ContentslistsavailableatScienceDirect
Labour Economics
journalhomepage:www.elsevier.com/locate/labeco
Education and criminal behavior: Insights from an expansion of upper secondary school ☆
Olof Åslund
a, Hans Grönqvist
b,∗, Caroline Hall
c, Jonas Vlachos
daInstitute for Evaluation of Labor Market and Education Policy (IFAU) and Department of Economics, Uppsala University, Sweden
bDepartment of Economics, Uppsala University, Sweden
cIFAU and Uppsala Center for Labor Studies (UCLS), Sweden
dDepartment of Economics, Stockholm University and the Research Institute of Industrial Economics (IFN), Sweden
a r t i c le i n f o
JEL classification:
K42 Keywords:
Education Delinquency
a b s t r a ct
WestudytheimpactonlongandshortruncriminalbehaviorfromalargescaleSwedishreformofvocational uppersecondaryeducation,extendingprogramsandaddingmoregeneraltheoreticalcontent.Thereformdirectly concernsagegroupswherecriminalactivityishighandindividualswhoareoverrepresentedamongcriminal offenders.Usingdetailedadministrativedataweshowthatthereformledtoareductioninpropertycrime,but nosignificantdecreaseinviolentcrime.Theeffectismainlyconcentratedtothethirdyearafterenrollment, whichsuggeststhatbeinginschoolreducestheopportunitiesand/orinclinationstocommitcrime.
© 2017PublishedbyElsevierB.V.
1. Introduction
Education can generate large private and social gains in many dimensions.Theliteraturenotonlydocumentsstrongassociationswith earnings, employment and other indicators of economic well-being, but also with health, family formation and crime (Oreopoulos and Salvanes, 2011). Having secured basic education for essentially all residents, many developed societies now focus reforms on how to extendandimprovesecondaryeducationforbroadergroupsofstudents thanthosehistoricallytargetedforadvancementintohighereducation (seeBenavotetal.,2006foranoverview).
Acommonlyheldviewisthatstudentsarenotsufficientlyprepared forhighereducationandlackthetypeofgeneralknowledgetypically required by thelabor markets of today. Consequently, andperhaps mostlyso forvocationaleducation, thereis acommontrendtoward secondary education curricula becoming more like primary school curriculawithabroaderrangeofsubjects,lessspecializationandmore integratedthemesinorder toprovide allstudentswithopportunities fortertiaryeducation(Sahlberg,2007).
But reforming secondary education in a way that satisfies these criteriamayleadtoexternalitiesthathavenotbeenfullyappreciated.In particular,suchpoliciesruntheriskofalsoincreasingdropoutratesand raisingyouthunemploymentbypotentiallymakingtheschool-to-work
☆ We are grateful to Randi Hjalmarsson, Helena Holmlund, Matthew Lindquist, Lance Lochner, Martin Lundin, two anonymous referees and seminar participants at IFAU, Tinbergen, Umeå University, University of Turku, Trondheim, the EALE conference in Turin, and the IFAU/UCLS workshop on economics of education and education policy for valuable comments.
∗Corresponding author.
E-mail addresses: [email protected] (O. Åslund), [email protected] , hans.gronqvist@sofi.su.se (H. Grönqvist), [email protected] (C. Hall), [email protected] (J. Vlachos).
transitionmoredifficult.1This potentialtradeoff ismadeparticularly starkbytheliteratureontheimpactofeducationoncriminalbehavior.2 Sincecrimetypicallyhassignificantnegativeexternalities,thesocial returns to education may be substantially higher than the private returns ifeducation has crimereducing effects. Onthe other hand, dropoutsfromsecondaryeducationarehighlyoverrepresentedamong those involved in criminal activities.3 Likewise, it is intuitive that difficultiestotransitfromschooltoworkmayaffectcriminalbehavior.
This paper studies the impact on crime by a major reform of Swedish upper secondary vocational education in the beginning of the 1990s. The reform extended vocational programs from two to three yearsand added more general theoretical contentto thecur- riculum. Theintervention took placefor age groupswherecrime is
1Bishop and Mane (2001) and Goodman (2012) find that higher requirements can improve labor market outcomes while other studies find that they increase dropout rates ( Dee and Jacob, 2007; Ou, 2009 ). Note that Clark and See (2011) do not find that tougher education standards have an impact, either on dropout rates or on earnings. Regarding the school-to-work transition, Hanushek et al. (2017) provide recent evidence on the trade- off between youth employment and long-term labor market flexibility by vocational and general secondary education.
2Hjalmarsson and Lochner (2012) provide a recent literature review.
3According to Harlow (2003) , 75% of US inmates did not complete high school. Our own calculations for Sweden reveal that about 50% of the individuals who were sentenced to prison in 2005 had not completed high school.
https://doi.org/10.1016/j.labeco.2017.11.007
Received 1 December 2016; Received in revised form 16 November 2017; Accepted 21 November 2017 Available online xxx
0927-5371/© 2017 Published by Elsevier B.V.
common,andprimarily affectedsegmentsof theyouthcohortswith particularlyhighcrimerisks.Ouranalysisismadepossiblebyaccess torichlongitudinal population-widemicrodataincludingdetails on allcriminalconvictionsinSwedish courts.Wetake advantageofthe factthatthereformwasimplementedgraduallyovertimeindifferent municipalities.AccordingtoHall(2012),thereformincreasedaverage educationalattainment,butalsoledtoasubstantialincreaseindropout ratesamonglow-performingstudents.
Wecomplementandaddtotheexistingliteratureinseveralways.
Mostimportantly,thenatureofthereformandtherichnessofthedata allowustoprovide oneofthefirstpieces ofevidenceonthecausal effectoncriminalactivityofmakingsecondaryeducationmoregeneral.
