IZA Discussion Paper No. 3112 - PDF document

IZA Discussion Paper No. 3112 October 2007 ABSTRACT Dismissals for Cause: The Difference That Just Eight Paragraphs Can Make * This paper provides evidence about the effects of dismissals-for-cause requirements, a specific component of employment protection legislation that has received little attention despite its potential relevance. We study a quasi-natural experiment generated by a law introduced in Portugal in 1989: out of the 12 paragraphs in the law that dictated the costly procedure required for dismissals for cause, eight did not apply to firms employing 20 or fewer workers. Using detailed matched employer-employee longitudinal data and difference-in-differences matching methods, we examine the impact of that differentiated change in firing costs upon several variables, measured from 1991 to 1999. Unlike predicted by theory, we do not find robust evidence of effects on worker flows. However, firm performance improves considerably while wages fall. Overall, the results suggest that firing costs of the type studied here decrease workers’ effort and increase their bargaining power. JEL Classification: E24, J64, J65 Keywords: employment protection legislation, worker flows, wages, firm performance Corresponding author: Pedro S. Martins School of Business and Management Queen Mary, University of London Mile End Road London E1 4NS United Kingdom E-mail: p.martins@qmul.ac.uk * I am grateful to Alcides Martins (Alcides Martins & Associados) for helping me with the very detailed Portuguese employment law. I also thank Dick Allard, Giulio Fella, Winfried Koeniger and Eshref Trushin for detailed comments and the ESRC (RES-062-23-0546) and for research support. All views and any errors in this paper are of my responsibility only. sensethatthoseringconstraintsmaystrengthenthebargainingpowerofincumbentworkersvis-a-vistheiremployers,whilethelattertendtoberegardedasthestrongerpartinthebargainingprocess.Inanycase,thelegalproceduressurroundingdismissalsforcausecaninducedeadweightlosses,asthecostsbornebyrmswhencarryingoutsuchdismissalsarenotrecoupedbytheworkersaected.Thosedeadweightlossescanthenresultininecientlylownumbersofseparationsand,indirectly,alsoininecientlylownumbersofhirings.Moreover,whileonecanarguethatemployersmaybeabletobargainwithworkersoveracompensationpaymentlargeenoughforthelattertoaccepttoquit,adverseselectionproblemsmaymakesuchapproachparticularlycostlyforrms(ifworkerstoberedhaveworseoutsideoptions,theyarelikelytodemandhighercompensationpayments).Furthermore,ifrmscircumventtheconstraintsimposedbyEPLbymakingcompensationpaymentstoworkersthatunderperformorthathavedisciplinaryproblems,thatmayundermineworkermorale(Bewley1999)andreducethelevelsofeortoftheremainingworkforce.3Inthispaper,weseektoassessthecausalimpactoftheadministrativeproceduresthatrestrictringsforcause.Asfarasweknow,wearethersttoexamineempiricallytheroleofthisspeciccomponentofEPL.Ouranalysisisbasedonaquasi-naturalexperimentthatoccurredinPortugal,acountrywellknownforitsverystrictEPL(Blanchard&Portugal2001,OECD2004,Boteroetal.2004).Specically,theexperimentresultsfromanewlawgoverningemployeedismissalsintroducedin1989,underwhichringconstraintswerereducedforallrms.However,rmsemploying20orfeweremployees(unlikelargerrms)wereexemptedfromanumberofadministrativerestrictionsregardingdismissalsforcause.Moreover,sinceuntilthentherewasnodierentiationinringconstraintsacrossrmsintermsoftheirsizeorothercharacteristics,onecansetupadierence-in-dierencesanalysis,bycontrastingtheoutcomesof`smaller'and`larger'rms,beforeandafter1989.Aswecandrawonparticularlydetailedpaneldata,wealsoconductadierence-in-dierencespropensityscorematchinganalysis,inordertominimiseanybiasdrivenbyun-observeddierencesandbydierencesinthecommonsupportacrossthetwormtypesandtheirworkers(Heckmanetal.1998).ThisistherststudyintheEPLliteraturethatadoptsthisspecicempiricalapproach.Finally,weprovideanexhaustiveanalysisoftheimpactsofthenewlaw,byconsidering 3Onecanarguethatthestringencyanddetailofthelegalproceduressurroundingdismissalsforcausealsodetermine,albeitindirectly,theminimumlevelofeortthatemployeesneedtoexerttokeeptheirjobs.3 Itisimportanttounderlinethatifanyformalaspectofthistime-consumingadministrativeprocedurewerenotpursuedandifthedismissedworkersubsequentlychallengedthelegalityofhis/herdismissal,thenthecourtwouldmostlikelydeclarethedismissalasillegal.Inthatcase,thecourtwouldalsoorderthermtoreinstatetheworkerandtopayhim/herallwagescorrespondingtotheperiodsincetheworkerwasunlawfullydismisseduntiltheworkerwasreinstated.Evenifthedismissalwasdeemedlegalorifthelegalityofthedismissalwasnotchallengedbytheemployee,thermwasstillalwaysobligedtopayseverancebenets.Thesebenetswereconsiderable,astheycorrespondedtoonemonthofpayperyearoftenure,withaminimumofthreemonthsofpay.Afterabouttenyearsofrelativeeconomicstagnationthatfollowedthe1974coupd'etat,Portugueseeconomicpolicyeventuallybecamemoremarketfriendly.Underthegovernmentsofthemid-andlate-1980s,severalreformsenvisagingmore exibleproductandfactormarketswereintroduced.Moreover,in1986,thecountrybecameamemberoftheEuropeanCommu-nity,afterwhichcapitalin