EconomicsofEducationReview28(2009)345–356 ContentslistsavailableatScienceDirect Economics of Education Review journal homepage: www.elsevier.com/locate/econedurev The French zones d’éducation prioritaire: Much ado about nothing? RolandBénaboua,FrancisKramarzb,∗,CorinneProstc aDeptofEconomics,320BendheimHall,PrincetonUniversity,Princeton,NJ08544-1013,USA bCrest-Insee,15bdGabrielPéri,92245Malakoff,France cInsee,18bddolphePinard,75014Paris,France a r t i c l e i n f o a b s t r a c t Articlehistory: WeprovideanassessmentoftheFrenchZEP(Zonesd’EducationPrioritaire),aprogram Received5June2007 startedin1982thatchannelsadditionalresourcestoschoolsindisadvantagedareasand Accepted2April2008 encouragesthedevelopmentofnewteachingprojects.Focusingonmiddle-schools,we firstevaluatetheimpactoftheZEPstatusonresources,theirutilization(teacherbonuses JELclassification: versusteachinghours)andkeyestablishmentscharacteristicssuchasclasssizes,school I21 enrolments,teachers’qualificationsandexperience,andstudentcompositionandmobility. I22 WethenestimatetheimpactoftheZEPprogramonfourmeasuresofindividualstudent I28 achievement:obtainingatleastonediplomabytheendofschooling,reaching8thgrade, Keywords: reaching10thgradeandsuccessattheBaccalauréat(thenationalexaminationattheendof Resourceallocation highschool).WetakeintoaccounttheendogeneityoftheZEPstatusbyusingbothdifference Expenditures indifferencesandinstrumentalvariablesbasedonpoliticalvariables.Theresultsarethe sameinallcases:thereisnoimpactonstudentsuccessoftheZEPprogram. ©2008ElsevierLtd.Allrightsreserved. 1. Introduction receivedextraresourcessuchasfunds,teacherhours,etc. These zones were originally meant to be temporary, but Manydevelopedcountrieshaveadoptedcompensatory theprograminsteadbecamepermanentandwassubstan- educationpoliciesthatdirectextraresourcestodisadvan- tiallyextendedinsuccessive“waves”throughoutthe1980s tagedschools.IntheUnitedStates,thishasbeenthecase and1990s.Initially,themaingoaloftheZEPswastofos- since1965withtheTitleIprogram.InGreatBritain,the terneweducationalprojectsandpartnershipswithlocal EducationPriorityAreasexperimentoftheearly1970sfol- actors that would help improve academic achievement. lowedthePlowdenreport,athoroughreviewofprimary Gradually,decreasingclasssizealsobecameanimportant educationinEnglandwhichhadaconsiderableimpact.This objective. The amount and nature of the extra resources programre-emergedin1997–1998astheEducationAction giventotheZEPswereneverspecified,however,northe ZonesandExcellenceinCities.Anumberofothercountries, actualprocedurebywhichprioritystatuswastobedeter- suchasPortugalandBelgium,havesimilarprograms. mined.Perhapsmosttellingly,tothisdaytheMinistryof InFrance,theideaofpriorityzonesfirstspreadamong Education’sbudgetstillcontainsnospecificlineitemfor teacherunionsinthe1970sandthenbecamepartofthe “priorityeducation”. political platform of the Socialist party, which came to As the first program to target schools and local areas powerin1981.Thus,in1982,anewprogram,the“Zones ratherthanprovideindividualfinancialaidtopoorstudents d’EducationPrioritaire”(PriorityEducationZones,hence- (usually very modest stipends), the ZEPs remain highly forth ZEP) was launched, under which selected schools controversial.Somearguethatanyformof“positivedis- crimination”iscontrarytothenationalideal,othersthatit serves both fairness and efficiency and should get much ∗ more resources, others yet that it has just been a waste Correspondingauthor.Tel.:+33141176033;fax:+33141176046. E-mailaddress:[email protected](F.Kramarz). of money. Most recently, in the wake of the riots which 0272-7757/$–seefrontmatter©2008ElsevierLtd.Allrightsreserved. doi:10.1016/j.econedurev.2008.04.005 346 R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 occurred in the poor suburbs of large cities in Novem- inparticularinthetimingofthechoiceofZEPs.Weshow ber 2005, the French government put forth a set of new that the vote shares obtained in different French regions measuresintendedtopromotegreaterequalopportunity bythemajorpartiesinthenational(legislative)elections through subsidized housing projects and reforms of the of1981and1988provideaplausiblyexogenoussourceof educationsystemandlabormarketregulation.Theflagship variationthatwecanusetoidentifytheZEPeffect. decisionintheeducationfieldistobeanewexpansionand OurresultsonthenatureoftheZEP“treatment”,atleast reformoftheZEPprogram. in term of quantifiable inputs, show that this treatment InspiteofthecentralplaceitoccupiesinFrenchedu- translatedintoacontinuousbutextremelyslowdecreasein cation policy and the political debate, there has been no classsizes(−0.2studentsperyearonaverage)andincrease systematicevaluationoftheimpactoftheZEPsonschool- inteachinghoursperstudent(+1.2%peryear);meanwhile, ingoutcomes.Inthispaper,weprovideanassessmentof the experience of ZEP teachers declined slightly, in spite theZEPsalongthreedimensions:theresourceseffectively of the bonuses offered. Our results on the change in the deployed,themobilityresponseofstudentsandteachers, compositionofthestudentbodyshowalargereductionin andtheoverallimpactonacademicachievement.Dueto enrollmentsfollowingaschool’sclassificationasZEP.This datalimitationswestudytheperiodthatcoversonlythe reduction did not come from an increase of the number firstphaseofthepolicy(1982–92)andfocusonitsimple- ofthestudentsleavingtheseschoolsbutfromadeclinein mentationinlowersecondaryschools. arrivals,anditdidnotleadtoanobviousdeteriorationin This evaluation presents particular challenges. In par- the socio-economic background of the students. Turning ticular, the precise nature of the intervention is not well finallytotheeffectsonstudentachievementoftheover- known: priority status is allocated according to a rather allZEPtreatment(includingfinancialresources,themore opaque,erraticadministrativeprocedureandittranslates qualitativeaspectsofthezones’educationalprojectsand intounspecifiedextraresourcesthatmayvaryfromzone theeffortandmobilityresponsesofstudentsandteachers), tozone,fromyeartoyear,orbeuseddifferentlybydiffer- theresultsfrombothourmethodologiesleadtothesame entschools.Togetherwiththelackofbudgetarydata,this conclusion: during our sample period, the impact of the leadsustodevotethefirstpartofthepapertoakindof ZEPsonthesubsequentacademicachievementofpupilsin “detectivework”aimedatdeterminingwhatitmeant,in