• No results found

Job displacement and crime: Evidence from Norwegian register data

N/A
N/A
Protected

Academic year: 2022

Share "Job displacement and crime: Evidence from Norwegian register data"

Copied!
15
0
0

Laster.... (Se fulltekst nå)

Fulltekst

(1)

ContentslistsavailableatScienceDirect

Labour Economics

journalhomepage:www.elsevier.com/locate/labeco

Job displacement and crime: Evidence from Norwegian register data

Mari Rege

a,b

, Torbjørn Skardhamar

b,c

, Kjetil Telle

b,d,

, Mark Votruba

e,b

aUniversity of Stavanger, Stavanger, Norway

bResearch Department, Statistics Norway, Oslo, Norway

cUniversity of Oslo, Oslo, Norway

dNorwegian Institute of Public Health, Oslo, Norway

eEconomics Department, Weatherhead School of Management, Case Western Reserve University, Cleveland, OH, United States

a r t i c le i n f o

JEL classification:

J12 J63 J65 K42 Keywords:

Crime Plant closure Plant downsizing Displacement

a b s t r a ct

WeestimatethejobdisplacementeffectoncriminalbehaviorforyoungadultNorwegianmenseparatedfrom theirplantofemploymentduringamasslayoff.Displacedworkersexperiencea20percentincreaseincriminal chargeratesintheyearofdisplacement,witheffectsdecliningthereafter.Effectsareparticularlylargeforprop- ertycrimes,consistentwiththeideathatdisplacedworkersturntoacquisitivecrimestoreplacelostearnings.

However,effectsarealsosizableforviolentandalcohol/drug-relatedcrimes,indicatingothermechanismsat work.Wefindstrongevidencethatdisplacementincreasescrimeeffectsthroughtheincreasedavailabilityof time,andsupportiveevidencethatpsychologicalfactors(mentaldistress,self-control)alsoplayarole.

1. Introduction

Alackof employmentandjobopportunitiesareoftenconsidered importantcausesofcriminalbehavior(Belletal.,2018;Fishbacketal., 2010).Recentworld-widerecessions,withparticularlyhighunemploy- mentratesamongtraditionallycrime-pronegroupslikeyoungandlow educatedmen,haveaccentuatedtheimportanceofunderstandingre- lationshipsbetweenworkandcrime(Hoynesetal.,2012;Hauserand Baker,2008).InthispaperweuseindividualleveldatafromNorway tostudyhowthecriminalbehaviorofemployedmenisaffectedbyjob displacement.

Thereisaricheconomicliteratureexploringthelinksbetweenla- bormarketconditionsandcrime.1Muchoftheempiricalworkdrawson USdatasourcestoestimatetherelationshipbetweenarea(usuallystate) unemploymentratesandcrime,withthegeneralfindingthatunemploy- menthasamodestbutstatisticallysignificantpositiveeffectonproperty crimerates,withlittleornoeffectonviolentcrimerates.2Thesefind- ingsareconsistentwithtraditionaleconomicrationalchoicetheories ofcrime,whichpredictthatareductioninlicitearningsopportunities increasestheallocationoftimetowardcrimeforprofit(Ehrlich,1973;

Correspondingauthorat:ResearchDepartment,StatisticsNorway,Akersv.26,0177,Oslo,Norway.

E-mailaddresses:[email protected](M.Rege),[email protected](T.Skardhamar),[email protected](K.Telle),[email protected](M.Votruba).

1 SeeforexampleMustard(2010)forarevieworBelletal.(2018)forarecentcontribution.

2 Mostoftheearlyworkinthisareasufferedfromendogeneityandattenuationbiasissues,thoughsomestudiesutilizingmorereliablyexogenousvariationhave emerged(Mustard,2010).Forexample,RaphaelandWinter-Ebmer(2001)andLin(2008)employinstrumentalvariable(IV)methodstoaddressmeasurementerror problemsandendogeneityofstateunemploymentrates,andtheyfindthataonepercentagepointincreaseinunemploymentraisespropertycrimerates4-6percent.

Becker,1968).Therelianceonaggregatedatahaslimitedtheability ofpreviousstudiestoinvestigatethemechanismsthroughwhichlabor marketconditionsmayaffectcriminalbehavior.Moreover,whenrely- ingonarea-levelvariation,typicallyinunemploymentratesacrossUS states,itishardtocrediblyidentifycausaleffectssinceanumberof otherfactorsco-varywithunemploymentrates.

We contributeto theexisting literature byusing individual-level crime datatoprovideindividual-levelestimates of theeffectsof job displacementoncrimeunderatransparentidentificationstrategy(sim- ilartothestrategyofe.g.Huttunenetal.,2018orBlacketal.,2015).

Specifically, we investigate the impact of job separation associated with mass layoffs on the displaced workers’ engagement in crime.

Workerssufferinginvoluntaryjoblossrepresentanimportantsubsetof individualsthroughwhichweakeninglabormarketsmightaffectcrime.

Focusingonjobseparationsassociatedwithplantmasslayoffsallowus toinvestigatetheimpactofinvoluntaryjoblosswhilecircumventingthe mostobviousformsofomittedvariablebias:theselectfiringofspecific workersbasedonunobservedattributes.Moreover,wecananalyzea richersetofcrimecategoriesthanothershave,includingalcohol/drug offenses and serioustraffic offenses,andwe can datecrimesto the

https://doi.org/10.1016/j.labeco.2019.101761

Received7July2018;Receivedinrevisedform28August2019;Accepted4September2019 Availableonline4September2019

0927-5371/© 2019TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBY-NC-NDlicense.

(http://creativecommons.org/licenses/by-nc-nd/4.0/)

(2)

day-of-weektheyarecommitted,allowingustodiscusshowdisplaced workers’variationintimeavailabilityonworkdaysversusweekends mayaffectcrime.OuranalysisdrawsonNorwegianregisterdatathat includearicharrayofsocioeconomicanddemographicvariablesfor theentireresidentpopulation,aswellasallcriminalchargesbrought againstanyresidentfrom1992through2008.Individualemployment spell recordsallow ustocalculate employmentcounts by plantand year, fromwhich we can identifyseparations andmasslayoffs. Our mainanalyticsampleconsistsof361,385differentmen,18–40years old,whowereemployedwithatleasttwoyearsoftenureinthebase- lineyear.Our difference-in-differences(DID)approach comparesthe evolutionofcriminalchargeratesina“treatedgroup” ofmaleworkers whowereseparatedfromtheirplantofemploymentduringaperiodof masslayoff (thedisplaced),totheevolutioninchargeratesof similar workers employedin plantsthatdid not undergoamass layoff (the comparisongroup).Pre-separationemploymentratesaresimilaracross the two groups, however pre-separation crimes rates are somewhat higherin thedisplacedgroup, necessitatingtheDIDapproach.3 Our estimatedeffectsareunbiasedundertheassumptionthatthedifference incrimeratesobservedpre-displacementwouldhavecontinuedinequal magnitudeintothepost-displacementperiodhadthedisplacementsnot occurred.Thefactthatpre-displacementcrimeratedifferencesappear stablethroughoutthepre-displacementperiodlendscredibilitytothis assumption.