owsincreasedsubstantially.Underthispositiveeconomiccontext,anewemploymentlawsoughttorevertor,atleast,attenuatetheveryrestrictiveconditionsgoverningtheterminationofpermanentcontractsdescribedabove.Afteraperiodofheatedpoliticaldebate,publicdemonstrations(includingageneralstrike),anddetailedscrutinybytheconstitutionalcourt-alleventswhichgeneratedconsiderableuncertaintyaboutwhethertheintendedreformswouldindeedcomethrough-anewlaw,Decreto-Lei64-A/89,nallycameintoforceattheendofMayof1989.4Thisnewlawsoftenedconsiderablythedismissalconstraintsfacedbyrms,namelybywideningtherangeofcircumstancesinwhicharmcouldreaworkeremployedunderapermanentcontract.Unlikeundertheoldlaw,itbecamepossibleforrmstoreaworkerbecauseofstructural,technologicalorbusiness-cyclereasons.However,whilethebenchmarkadministrativeprocedurerequiredfordismissalsforcauseremainedlengthyandcomplex,thenewlawallowedsmallrms(denedasthosermsemploying20orfewerworkers)tofollowamuchsimplerprocedure.Inparticular,outofthe12specicrulesthatlargerrmsneededtofollow(eachruleoutlinedinaseparateparagraphofarticle10ofthe1989law),onlyfourofthoserulesneededtobeconsideredbysmallerrms(article15).5Thisdierentiation 4CavacoSilva(1995)providesananalysisofthisandotherreformsintroducedinPortugalfromthemid-1980stothemid-1990s.5Theonlyexceptiontothisstreamlinedprocedureforsmallerrmswaswhentheworkerbeingdismissedwasaunionleader.Inthiscase,thebenchmark,12-paragraph-longprocedureapplied.5 relationshipsthanbefore1989)andtheeaseningofcollectivedismissals.Othernewlegaldiplomasinemploymentlawcoveredchildwork,healthandsafetypracticesandstrikes.Unlikewithdismissalsforcause,thetightertemporarycontractsdidnotchangeinadierentwayforrmsofdierentsizes.However,itisnotimpossiblethatsmallerrmshadadierentpercentageoftheirworkforcesmadeupoftemporarycontractsandthatmayconfoundtheassessmentofthemainreform.Ontheotherhand,collectivedismissalswereeasenedinaslightlydierentwayforrmsofdierentsizes.Firmsemployinglessthan50employeeswerefrom1989allowedtoconductacollectivedismissalinvolvingonlytwoemployees,whileforrmsemploying50ormoreemployeesacollectivedismissalrequiredthatatleastveworkerswerelaid-o.Inordertominimiseanybiasinourresultsrelatedtothechangesintemporarycontractsorcollectivedismissals,wefocusouranalysisoftheimpactofthecostsofdismissalonlyonrmsemployinganumberofworkers`suciently'closetothermsizethresholdofinterest(20).Moreover,weonlyconsiderrmswhichare`relatively'farfromthethresholdthatappliedforcollectivedismissals(50).Specically,inourbenchmarkresults,weconsideronlyrmsemployingbetween10and30workersuptoMay1989(whenthenewdismissalslawcameintoforce).Thisrelativelynarrowrangeofrmsizesalsoimpliesthatanybiasesrelatedtodierencesintheshareoftemporarycontractsacrossthetwotypesofrmsarelikelytobesmall.Moreover,ouruseofmatchingtechniques-describedbelow-alsohelpsouridenticationinthisregard,byrestrictingourcomparisontormsthatareeectivelycomparablealongalargesetofobservablevariables.Finally,therestrictionupontherangeofrmsizesweconsiderisalsoimportantinitself,eveniftherewerenoasymmetricchangesincollectivedismissals.Infact,theassumptionofcommontrendsforthetreatmentandcontrolgroupsislesslikelytoholdforawiderrangeofrmsizes.Thisandothermethodologicalissuesaredescribedinamoreformalwayinthenextsection.3IdenticationLetYDitbethepotentialoutcomeofinterestforindividuali(arm,inourcontext)attimethadtheybeeninstateD,whereD=1ifexposedtothetreatment(armemployinglessthan20workers)and0otherwise.Lettreatmenttakeplaceattimet(fromMay1989,inourcase).Thefundamentalidenticationproblemliesinthefactthatwedonotobserve,attime7 t,individualiinbothstates.Therefore,wecannotcomputetheindividualtreatmenteect,Y1it�Y0it.Onecan,however,ifprovidedwithaconvenientcontrolgroup,estimatetheaverageeectofthetreatmentonthetreated.Oneapproachisadierence-in-dierences(D-in-D)estimator(seeMeyer(1995)),inwhichoneusesanuntreatedcomparisongrouptoidentifytemporalvariationintheoutcomethatisnotduetothetreatment.However,inordertoachieveidenticationofthegeneralD-in-Destimatorweneedtoassumethattheaverageoutcomesfortreatedandcontrolswouldhavefollowedparallelpathsovertimeiftherehadbeennotreatment.Thisisknownasthetime-invarianceassumption,E[Y0it�Y0it0jD=1]=E[Y0it�Y0it0jD=0];(1)wheret0isatimeperiodbeforetheprogramimplementation.Theassumptionstatesthat,overtime,theoutcomevariableoftreatedindividuals(D=1),intheeventthattheyhadnotbeenexposedtothetreatment,wouldhaveevolvedinthesamefashionasactuallyobservedfortheindividualsnotexposedtothetreatment(D=0).Ifassumption(1)holds,theD-in-DestimateoftheaveragetreatmenteectonthetreatedcanbeobtainedbythesampleanalogsofbD-in-D=fE[YitjD=1]�E[YitjD=0]g�fE[Yit0jD=1]�E[Yit0jD=0]g:(2)Theexpressionabovesimplystatesthattheimpactoftheprogramisgivenbythedier-encebetweenparticipantsandnonparticipantsinthebefore-afterdierenceinoutcomes.Apotentialproblemwiththisapproachisthatthetime-invarianceassumptioncanbetoostringentifthetreatedandcontrolgroupsarenotbalancedincovariatesthatarebelievedtobeassociatedwiththeoutcomevariable(seeAshenfelter(1978)).Inthiscase,theD-in-Dsetupcanbeextendedtoaccommodateasetofcovariates,somethingwhichisusuallydonelinearly,takingintoaccounteligibilityspeciceectsandtimeoraggregateeects.Inthefollowingmodel,basedonasampleoftreatmentandcontrolunits:Yit=D+t+0Zit+DDt+"it;(3)whereDisasbeforeandrepresentstheeligibility-specicintercept,tcapturestimeor8 