Wethereby bringanotherpieceof evidencetothediscussion ofthe trade-offs involved when deciding between general and vocational secondaryeducation. Ourfindingsarethereforelikely tobe relevant outsidethespecificsettingthatweanalyze.Ourpaperisalsoamongthe firsttoprovidedesignbasedevidenceontheconsequencesofeducation policyatasomewhathighermargin:theuppersecondarylevel.Infact, thevastmajorityofpreviousstudiesconcernreformsaffectingcompul- soryschool.4Moreover,wecanstudybothlong-andshort-runeffects oncriminalbehavior,aswellasseparatebetweencrimescommittedon schooloperatingdaysandwhentheschoolisinrecess.Also,westudy asymmetriceffectsofthereformacrossdifferentstudentcharacteristics, amongthem the predicteddropout risk andprevious school perfor- mance.Taken together,thesesources of variationenableustoshed lightonthemechanismsthroughwhicheducationcanaffectcrime.
There areseveral theoreticalreasons astowhyincreased educa- tionalattainmentcouldaffect crime. Firstof all,educationtypically raises expected earnings from legitimate work, thereby making crime relatively less appealing (Lochner, 2004).5 Second, one can hypothesize that education directly affects preferences, potentially makingindividualsmoresocialized,future-oriented,andrisk-averse.6 Åkerlundetal.(2016)showthatindividualswithhighdiscountrates aremorelikelytoengageincrime.Thesefactorscouldthendecrease theprobability ofcrime. Third,merely beingin schoolmayinitself affecttheriskofcrimebyincapacitation;i.e.byparticipatinginedu- cationindividualshavelesstimetoengageincriminalactivities(e.g.
Luallen,2006).Ifthereishabitformationincrime,thensuchdecreases incriminalinvolvementmayleadtoevenlargercrimereductions in thelongrun.Ontheotherhand,schoolsalsoofferarenaspotentially increasingcontactsandtherebyfrictionsbetweenyouth,thustriggering certaintypesof crime(aso-called concentrationeffect).Educational reformcanalsoaffectthepeercompositionforagivenindividual,mean- ingexposuretocontextswithvaryingdegreesofcrimepropensity.The incapacitationandhumancapitalmechanismssuggestnegativeeffects ofeducationoncrime,whilepeerinfluencesareambiguousinsign.
We find robust evidence that increased accessto prolongedand moregeneraleducationformalevocationalstudentsleadtoareduction inpropertycrimebyabout20%,butnosignificantdecreaseinviolent crime.7Theseeffectsareconcentratedamongstudentsinthelowerhalf of thecompulsoryschoolGPA distribution.Wealso showthat these
4One exception is the study by Machin and Marie (2012) who examine the conse- quences of the expansion of the UK post-compulsory schooling system in the 1980s and 1990s. Since the study compares the evolution of crime rates across birth cohorts it is essentially using a time-series research design.
5Several studies have shown that better labor market opportunities reduce crime (e.g.
Grogger 1998; Gould et al., 2002; Machin and Meghir, 2004; Grönqvist, 2013 ). This effect is potentially counteracted if education raises the opportunities to commit crime and the returns to certain types of crime.
6In an influential study, Becker and Mulligan (1997) posit that people could learn to be more future-oriented. Perez-Arce (2011) demonstrates empirically that college students in Mexico who were randomly admitted from a pool of applicants were more patient than individuals in the control group, which indicates that education has an impact on time preferences.
7We follow the convention in the literature and focus on males because they are sub- stantially overrepresented among criminal offenders. But note that we also present some results for females.
effectsarepresentintheageinterval16–20butnotathigherages,and aremainlyconcentratedduringtheaddedthirdyearinschool.8These findingssuggestthatincapacitationisthemainmechanismbehindthe reductionincriminal behavior.However,sinceHall(2012)doesnot findanimpactonfutureearningsbythereform,theresultsshouldnot be read asarefutationof thehypothesis thatthealternativecostof crimeisalsorelevant.
Havingadegreefromuppersecondaryschoolisnegativelycorre- latedwithcriminalbehavior.Butourresultsrevealnoclearevidence thattheincreaseindropoutratesinducedbythereformincreasedcrime.
Onepossibilityisthatadecreaseincrimebyincreasedschoolingiscoun- teractedbyanincreaseincrimeduetothehigherlikelihoodofdropping outofschool.Inlinewiththiswefindthatamongthestudentspredicted tobemostlikelytodropout,theestimatedreformimpactiszero.
Thequasi-experimentalset-upofthereformallowsustodealwith thechallengeofdisentanglingthecausalcomponentoftheassociation between crimeandeducation.Obviously,anegativecorrelationmay arise for several reasons, e.g. family background or innate abilities making education more (less) attractive/likely and working in the opposite direction for the probability of engaging in crime. Despite a voluminous multi-disciplinary literature (see Lochner, 2010 or Rudetal.,2013foroverviews),relativelyfewstudieshavebeenable toachievecrediblecausalidentification.LochnerandMoretti(2004)is oneexception,findingthatschoolingdecreasestheprobabilityofincar- cerationusingvariationincompulsoryschoolinglawsacrossUSstates.
Machinetal.(2011)isanotherexample,concludingthateducationcan beanimportantmeasureforreducingcrime,basedontheirUKstudy.
RecentSwedishstudiesexploitingacompulsoryschoolreformalsofind a negativeimpact of compulsoryeducationon crime(Meghiretal., 2011;Hjalmarssonetal.,2015).Thesequasi-experimentalstudieshave focusedoncompulsoryschoolingreformsorpoliciesthataffectedthe earlystagestheeducationalsystem,whileweconsideruppersecondary school. Further, these studies link educational expansion to adult crime while our study directly concerns age groups wherecriminal activityis relativelyhigh,as wellasthefuture criminalbehaviorof these groups. Our studybears similarities with the recent work on NorwaybyBrugårdandFalch(2013).Usinghighschoolstructureand geographic information in anIV setting they show thathigh school achievementdecreasescrime.Comparedtoourstudy,thedatadonot allowBrugårdandFalch(2013)tostudythetimeprofileofcriminal activity,hencemakingitdifficulttodrawconclusionsconcerningthe mechanismsbehindthedecline.