the6thand7thgrade(age11and12)isnil. practice,foraschooltobecomeaZEP—intermsofinputs The paper is organized as follows. In Section 2, we suchasclasssize,weeklyteachinghours,qualificationsand describethegeneraleducationalcontextandthemainfea- seniorityoftheteachers,andsizeandcompositionofthe turesoftheZEPprogram.ThedataaredescribedinSection student body. Then, in the second part of the paper, we 3. Changes in schools’ characteristics when they become examine how the granting of ZEP status to a junior high ZEParethenexaminedinSection4,andtheimpactofpri- schoolaffecteditsstudents’educationaloutcomes,bothin ority zone status on student achievement is analyzed in thosegradesandbeyond.Thefactthattheprioritystatus Section 5. Section 6 offers a brief summary of our main may–inparticularthrougha“labeling”effect–inducea findingsandtheirpolicyimplications. mobilityresponseonthepartofteachersorstudents’fam- ilies(movingtoanotherdistrict,sendingtheirchildrento 2. AbriefdescriptionoftheEducationPriority privateschool,etc.)isanothersourceofcomplexityineval- Zones(“Zonesd’EducationPrioritaire”) uatingtheZEPs.Atthesametime,studyingtheelasticityof suchresponsesandassessingeducationaloutcomesfroma Firstestablishedin1982,theZEPs,oreducationprior- programinwhichtheypotentiallyoperateisbothinterest- ity zones, include both primary (1rst to 5th grades) and ingandimportant,becauseanyinterventionthatexceeds junior-highestablishments(6thto9thgrades),plusasmall thescaleofsmallcontrolledexperimentsislikelytotrigger numberofhighschools(10thto12thgrades).Initially,the suchbehaviors. ZEPscoveredapproximately8%ofjunior-highstudentsin Our analysis combines very rich student survey data thecountry.Theprogramwasoriginallymeanttobetem- withadministrativefilesonallFrenchschoolsandteach- porary,withthezonesestablishedforalimitedstatutory ers.Thefirstsourceconsistsoftwopanelsofstudentsthat termof4years.Overtime,however,theprogramwasnot providedetailedinformationontheirfamilybackground, only maintained but substantially expanded, with many early educational history, and major academic outcomes newzonescreatedin1989,1990,1994and1999.Itisnow throughoutsecondaryschooling.Thesecondsourceisan themainpolicyinFrancedirectedathelpingstudentsfrom originalpanelofschoolsthatweconstructed,usingexhaus- disadvantagedbackgrounds. tiveestablishmentfilesfromtheMinistryofEducation.We TheoperatingprincipleoftheZEPprogramistoprovide unfortunatelyhadtorestrictattentiontosecondaryschools additionalresourcestoschoolsinthemostdisadvantaged becauseoftheavailabledata,eventhoughtheZEPprogram zones and allow them to develop specific initiatives and alsoconcernsprimaryschools. educationalmethodstailoredtotheirstudents’needs.In WeaddresstheendogeneityintheallocationofZEPsta- 1982 the regional heads of the education administration tus using two methods, namely difference-in-differences werethusaskedtoselectzonesaccordingtosomevague (orestablishmentfixedeffects)andinstrumentalvariables. criteriaonthestudentpopulation.1Inlateryearsmanynew Thesetechniquesareappliedwithfourdifferentmeasures of academic achievement. Our instrumentation strategy reliesonthefactthatnationalpoliticalforcesinterferedin 1 Headsoftheadministrationwereconducedtouseindicatorssuch theprocessbywhichpriorityzonesweredesignated,and asparents’socialandprofessionalbackgrounds,parents’rateofunem- R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 347 zoneswerecreated,accordingtoaprocedurethatleftcon- ular, regional and local subsidies were not higher in ZEP siderablediscretiontotheregionalheadsoftheEducation schools. Ministry administration (“recteurs”) in making decisions The D400 million represented 1.2% of total expendi- that were loosely based on indicators such as the shares tures on teaching activities in primary and junior high amongparentsofblue-collarworkers,unemployedwork- schools, and were directed to 12% of the total student ers,andhigh-schooldropouts,thefractionoffamilieswith populationinthosegrades.Thismeansthataschool’sclas- at least one non-European member, and 3rd grade test sificationasZEPtranslatedin1998–99intoanextra10% scoresinanationalstudentevaluation. more resources per student. Looking more appropriately In1997nearly700,000studentsinprimaryschoolsand at instruction costs (leaving aside fixed operating costs, 400,000 in junior high schools benefited from “priority” etc.), which represent 77% of total costs on average, the treatment, representing respectively 11% and 15% of the increase was 13%.5 No such data is available for the ear- respectivetotals.2OnlyveryfewhighschoolsareZEPs.The lierperiodthatcorrespondstomostofourstudentsample. vastmajorityofZEPstudentsliveinmajorurbancenters. We can, however, use some of our estimation results to Agoodindicatorofpovertyisthefractionofstudentsnot come up with a reasonable ballpark estimate. Based on enrolledintheschool’scafeteria’splan,becausethisgener- the establishment data, we estimate that teaching hours allydenotesthattheirfamilycannotaffordtopayforthese perstudentincreasedbyabout1.2%peryearfollowingthe lunches.3Thisfractionisindeedhigherinthepriorityzones grantingofZEPstatusinthe1989and1990waves(relative (69%) than outside (40%), and reaches 80% in those ZEPs tonon-ZEPschools).Wealsoknowthattheteacherbonus locatedinmajorurbancenters.Similarly,theproportionof in 1990–91 was approximately equal to 2% of the aver- non-Frenchstudentsistypicallyhigher,exceeding35%in ageteacherwage.ThuswecanestimatethattheZEPlabel morethan10%oftheZEPschools.Concerningeducational broughtanextra4.4%intheteachingbudgetthatyear,with attainment,in199537%ofstudentsenteringthe3rdgrade theallocationbetweenbonuses(2%)andhoursperstudent (age8)inZEPschoolsdidnotpossess“basicreadingcompe- (2.4%). tencies”,whichisdoubletheproportionfoundinnon-ZEPs By comparison, the resources devoted to the EIC pro- (18%). gramintheUnitedKingdomamountedto£300millionin TheZEPstatusisassociatedwithextraresourcesforthe 2003(aroundD440–460million)andcoveredaboutathird selected schools, mostly in the form of additional hours of secondary school pupils (Machin, McNally, & Meghir, of instruction and bonuses for teachers and other per- 2004).IntheUnitedStates,TitleIprogramsprovided$11.7 sonnel. Here again, considerable discretion was given to billion in 2003 to schools with high concentrations of