** Secondarynotcompleted 1.163 0.758*** 0.855** Missing 1.383*** 0.645*** 0.769***FamilyComposition(Couple, no children, no dependents) Couple,nochildren,otherdependents 0.644*** 1.307** 1.252** Couple,children,otherdependents 0.589*** 1.313*** 1.473*** Oneparent,children,otherdependents 0.440*** 1.015 1.525*** Oneparent,nochildren,otherdependents 0.726 1.071 1.650*** Oneparent,nochildren,nootherdependents 0.794* 0.834 1.184 Loneperson 0.801* 0.963 1.022 Others 0.962 1.412*** 1.003



Table 5.3 - Effect of covariates on exit rates – Random effects logit model – hazard ratios (continued)

Variables FT PT OLFLastOccupation(Professional) Manager 0.910 0.544*** 1.158 Technician 1.220* 0.640*** 1.217* Community 0.480*** 1.357** 1.270** Clerical 1.004 0.702*** 1.112 Sales 0.686*** 0.858 1.141 Operator 0.959 0.576*** 1.071 Labourer 0.536*** 1.012 1.300*** Lastworkedmorethan2yearsago 0.035*** 0.125*** 3.370*** Firsttimelookingforwork 0.077*** 0.134*** 2.881*** Missing 0.116*** 0.107*** 21.200***State(NSW) Vic 0.884 1.163* 0.988 Qld 1.255*** 1.025 0.841** SA 0.771** 1.080 0.953 WA 1.413*** 1.096 1.096 Tas 0.742** 1.197 1.017 ACT 1.550*** 1.052 1.022 NT 2.277*** 0.936 1.039LanguageSpoken(English) Non-English 0.739*** 0.890 1.065YearofArrivalinAustralia(Before 2001) After2001 0.782* 1.279* 1.213**InitialUnemploymentQuarter(Quarter 1, 2008) Quarter2,2008 0.824 0.862 1.155 Quarter3,2008 0.817 0.908 0.966 Quarter4,2008 0.808** 0.862 0.931 Quarter1,2009 0.567*** 0.616*** 1.200* Quarter2,2009 0.730** 0.790* 1.234* Quarter3,2009 0.781* 0.791* 1.071 Quarter4,2009 0.704*** 0.919 0.999 Quarter1,2010 0.600*** 0.814* 0.944 Quarter2,2010 0.931 0.740** 1.060 Quarter3,2010 0.680** 0.944 0.970 Quarter4,2010 0.870 1.049 0.875TimeInterval 1 0.284*** 0.362*** 0.135*** 2 0.207*** 0.267*** 0.125*** 3 0.196*** 0.232*** 0.125*** 4 0.173*** 0.201*** 0.119*** 5 0.187*** 0.166*** 0.100*** 6 0.026*** 0.350** 0.090***AreaofUsualResidence (Balance of state/territory) CapitalCity 1.237*** 0.783*** 0.993Loglikelihood -6305.7 -6675.7 -7894.8Sigma 0.887 0.851 0.413Rho+ 0.193** 0.180 0.049AIC 12723.4 13463.5 15901.6BIC 13158.1 13898.2 16336.3Observations(n) 17369 17369 17369

Note:***p<0.01;**p<0.05;*p<0.10;+=likelihoodratiotestforrho=0.Referencecategoryisinbrackets.Robuststandarderrorswerecomputed.


6. Concluding remarks Buildingonthejob-searchtheoreticalframework,thispaperexaminesthetransitionsfromunemploymentusingtheABSLongitudinalLabourForceSurveyfile.Thefilecoversathree-yearperiod,fromthebeginningof2008totheendof2010,andincludesmorethan1.8millionrecordsfromaround150,000householdsobservedoveraperiodofuptoeightconsecutivemonths.