Wefindthatjobdisplacementleadstoasizableincreaseincrim- inal charge rates of about 20 percent in the year of displacement, withdeclining effectsin the subsequentyears. Jobdisplacement in- creasescrimeforallstudiedcrimecategories.Inarelativesense, es- timated effects are most pronounced for property crimes. Our esti- mateindicatesthatjobdisplacementraisesthelikelihoodofproperty crimesbyabout60percent intheyearofdisplacement.Therelative sizeofeffectsappearssmallerforothercrimecategories(violence,al- cohol/drug,serioustrafficviolations),butsignificanteffectsareesti- matedthroughout,andwithsimilar(thoughsmall)leveleffectsacrossall categories.4

Theeffectof displacementoncrimepresumablyoperates,atleast in part, throughworkers’ labor market detachment. Basedon ratio- nal choicetheories ofcrime (e.g.Becker, 1968),adisplacedworker hasincentivestoshifttheallocationoftimetowardillicitearningsop- portunities(i.e.propertycrime)sincedisplacementreduceslegalearn- ingsopportunities.Additionally,displacementlessenstheopportunity costofaworker’stimeduringtheperiodofunemployment(orunder- employment), withimplications for both property andnon-property crimes(Ehrlich,1973).Ouranalysisfindsthatdisplacementreducesem- ploymentearningsovertheimmediateyearsfollowingdisplacementby 10–15percent,anddisplacementsubstantiallyincreasesthelikelihood ofbeingunemployedorofworkinglessthanfull-time.Asinpriorstud- ies,theparticularlylargeincreaseinpropertycrimesprovidessupport forrationalchoicetheoriesemphasizingtheroleofearningreplacement asamotivationforcrimebythedisplaced.Ontheotherhand,thesignif- icanteffectsonnon-propertycrimesindicatebroadermechanismsare alsoatwork,includingapotentialrolefortimeavailability.

Ouranalysisshedsfurtherlightonthetimeavailabilitymechanismby exploitingdatawehaveontheexactdateeachrecordedcrimeoccurred.

Exceptforpropertycrimes,wefindmoredramaticincreasesincrimes committedonworkdays(Monday-Friday)thanonweekends.Thissug- geststhatnothavingtogotowork,associatedwithadeclineinstruc- tureddailyroutinesandareducedopportunitycostoftime,isanimpor-

3 Substantial effort was exerted in attempting to construct more finely- matchedsamplesofdisplacedandcomparisongroupworkers,sothatestimates couldbebasedonsampleswith(near)identicalpre-displacementcrimerates.

Theseeffortswereunsuccessful,butthosesamplesconsistentlyproducedDID estimatessimilartothosereportedhere.

4 The60percentincreaseinpropertycrimeratesappliestoamuchlower baselinerate.

tantchannelthroughwhichdisplacementaffectsnon-propertycrimes.

Theeffectsweobserveforviolentcrimesanddrug/alcohol-relatedof- fensesarealsoinlinewiththeoriesthathighlighttheimportanceof self-control,financialconcerns,frustrationandmentaldistressin de- terminingcriminalandcounterproductivebehavior(Manietal.,2013; Agnew,1992;GottfredsonandHirschi,1990).

Thesefindingsmakenovelcontributionstotheexistingempiricallit- eratureonjoblossandcrime.Wefindcredibleevidencethatdisplace- mentincreasesviolent(aswellasproperty)crimerates,afactthathas onlyweaksupportfrommostofthearea-levelstudies.Wefindcredible evidencethatdisplacementalsoaffectscrimesliketrafficoffencesand drugs/alcohol-offences,anareawherenoothercredibleevidencecur- rentlyexists.Thelargeeffectsonalcohol/drugcrimesmaybeparticu- larlynoteworthyintheeconomicsliterature,sincetheyarenotstraight- forwardlyexplainedbytherationalcrimetheoryandthuslikelyspeak topsychologicaleffectsofjobdisplacement.Ourday-of-weekanalysis is alsonoveltotheliterature andprovidesempiricalsupport forthe importanceoftimeavailability.

Theremainderofthepaperisstructuredasfollows.Section2dis- cusses theoreticalmechanismsthroughwhichplantclosurecould af- fect criminal behavior, and relates them tothe Norwegian context.

Section3presentstheempiricalstrategy,andSection4describesthe data.Section5presents ourresults,includingrobustnesschecks,and Section6exploresmechanisms.Section7concludes.

2. MechanismsandtheNorwegiancontext

IntheseminalrationalcrimemodelofBecker(1968),individuals commitcrimewhentheexpectedutilityfromdoingsoexceedstheex- pectedutilityofnotdoingso.WhileBecker(1968)wasprimarilyin- terested in optimallaw enforcement, a numberof economicstudies haveextendedhis modelof criminalbehavior(see,e.g., overviewin LevittandMiles,2007).Ofparticularrelevancearetheextensionsof Ehrlich(1973),whointroducesatimeconstraintwherebyindividuals dividetheirtimebetweenlicitandillicitactivities.

InsightsfromthemodelsofBecker(1968)andEhrlich(1973)suggest twocomplementarymechanismsthroughwhichinvoluntaryjoblosscan increasecriminalbehavior.Totheextentjobdisplacementreducesfu- tureearningsandemployment(Huttunenetal.,2011;Regeetal.,2009; Stevens,1997;Jacobsonetal.,1993),wewouldexpectdisplacedwork- erstoexperienceanincreaseinthemarginalutilityassociatedwithil- licitearnings(theearningsreplacementmechanism)andadecreaseinthe opportunitycostofspendingtimeinsuchactivities(thetimeavailabil- itymechanism).Thesemechanismswouldanticipateahigherlikelihood foracquisitivecrimeasaresultofdisplacement.

Theserationalchoice-basedmodelsprovidesomewhatweakerpre- dictionsfornon-acquisitivecrime,whichfailtocompensateforthere- ductioninlicitearnings.Nonetheless,non-acquisitivecrimesmaystill beaffectedbythereductioninthetimecosts.Criminologistsfrequently citetimeavailabilityasanimportantdeterminantofcriminalbehavior.

Felson(1998),forinstance,arguesthatindividualsmotivatedtocom- mitcrimecannotdosounlessanopportunityispresent.Lessstructured dailyroutinesandincreasedidlenessprovidegreateropportunitiesand lowertime-costsforcriminalactivity.Increasedidlenessmayalsoin- creaseone’sexposuretocriminogenicsettings,wherealcoholanddrugs maybepresentandsocialnormsagainstdeviantbehaviorareweaker (Hirschi,1969).These theories(including Ehrlich,1973) suggestthe crimeeffectsofdisplacementcouldextendtonon-acquisitivecrimes, andwouldpredictthoseeffectstobelargestondaysadisplacedworker wouldotherwisehavebeenworking.Thus,wewouldanticipatelarger effectsonworkdaycrimesthanonweekendcrimes.

Intheirwidelycited“general theoryofcrime”,thecriminologists GottfredsonandHirschi(1990)arguethattheassociationbetweenun- employment andcrimecan be explainedbyvariationin individuals’

capacityforself-control,whichaffectsindividuals’abilitytosucceedin

(3)

schoolandwork.5Theresourcemodelofself-controlpositsthattheca- pacityforself-controlislimitedandcanbedepletedbycognitiveand emotionalstrains,andthismodelhasfoundsupportintheexperimental psychologyliterature(e.g.InzlichtandScheichel,2012;Inzlichtetal., 2006; Baumeister et al., 1994).Empirical findings also suggest that joblossimposesstrainsandmentaldistressonaffectedworkers(e.g.

Marcus,2013;EliasonandStorrie,2009;Draganoetal.,2005;McKee- Ryanetal.,2005;VahteraandKivimaki,1997).Ifso,theresultingdi- minishmentofself-controlcouldresultincounterproductivebehaviors (Manietal.,2013).Therefore,mentaldistress/self-controlrepresentsa thirdmechanismthrough whichwe mightanticipate adisplacement effectoncrime,withparticularrelevanceperhapstonon-acquisitiveof- fenseslikeviolenceandalcohol/drugs.