the regional heads of the education administration with povertyandthisfundingrepresentedroughly5%ofthese respect to the amount and nature of the resources allo- schools’ total budget (Van der Klaauw, 2008). In terms cated to ZEP schools. The French education system is of overall financial scope, these three different national extremely centralized and the ZEP program was the first programs are thus more or less comparable. They dif- one whose aim was to give relative freedom of action fer markedly, on the other hand, in how the additional to local authorities. One of the drawbacks was that the resources are used. In the United Kingdom, the program information about the extra resources was not collected includes very specific components aimed at extending at the centralized level. The first year for which any learning opportunities through additional teachers, sup- kind of budgetary information is available is the school port units, or extra help for the most disadvantaged year1998–1999.ExtraresourcesdirectedtotheZEPsthat children.IntheUnitedStates,fundsareexplicitlytargeted year amounted to D400 million, of which D110 million todisadvantagedchildreninsidetheschool.6 took the form of bonuses paid to all employees of the InFrance,thenatureoftheinterventionwaslefttoeach schools – mostly teachers – while the rest was used to school’sdiscretion:theideaoftheZEPprogramwasjustto increasetotalteachinghoursandreduceclasssizes.4Every impulse new educational projects and partnerships with ZEP employee also benefited from a relative advantage localauthorities.Existingevaluationsarebasedonsimple in administrative promotion criteria. As far as non-wage comparisonsofaverageachievementbetweenZEPstudents financialresourcesareconcerned,therewasnopriorityfor and non-ZEP students (Meuret, 1994; Caille, 2001). Our ZEP schools (Jeljoul, Lopes, & Degabriel, 2001). In partic- paperthusdiffersfromtheseearlieranalysesthroughthe useofeconometrictechniquestoaddresstheendogeneity andselection-biasproblemsthatareofprimaryconcernin ployment,fractionofstudentswhoarenotnativeFrenchspeakersand fractionofstudentshavingrepeatedagrade(averylargefractionofFrench studentsrepeatedatleastonceinthoseyears). 5 Wecandecomposethis13%:13*(110/400)=3.6%wenttobonusesand 2 ThestatisticspresentedinthisparagrapharetakenfromtheMin- theremaining9.4%toincreasedhoursofteachingperstudent.Asacheck istryofEducation’sNoten◦98-15,“LesZonesd’EducationPrioritairesen onthiscalculation,directcomputationsbasedonthedataweobtained 1997–1998”. showthatthebonusinthatyearwasapproximately4%oftheaverage 3 InFrance,lunchesaresubsidizedforeverybodybuttherearenofree- teachersalary. lunchprograms.Someofthestudentswhodonothavelunchatschool 6 Machinetal.(2004)findapositive,albeitsmall,improvementinpupil havelunchathomebecausethefamilyincomeislargeenoughthatthe outcomemeasuresforMathematics(butnotforEnglish)andastrong mother(orfather)doesnotwork.Butthemajorityofstudentswhodonot reductioninabsenceswithinEiCschools.Onthecontrary,VanderKlaauw havelunchatschoolarethosewhosefamilycannotaffordtopayforthe findsthattheTitleIprogramdoesnotimprovestudentachievement.He lunches. interpretsthisfindingasrelatedtothemannerinwhichfundsarespread 4 Thus,aZEPbonusof1,046Euroswaspaidto96,000teachers,account- thinlyacrossalargenumberofschoolsandtotheapparentineffectiveness ingfor100outofthe110million. ofmanyTitleIfundedremedialeducationprograms. 348 R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 theliteratureonprogramevaluation,suchasthenonran- 3.2.2. The1989panel domallocationoftheZEPstatus. Thissampleincludes24,455studentswhoentered6th grade in 1989. The sampling scheme here has only one 3. Thedata level: the heads of all junior high schools were asked to include in the panel all students entering 6th grade in 3.1. TheFSEdatasets 1989 who were born on a specific day in each month (1/30th). These students were observed until the end of The FSE administrative files (Fichiers Standards their schooling, including higher education. The attrition Enrichis)oftheMinistryofEducationconstituteourmain ratewassimilartothatobservedinthe1980panel. sourceofschool-leveldata.Everyyear,allschoolprincipals Thesetwopanelsprovideextensiveinformationabout fill out a detailed questionnaire on the characteristics of thestudents.First,wehavefamilybackgrounddata:dateof eachgradeintheirestablishment.Thisincludesinforma- birth,sex,nationality,countryofbirth,numberofsiblings, tion on class size, nationalities of students, number of birthrank,occupationoftheheadofthefamily,parent(s) studentshavingrepeatedeachgrade,numberofstudents legallyresponsibleforthechild,andnumberofyearsspent havinglunchattheschool’scafeteria,andlanguagecourses innurseryschoolandinprimaryschool.Second,wehave chosen. Each statistic is measured by grade. These data schooling information for each year from 6th grade on: are exhaustive for the period 1987 to 1992 and cover grade, class size, foreign languages studied, lunch at the both public and private schools; we shall focus here on school’s cafeteria or not, financial aid received. For each public-sectorjuniorhighschools(6thto9thgrades).7 observation(student-year),wealsoknowtheidentification The FSE dataset can be extended to the period numberoftheschoolingestablishment. 1994–1999 by another data source, the IPES (Indicateurs Theonlyavailablecharacteristiconclassesisclasssize. pour le Pilotage des Etablissements Scolaires). Unfortu- Theinformationonschoolsconsistsoftheestablishment’s nately,notallthevariablesreportedtherearecompatiblein identification number, the educational administrative theirdefinitionswiththoseavailablefortheearlierperiod. region and whether it is a private or public establish- Wewillthereforemainlypresenttheresultsbasedonthe ment. Thanks to the identification numbers, we are able FSEdatasetandreportmorebrieflyonthoseobtainedfor tomatchthestudentpanelswiththetime-varyingZEPor thelongerperiod,whichareessentiallyidentical. non-ZEPstatus.Givenoursampleperiod,wehadstudents We also link these two establishment datasets with ineachofthethree“waves”ofZEPs—1982,1989and1990. filesfromtheEducationMinistry’sinformationsystemfor Wealsocomputedschool-levelvariablesbyaveragingstu- themanagementofteachers.Thissourceprovidesuswith dents’characteristicsbyestablishmentineachpanel. statisticsonteachercharacteristicsbyestablishment,such