From a methodological perspective, the paper implements the followingtechniques to dealwith the specific features of the data and of the analysis. First,theanalysis is restricted to thosewhobecomeunemployedduring theeight-monthinterviewperiod.Thisapproachavoidstherelianceonretrospectiveinformationandthemodelcomplexitiesinvolvedwithdealingwithleftcensoring/truncation.Second,tocapturethediscretenatureofthedurationdataandtodealwithleftcensoring,discretedurationmodelsareimplemented.Thisstrategyshiftsthefocusoftheanalysisfrommodelling a continuous randomduration variable to that of an analysis conductedontimeintervals.Third,toseparatelyconsiderthedifferentunemploymentexits,theanalysisadoptsthecompeting-risksframeworkandseparatelyexaminesthetransitionintothreedifferentexitstates:full-timeemployment,part-timeemployment,andoutof the labourforce.Finally, toaccountforunobservedheterogeneity,randomeffectmodelsareconsidered.

Fromanempiricalperspective–andalso in response to the twoquestionsposedintheintroduction–thefollowingcanbenoted.First,theresultsdifferbyagegroupswith olderworkers (aged 55-65) having significantly lower odds of exitingunemploymentintofull-timeemploymentandwithmuchhigheroddsofexitingthelabourforce.Inthecontextofdebatesonyouthunemploymentandmatureageworkersremaininginthelabourforcelonger, themarkedlyhigherexitrates toemploymentfortheyoungestgroup(aged20-24)compared,inparticular,totheoldestgroupareparticularlyrelevant.

Second,thehazardratesaresignificantlyaffectedbythepotentialdeterminantsof the probability of receiving a job offer, such as education, English languageproficiency,workexperience,and lastoccupation.Thesecharacteristics,asBorland(2000)indicates,areoftenusedasproxiesforskillorhumancapital.Inparticular,theresultsindicatethathighereducation,Englishlanguageproficiency,workexperience,andahigher-skilledlastoccupationareallassociatedwithhigheroddsofexitingintofull-timeemployment.Oneinterestingfindingisthatthosewholastworkedtwoyearsagoorlonger,orarefirsttimelookingforwork,havesubstantiallylowerhazardratesofexitingintoemploymentandsubstantiallyhigherratesofexitingthelabourforce.

Third, the variables which are likely to affect the reservation wage, inparticularmaritalstatusandfamilycomposition,areequallyimportantinexplainingthe heterogeneity across individuals. For males, beingmarried increases the oddsof exiting into full-time employment, but decreases the odds of exiting into part-time employment and leaving the labour force. For females, the opposite resultsare observed. Note also the markedly different results for one-parent families. Inparticular,whencomparedtotheothertypesoffamilies,loneparentswithchildrenhavethelowestoddsofexitingintofull-timeemployment.Theseresultspointtowardstheimportanceofconsideringbothtypesofvariableswhenanalysingthedurationof



unemployment,thoseaffectingthedemandofanindividual’slabouraswellasthoseaffectingthereservationwage.

Fourth, the resultsdiffersignificantlyacross regionswith theACTand theminingstatesofWA,NT,andQldbeingassociatedwiththehighestoddsofexitingunemployment into employment. Interestingly, a recent Productivity Commissionreportongeographicallabourmobilityfoundthatexactlythesamestatesandterritoriesare associated with the lowest unemployment rates and the strongest employmentgrowth,ProductivityCommission(2014).Theseandthepreviousresultscanbeuseful,forexample,intargetingthegroupswithlowprobabilityofleavingunemploymentforemploymentorthosewithhighprobabilityofexitingthelabourforce.

Fifth,thehazardrateseemstobeaffectedbyeconomiccycles.Mostnotably,the results point towards potential GFC effects. These effects are depicted by thesuddendropintheoddsofexitingintoanytypeofemploymentduringthelastquarterof2008andbythesuddenpeakofexitingthelabourforceduringthelastquarterof2008andthefirstquarterof2009.