aggregateeects,andZisavectorofcovariatesincludedtocorrectfordierencesinobservedcharacteristicsbetweenindividualsintreatmentandcontrolgroups,bDwouldcorrespondtotheD-in-Destimate.ThisestimatorcontrolsforbothdierencesintheZsandfortime-speciceects,butitdoesnotimposecommonsupportonthedistributionoftheZ0sacrossthecellsdenedbytheD-in-Dapproach.InordertoaddressthispossibleshortcomingofthestandardD-in-Dmethod,wecomple-mentitwithamatchingframework(Rosenbaum&Rubin1983),resultinginadierence-in-dierencesmatching(DDM)estimator(Heckmanetal.1997,1998).DDMhasbeenrecentlyreviewedandcomparedwithothermethodsbySmith&Todd(2005)andhasbeenshowntohavethepotentialbenetofeliminatingsomesourcesofbiaspresentinnon-experimentalset-tings,improvingthequalityofevaluationresultssignicantly.Moreover,DDMisparticularlyappropriateforouranalysisaswecandrawuponarichsetofcovariates,alldataarecompiledbythesameagencyandwecanalsousedataforcomparisongroupsfromthesamelocallabormarket(Heckmanetal.1997).Ingeneral,thefeasibilityofthematchingstrategyreliesonarichsetofobservableindividualcharacteristics,X,sothatthedistributionoftheindividualcharacteristicsimportanttotheevaluationexerciseisthesameinthedierence-in-dierencescells.Thematchingprocessthenmodelstheprobabilityofparticipationandmatchesindivid-ualswithsimilarpropensityscores.Moreover,thetimeinvarianceassumptionfortheDDMestimatorisnowE[Y0it�Y0it0jp;D=1]=E[Y0it�Y0it0jp;D=0];(4)wherep=Pr(D=1jX)isthepropensityscore.Whenestimatingthemeanimpactofthetreatmentonthetreated,thematchingestimatorrequiresaconditionalmeanindependenceassumption,EY0itjX;D=1=EY0itjX;D=0=EY0itjX;(5)andalsorequiresthatthereisanonparticipantanalogueforeachparticipant,implyingthatPr(D=1jX)1.TheDDMestimatortakestwoforms,dependingonthenatureofthedata,namelyre-peatedcross-sectionsandpaneldata.Inthelattercase,theoneemployedinthispaper,theestimatorinvolvesrstcalculatingthedierencesovertimeinthedependentvariableforeach9 observationandthenmatchingtreatmentandcontrolunitsusingpropensityscoreestimatesbasedon`before'-periodcharacteristics.Formally,bDDM=Eh(Y1t�Y1t0)�bE�Y0t�Y0t0jP
i;(6)wherebE(YjP)representstheexpectedoutcomeofindividualsinthecontrolgroupmatchedwiththoseinthetreatmentgroup.4DataThedatausedinthispaperarederivedfrom`QuadrosdePessoal'(QP),aparticularlyrichannualcensusofallrmsthatoperateinPortugalandthatemployatleastoneworker.Undertheregulationsofthiscensus,whichisadministeredbytheMinistryofEmployment,eachrmislegallyrequiredtoprovideextensiveinformationaboutitselfandalsoabouteachoneofitsworkersthatareemployedatthecensusreferencemonth(thereferencemonthisMarchupto1993andOctoberfrom1994onwards).Giventheextensivecoverageofthedata,theonlygroupsofworkersnotpresentinthedataaretheself-employedandthepublicsectoremployees,besidestheunemployed.Moreover,theperiodcoveredbythedataisalsorelativelylong,asthecensushasbeenongoingsince1982.8Thelonglistofvariablesavailableinthedataincludesuniqueidentiersforeachrm,foreachestablishmentandforeachemployee.Theseidentiersallowustofollowworkersovertime,eveniftheymovebetweenrms.Otherrm-levelvariablesaretheeconomicsector/industry(measuredattheve-digitlevel),region(upto400dierentunits),numberofemployees(constructedfromtheworker-leveldata),rmage,typeofownership(public,private/domesticorforeignowned),sales,andequity.Attheworker-level,thedatamakeavailableinformationaboutschooling,age(monthandyearwhentheworkerwasborn),gender,tenure(monthandyearwhentheworkerwashiredbytherm),occupation(ve-digitcode),wages,hoursworked,joblevel(atwo-digitvariable,comparableacrossrmsandovertime)andpromotions(monthandyearwhentheworkerwaslastpromotedintherm).Experienceisconstructedasage�education�6. 8Therehavebeentwodiscontinuitiesoverthisperiod:thereisonlyemployer-leveldatafortheyearof1990;thecensusdata(bothattheemployer-andtheemployee-level)arenotavailablefor2001.Overall,onaverage,between1982and2004,thereareapproximately2.5millionworkersandmorethan200,000rmsperyear.10 employees(seeSection2).Wealsodroprmsemployingexactly20workers,thethresholdlevelinthenewdismissalslaw,asitmayhavebeenunclearifsuchrmsbelongedtothetreatmentorthecontrolgroups(although,strictlyspeaking,rmswith20employeeswouldbeinthe`treatment'group).Furthermore,rmsizewillalsotypically uctuateovertime,evenifonlyslightly,whichwouldpossiblymakeitriskytoassignrmswith20employeestothecontrolortothetreatmentgroups.Ourdenitionofsizeisbasedonlyonpaidemployees,excludingothertypesofworkers(employers,unpaidfamilyworkersandotherresidualcategories).Oneconcernwhenselectingthesampleofinterestismeanreversionorthe`regressionfalacy'(Davisetal.1996a,b),asselectingrmsintothetreatmentorcontrolgroupsbaseduponsizeinasingleyearonlycouldbiasourresults.Infact,suchassignmentwouldprobablyimplythatsomermsinthesmall(large)sizecategorywouldcorrespondtormsthataretypicallyofalarger(smaller)sizebutthathadhadarelativelybad(good)yearinthatperiod.Thesermswouldthereforebelikelytoswitchbacktotheir`permanent'sizeafter1989,thuspotentiallydistortingouranalysis.Inordertoaddressthisproblem,weconstructasamplemadeupofrmsthatarelikelytohavereachedtheir`permanent'sizeby1989.Specically,werestrictoursampletormsthatremaininthesamesizecategory,between10and19workersorbetween21and30workers,overaperiodofthreeyearsupto1989.Initially,wend16,267rmsthatemploy10-19or21-30workersin1987,whilein1988and1989,theequivalentnumbersofrmsare17,565and18,964,respectively.Whenrestrictingthesampletormspresentinthedatainallthreeyearsandthatremaininthesamesizecategoryovertheperiod(i.e.thatarealways`small'or`large'from1987to1989),weobtain7,480dierentrms,ofwhich5,863are`small'.10Intermsoftheirobservablecharacteristics,somenoteworthydierencesbetweenthetwogroupsofrmsincludeworkerreallocationrates(anaverageof0.37forsmallerrmsandof0.33forlargerrmsinthe1989data)andhourlypay(2.76eurosperhourforsmallerrmsand2.96forlargerrms,againinthe1989data)-seeTableA.1foralistofdescriptivestatisticsforeachtypeofrms.The7,480rmsconsideredemployed122,062individualsin1989.Thisisalsotheyearinwhichthetotalnumberofemployeesofthoseselectedrmspeaks,althoughtheequivalentnumbersfor1987and1988areverysimilar(119,401and121,561,respectively)-seeTable 1044%ofthesermsarepresentinallyearsfrom1986to1999,theperiodwecoverinourdata.Thatisalsobyfarthemostcommontimepatterninthedata,asthesecondmostcommonpattern,comprisingrmsthatarepresentfrom1986to1993,includesonly4%ofallrms.12 Table1presentstheresultsconcerningthebalancingofcovariatesacrossthetreatmentandcontrolgroups,beforeandaftermatching.Wedisplaytheresultsforthemainvariablesmeasuredin1989andtheirlags,butnotthevaluesforthesquaredandcubictermsorforthelegaltype,regionandsectordummyvariables(resultsavailableuponrequest).Exceptfortwovariables(schoolingandhours),allvariablesindicateconsiderablereductionsintheabsolutevalueofbias,from50%to90%,aftermatching.Moreover,mostofthosevariablesalsoresultinthenon-rejectionoftestsofequalityoftheiraveragevalues,betweenthetreatedandthecontrolgroups,aftermatching.12Furthermore,wealsondthattheimpositionofthecommonsupportisnottoorestrictive,asonly22rmsareleftoutfromtheanalysisasaconsequenceofthatconstraint.Thedistributionsofthepropensityscoresacrossthecontrolandtreatmentgroupsarealsoverysimilar,althoughwendtheexpectedgreaterdensityofcontrolobservationsatlowerlevelsofthepropensityscorewhencomparedtotreatmentobservations.Thepseudo-R2'sfromprobitestimationsofthepropensityscoreonallthevariablesbeforeandaftermatching(Sianesi2004)alsoindicatethatthematchedsampleisconsiderablymorehomogeneousafterthematching(thepseudo-R2fallsfrom0.068to0.019).Overall,ourviewisthatthereisstrongevidencethatthematchingisofparticularlygoodstandards.5.2Jobandworker owsAfterestimatingthepropensityscoreandevaluatingthequalityofthematching,wenowassesstheimpactofthenewlawintermsofdierentrmoutcomes.Inthecaseofjob ows,theorydoesnotoerclearpredictions,ashiringsmaybegreaterorsmallerthanseparations.However,worker ows(hiringsand/orseparations)areexpectedtoincrease.Giventhatthecostsofdismissalsforcausefall,someworkersthatexhibitpoorlevelsofperformancebutthathavebeenemployeduntilthenduetotheprotectionoeredbythelawmaybedismissedoncethenewlawisinforce.Atthesametime,employerscanalsobeexpectedtohiremore,assuchmatcheswillnolongerbeasdiculttoterminateasbeforethe1989law.However,itisalsopossiblethatworkersthatexhibitpoorlevelsofperformancechangetheirbehaviourunderthenewcircumstancescreatedbythenewlaw,sothatseparationsdonotnecessarilyincrease.Thispossibilitycouldalsoweakentheexpectedincreaseinhirings.Ontheother 12Thecaseofschoolingisnotparticularlyimportantasthetestofequalityofitsaveragesisstillnotrejected.Theperformanceofhoursislessgood,asthebiasincreasesaftermatchingandtheequalitytestisrejected.However,theeconomicdierenceinthemeansisverysmall.14 hand,theeectsofthelawmaybe`toomicro'tobecapturedatthermlevelanalysisif,forinstance,onlyoneortwoworkersperrmtendtobehaveinsuchawaythattherearegroundsforadismissalforcause.Inordertotestthesedierentpredictions,weconsidertwoversionsofthenetjobcreationvariable,therstincludingtheyearof1990(forwhichonlyrm-leveldataareavailable)andthesecondnotincludingthatyear(i.e.computingnetjobcreationin1991basedonacomparisonbetween1989and1991).Hiringsandseparationsratesalsoignore1990:sincethereisnoworkerdataforthatyear,onecannotdecomposethenetjobcreationratein1990intermsofhiringsandseparations.Forinstance,hiringsin1991correspondtoallworkershiredafterMarch1989thatarestillemployedinthermbyMarch1991.From1992onwards,thetwoversionsofnetjobcreationcoincide(forinstance,netjobcreationin1992correspondstothedierenceinrmsizebetween1992and1991).Inourresults,wendevidencethatsmallrms,thosesubjecttoagreaterdecreaseintheirringcosts,experiencehighernetjobcreationgrowthratesin1990,therstyearaftertheintroductionofthenewlaw(seeTable2).TheATTstatisticforthisyearis0.034(t-ratioof4.3).However,thishighernetjobcreationratewaspartiallyreversedin1991,whentheATTstatisticisequalto-0.025(t-ratioof-2.8).Allremainingt-ratiostendtobelessthan2inabsolutevalue,exceptfor1993and1997,whenthenetjobcreationratesarealsopositive(ATTstatisticsof0.019and0.028,respectively).13ThesecondblockofestimatesinTable2referstothemeasureofnetjobcreationinwhichthecomparisonin1991takes1989asthebaseperiod(not1990asintherstblock).Inotherwords,theinformationavailablefor1990isignoredforthesakeofgreatercomparabilitywiththevariablesforwhichthereisno1990data(e.g.hiringsandseparations).Inthiscase,wendthattherewasnosignicantdierencebetweenthetwotypesofrms(ATT=-0.001;t-ratio=-0.1).Thisresultisnotsurprisingasintheprevioussetofestimatesthevaluesfor1990and1991wereofoppositesignandofsimilarmagnitudes.14Finally,Table2alsoreports 