TheNorwegiancontextmayaffect therelevanceof eachofthese theoreticalmechanisms.InNorway,strictrulesprotectemployeesfrom beingdismissed(AddisonandTeixeira2003),andjobdisplacementis rarelysuddenasworkers aretypicallyrequiredtoreceiveat least3 monthsofadvancenoticebeforebeingdismissed.Moreover,inthere- centdecades, Norwayhasbeen characterizedby lowunemployment rates,evenbyScandinavianstandards.In2007,thesurvey-basedun- employmentratewas2.5percent,comparedwith4.6percentforthe USand7.1percentfortheEuropeanUnion(OECD,2009).Withstrong demandforworkers,theeffectsofjobdisplacementmaynotresultin prolongedspellsofunemployment,oradeteriorationofnext-jobqual- ity,whichsuggeststhatwemightexpectmoredetrimental effectsof displacementoncrimeincountrieswithhigherunemploymentlevels.

Moreover,publicwelfareprogramsinNorwayaregenerousbyin- ternationalstandards.VirtuallyallNorwegianworkersarecoveredby thestate’smandatoryunemploymentinsuranceprogram.Thesizeofthe unemploymentbenefitsistypicallyaroundtwo-thirdsoftheearningsin thepreviouscalendaryear,anduntil2003atypicalreceiverwaseligi- bleforunemploymentbenefitsforuptothreeyears(thereafterupto 2years).Personsnotfindinganewjobwhentheunemploymentbene- fitsrunoutcangetbenefitsofthesamemagnitudebyparticipatingin medicalorvocationaltrainingprogramsorbyqualifyingfordisability pension.Thegenerouswelfarebenefitsavailablemayreducetheincen- tivestoengageincrimeforprofitcomparedwithothercountrieswhere theindividualeconomicconsequencesofjoblossaremoresevere.

Ontheotherhand,enforcementpoliciesarelesspunitiveinNorway thanintheUSandtheUK(Christie,2000),whichcouldlessenincen- tivestoavoidcrime. Prisontermsaresubstantiallyshorter for some typesof crimein Norwaythanin countrieslike US orUK, with re- markabledifferencesinincarcerationrates.TheUSincarcerationrate isabout751per100,000inhabitants(BJS,2009),whiletheUKrateis about140(EuropeanSourcebook,2006),andNorway’srateisabout91 (StatisticsNorway,2008).

However,foroffensesotherthanmurderandrobbery,6Norwegian convictionrates aresimilar tothoseof many otherOECDcountries.

Forexample,thetheftrateper100,000is2860forNorway,2182for theUSand3379fortheUK(UN,2008).TheInternationalCrimeVic- timSurveys,whichmightbeconsideredthemorereliabledatasource

5 Inameta-analysisPrattandCullen(2000)findconsistentassociationsbe- tweenindividuals’criminalbehaviorandmeasuredlevelsofself-control.An importantmethodologicalconcerninestimatingeffectsofjobdisplacementon crimeistoruleoutspuriousassociationsthatmightarisefromunobservedvaria- tionine.g.capacitiesforself-control(seeSections3and5.1).Crime,arrestsand incarcerationmightalsohavecausaleffectsonfutureemploymentopportuni- ties,forexample,ifstigmafromacriminalrecordorhumancapitaldepreciation fromincarceration,restrictsfutureaccesstomeaningfuljobs(Grogger,1995; Pager,2003;MocanandRees,2005;Kling,2006).

6 AsintheotherScandinaviancountries,Norwayhasamongthelowesthomi- cideratesintheWesternworld.Norwayhasamurderrateof0.71per100,000 inhabitants,comparedwithratesof5.62intheUSand1.41inEnglandand Wales.Therobberyrateper100,000inhabitantsis27.9inNorway,147.7in theUSand183.8intheUK(UN,2008).

forcross-nationalcomparisonsofcrimeprevalence,alsoindicatesthat crimeratesinNorwayaresimilartothoseofotherOECDcountries.Of the30countriesincludedinthestudy,Norwayisratedwithamedium victimizationrate,withlowerratesthanIreland(highestrate),England andWales(nexthighest)andtheUS,buthigherratesthane.g.France, GermanyandItaly(vanDijketal.,2008).Thecrimeandjusticeenvi- ronmentinNorwaymaythusbemoresimilartoothercountiesinthe Westernworldthansuggestedbytheincarcerationandhomiciderates.

Insummary,overthelastdecadesNorwegianresidentshavebeen facing lowunemploymentrates,generouspublicbenefits,lowhomi- cideandrobberyratesandatraditionoflesspunitivelawenforcement policies(Pratt,2008;Christie,2000).Theextentthatdisplacedwork- ersaremotivatedtoreplacelicitwithillicitearningscouldthushave beensmallerinNorwaythaninmanyotherWesterncountries,which maysuggestasmallereffectofdisplacementoncrime(especiallyprop- ertycrime)inNorwaythanelsewhere.However,itcannotberuledout thatsuchamoderatinginfluenceofwelfarebenefitsiscounteractedby lowerexpectedpunishmentinNorwaycomparedwithmanyotherWest- erncountries.Thiscontextualbackgroundshouldbekeptinmindwhen interpretingtheresults.

3. Empiricalstrategy

We estimate the effect of job displacement on crime using a difference-in-differences(DID)approach,whichcomparestheevolution incrimeratesinasampleofdisplacedworkerstothoseinasampleof similar7non-displacedworkers(ourcomparisongroup).

Workersare(potentially) includedin thedisplacementsampleif, inagivenyear,theworkersseparatefromtheirplantofemployment duringaperiodwhentheplantisundergoingamasslayoff.Forsuch workers,theyearofseparationisconsideredtheworkers’baselineyear, andwedeemtheseparationtobe associatedwithmasslayoff ifthe worker’splantexperiencedareductioninplantemploymentexceeding 30percent,eitherinthebaselineyearorineitherofthetwoadjacent years.8 This methodfor identifyingdisplacedworkers largelyresem- blesdefinitionsthathavepreviouslybeenappliedintheliterature(e.g.

Huttunenetal.,2018;Blacketal.,2015;DavisandvonWachter,2011; CouchandPlaczek,2010;Jacobsenetal.,1993).9Inanattempttoex- cludetemporaryormis-recordedseparations,suchascasesofworkers relocatingwithinthesamefirm,wealsorequiredevidencethatthesep- arationwaspermanent.Tooperationalizethis,weomitworkersfrom thedisplacedsampleiftheyhadreturnedtotheirbaselinefirmofem- ploymentbytheendofthesecondpost-displacementcalendaryear.

Incontrast, workersare(potentially) includedin thecomparison group, for that same baseline year, iftheir plant-of-employment at

7Asdescribedbelow,theworkersinbothsamplesaremaleswithatleast twoyearsoftenureintheirfirmofemploymentatthebeginningofaparticular baselineyear(aswellasmeetingotherinclusioncriteria).Thesamplesare, unfortunately,notsimilarinthepre-displacementcrimerates,necessitatingthe DIDapproachtakeninthispaper.

8Employmentcountsarebasedonfull-timeequivalents(FTEs)measuredat theendofeachcalendaryear.Massdownsizingeventsinourdataareoften markedbyseveralconsecutiveyearsofhighseparationrates,whichiswhywe associateseparationswithmassdownsizingevenwhenthemajorperiodofem- ploymentreductionwasayearremovedfromtheyearseparationoccurs.

9Therearealsosomesmallerdifferencesbetweenourapproachandthatof (e.g.)Jacobsenetal.(1993)orCouchandPlaczek(2010).First,theydefine masslayoff asa30percentdecreaserelativetothemaximumemploymentlevel oftheplantinthelast(6)yearsbeforethebaselineyear,whilewedefineitasa declineof30percentrelativetotheprecedingyear.Second,whiletheyrestrict thesampletoplantswithatleast50employees,werestrictourmainsample toplantswithmorethan10full-time-equivalents,andwhiletheyrequirejob tenureofatleast6yearsforinclusion,werequireonly2years.Lastly,because ofdataavailability,theyexcludeworkerswhodonotreceivepositiveearnings inalltheyearsoftheirdatawindow,whilewecanfolloweveryworkerthrough timeregardlessofearnings.

(4)

thestartof that yeardid not undergoamass downsizing10 andthe workerremainedemployedinthatplantthroughtheendofthatyear.

Again,thisdefinitionofcomparisongroupissimilartodefinitionsthat have previouslybeen applied in the literature(e.g. Huttunen et al., 2018; Blacket al., 2015; Davis andvonWachter, 2011; Couchand Placzek,2010;Jacobsenetal.,1993),wherethecomparisongroupis oftendefinedtocompriseworkerswhoareneverseparatedfromtheir plantofemployment.Thedefinitionsofthedisplacedandcomparison groupimplythatworkerswhoremainedemployedinplantsthatunder- wentamasslayoff areexcludedfromthesample,asareworkerswho separatefromnon-downsizingplants.

Todemonstrateourempiricalmodel, considerthesampleof dis- placedandcomparisonworkersconstructedforbaselineyear(b)1997.

Fortheseworkers,wecanobservecrimesovercalendaryears(t)1992–

2008or,equivalently,overrelativeyears𝜏=−5to𝜏=9(where𝜏=t-b).

Following the literature,we employ various specifications of a dis- tributedlagmodel,hereillustratedbyalinearprobabilitymodel(we willalsoapplylogitmodels):

Pr( 𝑐𝑖𝜏=1)

=𝛼𝑥𝑖+𝛾𝜏+

9 𝑘=−5

𝛿𝑘𝑑𝑘𝑖 +𝜀𝑖𝜏 (1)

where

ci𝜏indicatorthatworkericommitsatleastonecrimeinrelativeyear 𝜏(with𝜏=−5,−4,..,9forbaselineyear1997)

𝑥𝑖vectorofcontrolvariablesmeasuredatthebeginningofthebase- lineyear(seeAppendixBfordetails)

𝛾𝜏vectoroffixedeffectsassociatedwitheachrelativeyear 𝑑𝑖𝑘dummyvariablessettoonefordisplacedworkersinthekthrelative

year,otherwisezero(withk=−5,−4..,9forbaselineyear1997) 𝜀i𝜏errortermwithexpectationzero.

Themainparametersofinterestarethe𝛿kcoefficientswhichcap- turethedifferenceinthelikelihoodofcrimebetweenworkersinthe displacementandcomparisongroupineachrelativeyear,from5years precedingthedisplacement(ofthedisplacedworkers)to9yearsafter.If displacementincreasescrime,wewouldexpect𝛿0tobehigherthanthe 𝛿kinpre-displacementyears,i.e.𝛿5to𝛿2.Estimatesof𝛿kpertaining tosubsequentpost-displacementyears(𝛿1,𝛿2,..)allowustoexplore theextentthatthecrimeresponsetodisplacementfades(orpossibly increases)overtime.

Notably,wedonotregardestimatesof𝛿1as(strictly)pertainingto the“pre-displacementperiod” fortwomainreasons.First,asmentioned earlier,Norwegianworkersarenotifiedinadvanceofanimpendingdis- placement,andtheymayinmanyinstancesforeseeandpreparefortheir plantfailingwellbeforelayoff (Bastenetal.,2016).Totheextentknowl- edgeofanimpendinglayoff operatesoncriminalbehavior,asitmight undereithertheearningsreplacementormentaldistress/self-controlmech- anisms,estimatesof𝛿1wouldcapturethoseeffects.Furthermore,itis knownthatseparationdatesarenotrecordedwith100percentaccu- racy,withampleevidencethatseparationssometimesoccurredearlier thanwhatisrecordedinemploymentregistries.11Thiswouldalsocon- tributetousestimatingadisplacementeffectthatappearstopre-date thedisplacementevent.

Toproducespecificestimatesofthedisplacementeffect,werelyona standardDIDassumption:thatanydifferenceinpre-displacementcrime rateswouldhavepersistedifnot forthedisplacements thedisplaced workersexperienced.Econometrically,thisassumptionisimplemented

10 Eitherinthebaselineyear,ortheadjacent-to-baselineyears.

11 Forinstance,afairnumberofdisabilityprogramentrantsappeartostillbe employedfulltime(intheirpriorplant)forafewmonthsafterdisabilityentry.

Forthisreason,wealwaysexcludefromoursampleof“workers” personson socialbenefitsthatshouldhaveprecludedfulltimework.

bymodifyingourmodelasfollows:

Pr( 𝑐𝑖𝜏=1)

=𝛼𝑥𝑖+𝛾𝜏+𝑑𝑖+

9 𝑘=−1

𝛿𝑘𝑑𝑘𝑖 +𝜀𝑖𝜏 (2)

where𝑑𝑖isthefixedeffectassociatedwithbeinginthedisplacedgroup, andthetime-varyingeffectsofdisplacementareonlymodeledforrel- ativeyears−1goingforward.UndertheDIDassumption,the𝛿kcoef- ficientsprovidecausalestimatesofthecrimeeffectofdisplacementin eachyearrelativetothepre-displacementyears(−5through−2).

Thefactthatwehavepaneldataallowsustodefinedisplacement andcomparisongroupsformultiplebaselineyears.Tomaximizepower, wethereforestackthedataforeachofthebaselineyears,andrunre- gressionsonthepooleddata(seeHuttunenetal.,2018orBlacketal., 2015forasimilarprocedure).Baselineyear,relativeyearandcalen- daryeararethusdefined forallworkersinboththecomparisonand thetreatmentgroup,whichintroducesafewadditionalconsiderations.

First,itispossibleforsomeworkerstobedisplacedinmultiplebaseline years.Tosimplifymatters,weonlyconsiderthefirstbaselineyearin whichsuchaworkerisdisplaced.Second,itispossibleforaworkerto bedisplacedinonebaselineyearandbelongtothecomparisongroupin anotherbaselineyear.Toavoid“partly-treated” workersinthecompar- isongroup,wedonotallowaworkerwhoisinthedisplacementgroup tobeinthecomparisongroupofanybaselineyear.Third,togeneral- izethemodeltothepooleddata,andtoaccountforcommoncalendar yearshocks,weincludeindicatorstocapturethecalendaryeareffects.12 Finally,inallregressionsweclusterontheindividualworker,butwe alsoexplorehowtheestimatedstandarderrorsareaffectedbyrestrict- ingworkersinthecomparisongrouptobepresentinnomorethanone baselineyear.

Our estimationstrategyisa straightforwardgeneralization of the

“difference-in-differences” method,anditthusreliesonthecomparison grouptoaccountforchangesincrimeratesovertimethatwouldhave occurredintheabsenceofdisplacement.Thecrucialassumptionfora consistentestimateofthedisplacementeffectisthatthecrimeratesin thedisplacementandcomparisongroupswouldhaveevolvedsimilarly overtimeintheabsenceofthedisplacement.Thisassumptioncanbe tested,tosomedegree,bycomparingtheevolutionofcrimeratesinthe twogroupsduringthepre-displacementperiod.

Itmightbeworthdrawingattentiontoacoupleofdistinctionswith respect towhatconceptual effectsthisapproach doesnotattemptto estimate.First,itdoesnotestimatetheeffectofexposuretomasslayoff oncrime.Aslongastheplantdoesnotclosecompletely,anumberof workersareretainedintheplantduringandafterthemasslayoff.There areseveralstudiesthatlookattheaverageeffect(onvariousoutcomes) ofexposuretoplantdownsizingoverbothlaidoff andretainedworkers, andsomehavearguedthatthisaverageismorepolicyrelevantthanthe effectsspecifictolaidoff workers(e.g.Regeetal.,2011,2009).Indeed, somestudiesindicatethatadverseeffectsontheretainedworkerscould beassevere,orevenmoresevere,thantheadverseeffectsonthelaid off workers(VahteraandKivimaki,1997).However,whatweattempt toestimateinthecurrentstudyistheeffectonthedisplacedworkers only,neglectingpossibleeffectsontheworkerswhoremainintheplant throughandafterthemasslayoff.