Sixth, the results indicate that the probability of exiting unemploymentdepends on the length of unemployment spell with the baseline hazard functiondecreasingovertimeforallexittypes.Theresultspersistevenaftercontrollingforunobservedheterogeneity.Inthelightofthehysteresisofunemploymenttheory,theseresultsprovidesomeevidenceforpathdependenceofunemployment.

Theshapeofthebaselinehazardfunctionssuggests,ononehand,thatcontrarytothediscouragedjobseekereffect,theprobabilitiesofexitingintoemploymentaswellastheprobabilityofexitingintotheOLFstatusarelowerwithalongerunemploymentspell.Ontheotherhand,thediscouragedjobseekereffectcannotberuledout,asthedeclineinthehazardrates,ofexitingunemploymentintoemployment,couldalsobeattributed toadecline insearch intensity.Besides, theemployer’sperceptionmightalsobeafactor.Thisisaninterestingresultthatisworthyoffurtherinvestigation.

Finally, the results indicate that the groups in most need of assistance(specifically,forsecuringfull-timeemployment)includethosewholastworkedtwoyearsagoorlonger,thosewhoarefirsttimelookingforwork,singleparentfamilieswithchildren,andtheindividualsaged55-65years.Theseareinterestingresultsthatdeservefurtherexaminations.ItshouldalsobenotedthatasthetimeperiodcoveredbythisstudyisrelativelyshortandasitincludestheGFC,afurtherextensionwouldbetoexaminetheseresultsoveralongerandadifferenttimeperiod.


Appendices A1. Analysis results

Table A.1 - Hazard and survival functions for exiting unemployment

Hazard FunctionsTime Period Any Exit FT PT OLF0 - - - -1 0.607 0.171 0.173 0.2632 0.444 0.123 0.125 0.1963 0.380 0.107 0.102 0.1714 0.332 0.084 0.082 0.1665 0.286 0.081 0.064 0.1426 0.247 0.012 0.105 0.130

Survival FunctionsTime Period Any Exit FT PT OLF0 1.000 1.000 1.000 1.0001 0.393 0.829 0.827 0.7372 0.219 0.727 0.723 0.5933 0.136 0.650 0.649 0.4914 0.091 0.595 0.596 0.4105 0.065 0.547 0.558 0.3526 0.049 0.540 0.500 0.306

Figure A1.1 - Hazard function for the duration of unemployment by sex

0.0

Period

Haz

ard

Func

tion

0

0.4

0.8

2 4 5

0.6

0.2

1 3 6

Males Females



Figure A1.2 - Hazard function for the duration of unemployment by marital status

A2. List of variables Thissectionincludesthelistofvariablesusedinthemodels.

StateNSWVicQldSAWATasNTACTSex MaleFemaleAge Group20-2425-3435-4445-5455-65Marital Status MarriedNotmarriedOccupation ManagersandadministratorsProfessionalsTechniciansandtradeworkersCommunityandprofessionalserviceworkersClericalandadministrativeworkersSalesworkersMachineryoperatorsanddriversLabourersLastworkedmorethan2yearsagoFirsttimelookingforworkMissing

0.0

Period

Haz

ard

Func

tion

0

0.4

0.8

2 4 5

0.6

0.2

1 3 6

Married Not Married


A2. List of variables (continued)

Education Degree–BachelororPostgraduatedegreeTAFE–DiplomaorCertificateSecondaryschoolcompletedSecondaryschoolnotcompletedLanguage Spoken EnglishNon-EnglishYear of arrival in Australia (non-English speakers)Before2001After2001Initial Unemployment Quarter Quarter1,2008Quarter2,2008Quarter3,2008Quarter4,2008Quarter1,2009Quarter2,2009Quarter3,2009Quarter4,2009Quarter1,2010Quarter2,2010Quarter3,2010Quarter4,2010Family CompositionNote:Firstdigit:familySeconddigit:numberofparents(1–singleand2–couple)Thirddigit:whetherthefamilyhaschildrenunder15Fourthdigit:whetherthefamilyhasotherdependents0000–Loneperson1100–Oneparentfamilywithnochildrenandnootherdependents1101–Oneparentfamilywithnochildrenunder15andotherdependents1111–Oneparentfamilywithchildrenunder15andotherdependents1200–Couplefamilywithnochildrenandnootherdependents1201–Couplefamilywithnochildrenunder15andotherdependents1211–Couplefamilywithchildrenunder15andotherdependents9999–OthersTime Interval Note:thissplitstheperiodof8wavesintointervals123456(i.e.,thelasttwoperiodswerecombinedbecauseofthesmallsamplesizes)Area of Usual Residence CapitalcityBalanceofstate/territory