13Weconsiderinthistableallrmsthatarepresentinthedataforatleastoneyearintheperiod1990-1999.Thosermsarealsopresentinallyearsinthe`before'period,1987-1989,byconstructionofoursample.Thefactthatsomermsexitthemarketwhile,byconstruction,nonewrmsareaddedintooursampleexplainswhythenumberofrmsfalls,inourcasefromapproximately3,900to2,900(treatmentgroup)orfrom1,200to900(controlgroup).Moreover,each owrateisconstructedasacomparisonbetweenadjacentyears(e.g.of1993withrespectto1992)inthe`after'periodand1989withrespectto1988(forthe`before'period),notasthe owfrom1989toeachyearfrom1990onwards(e.g.thesumofthe owfrom1989to1993),whileno owsareregisteredforrmsthatleavethedata.14Theresultsfortheremainingyears(1992-1999)underthesecondtypeofnetjobcreationvariablearenotreported,astheyare,byconstruction,preciselythesameasthosefortherstnetjobcreationvariable.15 resultsforthehiringsandseparationsrates.Acrossthenine-yearperiodcoveredandforeachoneofthetwoworker owvariables,wendalmostnoevidenceofsignicantdierencesbetweenthetwotypesofrms.Theonlysignicantestimateindicateslowerseparationsratesforsmallerrmsin1993(ATT=-0.019;t-ratio=-2.2).Resultsfortheremaining owvariables(jobreallocation,workerreallocationandchurning)arenotreportedastheyarealsoinsignicant.Overall,ourconclusionfromtheresultsdisplayedinTable2isthatthereisonlysomemoderateevidenceofaslightincreaseinjobcreationinsmallerrms.Strikingly,thereisnorobustevidenceofincreasedworker ows(hiringsorseparations),unlikewhattheorypredicts.However,therearesomeissuesandcaveatsthatneedtobetakenintoaccountwheninterpretingthesendings,ontopoftheargumentdescribedaboveaboutthepossibleparticularlymicronatureoftheeects.Onecaveatisthatourdataaremeasuredannually,afrequencythatmaynotbesucientlyhightocapturetheworker- owadjustmentscarriedoutbyrms(Blanchard&Portugal2001).Thisproblemiscompoundedbythefactthatthereisnoinformationonhiringsorseparationsfor1990,althoughonlyifsuchmatchesareterminatedbeforeMarch1991.Anothercaveatisthatchangesinthevaluesinthedataaredriven,inpart,bybothvoluntaryandinvoluntaryseparations-andperhapsanincreaseinthelatter,assuggestedbytheory,coincideswithadeclineoftheformer.Thistrade-ocouldariseifthoseworkersthatbeforethenewlawwouldhaveleftnowbelievetheywillbemoreeasilyredattheprospectivenewjobandprefernottomove.Moreover,asinBaueretal.(forthcoming),ourstudycannotcontroldirectlyfordierencesacrossrmsintheirnumbersoftemporaryworkers.Itcouldbethatsmallerrmstendtoemployagreatershareoftemporaryworkersandarethuslessaectedbythereform.15However,aswecontrolcarefullyforthetenurelevelofeachrms'workforce,webelievethispotentialproblemisnotseriousinourcase.Furthermore,thisexplanationwouldalsobeatoddswiththendingsreportedinsubsection5.4(seebelow).Finally,thereisoneadditionaleectthatmayin uencetheabsenceofanyincreasesinworkerreallocation.Thiseectconcernsthenatureofthereformitself:ifarmexpandsabovethe20-employeethreshold,thermwillthenbecomesubjecttothemorecostlydismissal 15Moreover,someevidencebasedonsubjective,cross-sectiondatasuggeststhatmedium-sizermsaremorenegativelyaectedbyEPLthansmallerrms(Pierre&Scarpetta2006).Takingthisresultatfacevalue,thenanyincreasedlevelsofworkerreallocationinsmallerrmsinducedbytheeight-paragraphdierencemayhavebeenjustcancelledoutbysuchgreatersensitivityoflargerrmswithrespecttoEPL.16 long-termemploymentrelationship,workerswilltendtonditlessadvantageoustoinvestinrm-specicskills,leadingtoadeteriorationinrmperformance.However,ifworkersearnrentsattheircurrentjobs,thenworkereort(andthereforermperformance)mayinsteadincreasewhenringcostsfall.Asitbecomeslessunlikelythatworkerswilllosetheircurrentwagepremiums,ifeortisobservablebyemployers,theneortwillincreaseandrmperformancewillimprove(seealsothediscussioninAutoretal.(2007)).Similarly,improvementsinpersonnelmanagementwarrantedbythegreater exibilityofthenewlawcanalsobeexpectedtoleadtobetterperformance.Firmperformancemayincreasenotonly(ornotnecessarily)becauseworkersexertmoreeortbutalso(orsimply)becauseuncooperativeorevendisruptiveworkersarered.InordertoshedlightonthesecontrastingviewsabouttheroleofEPL,weexaminedierentmeasuresofrmperformancethatcanbeconstructedfromourdata(seethelastthreeblockofestimatesinTable3).Inourrstmeasure,weconsiderthelogarithmoftotalsales(2004prices)astheoutcomevariable,againinaDDMapproach.Wendthatallestimates,acrossthenine-yearperiodconsidered,arepositive,andfourofthemaresignicant.Thosesignicantestimatesarealsoparticularlylarge,rangingbetween8.3pp(1995)and12.8pp(1998),whiletheestimateswitht-ratiosbelow2rangefrom3.7ppand6pp.Moreover,thepatternofresultsisrelativelyunchangedwhenconsideringinsteadthelogarithmofsalesperworkerastheoutcomeofinterest(secondblockofestimates).Nowveoutofthenineestimateshavet-ratiosabove2,indicatingeectsbetween5.2ppand10.5pp.Weconsideroneadditionalmeasureofperformance,`surplusperworker',whichisdenedasthedierencebetweentotalsalesandthewagebillofeachrm.Thewagebilliscomputedbysummingallindividualwages,andmultiplyingthatsumby14(thenumberofmonthsofpayduetoeachworkerperyear,accordingtoPortugueselaw)andby1.2(correspondingtoemployerpayrolltaxesofapproximately20%).Wethentakeasourdependentvariablethelogarithmoftheratioofthesurplusbythenumberofworkers.Theresults(seethebottomofTable3)areagainconsistentwiththepreviousndings,indicatingthattreatedrmsunderwentapositiverelativeincreaseintheirperformance.Fouroutofthenineyearsproducelargeandsignicantincreasesinsurplus,whiletheremainingestimatesarealsopositiveand,inmostcases,marginallysignicant.1994isayearwhichneverexhibitsasignicantdierenceacrossthethreesetsofestimatesbut,asmentionedabove,thismaybe19 