A second strand of the quasi-experimental literature studies the contemporaneous link between education and crime among youths (JacobandLefgren2003;Luallen2006;Anderson2014;Berthelonand Kruger2011).9Thesepapersconcludethateducationreducescrimeby incapacitatingindividuals.Thissquareswellwithouranalysis,which expandsonthisresearchbyusingmoredetaileddataandstudyingthe long-runconsequencesbyfollowingindividualsforaperiodstretching atleast15years.
Therestof thepaperisoutlinedasfollows.Section2givessome institutional backgroundon crimeandthecriminal justicesystemin Sweden,andontheeducationalreformunderstudy.Section3describes thedataandtheresearchdesign.Section4presentstheempiricalanal- ysis, starting with the baseline results, then turning to investigate effectsbybackgroundcharacteristics,andfinallytryingtodisentangle themechanismsatwork.Section5concludes.
8We do however not find any clear differences in the effect of the reform on crime committed on weekdays versus weekends or for crime on school days compared to school breaks.
9Related is also Deming (2010) who finds that winners of school choice lotteries in middle or high school are less likely to be arrested and incarcerated seven years after the school assignment.
Fig. 1. Share of convicted persons by age relative to the national conviction rate.
Notes : The sample includes all men aged 16 and above. The year of observation is 2005.
2. Institutionalbackground
2.1. CrimeinSwedenandtheSwedishcriminaljusticesystem10
TheSwedishcrime rateis in linewiththatin manyotherOECD countries,atleastwhenconsideringafewofthemostcommontypes of crimes.11 Youths represent themost criminallyactive age group.
Fig.1plotstheshareofconvictedmalesin2005byagerelativetothe nationalconviction rate.Anumberabove(below)one indicatesthat theshareofconvictedmalesforthatagegroupishigher(lower)than theaverageforallagegroups.Itisclearthattheconvictionratepeaks alreadybeforeage20,andthenfallssharply.Thisisespeciallytruefor violentandpropertycrime.
InSweden, thegeneral courts deal with both criminal andcivil cases.Thegeneralcourtsareorganizedinathree-tiersystem:district courts,courtsofappealandtheSupremeCourt.Thedistrictcourtis thecourtoffirstinstance.Criminalcasesarenormallyinstitutedwhen apublicprosecutorinitiatesprosecutionproceedingsagainstasuspect bysubmittinganapplicationtoadistrictcourt.Thecourtrulesoncases afteramainhearingattendedbybothpartieswhostatetheirclaimsand othercircumstancesrelatingtothecase.Criminalcasesarenormally triedbyonejudgeandthreelayjudges.Thosewholacktheeconomic meanstotakeadvantageoftheirrightsareentitledtopubliclegalaid.
Theageof criminalliability is15.All individualsabove this age aretreatedinthesamejudicialsystem.Somespecialrulesdohowever applyforjuveniles.Themaindifferenceisthatcasesinvolvingyouths
10This section closely follows Grönqvist and Niknami (2014) .
11For instance, in 2006 the total number of assaults reported to the police per 100,000 inhabitants amounted to 845. The same year, official crime statistics from the US police revealed 787 recorded cases of assaults per 100,000 inhabitants, and the corresponding number for Canada was 738 ( Harrendorf et al., 2010 ). The number of reported burglaries per 100,000 persons was in Sweden 1,094. In the US and in Canada the equivalent numbers were 714 and 680, respectively.
aretobedealtwithpromptly.Juvenilesalsoreceiveadiscountontheir sentencedependingontheagewhentheycommittedthecrime.
2.2. TheSwedisheducationsystemandthereformunderstudy12
Aftercompleting nineyearsof compulsoryschooling,studentsin Sweden areentitledtouppersecondary education. Schooling atthe uppersecondarylevelisvoluntarybutthevastmajorityenrolls.Inthe end ofthe1980s,almost90%continueddirectlytouppersecondary school.Inrecentyears,theenrollmentratehasbeenashighas98%
(NationalAgencyforEducation,2008).Uppersecondaryschoolcom- prises severaldifferenteducationaltrackstowhichindividualsapply basedontheircompulsoryschoolGPA.13
Theuppersecondaryschoolsystemwentthroughamajorreform in the beginningof the 1990s. Before the reform, upper secondary educationconsistedofafewacademicandseveralvocationaltracks.
The vocational trackswere two years longand consisted mainlyof vocational training. Theacademictrackstypicallylasted three years andpreparedthestudentsforhighereducation.14Whiletheacademic tracksgranted eligibilityforuniversitystudies,graduates fromvoca- tionaltrackshadtosupplement theirdegreewithadditionalcourses beforetheycouldapplytoacollegeoruniversity.Intheyearsleading uptothereform,around45%oftheuppersecondaryschoolstudents wereenrolledinvocationaltracks.Construction,electricalengineering andcaringservicesrepresentedsomeofthemostcommontracks.Table A1liststhevocationaltracksthatexistedatthetime.
12This section draws heavily on Hall (2012) .
13Individuals who are older than 20 when they begin upper secondary education are not entitled to attend a general upper secondary school, but instead enter the adult education system. Within this system, both those who lack any upper secondary education and those who dropped out before graduating can finalize a degree. It is also possible to supplement a degree with additional courses. For more information on the adult education system, see Stenberg (2009) .
14There also existed a few two-year academic options as well as one four-year option.
University/College
Upper secondary
school:
3-year*
academic tracks (from age 16)
Compulsory school 9 years
(from age 7) Upper
secondary school:
2-year vocaonal
tracks (from age 16)
Upper secondary
school:
3-year vocaonal tracks, with
increased academic content (from age 16)
Upper secondary
school:
3-year academic
tracks
(from age 16)
Compulsory school 9 years (from age 7) University/College
Before the reform Aer the reform
Fig. 2. The Swedish school system before and after the reform.
Note: ∗) There was also a small proportion of two-year academic tracks as well as one four-year academic track. Regardless of track length, all academic tracks gave eligibility to university studies.
The general aim of the reform was to enhance the quality of educationand increasethe flexibilityof theuppersecondary school system.Thelargestchangesconcernedthevocationaltracks:Theywere prolongedfromtwo-tothree-yeartracks,andreceivedaconsiderably largeracademiccontent.These changesweremotivated bytheview thattoday’sworkingliferequiresmoregeneralknowledge,aswellas bythedesiretoenableallstudentstoenrollinuniversitystudies.As aresultof thereform,allstudentsgraduatingfromvocationaltracks attainedbasiceligibility15 foruniversitystudies.Fig.2illustratesthe Swedishschoolsystembeforeandafterthereform.
The reform was preceded by a nation-wide pilot period during 1988−1993inwhichnewthree-yearvocationaltracksweretriedout in several municipalities.16 The vocational tracks in the pilot had increasedacademiccontentcomparedtotheregulartwo-yeartracks.
While Swedish was the only general theoretical subject includedin all two-year tracks, the pilot tracks also contained English, Social Studies andanelective course.Math appearstobe by farthe most common choice of elective.17 Another difference between the two- andthree-year tracksis thatthe latterlocated a larger shareof the vocationaltrainingtoactualworkplaces.18
Thepilotcomprisedaround6000educationalslotsin1988,10,000 in1989,and11,200in1990.Ontopofthis,therewasaverylimited
15Note that ‘basic eligibility ’ does not mean eligibility to all university studies as some programs have special requirements.
16This extensive pilot scheme was the outcome of a thorough evaluation of the vo- cational upper secondary education conducted by a government appointed committee (ÖGY). See e.g. Prop. 1987/88:102 for a description of the pilot.
17The National Board of Education (1990a) reports that 86% of the students in 1988 chose to study Mathematics.
18About 40% of the extended time seems to have consisted of general theoretical courses for most tracks (own calculations based on information provided in Government Bill (1987/88 ): 102, p. 35–39). Compared to the pilot tracks, the three-year tracks that were implemented after the reform contained even more academic subjects and somewhat less training in workplaces.
‘pre-pilot’ in 1987whichonlycontained 500slots.19 Thenumberof three-year slotsthereby corresponded tobetween 1and20% of the total numberofslotsinvocationaltracks.Forthecurrent studyitis important to pointout that the total number of slots in vocational tracksdidnotexpandduetothepilot;ratheraclassinthepilotalways replacedaclassinacorrespondingtwo-yeartrack.Theadmissionrules forthepilottrackswerethesameasfortheregularvocationaltracks.
TheNationalBoardofEducationwasresponsibleforallocatingthe pilotslotsamongthedifferentvocationaltracksaswellamongthemu- nicipalities.Theallocationofslotsamongthedifferenttrackswasdone proportionally;thegoal wasthateach trackwouldreceivethesame shareofthree-yearvocationalslotsastheyreceivedoftwo-yearslots.
Therewerehoweversomedeviationsfromthisprinciple,e.g.trackswith asmallernumberofslotsweresomewhatoverrepresented.Thealloca- tiondecisionwasfurtherrestrictedbythefactthatinthebeginningof thepilotperiodnocurriculahadyetbeenpreparedforsomeofthethree- yeartracks.This meantthatallof the18 three-yeartracksavailable couldnotbeincludedinthepilotthefirstyears.TableA2liststhetracks thatwereincludedeachyearaswellastheirnumberofavailableslots.20 The government stipulated that the pilot be distributed between regionswithdifferentindustryandpopulationstructures.Italsostated that differentregionsshould participatetodifferent extents:insome regionsalloralargeshareofthevocationaltracksshouldbeprolonged to three-year tracks; in others only a few of the tracks should be prolonged.Themotivebehindtheserequirementswastogetanideaof howthemoreextensiveworkplacetrainingworkedindifferenttypes
19The 1987 tracks were somewhat different as they did not contain more extensive workplace training. The description of the implementation process below is based on SOU (1989 ):106 and refers to the actual pilot. There is no available documentation of the implementation of the pre-pilot scheme.
20The share vacant slots in the pilot varied from 0.03 in 1987 to 0.07 in 1990. The share vacant slots was in general somewhat lower for the pilot tracks than for the regular two-year vocational tracks.
0.1.2.3.4.5share of municipalities
1986 1987 1988 1989 1990
year
1-25% 3-year tracks 26-50% 3-year tracks 51-75% 3-year tracks 76-100% 3-year tracks
Fig. 3. Share of municipalities that participated in the pilot each year, and the extent of their participation.
Note : ‘% 3-year tracks ’ is the percent of all vocational tracks available in a municipality which were part of the pilot. Source : Fig. 1 in Hall (2012) . The calculations are based on the Upper Secondary School Application record.
oflabormarkets,aswellasofthestrainonthelocallabormarketifit wasimplementedonalargescale.Ontopofthesecriteria,theNational BoardofEducationtriedtoassesswhetherthelocallabormarketwould beabletoarrangetheextendedworkplacetraininginarelativelyshort time.Tojudgethistheyrelieduponrecommendationsfromemployer and union representatives in different sectors. This concern seems tohave meantthatsomeprioritywasgiventomunicipalitieswitha traditionofinvolvingworkplacetraininginthevocationaleducation.
The initiative to participate always came from the municipalities themselvesastheyhadtoapplyinordertobeconsidered.Theinterest waslarge;eachyearthemunicipalities’ demandforpilotslotsbyfar exceededthenumberofavailableplaces.
Around 70% of Sweden’s 284 municipalities offered vocational tracksat thetime.Studentslivingin theothermunicipalitieshadto attend school in a nearby municipality if they wanted to obtain a vocationaldegree.21Whenthepilotwaslaunchedin1988,about40%
of themunicipalities weregranted participation. In1990, the share hadincreasedtoabout52%. Theextenttowhich themunicipalities participated alsoincreased each year as more trackswere included inmunicipalitiesthatalready participated.Fig.3showstheshareof municipalitiesthatparticipatedeachyearaswellashowtheextentto whichtheyparticipatedvariedovertime.
Allthroughthepilotperiodmostparticipatingmunicipalitiesoffered bothtwo-andthree-yearvocationaltracks.Sometimesamunicipality wouldofferthesametrackbothasatwo-andasathree-yearoption.