References Allison,P.D.(2010),Survival Analysis Using SAS®: A Practical Guide,2nd(ed.),Cary

NC:SASInstituteInc.Arulampalam,W.(2001),‘IsUnemploymentReallyScarring?EffectsofUnemployment

ExperiencesonWages’,The Economic Journal,111(475),F585-F606.Arulampalam,W., Booth, A. and Taylor, M. (2000), ‘Unemployment Persistence’,

Oxford Economic Papers,52,24-50.AustralianBureauofStatistics(2013),LabourForce,Australia,Cat.no.2020,ABS,

Canberra.AustralianBureauofStatistics(2011),AustralianSocialTrendsSeptember2011,Cat.

no.4102.0,ABS,Canberra.Borland,J.andJohnston,D.(2010), ‘HowDoesaWorker’sLabourMarketHistory

AffectJobDuration’,WorkingPaperNo.6/10,MelbourneInstituteWorkingPaperSeries,UniversityofMelbourne,Melbourne.

Carino-Abello,A.,Pederson,D.andKing,A.(2000),‘LabourMarketDynamicsinAustralia.AnApplicationUsingthe1994-1997SurveyofEmploymentandUnemploymentPatterns’,Occasional Research Paper,Cat.no.6293.0.00.006,AustralianBureauofStatistics,Canberra.

Carroll,N.(2006),‘ExplainingUnemploymentDurationinAustralia’,The Economic Record,82,298-314.

Carroll, N. (2007), ‘Unemployment and PsychologicalWell-Being’,The Economic Record,83,287-302.

Chalmers, J. andGuyonne,K. (2001), ‘Moving fromUnemployment to PermanentEmployment: Could a Casual Job Accelerate the Transition?’, Australian Economic Review,34,415-36.

Chapman,B.andSmith,P.(1992),‘PredictingtheLong-TermUnemployed:APrimerfortheCES’inGregory,R.andKarmel,T.(eds.)Youth in the Eighties,263-281,Canberra.

CupplesL.A.,D’agostinoR.B.,AndersonK.andKannel,W.B.(1985),‘ComparisonofBaselineandRepeatedMeasureCovariateTechniquesintheFraminghamHeartStudy’,Statistics in Medicine;7,205-18.

Edin,P.-A.andGustavsson,M.(2008),‘TimeOutofWorkandSkillDepreciation’,Industrial and Labor Relations Review,61(2),163-180.

Feldstein,M. (1978), ‘The Private and Social Costs of Unemployment’,American Economic Review,PapersandProceedingsoftheNinetiethAnnualMeetingoftheAmericanEconomicAssociation,68(2),155-158.

Flinn, C.J. and Heckman, J.J. (1983), ‘Are Unemployment and Out of the LabourForce Behaviourally Distinct Labour Market States?’, Journal of Labour Economics,1,28-42.

Foley,C.M.(1997),‘DeterminantsofUnemploymentDurationinRussia’,DiscussionPaperNo.779,YaleUniversity.

Goldsmith,A.H.,Veum,J.R.andDarity,W.Jr.(1996),‘TheImpactofLabourForceHistoryonSelf-Esteemand itsComponentParts,Anxiety,Alienation, andDepression’,Journal of Economic Psychology,17,183-230.


Harding, A. and Kapuscinski, C. (1997), ‘Young Australians in Unemployment:DespairbyAnyOtherName’,DiscussionPaperNo.359,CentreforEconomicPolicyResearch,ANU,Canberra.