Second,itdoesnotestimatetheeffectofunemploymentoncrime.A numberoftheworkersseparatedfromtheirplantduringamasslayoff couldbedirectlyenteringanewjobinanotherplant.Indeed,someof these workersmaynotevenleaveinvoluntarily,butmayhave gota betterofferelsewherearoundthetimetheirplantdownsized.Whatwe estimateisthereforetheoveralleffectofjobseparationinassociation withamasslayoff overallseparatingworkers,includingthosewhogo directlyintoanotherjob,thosewhoundergoaperiodofunemployment,

12Wecouldhavealternativelyincludedfixedeffectsforbaselineyearwith identicalresults,asbaselineyear,relativeyearandcalendaryearareperfectly collinear.

(5)

andthosewhodropout oftheworkforcealtogether.Asdiscussedin Section2,themechanismsthroughwhichdisplacementincreasescrime arepresumablystrongerforthoseundergoingaperiodofinvoluntarily unemploymentfollowingthedisplacement.13

4. Data 4.1. Datasources

Toestimatetheeffectofjobdisplacementoncrime,wecombinetwo registerdatabasesprovidedbyStatisticsNorway.Thedatabasescanbe mergedusingauniquepersonalidentifierprovidedtoeveryNorwegian residentatbirthorimmigration.Thefirstdatabasecontainscomplete recordsofcriminalchargesforeveryNorwegianresidentoverthepe- riod1992–2009.Weutilizeoffensescommittedthrough2008toallow fora registrationlagbetween thetimean offenseis committedand thecharges.Thedatabasecontainsallseriouscrimes,butalsomisde- meanorslikedrunkdriving,excessivespeedingandshoplifting.Aper- sonisregisteredas“charged” ifthepoliceperformaninvestigationand concludethatthepersondidcommittherecordedcrime,andthecase isconsideredsolved.Theinvestigationmaybeinitiatedbythepolice receivingareportorbyanarrest.Theregistrationisindependentofthe furtheroutcomeofthecase(filingofformalcharges,prosecutionsor convictions).14Dateofcrimeanddetailedcodesof“offensetype” are alsoincludedonchargerecords.StatisticsNorwayhasconstructedsub- categoriesofcrimeandwerelyonthesedefinitionstoconstructcrime categoriesthatcorrespondtothoseusedbytheUSFBI(seeAppendix A).

The second database is called FD-trygd. It is a richlongitudinal databasewithrecordsforeveryNorwegianresidentfrom1992to2008 (formost variables), containingindividual demographic information (e.g.sex,age,maritalstatus),socio-economicdata(e.g.education,in- come),currentemploymentstatus,industryofemployment,indicators ofparticipationinNorway’swelfareprograms,andgeographicidenti- fiersofareaofresidence.

Inparticular,FD-trygdcontains recordsfortimingof employment

“events” since1995.Theseevents,capturedbyindividualanddate,in- cludeentriesintoandexitsoutofemployment,changesinemployment status(fulltime,parttime,minorparttime),andchangesinplantand firmofemployment.Theemploymentrecordsareconstructedbydata analystsatStatisticsNorwayfromrawemploymentspellrecordssub- mittedby employers,andverifiedagainstemployeewagerecordsto ensurethevalidityofeachspellandtoeliminaterecordspertainingto

“secondary” employmentspells.15

13 Itistemptingtoimaginethecausaleffectofunemploymentcouldbeinvesti- gatedbyusingmassdownsizingeventsasaninstrument.Toourmind,thisexer- cisemakesnosenseunlessweimagineunemploymenttobethesolemechanism throughwhichdownsizingaffectscriminalbehavior.Theresultswepresentin- dicatethisisn’ttrue.

14 Aproblemthatshouldbekeptinmindwhenmeasuringresultsfromany empiricalstudyofcrimeisthedifficultyinmeasuringlatentcriminalactivity.

Self-reportsofcriminalactivityshouldbeinterpretedcautiouslysincetheyare oftenimpossibletovalidateandsincetheextent oftruthfulself-reportingis loweramongsubjectswithanextensivecriminalrecordthanamongsubjects withlittleornocriminalhistory(Kirk,2006;MacDonald,2002;Farringtonetal., 2003;Hinderlangetal.,1981).Crimedatafromregistrieshavetheadvantage thatoffenderscannotchoosenottoberegistered.Themaindisadvantageof registerdataisthatcrimeswhicharenotreportedtothepolicearenotrecorded, andcrimesleft“unsolved” cannotbematchedtoaspecificindividual.

15 Ifanindividualwasemployedinmultipleplantsatagiventime,primary employmentwasdeterminedbyemploymentstatusandrecordedincomefrom eachsourceofemployment.Aplant’sidentifierisonlysupposedtochangeifat leasttwoofthethreefollowingconditionsaremetatthesametime:geograph- icalrelocation,changeofindustryandnewowner.Inreality,andespecially withinfirms,plantidentifiersmaychangeevenifalargeproportionofthesame employeesremainworkingtogether.Thoughsuchmeasurementissuesmayat-

Basedontheemploymentrecords,weconstructed plant-levelem- ploymentcounts attheendof years1995to2008.Thecounts were constructedasmeasuresoffull-timeequivalents(FTEs),withparttime andminorparttimeemploymentmeasuredas0.67and0.33FTEs,re- spectively.Excludedfrom thesecountswereanypersonidentifiedin FD-trygdasself-employedorreceivingassistancethatshouldhavepre- cludedfulltimework(rehabilitationpensions,disabilitypensions,etc.).

TheannualplantFTEwerethenusedtoidentifyseparationsthatwere associatedwithamassdownsizingasdescribedinSection3.

4.2. Defininganalyticsample

Ourmainanalyticsampleconsistsofmenbetween18and40yearsof ageatthebeginningofthebaselineyear.Werestricttonon-elderlymen becausecrimeratesamongwomenandoldermenaretoolowtoprovide estimateswithanyprecision(StatisticsNorway,2008;Freeman,1996; HirschiandGottfredson,1983).Moreover,tostudyeffectsoncrimeof jobdisplacement,meninoursamplewererequiredtohavehadrea- sonableattachmenttoanestablishedjob.Specifically,werestrictthe mainanalyticsampletomenwhowerefull-timeemployedpreceding thebaselineyear, excludingafewcases wherethemanreceivedas- sistancethatshouldhaveprecludedfulltimework,suchasdisability benefits.Wealsorequirethementohaveatleasttwoyearsoftenurein theplantatthebeginningofthebaselineyear,toensuredurableattach- menttoone’scurrentplantofemployment.Asaprecautionagainstthe plantdownsizingvariablebeingcorrelatedwithunobservedindividual determinantsofcrime,weexcludemenworkinginaplantwithlessthan 10FTEsatthebeginningofthebaselineyear.

Weconstructourmainanalyticsamplebyappendingthe10base- lineyeardatasets(1997–2006)together,yieldingapaneldatasetwith 10,526,937person-yearobservations.Thedatasetconsistsof361,385 differentmeninthetenbaselineyears,with83,974different menin thedisplacementgroupand277,411differentmeninthecomparison group.Asmentioned,thedisplacedmenarepresentinonebaselineyear only,whilemorethan90percentofthemeninthecomparisongroup appearinseveralbaselineyears.16Forallmenwecanobservecrimes overthe17years1992–2008,buttoavoidthatthepanelbecomeshighly unbalancedfortheearly(1997)andlate(2006)baselineyears,weonly usecrimedataforthe11relativeyears−5to5.17

4.3. Summarystatistics

Variablescapturingindividualandplantsocio-economiccharacteris- ticswereconstructedbasedonFD-trygdrecordspertainingtothebegin- ningofthebaselineyear.Anumberofthesevariables(x)areincludedas covariatesinourestimationmodels(seeAppendixBfordetails).Sum- marystatisticsarepresentedin Table1forourmainanalyticsample in thebaselineyear. About8percent18of thesamplewasdisplaced.