Alsoinmunicipalitiesthatonlyofferedeithertwo-orthree-yeartracks, students could in some cases have a choice of program length if a nearby municipality offered tracks of a different length. Hence, the pilotgeneratesasettingwheresomestudentsweregiventhechoiceof enrollinginamoreacademicthree-year,ratherthanaregulartwo-year, vocationaltrack. Thedegree towhichan individualhadthis choice dependedonwherehe/shelivedaswellasonwhichyearhe/shebegan uppersecondaryschool.
21Students generally attended a school in their municipality of residence, but if the track they desired to follow was not offered they could instead choose to attend in a nearby municipality.
3. Dataandresearchdesign
In this section we start by describing the data and present de- scriptivestatistics.Wethenpresenthowweexploitthepilotscheme discussedintheprevioussectiontoidentifyplausiblereformeffects.In Section3.4wediscusspotentialthreatstoidentification.
3.1. Data
Our dataoriginatefrom severaladministrative registerscollected andmaintainedbyStatisticsSweden.Theregisterscontaininformation ontheentireSwedishpopulationaged16andaboveeachyearfrom 1985to2007.ThesedatahavebeenlinkedtotheSwedishConviction RegisterkeptbytheNationalCouncilforCrimePrevention(BRÅ).We obtainedcompleterecordsofallcriminalconvictionsduringtheperiod.
This meansthat weareable tofollow allindividualsin oursample foratleast15years.Thedataincludeinformationoncrimetype,date of thecrime, aswellasthe sentenceruledby thecourt,andcovers convictionsinSwedishdistrictcourts(thecourtoffirstinstance).One convictionmayincludeseveralcrimesandweobserveallcrimeswithin asingleconviction.Theconvictiondataexcludesomeoffensessuchas speedingtickets,butincludee.g.drivingwithoutalicenseanddriving undertheinfluence.Insomecases,individualsmaybefoundguiltyof acrimewithoutbeingprosecutedorsentencedincourt.Thishappensif theoffenderisveryyoungorifhe/sheconfessestoalessseverecrime.
Althoughthesecasesarehandledbythedistrictattorneytheyarestill includedinourdata.
WeusetheUpperSecondarySchoolApplicationRecordtoobtain informationonwhenandwhereanindividualbeganuppersecondary school aswell as what track (type andlength) he/she started.This register is used to construct the sample of individuals, but also to acquire information on which educational tracks each municipality offeredeachyear.Basedonthisinformationwethendeterminedwhich municipalitiesparticipatedinthepiloteachyearandtheshareofthe availablevocationaltrackswhichconstitutedthree-yeartracks.
Our sample consists of individuals who began upper secondary schoolduring1986–1990.Inthebaselineanalysisweincludeallupper
secondary school enrollees, including those in academic tracks. As willbediscussedbelow,theresultsarequalitativelyverysimilarifwe insteadfocusonthosewhoenrolledinvocationaltracks.22Arestriction imposedisthatonlypilot trackswhichcorrespondedtotracksinthe regularsystemareincluded(seeTableA1).23 Wehavealsoexcluded someindividualswhowereyounger than15or olderthan20 when theyenrolledinuppersecondaryschool.Furthermore,andimportantly, welimitthemainanalysistomales.Resultsforfemaleswillhowever alsobediscussed.Crimerates aresubstantiallyloweramongwomen comparedtomen.Inoursample,thefractionconvictedforanytype of crimeup to 15 yearspost thestartof uppersecondary school is fourtimeshigher amongmen(27.9 comparedto7.0%).Oursample ofmalestudentsconsistsof116,787vocationaland107,654academic students,whichmakesatotalof224,441individuals.24
Weaugmentthedatasetwithinformationoneachparent’seduca- tion(measuredin1990),age,andwhetherbothparentshaveforeign background(definedasbornoutsideSweden).Forthefathers,wealso addedinformationonemploymentstatusandwageearnings(measured in1990),aswellascriminalconvictions(measuredintheyearthechild enrolledinuppersecondaryschool).Wealsohavedataonthestudents’ final gradepointaverage (GPA) from compulsoryschool. Regionof residenceisdefinedasthemunicipalityofresidenceinDecemberthe yearbeforeenrollmentinuppersecondaryschool.Thiswayweavoid thepossibilitythatmunicipalityofuppersecondaryschoolattendance maybeendogenouswithrespecttothelocationofthepilot.25
Themainadvantageofusingindividuallevelconvictiondataisthat wecaninvestigatewhetherthepotentialeffectoncrimediffersinsub- groupsofthepopulation.Wecenterongroupsathigherriskofcriminal involvement. We stratify individuals according to their compulsory schoolGPAaswellasaccordingtotheirpredicteddropoutrisk.
Criminalbehaviorisinthispaperinferredfromregisterinformation onconvictions.Themainadvantageofadministrativedatacompared to crime self-reports is that the latter is known to be plagued by underreporting andmeasurement error (McDonald2002). Thelarge samples availablein administrative registers alsoincrease statistical precision.Still, convictiondataarenotflawless. Apossibleobjection is that criminalbehavior is onlyobserved for individualswho have beenconvictedin court.One concerninthecontextof thispaper is thatpeoplewhoperform worsein schoolmaybemore likelytoget convicted conditional on actually having committed a crime. These individualsmayforinstancehavefewerresourcesavailablefordefense inacriminaltrial. Inthiscase ourestimateswouldbebiaseddown- wards.Thisis acaveatimportanttobearin mindwheninterpreting theresults.Note howeverthat thisis only aproblemifthis kindof selectionisnotpickedupbyourextensivesetofcontrolvariables.26
22In Section 3.4 we investigate whether the availability of pilot programs affected se- lection into vocational tracks.
23This restriction excludes students in the two smallest three-year tracks (Graphic and Handicraft); in total 176 persons.