We now turn to measures of academic achievement. asthenumberofyoungorexperiencedteachers,thediplo- Thepanelsunfortunatelydonotcontaininformationonthe masheld,etc.Finally,anothersetoffilesfromtheMinistry gradesreceivedinnationalexamsornationalevaluations ofEducationidentifywhichestablishmentsarelocatedin (theonlynationalevaluationinhighschoolisin6thgrade). apriorityzone.8 Wedoknow,however,whetherthestudentpassedorfailed anyexamthatsheorhetook.Wealsoknow,foreachyear, whether he or she moves up to next grade, repeats the 3.2. Thepaneldatasets grade,orexitstothevocationaltrack.Moreprecisely,the fourmeasuresofacademicachievementthatweuseare: Themainsourcesofindividualdatausedinouranaly- sisaretwopanelsofstudentscollectedbytheMinistryof Education. - Completionofschoolyearswithatleastonedegree(ver- susfinishingschoolingwithoutanydiploma). - Movingupto8thgrade(versusswitchingtoavocational 3.2.1. The1980panel trackattheendof7thgrade). This panel includes 20,961 students who entered 6th - Movingupto10thgrade(versusswitchingtoavocational grade (age 11) in 1980, and constitutes a representative trackafterthe7thorthe9thgrade). sampleof1/40ofallstudentsenteringjuniorhighschool - Successatthe“Baccalauréat”,theFrenchnationalexam inFrancethatyear.Thesamplingschemehastwolevels. attheendofhighschool,i.e.12thgrade(versusgoingto First,establishmentswerestratifiedaccordingtocitysize, avocationaltrackthatdoesnotleadtotheBaccalauréat, establishment size and sector (public or private school). orfailingtheBaccalauréat). Oneinfivejuniorhighschoolswasthusselected.Second, oneoutofeightstudentsinthoseschoolswasselectedand Thesechoicesaremotivatedbythefollowingobserva- thenfollowedacrossestablishmentsuntilthecompletion tions.First,reducingthenumberofstudentswholeavethe ofhisorhersecondaryschooling.Theattritionratewas9%, schoolsystemwithoutanydegreewasthemaintargetof reflectingdeparturesabroad,deaths,and(in80%ofcases) the Ministry of Education in general and of the ZEP pro- “unexplained”schooldepartures. graminparticular.9Moregenerally,thismeasurecaptures thebottompartoftheachievementdistribution.Second, 7 Throughoutthepaperwewilluse“year”insteadofthe“schoolyear”; forinstance,year1987correspondstotheschoolyear1987–1988. 8 Descriptive statistics are available in Benabou, Kramarz, & Prost 9 Amongstudentsentering6thgradein1980,23%hadnodegreeatthe (2005),BKPhereafter. endoftheirschoolingyears. R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 349 Fig.1. Chronology. inFrancemanystudentsarevirtuallyforcedtoswitchtoa into“ExtremeRight”,“RPRandUDF”(thetwomainright- vocationaltrack(seenasmuchlessdesirableandevenstig- wingparties),“OtherRight”,“Greens”,“CommunistParty”, matizing)attheendof7thor9thgrades.Forinstance,only “Socialist Party”, and “Other Left”. We grouped RPR and 46% of those entering 6th grade in 1980 continued their UDF together, since these two parties formed a coalition education in the regular track all the way to 12th grade. (singlelist)duringsomeoftheparliamentaryelectionsin Reachingthe8thand10thgradesarethereforeimportant the80s.VotinginFrenchparliamentaryelectionsinvolves milestonesintheschoolingprocess.Finally,successatthe two rounds, and as in Bertrand and Kramarz (2002) we Baccalauréatiskeybecauseitconditionsentryintotheuni- usedonlythefirstround.Finally,wefocusedontheyears versitysystemandthe“grandesécoles”(selectivecolleges). thatprecede1982,1989and1990(whennewZEPswere Fig. 1 describes how the time structure of our panels implemented),thatis,ontheelectionsthattookplacein mesheswiththechronologyoftheZEPprogram.In1981, 1981andin1988.Ourvariablesthusconsistofthefraction justbeforethelaunchingofthepolicy,moststudentsofthe ofvotesreceivedbythevariouspoliticalgroupsinevery 1980panelwerein7thgrade(exceptthosewhorepeated “département”,ineachofthesetwoelections. 6thgrade).Bycontrast,forthestudentsinthe1989panel, entryinto6thgradetookplaceas1989waveofZEPswas 4. TheimpactofZEPstatusonschoolcharacteristics ongoing.Andoneyearlater,whenmostofthemwerein 7th grade, the 1990 wave was launched. Because condi- 4.1. Numberofstudentsandclasssize tionsprevailinginthe6thand7thgradesstronglyaffect students’futureschoolingoutcomes,andbecausemobility Using the FSE data for every year from 1987 to 1992, betweenestablishmentsislow,theZEPindicatorweshall weanalyzethechangesinschoolcharacteristicsthatoccur useisequalto1ifthestudent’sestablishmentinhisorher whentheZEPstatusisgranted.Wedistinguishtheimpact 7thgradeislocatedinaZEP,andequalto0ifnot.10Notethat ofthiseventonbothlevelsandtrends,byestimatingthe thisindicatoristime-varying,sincesomestudentsinthe followingmodel 1980paneldotheir7thgradeinschoolsthatwillbecome ZEPsonlyin1982,1989or1990,whilesomeinthe1989 Yj,t =ıt+ıj+1j∈zep89[(cid:2)zep891t≥88+(cid:2)czheapn8g9e1t≥89 phaavneeladcoquthireeidr7ptrhiogrriatydestiantuthse. verysameschoolsafterthey +(cid:2)tzreepn8d9T89]+1j∈zep90[(cid:2)zep901t≥88+(cid:2)zep901t≥89 +(cid:2)czheapn9g0e1t≥90+(cid:2)tzreepn9d0T90]+εj,t, (1) 3.3. Thepoliticaldataset whereY correspondstocharacteristicsforschooljinyear j,t The last set of data that we use is political variables, t,ıtisanindicatorforyeart(yearfixedeffect),ıjanindi- which will provide instruments to address the potential catorfortheestablishment(schoolfixedeffect),1j∈zep89is selection bias in the determination of the priority zones. equalto1iftheschooljbecomesZEPin1989,1t≥(cid:4)isequal to1iftheyeartisgreaterorequalto(cid:4),T isalineartrend Thesharesofthevotereceivedbyeachpoliticalpartyin 89 startingin1989,andεisani.i.d.residual.Thevariablesin everyparliamentaryelectionarecollectedattheregional thesecondlineof(1)aredefinedsimilarly. level (“département”) by the CEVIPOF (“Centre d’Etudes Sinceyeardummyvariablescontrolforevolutionscom- de la Vie Politique Franc¸aise”). To reflect the main fea- mon to all establishments and school dummy variables tures of the French political system we aggregated them control for any difference in levels between them that alreadyexistedin1987,thecoefficients(cid:2)zep89 and(cid:2)zep90 change change capture the change in the level of Y specifically asso- j,t 10 Ourresultsarerobusttousinginsteadanindicatorequalto1ifthe ciated to a change in status. Similarly, (cid:2)zep89 and (cid:2)zep90 studentwasinaZEPineitherthe6thor7thgrade,duetothelowinter- trend trend establishmentmobilitybetweenthesetwoclasses. capture the linear trends specific to ZEP establishments 350 R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 Table1 Students,teachersandZEP. Numberof Numberof Numberof Numberof Numberofweekly Shareofyoung Shareof students studentsperclass teachers teachersper hoursperstudent teachers non-certified student teachers Zep89*1988 −6.7(4.7) 0.2*(0.1) −0.6(0.4) 0.000(0.001) 0.008(0.010) −0.006(0.008) −0.002(0.005) dummy Zep89*1989 −3.9(5.3) −0.2(0.2) −0.3(0.5) 0.000(0.001) 0.012(0.012) −0.014*(0.009) 0.010*(0.006) dummy(zepeffect, level) Zep89*Trend −6.5**(1.5) −0.2**(0.0) 0.1(0.1) 0.001**(0.000) 0.011**(0.003) 0.013**(0.002) −0.003*(0.002) startingin1989 (zepeffect,trend) Zep90*1988 0.3(4.0) 0.3**(0.1) 0.1(0.4) 0.000(0.001) −0.003(0.009) 0.000(0.007) 0.005(0.004) dummy Zep90*1989 −5.2(4.0) −0.2(0.1) −0.3(0.4) 0.000(0.001) 0.008(0.009) 0.008(0.007) 0.004(0.004) dummy Zep90*1990 −6.1(5.1) 0.0(0.2) −0.2(0.5) −0.001(0.001) −0.004(0.011) −0.014(0.008) −0.007(0.005) dummy(zepeffect, level) Zep90*Trend 0.5(2.0) −0.2**(0.1) 0.4*(0.2) 0.001**(0.000) 0.015**(0.004) 0.009**(0.003) 0.012**(0.002) startingin1990 (zepeffect,trend) R-Square: 0.97 0.75 0.96 0.82 0.79 0.66 0.63 Source:FSEfiles,1987–1992,4743establishmentsperyear.StandardErrorsinparentheses(*:significantatthe10%level,**:−5%level).Establishment fixedeffectsandyearfixedeffects. aftertheybecameZEP.BecauseaZEP-specifictrendcould 1995and1999.Again,thedecreaseinthenumberofstu- pre-exist,thecoefficients(cid:2)zep89,(cid:2)zep90and(cid:2)zep90capture dentsisessentiallyconcentratedon6thand7thgraders.12 88 88 89 anychangesinY thatstartedin1988(respectively,in1988 The second and important result in Table 1 is that j,t or1989)betweentheestablishmentsthatbecameZEPin thereductionsinclasssizeassociatedtoZEPstatuswere 1989(respectively,in1990)andtheothers. quite small: on average, a decrease of 0.2 students per InTable1weonlyreport(cid:2)zep89,(cid:2)zep90,(cid:2)zep90,(cid:2)zep89, class per year. On the other hand, class size in 1987 was 88 88 89 change (cid:2)zep90, (cid:2)zep89 and (cid:2)zep90, together with their estimated alreadysmallerinschoolsthatbecameZEPin1990(24.0 change trend trend versus24.4fornon-ZEP).Theseschoolswerethusproba- standard deviations. The estimation is carried out over blyalreadyidentifiedandbenefitingfromextraresources 4743juniorhighschoolsperyear.Amongthose,138estab- evenbeforeacquiringprioritystatus.Theanalysisoverthe lishmentsbecameZEPin1989(weshallrefertothemas longer period 1987 to 1999 shows a continuation of this ZEP-89)and365othersacquiredthestatusin1990(ZEP- modestandslowdecrease.Overthecourseof10years,the 90). averageclasssizedecreasedby2studentsfortheZEP-89 In1987,ZEPestablishmentsareonaverage7–9%larger and by about 1.5 students for the ZEP-1982 and ZEP-90, thanothers.Table1revealsthattheaveragenumberofstu- whiletheaverageclasssizeinnon-ZEPsremainedstable. dentsdecreasedby7peryearinthoseschoolsthatbecame OurresultsarethusconsistentwiththeMinistry’sesti- ZEPin1989.Themostlikelyexplanationisthatofanavoid- mate that ZEP junior high schools in the late 90s had 2 ancestrategyonthepartofsomefamilies,evenbeforethe students less per class than non-ZEP,13 but at the same statuschange,aviewthatfindfurthersupportinananaly- timetheyrevealamuchlesswell-knownfact:theunderly- sisofstudentmobility,whichconfirmthatthereductionin ingprocesswasextremelyslowandactuallybeganbefore thenumberofstudentswasprimarilyduetoreducedentry thegrantingofprioritystatus.Inanycase,eventheten- intothoseestablishments,ratherthantoincreasedexit.11 yearreductioninclasssizeinthepriorityzoneswasatbest ThecoefficientsfortheZEP-90aregenerallynotsignifi- modest,especiallyinlightoftheintendedgoalofimproving cant,butthisisnotsurprisinggiventhattheFSEdataends educationalconditionsofstudentslivinginsociallydisad- in 1992, making it difficult to identify structural breaks. vantagedareas. And indeed when we analyze the full 1987–1999 period usingbothFSEandIPESdata,thepreviousresultsarelargely 4.2. Sizeandcompositionoftheteachingstaff confirmed.ThenumberofstudentsinZEP-89schoolscon- tinueditsgradualdecrease,relativetothegeneraltrend, Table1showsthatthenumberofteachersinZEPschools until 1999. Schools that became ZEP in 1982 lost about didnotincreasemorethaninotherestablishments,except 4 students per year between 1987 and 1999, and those forasmallpositivetrendintheZEP-90.14SinceintheZEP- becameZEPin1990lostabout6studentsperyearbetween 12 Theresultsfortheperiod1987–1999areavailablefromtheauthors 11 Theanalysisofmobility,notshownhere,wasdonethroughlinear uponrequest. probabilitymodelsofleavingaschool,usingthestudentpanelstoanalyze 13 Seefor1997,MinistryofEducation,noten◦98-15. anindividual’sprobabilityofleavinghisorherschoolforanotheronein 14 Overthelongerperiod1987to1996,thenumberofteachersactually thesameeducationaladministrativeregion(seeBKP). decreasedbyoneperyearfortheZEP-1982andZEP-1989,andremained R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 351 Table2 SocialcompositionoftheschoolsandZEP. OccupationoftheHeadoftheFamily: Nationality:African,Asian Schoolinginformation: Nolunchatthe canteenoftheschool White-collarworker Skilledblue-collarworker EstablishmentbecameZEPin1982 −0.0226(0.0244) 0.0355(0.0279) 0.0888**(0.0142) 0.0228(0.0326) EstablishmentbecameZEPin1989 −0.0154(0.0344) 0.0596(0.0394) 0.0600**(0.0201) 0.0526(0.0460) EstablishmentbecameZEPin1990 −0.0618**(0.0294) 0.0901**(0.0336) 0.0454**(0.0172) 0.0466(0.0393) R-square: 0.1279 0.1648 0.2252 0.2929 Sources:1980and1989panels.28,713observations.Regressionsinclude3200establishmenteffects.Therelevantschoolistheonewherethestudentisin hisorher7thgrade.StandardErrorsinparentheses(*:significantatthe10%level,**:−5%level). 