Theaverage agein thesampleis about34 years,and38percent of

tenuateourresultssomewhat,themosttypicalcasesofrestructuringshouldbe capturedbyutilizingfirmidentifiersindefiningpermanentdisplacement(see Section3).

16Onaverage,aworkerinthecomparisongroupispresentin3.4baseline years.Asmentioned,wewillalwaysclusterontheindividualworker,andwe alsoexplorehowtheestimatedstandarderrorsareaffectedifweallowworkers inthecomparisongrouptobeincludedinonlyonebaselineyear.

17Thisensuresthatthepanelisfullybalancedinthefiveyearspriortothe baselineyear(1997-5=1992)anduptorelativeyear+2(2006+2=2008),while forrelativeyear+3andafteritbecomesunbalanced(sincewedonothavecrime dataafter2008).

18Asmentioned,workersinthecomparisongrouparepresentin3.4baseline yearsonaverage.Thisimpliesthattherateofuniquemeninourmainanalytic samplewhoweredisplacedismuchlargerthan8percent,itis24percent.Re- callthatallthemenwhowerenotseparatedfromaplantinassociationwitha masslayoff (orwhoe.g.workedinplantswithlessthan10FTE,cf.theexclu- sionrestrictionsdescribedabove)areexcludedfromourmainanalyticsample,

(6)

Table1

Summarystatistics.

Variable All Displaced group Comparison group Difference

Displaced 0.08

Age 33.8 33.3 33.9 0.54

(4.7) (4.9) (4.7)

Compulsory school only 0.07 0.08 0.07 0.01

High school only 0.65 0.65 0.65 0.00

More than high school 0.28 0.27 0.28 0.01

Educ. Missing 0.00 0.00 0.00 0.00

Earnings 354,600 354,500 354,600 0.81

(184,200) (197,200) (183,000)

Tenure 5.6 4.7 5.6 0.99

(3.1) (2.9) (3.1)

Married 0.39 0.37 0.39 0.02

Children (below 18) 0.55 0.52 0.55 0.03

FTE of plant 298.0 224.3 304.8 80.5

(721.8) (475.8) (739.6)

Crime in baseline year 0.019 0.028 0.018 0.009

# observations 1,019,940 83,974 935,966

Notes:Variablesaremeasuredatthebeginningofthebaselineyear(operationalizedasthe endoftherelativeyear−1)unlessotherwisespecified.Standarddeviationsinparentheses. andindicatethatthevariableissignificantlydifferentacrossthegroupofdisplacedand comparisonworkersatthe5and1percentlevel(two-sidedt-test).

Fig.1. Proportionfull-timeemployedaroundthebaseline year(+/-5years).

themeninthesampleweremarried.Thedisplacedandthecompar- isongroupdifferonobservables,butingeneralthemagnitudeofthe differencesisquitesmall.Thedisplacedareabouthalfayearyounger thanthemeninthecomparisongroup,andaresomewhatlesseducated, buttheir(pre-displacement)earningsaresimilar.Wealsoseethatthey hadsomewhatshortertenureandthattheyworkedinsmallerplants.

Fig.1showsthedevelopmentovertimeintherateoffull-timeemploy- mentforthetwogroups.Byconstructionof thedataset,everyone is requiredtobefull-timeemployedatthestartofthebaselineyear,and asexpected,weseethatfull-timeemploymentdropssubstantiallyfor thedisplacedinthebaselineyear;before itstarts toconvergetothe comparisongroup. Overall,thedisplacementandcomparisongroups arefairlysimilar,butTable1showssomedeviationswhichindicatethe

implyingthatfewerthan24percentofallemployedmeninNorwayexperienced aseparationinassociationwithamasslayoff overtheperiod.

needtoconsiderrobustnesstocontrollingforpre-existingdifferences acrossthetwogroups.

5. Empiricalfindings 5.1. Mainresults

The twothick linesin Fig.2 show theevolution of crime rates, relative tothe baseline year,for the displaced(solid) andthecom- parison(dashed)group.Weseethatthecrimerateofthedisplacedis generallyabovethatofthecomparisongroup,butthetrendincrime forthecomparisongroupissimilartothetrendforthedisplacedduring the pre-displacementperiod. Thisis illustratedwiththe thindashed line,whichiscalculatedbyaddingthemeanpre-displacement(𝜏<−1) differenceincrimeratestothecrimerateofthecomparisongroupin every year.Itisevident thattherelativecrimerateofthedisplaced increases aroundthe timeofdisplacement, whilethere isno similar

(7)

Fig.2. Proportioncharged ofcrime aroundthebaseline year(+/-5years).

increase forthecomparison group aroundthebaseline year. Thisis whatwewouldexpectifjobdisplacementresultsinmorecrime.

Table2presentsregressionresults.Model1showsOLSregression resultsforEq.(1)withnocontrolvariables,whichsimplyprovidesthe differencesincrimeratesbetweenthedisplacedandcomparisongroups, i.e.thedifferencebetweenthetwothicklinesinFig.2.Weseethatthe differenceinthecrimeratefluctuatesaround0.6–0.7percentagepoints overyears𝜏=−5to𝜏=−2.Thenthedifferencerisesto0.9percentage pointsinthebaselineyear(𝜏=0),beforedeclininginsubsequentyears.

ControllingforageandcalendaryearfixedeffectsinModel2reduces eachofthepointestimatessomewhat,butthedifferentialchangefrom thepre-displacementtopost-displacementyearsisslightlylarger.

Model1and2andFig.2showthatthedisplacedhaveahighercrime ratethanthecomparisongroupovertheyearsprecedingthedisplace- ment,indicatingthatthedisplacedaremorecrime-proneirrespective ofanyexposuretojobdisplacement.19AsdiscussedinSection3,this isnotaconcernforourdifference-in-differencesidentificationstrategy aslongasthecrimerateinthedisplacedgroupwouldhaveevolved similarlyovertime(intheabsenceofdisplacement)asitdoesforthe comparisongroup.Inthisrespect,itisreassuringthatthetrendincrime ratesissimilarforthedisplacementandcomparisongroupsintheyears precedingdisplacement.

Toobtainadifference-in-differencesestimator,weincludeinModel 3adummyvariableidentifyingdisplacedworkersandomitthedisplace- menttermspertainingtothepre-displacementperiod,asinEq.(2).In doingso,weeffectively“differenceout” themeanpre-displacementdif- ferenceincrimeratesobservedacrossthetwogroups.Ourestimates inModel3thereforecapturetheeffectofdisplacementundertheas- sumption that pre-existing differences in crime rates across the two groupswouldhaveremainedunchanged(conditionalontheincluded covariates)in theabsenceof displacement. Ourestimates indicate a pre-displacementdifferenceincrimesratesof0.44percentagepoints betweenthedisplacedandcomparisongroup(conditionalonageand calendaryeardummies).Theestimatedeffectofdisplacementoncrime ratesinthebaselineyearis0.38percentagepoints.Thiseffectestimate

19 Therecouldbeselectionattheplantlevel,forexampleifplantswithos- cillatingemploymentstocksareonlyabletoattractmorecrime-proneworkers.