24Around 8% of the original sample has been excluded due to missing information on some of the variables. For most of these we lack information on location of residence and compulsory school GPA. It turns out that criminal activity is higher in this group. For instance, while the overall conviction rate is 28% in our sample 32% of the individuals that we exclude were convicted. The same numbers for property crime and violent crime are: 9% versus 15% and 4.4% versus 6.9%. Thus, while this group constitutes a relatively small share of the population it is slightly overrepresented among criminals. Our results may therefore not generalize to the full population.
25It is unlikely that students would move already during compulsory school in order to take advantage of the pilot tracks, especially as it was already possible to apply to upper secondary schools in municipalities other than one’s own. Moreover, the decision of where to locate the new available pilot slots each year was not taken until during the following spring, i.e. after the point in time when we measure municipality of residence (see SOU, 1989 :106 for details on the implementation process).
26In their study of the effect of education on crime as measured by arrests, Lochner and Moretti (2004) raise a similar concern. Using data on self-reported crime they conclude that for this to be a problem, education must substantially alter the probability of being arrested conditional on criminal behavior.
3.2. Descriptivestatistics
Table1presentsdescriptivestatisticsformalesattendingtwo-year andthree-yearvocationaltracks,andforthoseenrollinginacademic programs.Regardlessoftrack,thevastmajoritystartsuppersecondary educationatage16.Thetableclearlyshowstheexpectedpatternthat academic students have more favorable background characteristics andstudycredentials.CompulsoryschoolGPAsaremuchhigher,and mothers and fathers have substantially more education. Since the three-year vocationaltrackswerenot onlylongerbutalsocontained more generaltheoreticalcontent,itis notsurprisingtofind thatthe studentstakingtheseprogramsarealsosomewhatpositivelyselected inthesedimensions.However,thedifferencesrelativetothetwo-year enrolleesaremodest,especiallycomparedtotheacademicsstudents.In particular,thereisnosignificantdifferenceinthecriminalbackground offathersbetweenthesetwogroups.
Thelowerpanelofthetableexhibitsthefractionswithatleastone crimeconvictionwithin15yearspoststartinguppersecondaryschool.
Thelevelsfor“anycrime” arestrikinglyhigh.Oneinthreeofthosein vocational programs,andonein fiveof thosein academicprograms have aconviction.Propertycrimesareroughly twiceas commonas violentcrimes,eventhough alsothelatterarerelativelyfrequent.A closerlook(seeTableA3)atthetypesofoffensesinthedata,reveals that about 30%of the convictionsare forviolations of trafficlaws:
drivingundertheinfluence,drivingwithoutalicenseetc.About23%
oftheconvictionsconsidertheft(shoplifting,burglary,robbery).
3.3. Exploitingthepilotschemeasapolicyexperiment
Forthebaselineresults,wewishtoestimatethefollowingregression model
𝐶𝑜𝑛𝑣𝑖𝑐𝑡𝑒𝑑≤𝑖𝑗𝑠15𝑦𝑒𝑎𝑟𝑠=𝛼+𝛽3𝑦𝑒𝑎𝑟𝑝𝑟𝑜𝑔𝑟𝑎𝑚𝑖𝑗𝑠+𝛾𝑋𝑖+𝛿𝑗+𝜃𝑠+𝜀𝑖𝑗𝑠 (1) where iindexes individual, jmunicipality of residence,and s upper secondaryschoolstartingyear;Convictedijsisanindicatorequaltooneif theindividualhasbeenconvictedforacrimecommittedwithin15years ofstartinguppersecondaryschoolandzerootherwise;3yearprogramijs isadummywhichtakesthevalueoneiftheindividualchosetoenroll in a three-year (or longer) track, and zero if he/she enrolled in a two-yeartrack;Xi isavectorofindividualandfamilycharacteristics (compulsoryschoolGPA,sex,ageatenrolment,immigrantbackground, eachparent’shighesteducationlevel,eachparent’sage,eachparent’s wageearnings,fatherconvicted,andwhetherbothoftheparentshave immigrantbackground);𝛿j and𝜃s representmunicipalityofresidence anduppersecondaryschoolstartingyearfixedeffects;ɛijsis anerror term.Theparameterofinterestis𝛽whichideallygivesthecausaleffect ofenrollinginalongertrack.
Even though Eq. (1) includesa rich setof covariates aswell as controlsforability,asmeasuredbycompulsoryschoolGPA,onecould stillbeconcerned thatstandardOLS estimatesmaybe biaseddueto non-randomselectionintoeducation.Thereispotentiallyalargenum- berofunobservedfactorsincludedinɛijswhichcouldbecorrelatedwith an individual’schoiceof educationaltrack.Forinstance, individuals withhighcareeraspirationsorlowdiscountratesmaybemorelikely toinvestineducationandmayalsohavealowerriskof committing crime.Itisalsopossiblethatinvestmentsineducationandcrimeare determinedjointly,makingtheanalysissusceptibletoreversecausality (seeHjalmarsson,2008).
Toaccountforendogenousschoolingchoicesweexploit variation across regions over time in the implementation of the pilot which preceded the reform. Aspreviously mentioned, thepilot gavesome students the choice of attending a three-year rather than a regular two-yearvocationaltrack.Theextenttowhichapersonhadthischoice depended jointly on:(i)which yearthestudentfinished compulsory school,and(ii)thestudent’smunicipalityofresidence.Wearguethat this plausibly exogenous variation, conditional on upper secondary
Table 1
Descriptive statistics. Males only.
Vocational students Academic students 2-year tracks 3-year tracks Individual background characteristics:
Age at enrolment in upper secondary school 16.10 16.09 16.02
GPA compulsory school a 25.57 26.55 64.35
Parental characteristics:
Both parents foreign born 0.070 0.052 0.058
Mother’s age 42.16 42.21 43.13
Father’s age 45.12 45.20 45.65
Mother completed upper secondary education 0.591 0.639 0.791
Mother completed post-secondary education 0.117 0.137 0.380
Missing data on mother’s education 0.020 0.016 0.015
Father completed upper secondary education 0.505 0.554 0.746
Father completed post-secondary education 0.094 0.113 0.353
Missing data on father’s education 0.056 0.043 0.043
Father’s wage earnings b,c 135,764 139,144 193,354
Father employed b,c 0.888 0.896 0.917
Father convicted of any type of crime b 0.020 0.019 0.011
Outcome variables:
Convicted of any type of crime, ≤ 15 years post school start 0.363 0.344 0.190 Number of convictions, ≤ 15 years post school start 0.894 0.750 0.286 Convicted of violent crime, ≤ 15 years post school start 0.067 0.059 0.019 Convicted of property crime, ≤ 15 years post school start 0.124 0.110 0.050
Number of observations 103,621 13,166 107,654
Notes : a) The GPA’s are percentile ranked by year of graduation. b) In year 1990. c) Missing values are replaced with zeros.
schoolstartingyearandmunicipalityofresidence,isavalidinstrument forthelengthofthechosentrack.27
Morespecifically,ourinstrumentisthedegreetowhich theindi- vidual’smunicipalityofresidenceparticipatedinthepilotbythetime he/she beganupper secondary school, asmeasured by theshare of the available vocational tracks which constituted three-year tracks, i.e. 𝑁−1∑𝑁
𝑙 1{𝑇𝑟𝑎𝑐𝑘𝑙>2𝑦𝑒𝑎𝑟}.28 Ideally, the instrument would be measured as the share of availableslots in vocational tracks which representedthree-yeartracks, butsuch dataarenotavailable atthe municipalitylevel.Assumingthattheinstrumentisuncorrelatedwith anyunobservedvariablesaffectingtheconditionaloutcomesofinterest, andthatithadnodirecteffectontheoutcomes otherthanthrough influencingwhetherthepersonenrolledinatwo-orathree-yeartrack, aninstrumentalvariables(IV)estimatorof𝛽isconsistent.Intheem- piricalanalysisweprovideseveralpiecesofevidenceinsupportofthe validityoftheseassumptions.Iftheeffectofenteringathree-yeartrack variesacrossindividuals,andifthereisalsosortingongains,theIV estimateshouldbeinterpretedastheeffectforindividualswhoonthe marginareinducedtoselectathree-yeartrackbecauseofthepilotand themarginthatvarieswiththeinstrument(e.g.BjörklundandMoffitt 1987;HeckmanandVytlacil2005).Forthisinterpretationtobecorrect, increasedavailabilityofpilottracksinamunicipalitymustneverhave reducedparticipationin three-yeartracksamongthoselivingin that municipality(monotonicityassumption).Table4suggeststhatthisis indeedthecase;thefirststagerelationshipisaround0.35forthefull sampleandthe relationshipis positive andstrongin all subsamples examined.29
Itisimportanttonotethatthedesignofthepilotgeneratesasetting wheresomestudentsweregiventhechoiceofenrollinginathree-year ratherthananordinarytwo-yearvocationaltrack.Becauseindividuals
27This identification strategy has previously been used by Hall (2012,2016 ) to investi- gate the effect of the reform on educational attainment and labor market outcomes, and by Grönqvist and Hall (2013) to investigate the effect on teenage childbearing. Similar strategies have also been used in other studies; see e.g. Duflo (2001) .
28The instrument is zero for municipalities not offering any vocational tracks. In Section 4.3 we discuss some alternative definitions of pilot intensity.
29Although not reported, we also estimated the first stage by parental education, em- ployment and convictions and find support for the monotonicity assumption also in these sub-samples.
areallowedtodropoutofschool,theparameterweestimateisthus notnecessarilythesameasinstudiesoncompulsoryschoolingreforms whereindividuals areforced tostay inschool. However,as long as countriesdecide to keeptheir highereducation voluntary, we think thatthisparameterisrelevantforpublicpolicy.Wediscusseffectsof thereformontheriskofdroppingoutinSection4.4.Itisalsoworth mentioningthatjustasforstudiesexploringcompulsoryschoolingre- formsourestimatesreflectthecombinedeffectofprolongingeducation andchangesinthecurriculum.
3.4. Issuesrelatedtotheidentificationstrategy
As described in Section 3.3, our empirical approach builds on the arguably exogenous variationgiven by thepilot scheme for the possibilityofenrollinginathree-yearratherthanatwo-yearvocational track. Twopotentialissuesariseimmediately:(i)pilot intensitymay affectthechoiceofwhethertoenrollinuppersecondaryschoolatall and/orthechoicebetweenvocationalandacademicprograms,andthus theselectionofindividuals;(ii)pilotintensitycouldbeendogenousto thecharacteristicsofthestudentpopulation.
Table2providessomeevidencerelatedtothefirstpoints.Itdisplays estimatesoftheeffectofpilotintensityontheprobabilityofenrolling in upper secondary schoolandchoosing avocational asopposed to anacademictrackrespectively(usingthesamesetofcovariatesasin themainanalysis).Noneoftheestimates,withorwithoutcovariates, in the overallsampleor in subgroups, arestatistically significantat conventionallevels.Thissuggeststhatthistypeofselectionisprobably notabigconcern.ForthosewiththehighestcompulsoryschoolGPA, there issometendencytochoosevocationaltracksmoreoftenwhen there are more three-year vocational programs. Therefore, in the baseline analysis, we include all upper secondary school enrollees, ratherthanlimitingthesampletovocationalstudentsonly.
The second point raised above is addressed in Table 3, which presentsresults fromregressionsofindividualcharacteristicsonpilot intensity. Overall, the results do not indicate that pilot intensity is to any substantial degree correlated with student characteristics. A few estimates areindeedstatistically significant, butthe association is smallina quantitativemeaning.Forexample,thosewhowere17 insteadof16yearsoldwhentheyenrolledinuppersecondaryschool, met on average a 0.35percentage points lower share of three-year
Table 2
The effect of pilot intensity on the probability of enrolling in upper secondary school, and of enrolling in a vocational rather than an academic track (males only).
Upper secondary school enrollment Vocational track
(1) (2) (3) (4)
A . Entire sample
Pilot intensity in municipality of residence 0.014 (0.023) 0.006 (0.020) 0.005 (0.013) 0.010 (0.013)
Mean of dependent variable 0.845 0.845 0.520 0.520
Number of observations 163,531 163,531 224,441 224,441
B . GPA: below average
Pilot intensity in municipality of residence 0.024 (0.037) 0.008 (0.034) − 0.001 (0.020) − 0.001 (0.019)
Mean of dependent variable 0.743 0.743 0.841 0.841
Number of observations 84,945 84,945 106,482 106,482
C . GPA: at least average
Pilot intensity in municipality of residence 0.003 (0.014) 0.003 (0.014) 0.011 (0.016) 0.022 (0.015)
Mean of dependent variable 0.961 0.961 0.231 0.231
Number of observations 77,947 77,947 117,959 117,959
Covariates included No Yes No Yes
Notes : Each cell represents a separate regression. In addition to municipality of residence and upper secondary school starting year fixed effects, Col. 2 and 4 include controls for: compulsory school GPA (quadratic), age at enrolment (dummies), each parent’s educational attainment (3-levels), whether both parents are foreign-born, each parent’s age (linear), missing data on parents ’ education, the father’s employment status, and the father’s earnings (linear). Col. 4 also controls for age at enrolment (dummies) and for whether the father has been convicted of crime. “Pilot intensity” is measured as the share of available vocational tracks in the municipality of residence at the time of enrollment which constituted three-year tracks. Robust standard errors in parentheses allow for clustering by municipality of residence. ∗/ ∗∗denotes significance on the 10/5 percent level. In Panel B–C, the students are divided into sub-groups based on the grade distribution among the male upper secondary school students. The sample used in Columns (1) and (2) consists of all students who finished compulsory school during 1988 − 1990 (the register for compulsory school completion begins in 1988, hence this restriction).
programs, i.e. a marginal difference relative to the average pilot intensityof11.2%.Moreover,theF-statisticsofthefullmodelsareboth statistically insignificant, although the p-value for upper secondary schoolenrollmentisjustslightlylargerthanthe10%significancelevel.
4. Results
This section presents theresults from theempirical analysis. We begin by showing estimates of the first stage relationship, i.e. the relationshipbetween thepilot intensityinan individual’shome mu- nicipalityandthechoiceoftrack.Thereafter,wediscusstheeffectsof thereformoncrimeingeneralandoncertaintypesofcrimes,inthe overallpopulationunderstudyaswellasindifferentsubgroups.This presentationis followedbysomerobustnesschecks.Thenweturnto investigateissuesrelatedtopotentialunderlyingmechanisms.
4.1. Effectsofthereform
Table4displaystheestimatedfirststagerelationshipfordifferent subgroupsofstudents.Thetableshowsthatthefirststagerelationshipis strongbotheconomicallyandstatistically.Forinstance,theestimatein column(1)suggeststhatincreasingtheshareoftheavailablevocational tracks which constituted three-year tracks by 10 percentage points in a student’shome municipality, raises theprobability of enrolling in athree-yeartrackby3.56percentagepoints.Alternatively,a one (within-municipality)standarddeviation(.124)increasein theshare ofthree-yeartracksraisestheprobabilityof enrollinginathree-year trackby 4.4percentage points.Although not shown,the coefficient fortheinstrumentisalmostunchangedwhenincludingcovariates.The F-testclearlyindicatesthatourinstrumentisnotweak(cf.Staigerand Stock,1997).We can seethat females,studentswithabove average compulsoryschoolGPA,andthosewiththelowestpredicteddrop-out riskrespondedlesstoaccesstothelongerprogram.
Table 5presents thebaseline results on theassociation between enrollmentin a three-year (orlonger), instead of a two-year,upper secondaryschooltrackandtheprobabilityof havingbeenconvicted ofacrime.Column(1)containsthetotalsamplewithin15yearsafter enrollment;columns(2)–(4)presentestimatesforfinerageintervals;
andcolumn(5)reportsresultsusingthetotal numberof convictions
ratherthancrimeasabinaryoutcome.Thefirstpanelinvestigatesthe probability of any typeof crime,whereas thelower panelsconsider violent andproperty crimes respectively. Weinclude three different estimators:OLS(regressingtheoutcomeonthepotentiallyendogenous enrollment variable); IV (using pilot intensity as an instrument for track choice); and Reduced form (regressing the outcome on the instrument).
In column (1), the OLS estimates suggest very little correlation betweencrimepropensityandenrolmentinalongerandmoregeneral upper secondary school program. There is a statistically significant association with violent crime, but it is small in economic terms.
However,theIVestimateforpropertycrimesuggestsasizableimpact of the reform: 4.6percentagepoints relative toan averageof 8.8%
means a decrease of over 50%. Comparing instead to the higher and arguably more relevant baseline of 12.2% among vocational students, the decrease is still37%. Assuming a LATEinterpretation of thecoefficient,itseemsthatthoseinducedtoenrollinthree-year programsbyhighercoverage intheirhome municipalityatthetime of enrollment, exhibit lower degrees of property crime due to the reform.
We thinkitis plausibletoassume thatifthe pilotaffected indi- vidualcrimerates,theeffectworkedthroughtheamountandtypeof education given toparticipating individuals(i.e. theIV approach is justified).Nevertheless,Table 5alsopresents reduced-formestimates fromregressingconvictionprobabilitiesonpilotintensitydirectly.This modelisrelevantifoneisinterestedinthetotaleffectofofferinglonger andmoregeneraleducationtoagreatershareofagivenpopulation.
The pointestimate for theentire samplesuggeststhat offeringonly three-yearvocationalprogramsasopposedtonothree-yearvocational programsleadstoa1.6percentagepointdropinpropertycrimeamong thestudents,i.e.areductionofaround18%.
Turning to columns (2)–(4) it is quite clear that the effects on crimeareconcentratedtotheages16–20.Therewefindasubstantial reductioninoverallaswellasinproperty crime.Forthehigherage intervalstherearenosignificantnegativeeffects.Wewillreturntothis patterninthediscussionofmechanismsbelow.
Whenusingthenumberofconvictionsastheoutcome,column(5) shows alarger discrepancybetween OLS andIV.TheOLS estimates