andthendisappearfromthesample(possiblybecausetheybecomebankrupt),consequentlybeingignoredintheresults.Byconsideringnowonlyrmsthatarealwayspresentinourdataset,weruleoutsuchcensoringbias.WepresentthendingsforthismorerestrictivesampleselectioninTable4(jobandworker ows)andTable5(wagesandrmperformance).First,weobservethatthenumberofrmsdeclinesfrombetween5114in1990and3816in1999,whenconsideringallrm-years,to1671,whenonlyconsideringrmspresentinallyears,from1987to1999,indicatingthatthesampleofrmsusedforeachyearbeforecanvaryreasonably.Second,andmoreimportant,theresultsindicatethatourbenchmarkndingsarerobust.Therearenoqualitativedierencesintermsofjoborworker ows(includingthe1990spike,whichisstilldiscernible)whencomparedtothebenchmarkresultsofTables2and3.Moreover,wagesalsodecreasesignicantlyandrmperformancealsoundergoesasignicantincrease,atverysimilarmagnitudes,althoughinthelattercasetheprecisionoftheestimatesfalls.Whilethedecreaseinprecisionmayinpartindicatethatrmselectionmatters,webelieveamoreimportantreasonisthesmallersamplesize.Inanycase,itisimportanttounderlinethatthequalitativeresultsareunchangedandthatseveralpointestimatesremainstatisticallysignicant.Inoursecondrobustnesstest,wereestimatethemainrm-levelresultsbutusingadierentmatchingalgorithm.Insteadofkernelmatching,wenowconsidernearestneighbourmatching(venearestmatches).Theresults(jobandworker owsinTable6andwagesandrmperformanceinTable7)areagainverysimilartothoseobtainedinthemainestimates,bothinqualitativeandinquantitativeterms.Asbefore,wendinsignicantdierencesin ows(exceptfortheyearof1990),asignicantdecreaseinwagesandasignicantincreaseinourdierentproxiesofrmperformance.Ourthirdrobustnessanalysisisbasedonconsideringan`articial'rmsizethresholdabovetheonedeterminedbythelaw(20employees).Theconcernthisstrategyseekstoaddressisthatthendingsreportedabovemaybedrivenbydierencesinrmsizeonly,insuchwaythattheyarenotrelatedtoanyrealimpactofthelaw.Forinstance,althoughmostresearchdocumentsaverystrongpositiverelationshipbetweenrmsizeandwagelevels(seeOi&Idson(1999)forasurvey),itisalsopossiblethatwagegrowthishigheratbiggerrms.Suchpossibilitycouldexplainourndingsaboutwages,astheyarebasedoncontrastingwagegrowthovertimebetweentwogroupsofrmswofdierentsizes.21 Weimplementthisfalsicationexercisebycreatinganewdataset,includingonlyrmsthatemploy,intheperiod1987-1989,eitherbetween20-29workers(thenew`treatment'group)or31-40workers(thenew`control'group),i.e.assumingthatthesizethresholdin-dicatedbythelawwas30.Intheresultsbasedonthisnewdataset,wend,asbefore,thatthereisaspikeinjobcreationin1990,althoughtherearenosignicantdierenceswhenconsideringonlythedierencebetween1991and1989(seeTable8).Theresultsforhiringsandseparationsarealsoverysimilartothoseobtainedwhenconsideringthecorrectrm-sizethreshold(20).Theseresultsthereforesupportourpreviousconclusionthattherearenosignicantdierencesinducedbythelawregardingemploymentandjobandworker ows(andthatthe1990spikeisspurious).However,whenexaminingtheeectsintermsofwagesandrmperformance(seeTable9),wenowfailtoobserveanysignicantdierencesbetween`small'and`large'rms.Thiscontrastsdramaticallywiththendingsbasedonthethresholdestablishedinthelaw(20employees),fromwhichwedocumentedsignicantdecreasesinwagesandsignicantincreasesinrmperformance.Thedierencesbetweenthetwosetsofresultsareparticularlyclearinthecaseofrmperformance,inwhichmanypointestimatesareevennegativewhenconsideringthearticialthreshold.ThecontrastbetweenthetwosetsofndingsthereforelendsadditionalcredibilitytoourresultsinSection5andtheirinterpretationascausaleectsofthelawreform.Finally,weaddressthepossibilitythatbyMarch1989(thereferenceperiodinourdata)somermswerealreadyrespondingtothereform,implyingthat1989cannotbeconsideredasthelastyearofthe`before'period.Asmentionedbefore,wedonotbelievethistobethecase,astherewasstillgreatuncertaintyatthetimeconcerningthespeciccontentoftheemploymentlaw,nottomentionifanyreformwouldgoaheadatall.Inanycase,inordertoexaminethispossibility,weconsider1988asthelastyearofthe`before'periodandkeep1990and1991astherstyearsofthe`after'period.Againweobtainthesamequalitativeresultsasinthebenchmarkcase,althoughtheresultsaboutrmperformancetendtobeslightlylesssignicantintherstyears(resultsavailableuponrequest).18 18Wehavealsoreplicatedourmainanalysisbuttrimmingtherangeofrmsizescovered,inthespiritofaregression-discontinuityapproach(Hahnetal.2001).Whenconsideringonlyrmsemployingbetween15and19workersin1987-1989andrmsemployingbetween21and25workersinthesameperiod,wendthesamequalitativeresultsasthosepresentedinSection5.However,asthesamplesizebecomesconsiderablysmaller(lessthan700treatedrmsandlessthan400controlrms)and,inrelativeterms,thescopeformisclassicationincreases(e.g.rmsemploying19workersinMarch1989maybeemploying21workersinJune),theestimatestendtobesomewhatlesssignicantthaninthebenchmarkresults(resultsavailableuponrequest).22 isalsolikelytotransferbargainingpowerfromemployeestotheiremployers,whichwouldexplainthedeclineinwagesfoundinourresults.ReferencesAcemoglu,D.&Angrist,J.D.(2001),`Consequencesofemploymentprotection?ThecaseoftheAmericanswithDisabilitiesAct',JournalofPoliticalEconomy109(5),915{957.Ashenfelter,O.(1978),`Estimatingtheeectoftrainingprogramsonearnings',ReviewofEconomicsandStatistics60(1),47{57.Autor,D.H.(2003),`Outsourcingatwill:Thecontributionofunjustdismissaldoctrinetothegrowthofemploymentoutsourcing',JournalofLaborEconomics21(1),1{42.Autor,D.H.,Donohue,J.J.&Schwab,S.J.(2006),`Thecostsofwrongful-dischargelaws',ReviewofEconomicsandStatistics88(2),211{231.Autor,D.,Kugler,A.&Kerr,W.(2007),`Doemploymentprotectionsreduceproductivity?EvidencefromU.S.states',EconomicJournal117(June),F189{F217.Bauer,T.K.,Bender,S.&Bonin,H.(forthcoming),`Dismissalprotectionandworker owsinsmallestablishments',Economica.Bertola,G.(1990),`Jobsecurity,employmentandwages',EuropeanEconomicReview34(4),851{879.Bertola,G.&Rogerson,R.(1997),`Institutionsandlaborreallocation',EuropeanEconomicReview41(6),1147{1171.Besley,T.&Burgess,R.(2004),`Canlaborregulationhindereconomicperformance?evidencefromindia',QuarterlyJournalofEconomics119(1),91{134.Bewley,T.(1999),WhyWagesDon'tFallDuringaRecession?,HarvardUniversityPress,Cambridge.Blanchard,O.&Portugal,P.(2001),`Whathidesbehindanunemploymentrate:ComparingPortugueseandU.S.labormarkets',AmericanEconomicReview91(1),187{207.Blanchard,O.&Tirole,J.(2003),Contoursofemploymentprotectionreform,MITWP03-35.24 Kugler,A.&Pica,G.(forthcoming),`Eectsofemploymentprotectiononworkerandjob ows:Evidencefromthe1990Italianreform',LabourEconomics.Kugler,A.D.&Saint-Paul,G.(2004),`Howdoringcostsaectworker owsinaworldwithadverseselection?',JournalofLaborEconomics22(3),553{584.Lazear,E.P.(1990),`Jobsecurityprovisionsandemployment',QuarterlyJournalofEco-nomics105(3),699{726.Leonardi,M.&Pica,G.(2007),Employmentprotectionlegislationandwages,IZADP2680.Leuven,E.&Sianesi,B.(2004),`PSMATCH2:StatamoduletoperformfullMahalanobisandpropensityscorematching,commonsupportgraphing,andcovariateimbalancetesting',StatisticalSoftwareComponentss432001.Marinescu,I.(2007),Shorteningthetenureclock:TheimpactofstrengthenedU.K.jobsecuritylegislation,UniversityofChicago,Mimeo.Martins,P.(2007),Dismissalsforcause:Anindividual-levelanalysisofdisplacementandwageeects,QueenMary,UniversityofLondon,Mimeo.Meyer,B.D.(1995),`Naturalandquasi-experimentsineconomics',JournalofBusiness&EconomicStatistics13(2),151{162.OECD(2004),Employmentprotectionregulationandlabourmarketperformance,in`Em-ploymentOutlook2004',OECD,Paris.Oi,W.Y.&Idson,T.L.(1999),Firmsizeandwages,inO.Ashenfelter&D.Card,eds,`HandbookofLaborEconomics',Vol.3ofHandbookofLaborEconomics,Elsevier,chap-ter33,pp.2165{2214.Oyer,P.&Schaefer,S.(2000),`Layosandlitigation',RANDJournalofEconomics31(2),345{358.Pierre,G.&Scarpetta,S.(2006),`Employmentprotection:Dorms'perceptionsmatchwithlegislation?',EconomicsLetters90(3),328{334.Rosenbaum,P.&Rubin,D.(1983),`Thecentralroleofthepropensityscoreinobservationalstudiesforcausaleects',Biometrika70(1),41{55.26 Table1:Analysisofbalancingproperties(rm-levelanalysis) Mean%reductt-testVariableSampleTreatedControl%biasbiastp�[t] SchoolingUnmatched5.61685.6442-1.5-0.480.633Matched5.61625.6674-2.9-87.2-1.30.193ExperienceUnmatched23.4624.104-9.5-2.90.004Matched23.44323.4160.495.80.170.866TenureUnmatched7.53018.5899-23.4-7.290Matched7.54247.37353.784.11.740.082FemaleUnmatched0.339580.34491-1.8-0.560.577Matched0.339590.337180.854.80.370.71PayUnmatched2.7192.9566-16.7-5.370Matched2.71082.781-4.970.5-2.480.013HoursUnmatched168.98168.750.90.290.769Matched169.01166.719.5-911.14.130JoblevelUnmatched5.72595.650211.23.350.001Matched5.7285.7405-1.983.4-0.810.42ForeignUnmatched0.009370.01178-2.4-0.760.445Matched0.009410.008470.961.10.460.645FirmageUnmatched1970.51965.919.57.040Matched1970.51970.7-0.995.3-0.580.56LagschoolingUnmatched5.58615.57120.80.250.801Matched5.58725.6554-3.7-357.8-1.670.096LagexperienceUnmatched23.25723.914-9.8-3.010.003Matched23.23422.8485.841.22.60.009LagtenureUnmatched7.30988.3859-23.7-7.410Matched7.31987.12964.282.31.950.051LagfemaleUnmatched0.334370.33799-1.2-0.380.706Matched0.334160.333280.375.70.140.893LagpayUnmatched2.63142.8861-17.9-6.040Matched2.63232.7078-5.370.4-2.760.006LaghoursUnmatched170.94170.760.70.230.816Matched170.95168.878.4-10273.620LagjoblevelUnmatched5.74035.657123.610Matched5.74165.7577-2.380.6-1.020.308 Notes:Source:Author'scalculationsbasedonQuadrosdePessoal.TreatmentandControlindicatesthenumberofrmsinthetreatmentandcontrolgroupsrespectively(rmswith10to19employeesin1989areinthetreatmentgroup;rmswith21to30employeesin1989areinthecontrolgroup).Allvariablesarerm-levelaveragesbasedonthecharacteristicsofrmsin1989(lagscorrespondto1988information).28 Table3:Eectsonwagesandrmperformance(rm-levelanalysis);Matchingmethod:kernel VariableYearATTt(ATT)TreatedControl Wages1991-0.016-2.0399112111992-0.022-2.2401011981993-0.024-2.4396611991994-0.015-1.4398012071995-0.011-1.1381311521996-0.024-2.1363310801997-0.035-2.9356710791998-0.033-2.9339910221999-0.012-1.033601001Totalsales19910.0451.7333698719920.0461.7332098119930.0902.8327298019940.0371.1313694519950.0832.3295387319960.0571.5283283919970.0601.4272481219980.1282.9267379019990.1072.52635791Salesperworker19910.0461.8333698719920.0522.0332098119930.0622.1327298019940.0020.1313694519950.0652.0295387319960.0431.4283283919970.0250.7272481219980.1052.9267379019990.0872.52635791Surplusperworker19910.0692.2322295719920.0852.6319895919930.0571.7312893319940.0130.3301291019950.0681.8285084519960.0541.4272180619970.0330.8262978219980.1673.8257975719990.1503.42546770 