89thenumberofstudentswentdownslightlymorethan Theresultsestablishedsofarleadtotwomainconclu- elsewhere,thenumberofteachersperstudentroseslightly sions. First, the extra resources allocated to ZEP schools after 1989. The annual rate of increase was again small, intermsofadditionalteacherslotsandextrahourswere whichisconsistentwiththeresultsonclasssize. quite limited. Second, the more substantial bonuses and It is worth noting that the extra teaching hours promotion incentives granted to ZEP teachers (indepen- attributed to ZEP schools would not necessarily have dentlyoftheperformanceoftheirstudents)didnothelpin translated into reductions in class size. These additional stabilizingtheteachingstafforimprovingitsskillcomposi- man-hours(orwoman-hours)canalsobeusedtodivide tion.Inthe“market”forteachers(internaltotheEducation classesintosubgroupsforsomedisciplines,ortoaddreme- civil service), the ZEP wage premium was not nearly dial classes given in small groups. One way of capturing sufficient to compensate for the adverse “hedonic” and suchusesistocomputetheweeklynumberofhoursper signalingcharacteristicsassociatedwithteachinginsuch teacherandperstudent.Thisratioincreasedwhenaschool schools. acquiredprioritystatus,by0.011hperweekintheZEP-89 andby0.015hintheZEP-90.Thus,onceagain,theincrease 4.3. Socialcompositionofschools wasveryslowandmodest—inpercentageterms,0.8%and 1.2%peryearrespectively.Sincethereductioninclasssize WenowcomparethesocialcompositionofZEPestab- was0.8%peryear,itthusaccountedformostoftheincrease lishmentstothatofnon-ZEPones,bothbeforeandafterthe in hours per student, with only a much smaller fraction former’schangeinstatus.Theevolutionofthesocialmix usedinotherways. isofinterestbothperseandbecauseofitspossibleimpact While class size and hours show only very mod- onstudentperformancethroughpeereffects.16 estchanges,thesignificantresourcesallocatedtoraising Theestimationissimilartothepreviousone,butnow teacherpayinZEPscouldhaveledtoanimprovementin onthepanelsofindividualstudentdata.Onlytwoyearsare thequalificationsoftheprofessorialstaff.Infact,ouresti- compared:1981,whenstudentsfromthePanel1980arein matesrevealexactlytheopposite.Onestandardmeasure their 7th grade and 1990, when students from the Panel of quality is experience: Table 1 shows that the fraction 1989areinthatsamegrade.Thesestudentsaretheones ofyoungprofessors(lessthan30yearsold)inZEPschools whowillbeusedinSection5toestimatetheZEPeffect. goesupslightlyshortlyafterthestatuschange(theincrease Themodelis: appears in 1990 for the ZEP-89). An optimistic interpre- tationofthisevolutionwouldbethatyoungteachersare 1characteristic(i)=ıi∈80panel+ıj(i)+(cid:2)82·1j(i)∈zep82·1t(i)≥82 moredynamicorbetterabletorelatetothechildren,even +(cid:2)89·1j(i)∈zep89·1t(i)≥89 though they have less professional experience. Unfortu- nately,ourresultsonprofessionalqualificationsshowthat +(cid:2)90·1j(i)∈zep90·1t(i)≥90+εi thefractionofteacherswithouttenureandholdinglesser where 1 (i) is a dummy variable for a student diplomas(“maîtres-auxiliaires”,whodonothavethereg- characteristic characteristic(forexample,occupationoftheheadofthe ularteachingcertificate)alsoincreasedslightlyinschools family=executive)ofstudenti;j(i)andt(i)denoterespec- thatbecameZEPin1990.15ItisworthnotingthatinFrance, tively the establishment and year where and when the schools do not hire their teachers. Job assignment are studentwasinhisorher7thgrade;ı isanindicatorfor insteaddeterminedatthenationallevel,throughasystem j(i) theestablishmentwherethestudentwasinhisorher7th inwhichteachersexpresspreferredchoicesandpriorityis grade andı an indicator equal to 1 if the student roughlydeterminedaccordingtoseniority(tenurelength). iε80panel Hence,moreexperiencedteachersareabletochoosethe schoolstheywant,whereasyoungeronestendtoendupin 16 InSection5,wewillestimatetheimpactofZEPstatusonstudent less“desirable”establishments. achievementwithadifference-in-differencesapproach.Thankstothe paneldatasets,wewillcomparestudentsbeinginaZEPschooltothose beinginanon-ZEPschool,withschoolvariablesandschoolfixedeffects stablefortheZEP-90.Asexplainedearlier,theseresultsarenotreported controllingforstabledifferencesbetweenestablishments.Butifthereis herebutareavailablefromtheauthors. deteriorationinthesocialcompositionofZEPschoolsrelativetotheoth- 15 Othermeasuresofskills,suchastheproportionofthosewiththe ers,ourschoolvariablesmaynotbesufficienttocontrolforthatandthe highestteachingdiploma(“agrégation”),remainedvirtuallyunchanged estimatedZEPeffectmayunderestimatethe“pure”impactoftheZEP (resultsnotreportedhere). treatment(effectoftheextraresourcesandeducationalprojects). 352 R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 belongedtothe1980panel.Indicators1j∈zep82,1t≥82,etc., tion)couldhavediscouragedeffortbybothprofessorsand aredefinedinthesamewayasbefore. students,leadingtodeteriorationinschoolperformance. Table2presentstheresults.Itseemsthatthecomposi- Finally,ifteacherexperienceimprovesstudentattainment, tionofthestudentsdidnotchangemuchintheZEPschools, theincreaseinthefractionofyoungerteacherscouldhave exceptregardingstudents’nationalities.In1982,theZEP lowered the educational achievement of some or all the schoolsalreadyhadalargershareofAfricanandAsianstu- students. dents than the other schools. This share increased more ToassessthevalueoftheZEPprogram,onetherefore thanelsewherebetween1982and1990(Table2).Butitis needstoquantifytheoverallimpactofgrantingprioritysta- worthnotingthataccordingtoseveralstudies,17 children tustoaschoolonstudents’academicachievement.Thisis fromimmigrationhaveinFrancesimilarorbetterachieve- thepaper’ssecondmainobjective,towhichwenowturn. mentthanotherchildren,allotherfactorsbeingequal.Our resultsontheachievementregressions,inSection5,lead 5.1. Estimationstrategy tothesameconclusions. There is no strong evidence of deterioration of the Itwouldclearlybeinappropriatetosimplyregressindi- socialcompositioninZEPschools.Concerningtheparents’ vidual student performance on personal characteristics occupations, the only significant change is more skilled plus a ZEP indicator, and thus compare mean outcomes blue-collar workers and less white-collar workers in the betweenZEPandnon-ZEPstudents.Indeed,theZEPvari- schoolsthatbecameZEPin1990.Thereisalsonochange ablecertainlyisendogenous,evengivenallourindividual in the fraction having lunch at the school cafeteria.18 As controls.Forinstance,prioritystatuscouldhavebeenpref- mentionedearlier,inFrancechildrenwhoareenrolledin erentiallygrantedtothoseestablishmentswhereschooling the cafeteria plan typically come from more advantaged outcomesweretheworseordeterioratingthemostrapidly, backgroundsthanthosewhodonot. orconverselytothoseamongthe“difficult”zonesdeemed the most likely to succeed. To deal with this prob- 5. TheimpactofZEPstatusonindividualschooling lem,weusebothdifference-in-differences(implemented achievement through establishment fixed