Therecouldalsobeselectionattheindividuallevel,forexampleiffirmsarelay- ingoff morecrime-pronemenfirstinassociationwithmasslayoffs.Wereturn totheempiricalrelevanceofthesepotentialsourcesofbiasbelow.

ishardlyaffectedbyaddingaricharrayofcontrolvariables(Model4) orindividualfixedeffects(Model5),butwenotethattheindividual fixedeffectsmodelrevealsamoreclear-cutdeclineintheeffectofdis- placementoncrimeintheyearsafterdisplacement,andnostatistically significanteffectremainsafter4years.20

Thedependentvariableisdichotomouswithameanclosetozero, whichsuggestthatthelogitmodelis,forexample,moreefficientthan OLS.Models6–8presentestimatedoddsratiosfromlogitmodelsthat correspondtotheOLS Models2–4.21 Fromtheimpliedmarginal ef- fects(reportedinsquarebrackets)weseethatthelogitandtheOLS modelsproducesimilarestimates,and,moreimportantly,thatthetime patternofthelogitestimatesalsoindicateapositiveeffectofjobdis- placementoncrime.TheresultsinModel4(OLS)andinModel8(Logit) indicatethatjobdisplacementincreasestheprobabilityofcommitting crimebyabout20percentinthebaselineyear,22withestimatedeffects thatweakeninsubsequentyears.Inthefollowing,wewilluseModel8 asourmodelofreference.

5.2. Robustnessofmainresults

Table 3presents several robustnesschecks. First,theresults pre- sentedinModel1arefromtheexactsamemodelastheonepresented inModel8ofTable2,butnowweonlyreporttheestimatedcoefficients forrelativeyears−1to+2andthedummyindicatingthattheworker isinthedisplacementgroup.

Oneconcernwithourreferencemodelisthatthesamemancanbe presentinthecomparisongroupinseveralbaselineyears.Whilethis

20Inthesubsequenttableswewillrestrictattentiontotheestimatesfor−1 to+2.Wedothisforthreereasons.First,pointestimatesthatmighthavebeen somewhatdifferentornotsignificantintheindividualfixedeffectsmodelare notreported.Second,pointestimatesthatmightbebiasedduetounbalanced panel(thepanelbecomesunbalancedfrom+3;seefootnote17)arenotreported.

Third,itsuccinctlyconveysthepointestimatesofmaininterest.

21Negativebinomialmodelstookanexcessivelylongtimetoconverge,and oftenfailedtoconvergealtogether.Formodelsthatdidconverge,liketheone correspondingtoModel7ofTable2,resultswerequalitativelythesameasthose reported.

22Dividingthemarginaleffectestimate(0.38percentagepoints)fromModel4 bythebaselinecrimerate(1.96percent)yieldsarelativeeffectof19.4percent.

ThemeanmarginaleffectimpliedfromthelogitestimateofModel8(provided insquarebrackets)producesanearlyidenticalrelativeeffectestimate.

(8)

Table2

Mainresults:effectoncrimeofbeingdisplacedinrelativeyear0(baselineyear).

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Dependent variable: Any crime in the given relative year

Displaced (dummy) 0.0044 0.0038 1.1933 1.1713

(0.0004) (0.0004) (0.0163) (0.0160)

[0.0035] [0.0031]

Estimate of effect of displacement in given relative year

5 0.0061 0.0038 1.1522

(0.0006) (0.0006) (0.0244)

[0.0028]

4 0.0062 0.0041 1.1765

(0.0006) (0.0006) (0.0258)

[0.0031]

3 0.0073 0.0054 1.2446

(0.0006) (0.0006) (0.0279)

[0.0043]

2 0.0060 0.0043 1.2108

(0.0006) (0.0006) (0.0284)

[0.0038]

1 0.0077 0.0064 0.0020 0.0019 0.0017 1.3265 1.1115 1.1085 (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0308) (0.0263) (0.0264)

[0.0055] [0.0021] [0.0020]

0 0.0093 0.0082 0.0038 0.0038 0.0035 1.4367 1.2039 1.1987 (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0329) (0.0285) (0.0286)

[0.0071] [0.0036] [0.0035]

1 0.0089 0.0080 0.0036 0.0035 0.0032 1.4319 1.1999 1.1912 (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0332) (0.0293) (0.0293)

[0.0070] [0.0036] [0.0034]

2 0.0079 0.0071 0.0026 0.0026 0.0022 1.3902 1.1650 1.1537 (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0331) (0.0293) (0.0292)

[0.0065] [0.0030] [0.0028]

3 0.0072 0.0065 0.0020 0.0020 0.0015 1.3635 1.1426 1.1323 (0.0006) (0.0006) (0.0006) (0.0006) (0.0006) (0.0345) (0.0304) (0.0303)

[0.0061] [0.0026] [0.0024]

4 0.0068 0.0061 0.0017 0.0017 0.0010 1.3593 1.1390 1.1261 (0.0006) (0.0006) (0.0007) (0.0007) (0.0007) (0.0371) (0.0326) (0.0324)

[0.0060] [0.0026] [0.0023]

5 0.0071 0.0065 0.0020 0.0021 0.0012 1.3873 1.1625 1.1548 (0.0006) (0.0006) (0.0007) (0.0007) (0.0007) (0.0411) (0.0359) (0.0359)

[0.0064] [0.0030] [0.0028]

Estimation model Covariates included (in addition to dummies for crime in comparison group in relative years)

OLS OLS OLS OLS OLS FE Logit Logit Logit

No controls No controls except dummies for age and calendar year

No controls except dummies for age and calendar year

All observed controls given in Appx. B

No controls except dummies for age, calen-dar year and individual fixed effects

No controls except dummies for age and calendar year

No controls except dummies for age and calendar year

All observed controls given in Appx. B

Mean of dependent variable in comparison group

0.0196 0.0196 0.0196 0.0196 0.0196 0.0196 0.0196 0.0196

R-squared 0.0006 0.0104 0.0104 0.0146 0.1696

N 10,526,937 10,526,937 10,526,937 10,526,937 10,526,937 10,526,937 10,526,937 10,526,937

Note:Estimatesofhowmuchhigherthelikelihoodofcrimeisamongthedisplacedthanthecomparisongroup(andthepre-displacementperiodofthedisplacedfor Models3–5and7–8)inthegivenrelativeyear(0indicatesthebaselineyear).Odds-ratiosreportedforthelogitestimationswithimpliedmeanmarginaleffectsin squarebrackets,andmarginaleffectsreportedforOLSregressions.anddenotesignificanceatthe5and1percentlevels.Robuststandarderrorsinparentheses correctedfornon-independentobservationsforthesameindividual.

shouldnotbiasthepointestimates,itraisesconcernsthattheestimated standarderrorsaretoosmall(butrecallthatthisconcernislimitedby thefactthatwealwaysclusterontheindividuallevel).InModel2of Table3we haverandomly selectedno morethanone baselineyear foreachworkerinthecomparisongroup.23Asexpected,thisproduces

23 Tocreateasamplerepresentativeofouroriginalcomparisongroupsam- ple,thiswasdoneasfollows.First,eachcomparisongroupworkerhadann/10

similarpointestimatesasinourreferencemodel,butthesamplesize drops substantiallyandtheestimated standarderrorsbecomebigger.

probabilityofbeingincludedinthesample,wherenrepresentsthenumberof timestheworkerwasincludedintheoriginalcomparisongroupsample.(Recall, workersinthecomparisongroupcouldbeincludedforupto10baselineyears).

Next,forincludedworkers,oneoftheirrecordswasrandomlychosenforinclu- sion.Ifwehadomittedthefirststep,therestrictedcomparisongroupsample wouldhavebeenover-representedbyworkerswithlessconsistentemployment.

(9)

Table3

Robustnesschecksoftheeffectoncrimeofbeingdisplacedinrelativeyear0(baselineyear).