Notes:Source:Author'scalculationsbasedonQuadrosdePessoal.ATTreferstotheaveragetreatmentonthetreatedintermsoftheoutcomevariablesconsid-eredandattheyearunderanalysis.t(ATT)denotest-ratiosbasedonanalyticalstandarderrors.Theoutcomevariableismeasuredbythedierencebetweenthevalueofthevariableintheyearunderanalysisandthebaseyear,1989.Seemaintextfortheformaldenitionofeachvariable.TreatmentandControlindicatesthenumberofrmsinthetreatmentandcontrolgroupsrespectively(rmswith10to19employeesin1989areinthetreatmentgroup;rmswith21to30employeesin1989areinthecontrolgroup).30 Table5:Robustness-Eectsonwagesandrmperformance(rm-levelanalysis;onlycontinuingrms);Matchingmethod:kernel VariableYearATTt(ATT)TreatedControl Wages1991-0.013-0.912454191992-0.040-2.012484191993-0.024-1.312474181994-0.023-1.312454191995-0.036-2.012444141996-0.050-2.312394131997-0.039-1.812434161998-0.073-3.812414181999-0.064-3.51241419Totalsales19910.0871.9110236819920.0671.5109837019930.0761.5110836919940.0581.2110037219950.0571.1109937219960.0571.1109936719970.0821.4109436319980.0841.4109736719990.1171.71034355Salesperworker19910.0791.8110236819920.0902.0109837019930.0691.4110836919940.0491.0110037219950.0491.0109937219960.0761.5109936719970.0951.8109436319980.1112.1109736719990.1132.01034355Surplusperworker19910.0971.8106836019920.0991.7105636319930.0731.3105635519940.0631.1105436419950.0661.1105336819960.1071.9105535619970.1001.5105335519980.1612.4105335919990.1302.0996348 Notes:Source:Author'scalculationsbasedonQuadrosdePessoal.ATTreferstotheaveragetreatmentonthetreatedintermsoftheoutcomevariablesconsid-eredandattheyearunderanalysis.t(ATT)denotest-ratiosbasedonanalyticalstandarderrors.Theoutcomevariableismeasuredbythedierencebetweenthevalueofthevariableintheyearunderanalysisandthebaseyear,1989.Seemaintextfortheformaldenitionofeachvariable.TreatmentandControlindicatesthenumberofrmsinthetreatmentandcontrolgroupsrespectively(rmswith10to19employeesin1989areinthetreatmentgroup;rmswith21to30employeesin1989areinthecontrolgroup).32 Table7:Robustness-Eectsonwagesandrmperformance(rm-levelanalysis);Matchingmethod:venearestneigh-bours VariableYearATTt(ATT)TreatedControl Wages1991-0.021-2.3387011831992-0.028-2.5389711701993-0.029-2.6385411701994-0.021-1.7387211781995-0.014-1.2370411241996-0.032-2.4353010541997-0.048-3.5346410521998-0.040-3.132979951999-0.018-1.43261975Totalsales19910.0541.7325996719920.0431.3324296419930.0942.5319796219940.0451.2306292819950.0852.0287785419960.0851.9276182019970.0851.7265879319980.1392.8261277419990.1242.62574778Salesperworker19910.0491.6325996719920.0401.3324296419930.0511.5319796219940.0050.1306292819950.0611.6287785419960.0641.8276182019970.0340.9265879319980.1152.8261277419990.0982.52574778Surplusperworker19910.0591.7315393719920.0721.9312594219930.0511.3305891619940.0160.4294489319950.0671.5278082819960.0791.8265478719970.0420.9257076419980.1713.4252374419990.1713.42491757 Notes:Source:Author'scalculationsbasedonQuadrosdePessoal.ATTreferstotheaveragetreatmentonthetreatedintermsoftheoutcomevariablesconsid-eredandattheyearunderanalysis.t(ATT)denotest-ratiosbasedonanalyticalstandarderrors.Theoutcomevariableismeasuredbythedierencebetweenthevalueofthevariableintheyearunderanalysisandthebaseyear,1989.Seemaintextfortheformaldenitionofeachvariable.TreatmentandControlindicatesthenumberofrmsinthetreatmentandcontrolgroupsrespectively(rmswith10to19employeesin1989areinthetreatmentgroup;rmswith21to30employeesin1989areinthecontrolgroup).34 Table9:Robustness-Eectsonwagesandrmperformance-Articialthreshold(rm-levelanalysis,20-40employeesin1989);Matchingmethod:kernel VariableYearATTt(ATT)TreatedControl Wages1991-0.008-0.712124941992-0.005-0.411984891993-0.027-1.911924821994-0.048-3.112134991995-0.021-1.511514791996-0.013-0.810904491997-0.019-1.210774501998-0.018-1.110194341999-0.026-1.5999422Totalsales19910.0150.4100641819920.0080.29984171993-0.090-2.199941719940.0140.396840719950.0280.689737919960.0250.585936819970.0390.682736219980.0140.28073451999-0.031-0.5806352Salesperworker19910.0060.210064181992-0.012-0.39984171993-0.087-2.29994171994-0.016-0.49684071995-0.018-0.48973791996-0.021-0.585936819970.0020.08273621998-0.043-0.88073451999-0.077-1.6806352Surplusperworker19910.0401.09764121992-0.020-0.59744061993-0.113-2.59523981994-0.059-1.29293911995-0.054-1.18693641996-0.031-0.58273581997-0.016-0.37943461998-0.018-0.37773371999-0.151-2.6783342 Notes:Source:Author'scalculationsbasedonQuadrosdePessoal.ATTreferstotheaveragetreatmentonthetreatedintermsoftheoutcomevariablesconsid-eredandattheyearunderanalysis.t(ATT)denotest-ratiosbasedonanalyticalstandarderrors.Theoutcomevariableismeasuredbythedierencebetweenthevalueofthevariableintheyearunderanalysisandthebaseyear,1989.Seemaintextfortheformaldenitionofeachvariable.TreatmentandControlindicatesthenumberofrmsinthetreatmentandcontrolgroupsrespectively(rmswith10to19employeesin1989areinthetreatmentgroup;rmswith21to30employeesin1989areinthecontrolgroup).36