effects) and instrumental variables. Ourresultssofarshowthatthefinancialsupportgiven Thefirstmethodexploitsthefactthat,thankstoourtwo toZEPschoolswasfarfromnegligible,butalsothatthese panels, we can compare students who went through the funds were sprinkled across many establishments, with- samegradeinthesameschool,butwithsomeattending out any apparent targeting towards the potentially most before it became a ZEP and others nine years later, after efficient inputs or towards those students most likely to it had acquired priority status. The idea is then to sub- benefitfromtheseextrainputs.Thusthedecreaseinclass tractfromthedeviationbetweenZEPandnon-ZEPschools size was small and progressive, the number of teachers estimatedonthe1989panelthecorrespondingdeviation hardlyincreasedandtheirqualificationremainedatbest estimated on the 1980 panel, but with the ZEP indicator unchanged. replaced there by a “future ZEP” dummy, equal to 1 if a The ZEP “treatment”, however, is a potentially much school was part of the ZEP “wave” of 1989 or 1990. This more complex object than a simple change in financial differenceindifferencesapproachcontrolsforanyunob- resourcesorteachinghours.First,theofficialgoaloftheZEP servedfactorsaffectingstudentperformanceinthepriority programwasalsotoprovidethemeansforschoolstocre- zones(relativetonon-ZEPones)thatalreadyexistedprior ateneweducationalprojectsandconnectmorecloselywith tothestatuschange,andthereforeyieldsanunbiasedesti- local institutions such as municipalities. These projects mate of the reform’s impact (assuming stability of the weresupposedtohaveapositiveimpactontheacademic unobserved heterogeneity and distribution of errors.) In achievementofZEPstudents.Second,thesizeablebonuses the linear model, this method is implemented both very andcareerimprovementsofferedtoteachersinZEPschools simplyandmoregenerallybyrunningasingleregression couldhavecontributedtoimprovingtheirmotivation,19or that includes establishment-specific fixed effects, which allowedtheEducationMinistrytoselectteachersforthese control for unobserved stable heterogeneity across all schoolsfromahigherqualitypoolofapplicants(inways schools. notreflectedbyseniorityandtenureindicators).Third,on Oursecondmethodfordealingwithpotentialselection thenegativeside,anadversesignalingeffect(stigmatiza- biasesusesinstrumentalvariables,describedinsubsection 3.3. Inallthesecases,ourestimatesdifferentiatebetween 17 SeeforinstanceCailleandVallet(1995)andGary-Bobo,Prieto,and thethree“waves”ofZEPs,thuscapturingapotential“dura- Picard(2006). tioneffect”:forinstance,someestablishmentsthatwere 18 ThisinformationisalsoavailableintheFSEdata.Thesameestimation ZEPsin1990acquiredthatstatusin1982,whereasothers asin3.1and3.2showsthatthefractionofstudentshavinglunchatthe receiveditonlyin1989. schoolcafeteriadecreasedslightlyintheZEPschoolsafter1989and1990 comparedto1988andtotheotherschools(seeBKP).Thisdecreasemay havefollowedanincreasebetween1982and1988. 5.2. Difference-in-differencesestimates 19 Sincethesebonuseswerenotconditionedonstudentachievement oranyotherperformancemeasuretheycouldnothavehadanystan- Inthissectionweestimatetheeffectofthe“ZEPtreat- dardincentiveeffect.Theycouldstill,however,haveenhancedteachers’ ment” using the difference-in differences technique in a “intrinsicmotivation”byelicitingfeelingsofreciprocityorprofessional pride. linearprobabilitymodelwithfixedeffects.Thus,astudent’s R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 353 academicachievementismodeledasfollows: Table3 Linearmodelwithestablishmentfixedeffects:ZEPvariable. 1achievmt(i)=˛Xi+ıi∈80panel+ıj(i) Gettingonedegree Movingupto8th-grade Coefficient Stderr Coefficient Stderr +(cid:2)82·1j(i)∈zep82·1t(i)≥82 ZEPin1982 −0.0428 0.0299 −0.0052 0.0293 +(cid:2)89·1j(i)∈zep89·1t(i)≥89 ZZEEPPiinn11999809 −00..00003608 00..00346264 −00..00132369 00..00336934 +(cid:2)90·1j(i)∈zep90·1t(i)≥90+εi (3) R2 0.2077 0.2556 Nbofobs 27831 28713 where1 (i)isanindicatorforachievement(moving achievmt Movingupto10th-grade Successatbaccalauréat upto8thgrade,to10thgrade,gettingatleastonedegree, successattheBaccalauréat)ofstudenti;j(i)andt(i)denote Coefficient Stderr Coefficient Stderr respectivelytheestablishmentandyearwhereandwhen ZEPin1982 −0.0046 0.0330 −0.0200 0.0338 thestudentwasinhisorher7thgrade;ı isanindica- ZEPin1989 0.0561 0.0497 0.0212 0.0457 j(i) torfortheestablishmentwherethestudentwasinhisor ZEPin1990 −0.0171 0.0432 −0.0494 0.0443 her 7th grade and ıiε80panel an indicator equal to 1 if the R2 0.3272 0.3179 student belonged to the 1980 panel. The vector X con- Nbofobs 27831 27831 i tainsthestudent’sindividualandfamilycharacteristicsas Sources:1980and1989panels.Regressionsincludeabout3200establish- wellasestablishmentvariables,measuredastheaverage mentfixedeffects,inadditiontotheindividualcharacteristicsandschool oftheindividualvariablesoverthosestudentsinthepanel variables(averagesofindividualcharacteristicsbyschooloneachpanel). Therelevantschoolistheonewherethestudentisinhisorher7thgrade. studyinginthesameschoolj(i).Thisaggregationisdone StandardErrorsinparentheses(*:significantatthe10%level,**:−5% separatelyforeachpanel,sotheseestablishmentvariables level). aretime-varying.Finally,theindicators1j∈zep82,1t≥80,etc., aredefinedinthesamewayasbefore. Turning to the variables of central interest,22 we see Eq. (3) makes clear how the coefficients (cid:2) , (cid:2) and 82 89 fromTable3thattheZEPcoefficientsareneversignificantly (cid:2) , which capture the effects of a change to ZEP status, 90 different from 0, irrespective of the measure of achieve- areidentifiedbyusingourtwopanelssimultaneously.For mentused.Thesenon-significantZEPeffectsshowthatthe instance,foragivenestablishmentthatbecameZEPin1982 granting of “priority” status to their school did not help theindicatorfunction1j(i)∈zep82·1t(i)≥82goesfrom0forthe students,oncepre-existingdifferencesbetweenestablish- studentsinthe1980panelwhoattendedthatschoolto1 mentsarecontrolledfor.Thisisoursecondmainfinding. forthoseinthe1989panelwhofollowedthemtherenine yearslater. 