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Dependent variable: Any crime in the given relative year

Estimate of effect of displacement in given relative year

1 1.1085 1.1052 1.0735 1.0784 1.1317 1.1282 1.0959

(0.0264) (0.0377) (0.0368) (0.0254) (0.0497) (0.0373) (0.0268) [0.0020] [0.0022] [0.0014] [0.0016] [0.0019] [0.0021] [0.0018]

0 1.1987 1.2011 1.1785 1.1336 1.1124 1.1993 1.1930 1.1628 (0.0286) (0.0412) (0.0400) (0.0266) (0.0503) (0.0400) (0.0294) (0.0307) [0.0035] [0.0041] [0.0031] [0.0026] [0.0016] [0.0032] [0.0034] [0.0027]

1 1.1912 1.1879 1.1441 1.1380 1.2257 1.1893 1.1927 1.1766 (0.0293) (0.0414) (0.0402) (0.0275) (0.0546) (0.0411) (0.0303) (0.0317) [0.0034] [0.0038] [0.0026] [0.0027] [0.0031] [0.0031] [0.0034] [0.0029]

2 1.1537 1.1788 1.1088 1.1042 1.2058 1.1464 1.1466 1.1480 (0.0292) (0.0423) (0.0400) (0.0275) (0.545) (0.0407) (0.0300) (0.0316) [0.0028] [0.0037] [0.0020] [0.0021] [0.0028] [0.0024] [0.0026] [0.0025]

Displaced (dummy) 1.1713 1.1575 1.1313 1.1240 1.1568 1.1861 1.1916 1.1399 (0.0160) (0.0203) (0.0211) (0.0149) (0.0282) (0.0226) (0.0170) (0.0161) [0.0031] [0.0033] [0.0024] [0.0024] [0.0022] [0.0030] [0.0034] [0.0023]

Sample redefinitions Reference model (i.e.

Model 8, Table 2 )

Comparison group is a random draw from main sample, which ensures that an individual is never present in more than one base-line year (see Section 5.2 . for details)

Displaced if

PDR > 0.9

(instead of PDR > 0.3)

Workers in comparison group remain in plant through

1 (instead of through baseline year)

Tenure 5 (instead of tenure 2)

Plant size 50 (instead of plant size 10)

Comparison group if PDR 0.1 (in- stead of PDR 0.3)

Excluding workers who committed crime in t-1 (instead of in-cluding them)

Mean of dependent variable in comparison group

0.0196 0.0198 0.0196 0.0212 0.0152 0.0179 0.0191 0.0180

N 10,526,937 1,842,131 10,093,950 12,931,965 5,600,492 6,147,571 7,384,630 9,321,536

Note:Estimatesoftheeffectofdisplacement(inthebaselineyear,denoted0)oncrime(dummy)inthegivenrelativeyear.Estimatedusinglogitmodels(odds-ratios reported,withimpliedmeanmarginaleffectsinsquarebrackets).AllcovariatesdescribedinAppx.Bincludedinallmodels(coefficientsforthemandforeffect estimatesinyears3–5notreported).anddenotesignificanceatthe5and1percentlevels.Robuststandarderrorsinparenthesescorrectedfornon-independent observationsforthesameindividual.

Thepoint estimates,however,remain significantat the onepercent level.

Anotherconcernisthatlessproductiveworkersmightbethefirstto belaidoff inassociationwithmass-layoffs.Totheextentthatlaidoff workerscommitmorecrimeirrespectiveofdisplacement,this would notbiasoureffectestimateoftheoveralleffectofjobdisplacementon crime(sincewecontrastcrimeratesafterdisplacementwithratesofthe samemenbeforedisplacement).Biascouldarise,however,ifthecrimi- nalbehaviorofsuchworkerswasmoreresponsivetodisplacement.We cangetanideaofthispossiblebiasbyrestrictingthesampletoworkers separatedfromplantsthatclosed,sinceclosingplantsarenotretaining employeesandthushavenodiscretionwithrespecttowhomtolay- off.InModel3ofTable3werestrictthedefinitionofthedisplacedto workersseparatedfromaplantthatdownsizedbymorethan90percent (andthecomparisongroupremainsthesameasbefore).Aswecansee, thisreducesthepointestimatessomewhat,suggestingsome differen- tialselectionofmorecrime-proneworkersinourmaindisplacedworker sample.Nonetheless,theestimatesremainlargeandhighlysignificant.

InModel4weremovetherequirementthattheworkersinthecom- parisongroupremainintheplantthroughoutthebaselineyear.This requirementcouldgenerateselectionoflesscrime-proneworkers(on unobservables)intothecomparisongroup.Removingthisrequirement alsoimplies,however,thatthecomparisongroupcannowincludework- erswhoareseparatedfromaplantinassociationwithasmallerdown- sizing(e.g.withmasslayoffsof29percent).Onemayarguethatthis resultsinsomepartlytreated(i.e.separatedinassociationwith29per-

centdownsizing)workersendingupinthecomparisongroup,thereby attenuatingtheeffectestimate.24 Inlinewithwhatwewouldexpect fromtheattenuationstory,wesee fromModel4thattheeffectesti- mateofthebaselineyeardeclinessomewhatunderthisrestriction,but itremainssignificant.

InModels5and6wecheckforrobustnesstothesampleselection criteriarelatedtoincreasingtherequirementsfortenureandplantsize.

Theeffectestimateforthebaselineyearissomewhatlowerwhenwe requiretenureofatleast5years(insteadof2years),buttheeffectesti- matesarelargerinyears1and2(seeModel5).Restrictingtoplantswith atleast50employees(insteadof10)produceseffectestimates(Model6) thatarealmostidenticaltotheestimateofourreferencemodel.Finally, wecheckthattheresultsarenotsensitivetothedownsizingrequirement ofthecomparisongroup.Inthemainspecificationwerequiredthatthe plantofemploymentdidnotdownsize30percentormorearoundthe baselineyear,whileinModel7wehavechangedthisrequirementto

24AsnotedinSection3,previousstudieshavetypicallyrequiredthatthework- ersinthecomparisongroupremainintheirplantofemploymentthroughout theobservationwindow(forusthatcouldbethrough+5).Theadvantageofthis requirementisthateffectestimatesarenotattenuatedbythepresenceinthe comparisongroupofworkerswhoarelaidoff inassociationwithdownsizing eventsafterthebaselineyear(“partlytreated”).Thepossibledisadvantageis thatthecomparisongroupthencomprisesverystableworkerswhomayexhibit differenttrendsincriminalbehavior(e.g.steeperdeclinesovertime),which mightresultinupward-biasedeffectestimates.

Referanser

RELATERTE DOKUMENTER

The Severity of Behavioral Changes Observed During Experimental Exposures of Killer (Orcinus Orca), Long-Finned Pilot (Globicephala Melas), and Sperm (Physeter Macrocephalus)

interview that, “Even if problematic, the Pakistani leadership has realised it has an internal problem it needs to control.” 4 While the Afghan government has repeatedly

228 It further claimed that, up till September 2007, “many, if not most, of the acts of suicide terrorism and attacks on the Pakistani Armed Forces since the Pakistan Army's

Keywords: gender, diversity, recruitment, selection process, retention, turnover, military culture,

The system can be implemented as follows: A web-service client runs on the user device, collecting sensor data from the device and input data from the user. The client compiles

This report documents the experiences and lessons from the deployment of operational analysts to Afghanistan with the Norwegian Armed Forces, with regard to the concept, the main

The increasing complexity of peace operations and the growing willingness of international actors to assume extended responsibil- ity for the rule of law in often highly

Overall, the SAB considered 60 chemicals that included: (a) 14 declared as RCAs since entry into force of the Convention; (b) chemicals identied as potential RCAs from a list of