Holzer, H.J. (1986), ‘Black Youth Nonemployment: Duration and Job Search’, inFreeman, R.B. and Holzer, H.J., ‘The Black Youth Employment Crisis’,UniversityofChicagoPress,23-73.

Hubbard,R.,Garnett,A.,Lewis,P.andO’Brien,A.(2014),Macroeconomics,3rd(ed.),PearsonEducation,Sydney.

Jackson,P.R. andWarr,P.B. (1984), ‘Unemployment andPsychological Ill-Health:TheModeratingRoleofDurationandAge’,Psychological Medicine,14(3),605-614.

Junankar,P.N.andKapuscinski,C.(1991),‘TheIncidenceofLongTermUnemploymentinAustralia’,Australian Bulletin of Labour,17,325-352.

Kroft,K.,Lange,F.andNotowidigdo,M.J.(2013),‘DurationDependenceandLabourMarket Conditions: Evidence from a Field Experiment’, The Quarterly Journal of Economics,128(3),1123-1167.

Lancaster, T. (1990), The Econometric Analysis of Transition Data, CambridgeUniversityPress.

Lippman, S. and McCall, J. (1976a), ‘The Economics of Job Search: A Survey’,Economic Inquiry,14,347-368.

Lippman, S. and McCall, J. (1976b), ‘The Economics of Job Search: A Survey’,Economic Inquiry,14,155-189.

Long,K. (2009), ‘UnemploymentDurations: Evidence from theBritishHouseholdPanelSurvey’,Economic & Labour Market Review,3(10),48-54.

Morrison, P. and Berezovsky, O. (2001), ‘LabourMarket Risk and the Regions:Evidence from the Gross Labour Flows’ inMartin, R. andMorrison, P.(2003 eds.),Geographies of Labour Market Inequality, TSO Publishing,Routledge:UK.

Mortensen,D.T.(1970),‘JobSearch,theDurationofUnemployment,andthePhillipsCurve’,The American Economic Review,60(5),847-862.

Muthén,B.andMasyn,K.(2005),‘Discrete-TimeSurvivalMixtureAnalysis’,Journal of Educational and Behavioural Statistics,30(1),27-58.

Productivity Commission (2014), ‘Geographic LabourMobility’,Research Report,Canberra.

Rotaru,C.I.(2014),‘DiscreteChoicePanelDataModellingUsingtheABSBusinessLongitudinalDatabase’,Methodology Advisory Committee Papers,Cat.no.1352.0.55.139,AustralianBureauofStatistics,Canberra.

Sen,A.(1997),‘Inequality,UnemploymentandContemporaryEurope’,International Labour Review,136(2),155-171.

Singer,J.D.andWillett,J.B.(2003),Applied Longitudinal Data Analysis: Modelling Change and Event Occurrence,OxfordUniversityPress,NewYork.

Singer,J.D.andWillett,J.B.(1993),‘It’sAboutTime:UsingDiscrete-TimeSurvivalAnalysistoStudyDurationandtheTimingofEvents’,Journal of Educational Statistics,18(2),155-195.

Tansel,A.andTasci,H.M.(2010),‘HazardAnalysisofUnemploymentDurationbyGenderinaDevelopingCountry:TheCaseofTurkey’,Labour,24(4),501-530.



Trivedi,P.andHui,W.(1987),‘AnEmpiricalStudyofLong-TermUnemploymentinAustralia’,Journal of Labour Economics,5,20-42.

Watson,I.(2008),‘LowPaidJobsandUnemployment:ChurningintheAustralianLabourMarket, 2001 to 2006’,Australian Journal of Labour Economics,11,71-96.

Watts,M.J.andMitchell,W.F.(2000),‘TheCostsofUnemploymentinAustralia’,The Economic and Labour Relations Review,11(2),180-197.

transitioning out of unemployment: analysis using the...

Documents