5.3. Instrumentalvariablesestimation Theregressionscorrespondingtoreaching8thgradeare carriedoutover17,279studentsenrolledin2099establish- ToaddresstheendogeneityoftheZEPstatus,wealso ments in the 1980 panel and 11,435 students enrolled in estimate the model using instrumental variables, based 3031establishmentsinthe1989panel.20Theidentification on the shares of the vote received by the various parties of the ZEP coefficients relies on the 1944 establishments (or coalitions) in the first round of the 1981 and 1988 present in both panels, with 93 of these becoming ZEP parliamentaryelections.Severalelementsrevealthatthe in1982,40in1989and62in1990.Thereferenceestab- grantingofZEPstatuswasindeedinfluencedbypolitical lishment is that of the first 7th grade of each student. considerations,ontopofeducationalones.First,establish- Whenestimatingtheotherthreeequations–forobtaining mentsinadesignatedgeographicalzonecouldoptoutand, adegree,movingupto10thgradeandsuccessattheBac- in the first years of the program, the Communist Party calauréat–thesampleconsistsof16,816studentsenrolled gave instructions to its mayors to refuse the ZEP status in2051establishmentsforthe1980paneland11,016stu- in their cities, as it saw it as stigmatizing. Second, and dentsenrolledin3009establishmentsforthe1989panel. surprisingly,priorityzoneswereinitiallyconcentratedin TheidentificationoftheZEPcoefficientsnowrelieson1891 onlyahandfulofregions,especiallyinSeineMaritimeand establishmentscommontobothpanels,with93establish- Aquitaine,whicharefarfrombeingparticularlypoor.By mentsbecomingZEPin1982,40in1989and61in1990.In contrast,therewerenopriorityzonesinMarseillesuntil thesecases,thereferenceestablishmentisthatofthelast the“wave”of1990,eventhoughitisFrancethird-largest 7thgradeofeachstudent.21 city and includes some of its most disadvantaged areas. Andeventhen,therewereoddpriorities:asmanyas29% ofthejuniorhighschoolslocatedintheNièvre“départe- ment” were granted ZEP status, as were 19% of those in 20 We excluded the observations from Corsica from our sample, in ordertomaketheregressionsmorecomparabletothoserunlaterwith the Ariège “département”—two rural regions with obvi- instrumental-variables,inwhichthepoliticalvariablewasnotavailable ously much less need for this type of program. Analysts forthatregion(duetothefactthatasingle“département”waslatersplit dulynotedthattheNièvrewaspresidentMitterand’selec- intotwo).LeavinginCorsicahasnoeffectontheresults,however. 21 Theideaherewastominimizethetimeintervalbetweenthe7th gradeZEP/non-ZEPtreatmentandthe10thgradeorBaccalauréatout- come.Usingthefirst7thgradeinsteadmakesnodifferencetotheresults, 22 Resultsofindividualandestablishment-levelvariablesareshownin however. BKP. 354 R.Bénabouetal./EconomicsofEducationReview28(2009)345–356 Table4 InstrumentationoftheZEPindicators. Zepin1989 Zepin1990 Coef. Sterr Coef. Sterr Votesduringparliamentaryelections: Extreme-right −0.0948** 0.0373 0.4238** 0.0432 RPR-UDF −0.0213 0.0421 0.0944* 0.0487 Otherright 0.1609** 0.0476 −0.1986** 0.0550 Otherleft −0.3355** 0.0793 −1.1475** 0.0916 Communistparty −0.3632** 0.0458 −0.1434** 0.0529 Greens 0.0783 0.0977 0.9149** 0.1129 Extreme-left 0.1846** 0.0590 0.3613** 0.0682 F-statforthesignificanceoftheinstruments 25.24 63.97 (p-value) (p<0.01) (p<0.01) Sources:1980and1989panels.27,831observations.Regressionsincludeabout3200establishmentfixedeffects,inadditiontotheindividualcharacteristics andschoolvariables(averagesofindividualcharacteristicsbyschooloneachpanel).ThepoliticaldataareCevipofdata,fortheyears1981and1988.The estimationisdonewithalinearprobabilitymodel.TheFisherstatisticsisthesignificancetestofthepoliticalvariables.Therelevantschoolistheonewhere thestudentisinhisorher7thgrade.StandardErrorsinparentheses(*:significantatthe10%level,**:−5%level). toralstronghold,andAriègethatofLionelJospin,Minister cationoperates),whereeducationaloutcomesandpolitical ofEducationbetween1988and1992.EvenLionelJospin conditionscouldbequitecorrelated;(ii)thecontrolvari- recognizedthattheattributionoftheZEPstatuswaslinked ables include school fixed effects, which will absorb in to political considerations: in a June 1998 speech at the particularanyfixeddifferencesinthepopulationcompo- “AssisesnationalesdesZEP”inRouen,whenhewasnow sitionofa“department”thatcouldaffectbothitspolitical PrimeMinister,hecriticizedthefactthat39%ofjuniorhigh outcomeandthe(relative)performanceofitsmostdisad- schoolstudentsinNièvrewereinaZEPschoolagainstonly vantagedschools. 19%ofstudentsinSeine-Saint-Denis,averypoor“départe- The first-stage regression corresponds to the linear ment”wheretheriotsstartedinNovember2005. probabilitymodel Our instrumentation strategy relies on two hypothe- 1 =X(cid:6)ˇ+(cid:2)P +ı +ı +ε (4) ses.First,politicalfactorsmustaffectthedeterminationof iεzep89 i j,t(i) iε80panel j i whereZEPschoolsarelocated—aswejustconfirmed.Sec- wherePj,t(i)denotesthesharesofthedifferentpartiesin ond,thepoliticalvariablemustbeuncorrelatedwiththe theparliamentaryelectionsthattookplacein1981(resp. (differential)performancestudentsinZEPschools,condi- in1988)ifthestudentbelongedtothe1980panel(resp. tionalonalltheotherexogenousregressors.Tounderstand the1989panel),theshareoftheSocialistPartybeingthe whythisisaplausibleassumption,itisimportanttonote omittedvariable.Theinclusionoftheestablishmentfixed that:(i)thepoliticalvariableismeasuredatthelevelofa effectsisagainallowedbythefact(withtheestimationper- “département”,whichisamuchlargerentitythanthatof formedonbothpanelsofstudents)boththeZEPvariables schooldistricts(thelevelatwhichtheZEP/non-ZEPclassifi- andthepoliticalinstrumentsaretime-varying. Table5 Linearmodelwithinstrumentalvariables. Gettingonedegree Movingupto8th-grade Coefficient Stderr Coefficient Stderr ZEPin1989 0.1622 0.5195 −0.0546 0.5177 ZEPin1990 0.0106 0.3079 −0.1541 0.2915 R2 0.1001 0.1354 Nbofobs 27831 28713 (cid:2)2over-identification 3.2728 2.4089 test(p-value) (p=0.86) (p=0.93) Movingupto10th-grade Successatbaccalauréat Coefficient Stderr Coefficient Stderr ZEPin1989 0.0062 0.6199 0.3879 0.6293 ZEPin1990 −0.0608 0.3674 −0.5339 0.3730 R2 0.2161 0.1798 Nbofobs 27831 27831 (cid:2)2over-identification 2.3147 5.9496 test(p-value) (p=0.94) (p=0.55) Sources:1980and1989panels.Regressionsincludeabout3200establishmentfixedeffects,inadditiontotheindividualcharacteristicsandschoolvariables (averagesofindividualcharacteristicsbyschooloneachpanel).TheZEPvariableisinstrumentedbytheschoolvariablesandthepoliticalvariables.The relevantschoolistheonewherethestudentisinhisorher7thgrade.StandardErrorsinparentheses(*:significantatthe10%level,**:−5%level).
Description: