how decentralizing school finance may narrow … · joonho lee and bruce fuller june 2017 how...
TRANSCRIPT
UNIVERSITYOFCALIFORNIA,BERKELEYInstituteofHumanDevelopmentWorkingPaperJoonHoLeeandBruceFullerJune2017
HOW DECENTRALIZING SCHOOL FINANCE MAY NARROW ACHIEVEMENT GAPS – UnevenProgressinCaliforniaafter$41Billion
DecentralizingFinance,OrganizationalChangeandAchievement–2
Summary
Severalgovernorshavemovedtodistributeschooldollarsbasedonpupilweights,recognizingthatdisadvantagedstudentsrequirefreshresourcestoreachstateproficiencybars.Thispolicystrategy,afterwinningbroadsupportinCalifornia,balancesadequatedollarsforallstudentswiththeadditionalcostofhoistingchildrenheldbackbyfamilypovertyorlimitedEnglish.California’srenditionofweighted-pupilfunding,approvedin2013,providesconcentrationgrantstodistrictsinwhichatleast55percentofenrollmentconsistsofdisadvantagedchildren.Localdistrictsenjoydiscretionoverwhetherthenewfundsgeneratedbythesechildren–equaling$41billiontodatewhensupplementalgrantsareincluded–gotoschoolsthatservethem.WefindthatschoolsindistrictsreceivingconcentrationgrantsduringtheinitialtwoyearsofLocalControlFundingdidengageinorganizationalchangethatparallelsgainsinpupilachievement,comparedwithschoolsinalmostidenticaldistrictsnotreceivingconcentrationgrants.ThesebenefitswerelargelyexperiencedbyLatinostudentsandnotbyothergroupsatsignificantlevels.Weexploitaregression-discontinuitydesign,capturingthislargenaturalexperimentandallowingforstrongercausalinferences.
SpecialthankstoSophieFanelli,MaryLouFulton,andPeterRiverawhohavegenerouslysupportedourworkthroughtheCaliforniaCommunityFoundation,CaliforniaEndowment,andtheStuartFoundation.GregDuncan,BettyMalen,LarryPicus,andJesseRothsteinofferedwonderfullyhelpfulcommentsonearlierdrafts.RyanAndersonandEdgarCabralansweredmanyquestionabouttheinnerworkingsofthefundingformula.SophiaRabe-Hesketh,ourgenerouscolleagueatBerkeley,contributedtothisanalysis.EarlierpiecesofthispaperwerepresentedinBerkeley,Sacramento,andSanAntonioattheAmericanEducationalResearchAssociation.ThanksgotoDavidPlankandPolicyAnalysisforCaliforniaEducationforfacilitatingdiscussionswithpolicythinkers.
HowDecentralizingSchoolFinanceMayNarrowAchievementGaps–UnevenProgressinCaliforniaafter$41Billion
DecentralizingFinance,OrganizationalChangeandAchievement–3
HowDecentralizingSchoolFinanceMayNarrowAchievementGaps–UnevenProgressinCaliforniaafter$41Billion
Statepolicymakershaveturnedtoweightedpupilformulae,goingbackfourdecades,toadvancesufficientandequitablefundingforpublicschools.Manystateshadmovedawayfromrelianceonpropertytaxtofinanceschoolsbythe1980sinthewakeofCalifornia’sSerranodecisionandsimilarcases.Thislitigationhadrevealedgrossdisparitiesinperpupilspendingamongdistricts.Toequalizestatesupportbetweenrichandpoordistricts,ortobetterservespecificgroups,moststatescentralizedfunding,thenregulatedabevyof“categoricalaid”programsfromtheirstatecapital.Thisreformagendadramaticallynarroweddisparitiesinperpupilspendingamonglocaldistricts,whethertheyenjoyedampletaxbasesornot(Wirt&Kirst,1972;Odden&Picus,2014).
Butasgovernorsandlegislaturesputinplaceaggressiveschoolaccountabilityregimes
throughthe1990s–expectingallstudentstoreachrigorousproficiencystandards–thedifferentialcostofliftinglow-achievingstudentscameintofocus.Severedeclinesinschoolspending,especiallyinthewakeoftheGreatRecession,alsopromptedtheadditionalquestionofwhetherlocalschoolswereadequatelyfunded,andthenwhetherallstudentsenjoyedawell-resourcedopportunitytolearn
Weighted-pupilfunding(WPF)hasgainedappealamongavarietyofpolicyactivists,under
whichstudentsfromdisadvantagedbackgroundsareassignedalargerweightinthestatefundingformula.Thisresultsinadditionaldollarsfordistricts(notnecessarilyschools)thatservehigherconcentrationsoftheseyoungsters.Basegrantstodistricts,whereallpupilsareweightedequally,alongwithsupplementalgrantsweightedforpoorstudents,servetobalancethetandempolicyaimsofadequacyandequity(Augenblick,Myers,&Anderson,1997;Leppert&Routh,1980).SeveralstateshadcreatedWPFfinancestructuresbythe1990s,includingFlorida,Oklahoma,andTexas(King,1983;Berne&Stiefel,1979).
Long-termdatasuggestthatpro-equityfinancereformsoverallhavecontributedtogainsin
studentachievement,evenliftingtheperformanceofpoorchildren(Jackson,Johnson,&Persico,2015;Lafortune,Rothstein,&Schanzenbach,2016).Theefficiencywithwhichequityreformshaveraisedachievementornarroweddisparitiesremainsakeyquestion(Hanushek&Woessmann,2017).NordoweunderstandwhetherWPFstrategies–bankingondecentralizedgovernance–leadtoorganizationalchangesinsideschoolsthatstrengthenpupilengagementandlearning.AdvocatesoftenremainsatisfiedwiththeredistributiveeffectsofWPFreforms,whethertheyalterschoolsinwaysthatelevatelearning,ornot.
WefocusononeelementofCalifornia’srenditionofWPFreform,whatGov.JerryBrown
cametocalltheLocalControlFunding(LCF)Formula.Approvedinthesummerof2013,itrecasthowlocaldistrictsweretobefunded,kicking-injusttwomonthslater.Stateaidtiedtoweightedpupils–thosefrompoorornon-Englishspeakingfamilies,andthoseinfostercare–grewby$41billionduringtheinitialfouryearsofimplementation(California,2016).Thispaper
DecentralizingFinance,OrganizationalChangeandAchievement–4
estimatestheextenttowhichoneportionofthefundingapparatus–about$3.3billioninannualconcentrationsgrants–areassociatedwithschool-levelchangesthatmayshapestudentlearning,alongwithactualdifferencesinpupiltestscores,comparedwithalmostidenticaldistrictsthatdidnotenjoythisextrafunding.
UndertheCaliforniareform,thestateawardsconcentrationgrantstolocaldistrictswhen
morethan55%oftheirenrollmentsconsistofpupilsintheweightedcategories(poor,Englishlearners,orkidsinfostercare).Weexploitthisdiscretecutpointtoestimatechangesinschoolorganizationsand,inturn,pupilachievementlevelsindistrictsjustabovethe55%thresholdandthosejustbelowbutotherwisealmostidenticalintheirpupilcomposition,whichcouldconfoundthediscreteinfluenceofreceivingthisextrafunding.Notethattheadditionalfundingbecomessubstantialonlyasadistrict’senrollmentreflectsmanymoredisadvantagedstudentsabovethe55%line.
Inshort,wefindthatdistrictsreceivingconcentrationgrants,duringtheinitialtwoyearsof
LCFimplementationassignedsignificantlyfewerinstructionalperiodstoteachersandprovidedmoreadvancedcoursesatthehighschoollevel,comparedwithdistrictsjustbelowthecutoffforreceivingthesegrants.Thosewinningtheextrafundingdidnotmovetowardsmallerclassesonaverage.Studentperformanceofpupils,grades3-8,alsorangedhigherwhenattendingschoolsindistrictsthatreceivedconcentrationgrants.ThesebenefitswereexperiencedbyLatinostudents,butnotbyothergroupsatstatisticallysignificantlevels.Futureworkwillformallyassesstheextenttowhichtheseandotherschool-levelorganizationalchangesmediateobservedgainsinachievement.
AHazyTheoryofOrganizationalChange
WhytheRiseofWeighted-PupilFunding?
AffectionforWPFstrategiesstemsfromtheconfluenceofthreepolicydevelopments.First,there’stherecognitionbypolicymakersthat–aftersettingdemandinglearningstandardsorproficiencybars–itwillcostmoretoliftmanypoorchildrenoverthesehurdles.Childrenmaybedisadvantagedbyfamilypoverty,language,orunstablehomessituations.Growingupinneighborhoodswithconcentratedpovertymayconstrainthecapacityofchildrentoengageandgrowwithinclassrooms(Bryk,Sebring,Kerbow,Allensworth,&Easton,2010).“Becausenotallstudentscometoschoolwiththesameindividual,family,orneighborhoodadvantages,someneedmoreresourcesthanotherstomeetagivenachievementstandard,”asarguedbyinitialdesignersofCalifornia’sLCFreform(Bersin,Kirst,&Liu,2008:5).
Second,theaccretionofcentrallyregulatedcategoricalaidfellintodisfavorbythelate
1990s.Californiaofficials,forinstance,monitoredover65separatefundingstreams,mostrequiringbureaucraticoversightatstateanddistrictlevels.Onesurveyofschoolprincipalsdetailedthesubstantialtimespentsimplycompletingbudgetsandforms,tryingtokeeppacewithdisparateregulationstiedtomultiplefundingstreams(Fuller,Loeb,Arshan,Chen,&Yi,2007).InCalifornia,Gov.Schwarzeneggerpushedthroughlegislationin2009toconsolidate
DecentralizingFinance,OrganizationalChangeandAchievement–5
manyofthestate’scategoricalprograms,amovewelcomedbylocalschoolboardsandunionleaders(placingmoredollarsonthebargainingtable),whileworryingequityadvocateswhofeareddiminishedaccountabilityoverhowtargetedfundswouldbespentlocally(Fuller,Marsh,Stetcher,&Timar,2011).Theoriginalargumentforstate-runcategoricalaidwasthatdollarsforpoorchildrenmightotherwisegotoschoolssituatedinpoliticallystrongercommunities.
Theconsolidationofcategoricalaidintoblockgrantsmirrorsathirdpolicyshiftthat’s
gainedmomentumoverthepasthalf-century:delegatinggreaterauthoritytoschoolprincipalsoverbudgetandpersonnel.Risingdistrustofeducationbureaucracies,alongwithfaithinschool-levelcontrol,hadalreadybeenexploitedbycharterschooladvocates,includingearlyadvocatesofneighborhoodcontrol,largelyonthepoliticalLeft(Finn,Manno,&Wright,2016;Fuller,2015).TwoCaliforniadistricts,OaklandandSanFrancisco,hademployedweighted-pupilformulaetomoreprogressivelydistributedollarstoschools–onefacetofdecentralizedfiscalcontrolouttoprincipals.Resultsintermsofalteringschoolstaffingorquality,however,provedmixed(Chambers,Shambaugh,Levin,Muraki,&Poland,2008).
Still,thedecentralizingofschoolfinanceandbudgetdecision-makingdowntoschool
districtsfeelsconsonantwiththeshifttowardschool-sitemanagement.Atthesametime,designersofCalifornia’sWPFreformappliedstudentweightsonlytoallocationsfromSacramentoouttothestate’snearlyonethousanddistricts.NothinginGov.Brown’sinitiative–beyondhisrhetoricandthereform’spro-equityspirit–requiresdistrictstoallocatenewdollarstoschoolsthatservehigherconcentrationsofdisadvantagedstudents.Legalactionshavebeenfiledinthreemajorurbandistricts,claimingthatnewdollarshavebeendivertedfromtheintendedstudents.
WhenDoesDecentralizedFinanceTouchSchools?
WPFfinancingunfoldshighabovetheinnerworkingsofschools.Thisfundingstructuredoesofferasimplewayofgrantingresourcestodistrict-levelleaders,theactorsthatgaindiscretionunderCalifornia’sLCFinitiative,notnecessarilyprincipalsorschool-levelleaders.OnesageactivistpunctuatesthispivotalfacetofCalifornia’sreform,askingwhetheritoperatesasadumptruckorabackpack.Thatis,doesthestatedriveuptoaschooldistrictofficeanddumpnewdollarsontotheloadingdock,noquestionsasked,ordidpolicymakersbelievetheywerestrappingfundsontothebacksofdisadvantagedstudents,withdollarsflowingtotheirparticularschools?1
NothinginCalifornialaworregulationsunderLCFrequiresthebackpack.Schoolboardsdo
notneedtotakeintoaccountpupilweightswhendistributingdollarsamongschoolswithintheirdistrict.Somefeelethicallyobligedtodoso,butthat’snottheletterofthelaw.Indeed,between-schooldistributionsshapedbyweightingpupilshaveyieldedmixedresults.Hawaii’sstatewidereformdidresultinlargerallocationstoschoolsthatservedgreatersharesofdisadvantagedstudents(Levinetal.,2013).ButanambitiousWPFexperimentinPrinceGeorge’sCounty,Maryland,unearthedonlyslightredistributionbasedonstudentattributesafterthedistrictmovedtoboostper-pupilspendingamongschools.Thedistrict“unlocked”
DecentralizingFinance,OrganizationalChangeandAchievement–6
onlycertainschool-levelpoststhatotherwiseremainedcentrallycontrolled.Overallprogresstowardequalizingspendingperpupilwasslight(Malen,Dayhoff,Egan,&Croninger,2015).
HawleyMilesandRosa(2006)similarlyfoundthatthedevilliesinparticulardetails,after
studyingWPFprogramsinCincinnatiandHouston:theshareofdistrictbudgetstowhichpupilweightsareapplied,fine-grainelementsoftheformula,andprior,ofteninstitutionalized,waysofdistributingteachingpostsandfungibledollars.Between-schoolallocationsinNewYorkCityunderformermayorMichaelJ.Bloombergdidbecomeweightedtowarddisadvantagedstudents,butonlywhenfederalTitleIandothernon-staterevenueswereincluded.Stateandcitydollarswerenotprogressivelyallocatedandmoreexperiencedteachersweredisproportionatelyassigned(orseniorityruleseasedmigration)toschoolsservingfewerpoorstudents(Rubenstein,Schwartz,Stiefel,&Amor,2007).
California’sLCFlegislationdoesrequirethateachlocaldistrictestablishabaselinelevelof
supportin“supplementalservices”provideddisadvantagedpupil(thatis,priortopassageofLCF).Then,districtsreceivingso-calledsupplementalorconcentrationgrantsmustincreasespendingonweightedstudentsinproportiontotheamountofnewdollarsgeneratedbythesepupils(theproportionalityrequirement;California,2013).Somedistrictsrespondedbyover-estimatingwhattheyhadearlierspentonweightedpupils,byasmuch$450millionperyearinthecaseoftheLosAngelesUnifiedSchoolDistrict(LAUSD),laterchallengedbythestateDepartmentofEducation.Thisallowsthedistricttoallocatefewernewdollarsforservicesthatbenefittheweightedstudentsthanthelevelrequirediftheyhadaccuratelyreportedtheirpre-LCFbaselinelevelofsupport(UnitedWay,2017).
Wealsoarriveatthetheoreticalvagariesofdecentralizedfinancing,namelythelackof
specificityinhowdistrictflexibility,blendedwithextrafundingforcertainstudents,willalterschoolorganizations.UnderWPFregimes,stateofficialsoftenreasserttheirfaithinlocalschoolboards,thecapacityofdistrictleaderstocraftanddiscerneffectiveschoolpractices.ButintheCaliforniacase,thesedistrictofficialsarenotrequiredtomovenewfundingtoschoolsthatservedisadvantagedstudents.Noristhereanyrequirementthatdistrictsreporthardevidenceonwhatschool-levelchangesarepredictiveofachievementgains.(Tobefair,state-runcategorical-aidprogramsoftenmetthefirstcriterionyetfellmiserablyshortonthesecond.)
ThemaldistributionofCalifornia’slargeinfusionofnewfunding–oratleasterratic
channelingofpupil-weighteddollarstotheirschools–maystemfromcollateralfiscalpressuretobackstopemployeebenefitsandpensionliabilities,fixedfacilitycosts,andongoingpressurefrommiddle-classandaffluentparentswithindistrictstoprotecttheirschools.ResearchinsideLAUSDalsorevealsalackofanalyticcapacityinsidethisparticularheadquarterstodeterminewhichschoolsbenefitfromnewstatefunding,areticencetoevenaskthequestion(Partnership,2017;UnitedWay,2017).Whatevertheunderlyingforces,thelackofprogressivedistributionoffundingouttoschoolsservingthekidsthatgeneratethenewmoniesmanifestsoneweaklinkintheWPFtheoryofaction.Ifdistrictsfailtoprogressivelytargetdollarsonhigh-needsschools,howcanweightedfundingnudgeorganizationalchangesinsideschoolsthatareproximaltoteachingandlearning?
DecentralizingFinance,OrganizationalChangeandAchievement–7
TheCaliforniaCase
Thestate’sLCFreformsweptasidescoresofcategoricalaidprogramsinthesummerof
2013,retainingaboutsevenmajorprograms.Italsoreplacedearlierlocalrevenuelimits(putinplacebyProposition13in1978)withanelegantlysimpleWPFprogram.Gov.Brown’slegislationsetforththreefundingtiers.Thebasegrantprovidesequaldollarsperpupil,varyingbygradelevels,inamountssetatabout$6,900perK-6student,$7,200formiddleschoolpupils,and$8,300foreachhighschoolstudent,adjustedforinflationeachyear(California,2013).Supplementalgrantsprovidedistrictsanadditional20%foreachstudentfromalow-incomefamily,designatedEnglishlearners,andthoseinfostercare.Concentrationgrantsfurtherboostper-pupildistributionsbyanamountequalto50%ofthebasegrant,kicking-inforthefirsttargetedstudentafterthedistrictreaches55%ofitstotalenrollmentfallingintooneoftheweightedcategories.2
Atthesametime,districtsonlyreceivethe50%additiontotheirbasegrantforeach
studentenrolledoverthe55%leveloftargetedstudentsinthedistrict.So,adistrictthathypotheticallyenrollsjust10studentsinoneoftheweightedcategoriesabove55%ofallweightedstudentswouldreceiveverylittleadditionalfunding.Thusouranalysisfocusesonwhatanalystscallthe“intent-to-treat”–thatis,alldistrictsoverthe55%cutpointreceivesomelevelofadditionalfunding,nomatterhowlargeorsmalltheiraugmentation.Futureworkwillcomparethepresentresultswiththeestimatedeffectofactualdollaraugmentationstiedtoreceiptofconcentrationgrants.
RisingSchoolSpending,withTargetingonPoorChildren
TheLCFreformwentintoeffectsoonafterbeingsignedbytheGov.Brownforthe2013-14schoolyear.“Wearebringinggovernmentclosertothepeople,totheclassroomwhererealdecisionsaremade,anddirectingthemoneywheretheneedandthechallengeisgreatest,”hesaid(Brown,2013).California’sresurgingeconomyandtheconstitutionallyrequiredallocationforK-12educationwasalreadyspurringnewspendingonpublicschools.California’sset-asideforschoolsequaled$63.6billionby2016-17,about88%flowingtolocaldistrictsthroughtheLCFmechanism,theremaining12%viasurvivingcategoricalprograms(California,2016).
Overallspendingreboundedtojustabovepre-recessionlevelsby2016-17afteradjusting
forinflation,reaching$10,657perpupil(Figure1,California,2016).ThislevelstillplacesCaliforniainthebottomthirdofallstatesnationwide.ThebulkofLCFfundingistiedtothenon-weightedbasegrant,equaling$45.3billioninthesameyear,comparedwith$5.7and$3.3billioninsupplementalandconcentrationgrants,respectively.Still,districtswithlargesharesofdisadvantagedpupilsalreadyenjoythebiggestgainsinstatesupport;thiswillcontinueinto2018-19whenLCFwilllikelybefullyimplemented.Districtswithenrollmentsthatincludemorethan25%weightedstudentswillseeper-pupilspendingrise5%oversixyears,comparedwithaboutaone-thirdgaininspendingfordistrictsinwhich80%ofallstudentsfallintoatleastoneofthedisadvantagedcategories(EdSource,2016).
DecentralizingFinance,OrganizationalChangeandAchievement–8
Figure1.Post-recessionrecoveryinperpupilspending,Californiaschools
TheLCFinitiativeintendstowidencivicparticipationinlocalbudgetdiscussions,aprocessresultinginathree-yearLocalControlAccountabilityPlan(LCAP).Districtleadersmustconsultavarietyofstakeholdersanddeviseablueprintthatspecifiesspendingpatternsthataimtoadvancethestate’seightprioritiesandimproveservicesfordisadvantagedstudents(Wolf&Sands,2016).3Districtsdonothavetoreportonthebetween-schooldistributionofdollars.
DoesLocalControlFundingImproveSchools?
StudiesofLCFimplementationhavereliedmostlyoninterviewswithdistrictofficialsorschoolprincipals.TheGovernor’sOfficeandstateBoardofEducationhaveshownlittleinterestinsupportingrigorousevaluationoftheirmassiveexperiment(Baker,2013).4SincetheinceptionofLCF,twoteamshaveconductedqualitativeresearchinsidedistrictofficestolearnaboutimplementation.Theyfindthatdistrictleadersdoexercisefiscaldiscretionbywideningcivicparticipationinbudgetdiscussions,andavarietyofneweffortsareoftenmountedtobetterengagedisadvantagedstudents(Fuller&Tobben,2014).Districtofficialsexpressconfusionoverwhether,andifsohow,toallocatesupplementalandconcentrationgrants,includinghowtoimprovepedagogyorsupportsforweightedstudents(Wolf&Sands,2016).SomedistrictleadersreportseeingtheLCAP’snudgeforparticipatorybudgetasacomplianceexercise,notonethatspursinventivethinkingorseriousassessmentofwhat’sworkinginsideschoolstoliftachievement.
Concernhasspread,fouryearsintoLCFimplementation,overwhetherdollarsgenerated
throughsupplementalandconcentrationgrantsarebeingdirectedtoschoolsthatservelargerconcentrationsofweightedstudents.Onelargedistrict,FresnoUnified,soughttousesupplementalandconcentrationgrantdollarstohelpfundteachersalaryincreases,evena
DecentralizingFinance,OrganizationalChangeandAchievement–9
gunshottrackingprogrampushedbythecitypolicedepartment.ThisdiversionofLCFfundswasdiscouragedbythestateschoolschieffollowingalegalcomplaintfiledbytheACLU(Fensterwald,2015).Thislocalboardreverseditsearlierdecision.Thestate’schiefeducatorhasalsoruledthatLAUSDoverestimatedhowmuchitspentonweightedstudentspriortothegovernor’sreform,allowingthedistricttounderspecifybyabout$245millioninwhatit’slegallyrequiredtospendannuallyforservicesaimedatdisadvantagedpupils(Kholi,2016).
ResearchQuestionsandAnalyticStrategy
Overall,littleisknownabouttheextenttowhichtheLCFreformaltersthesocial
organizationofschools,especiallyinwaysthatelevatestudentlearningovertime.Figure2advancesasimplecausalsequencebywhichtargetedaid–ifallocatedtoschoolsthatservedisadvantagedstudents–mightspurorganizationalchange.Weassumethatdistrictsvaryintheextenttowhichtheyprogressivelydistributenewdollarstoschoolsthathostgreatersharesoflow-achievingpupils(detailedforLAUSDinUnitedWay,2017).Racialorclassinterestsinsidedistricts,orthelackoftechnicalcapacity,mayacttopreserveinstitutionalizedwaysinwhichteachersanddollarsareallocatedamongschools(Fuller,Marsh,Stetcher,&Timar,2011;HawleyMiles&Rosa,2006;Shipps,2006).
Figure2.Causallogicforhowtargetedfundingmayaffectschoolorganizations,mediatingachievementeffects
Whendistrictleadersdoallocatenewdollarstoschoolswithgreatersharesofweightedpupils,learninggainscanonlyresultiforganizationalchangeorgainsinteacherqualityresultfromthesefreshallocations.Suchorganizationalchangesmustbeproximaltostudentengagementorlearning.GiventhedataavailableinCalifornia,wecanassesswhetherdistrictsreceivingadditionalfundingtendtoreduceaverageclasssize,enrollmorehighschoolpupilsinrigorouscourses,orimproveworkingconditionsbyassigningteacherstofewerinstructionalperiodseachday.What’skeyisthatsuchchangesinthesocialorganizationofschoolingorthequalitiesofteachersmustpresumablybeobservedbeforeachievementgainscanbeexpected.
Wefocusonwhetherorganizationalshiftsorachievementdifferencescanbeobservedin
whendistrictssurpassthe55%thresholdintermsofdisadvantagedpupilenrollments.Wecomparethesewinnersofconcentrationgrantstodistricts,almostidenticalintermsofpupildemographics,butfallingshortofthe55%cutoff.Thisnaturalexperimentallowsforaregression-discontinuitydesign(RDD)inwhichconfoundingfactorsamongdistrictsareheld
Newschoolfundingfordistrictsservinglow-achievingpupils
èSchool-levelstaffingororganizationalchanges(proximaltostudentlearning)
ì
Gainsinpupillearning
(Unobserved)confoundingfactors–familybackgroundandneighborhoods
è
DecentralizingFinance,OrganizationalChangeandAchievement–10
constant.Theonlyfactorthat’schangingindiscretefashionisthe“intent-to-be-treated,”thereceiptofconcentrationgrants.
ThiscaptureshowLCFpolicyreflectsatreatmentinbinaryfashion,beloworabovethe55%
cutpoint.Theactualtreatmentismorecontinuousinthatdistrictsreceivetheadditionfundingforeachpupilenrolledabovethe55%compositionoftargetedstudents(seeAppendix).Ourintent-to-treatapproachyieldsinsightintothenetimpactofallofpossiblechangespotentiallyinducedbyassigningdistrictstothetreatmentgroup.Futureworkandotherresearcherswillexaminethatdiscreteeffectofrisingperpupilspendingabovethe55%cutpoint.
Thestatisticalanalysisspecificallyinformstheseempiricalquestionsfortheconcentration-
grantportionofCalifornia’sLocalControlFunding–RQ1.Howdoschooldistrictscomparedescriptivelywhenfallingoneithersideofthe55%threshold,triggeringconcentrationgrants,basedontheirshareofstudentswhoaredisadvantagedandweighted?5RQ2.Doschoolsdisplayorganizationaldifferencesthatareproximaltostudentlearningwhensituatedindistrictsreceivingconcentrationgrants,comparedwithschoolsinotherwiseidenticaldistrictsthatdonotreceiveconcentrationgrants?WefocusonorganizationalfeaturesthatallschoolsmustreporttothestateDepartmentofEducation,including(a)thenumberofnewlyhiredteachers,(b)countofclassassignmentsforteachers,and(c)thepercentageofhighschoolcoursesmeetingA-Gcollege-entryrequirementsinCalifornia.RQ3.Dostudentsdisplaystrongerachievementonstandardizedtestswhensituatedindistrictsreceivingconcentrationgrants,comparedwithpeersinotherwisesimilardistrictsthatdidnotreceiveconcentrationgrants?Welookatbothendsofthetest-scoredistribution,includingtheshareofstudents,enrolledingrades3-8,fallingbelowstandardfortheirgradelevel,andthoseexceedingthestandard,inmathematicsandEnglishlanguagearts.
Wealsoidentifywhichgroupsofstudentsbenefitmostfromtheinfusionofconcentrationgrants,andwhetherdifferinggainsmayreflectanarrowingofachievementgaps,definedalonglinesofraceorethnicity.Ideally,concentrationgrantsnarrowlearningdisparitiesbetweendisadvantagedandbetter-offchildrenstatewideandamonggroupswithindistricts.
DecentralizingFinance,OrganizationalChangeandAchievement–11
Methods
DataToinformthesequestionswedrawdatafromthreemainsources:the2013-14LCF“fundingsnapshots,”2014-15LCF“stateprioritiessnapshots”,and2014-15staffassignmentandcourse-takingdatasubmittedbydistrictstotheCaliforniaDepartmentofEducation(CDE).Thesedatareportsarenewlyrequiredunderthe2013financelegislation,offeringawealthoffundingandschool-performanceinformation,postedathttp://www.cde.ca.gov/ds.Wemergedthesedatatoprovidecompleteaccountsfor949districtsstatewide,allowingustotestfororganizationalandachievementeffectsobservedamongdistrictthatreceivedconcentrationgrants.
LCFfundingsnapshotdata.ThisincludesLCFfundingallocations,alongwithcountsofdifferingtypesofweightedstudentsfromwhichfundinglevelsarederivedforthe2013-14fiscalyear.Thesedataspecifythepercentageofunduplicatedcountsofweightedstudentsforeachdistrict,thoseclassifiedbydistrictsasEnglishlearners,comingfrompoorfamilies,andthoseinfostercare.
Ourassignment-to-treatmentvariable–thepercentageofadistrict’senrollmentconsisting
ofunduplicatedweightedpupils(UPP)–iscalculatedbydividingeachdistrict’sweightedpupilcountbytotaldistrictenrollment.TheUPPisusedindeterminingeligibilityforaconcentrationgrant,whenexceeding55%,thencalculatingtheconcentrationgrantlevel.Thismakesforaclearassignmentvariableforourregressiondiscontinuitydesign.Figure3showsthedistributionofUPPsharesamongallCaliforniaschooldistrictsinthreerecentyears.
LCFFstateprioritiessnapshotdata.Currentlawrequiresdistrictstoannuallyupdatetheir
LCAP,whichmustincludedataon26elements,includingmeasuresofstudentachievement,studentengagement,andschoolclimate.TheCDEthencompilestheLCAPdatacomingfromdistricts,generatingtheso-calledstateprioritiessnapshotdata.Wedrawmeasuresofstudentachievementforeachdistrictfromthegeneral2014-15CDEfile,utilizingstandardizedtestsadministeredtostudentsinspring,2015.
Thesestudentoutcomes,aggregatedtothedistrictlevel,include(a)thepercentofstudents
whopassedanAdvancedPlacement(AP)Examwithascoreof3orhigher,(b)thepercentofstudents,grades3to8,whoscoredStandardExceededinthestate’sSmarterBalancedexamsinmathematicsandEnglish-languagearts(ELA),and(c)thepercentofthirdtoeighth-gradepupilswhoscoredStandardNotMet.
AlthoughthetestresultsforELAandmathematicsarecategorizedbyfourlevelsof
performance(StandardNotMet,StandardNearlyMet,StandardMet,andStandardExceeded),wefocusonthehighestandthelowestachievementranks,testinghowconcentrationgrantsmayberelatedtopupilperformanceatthelowandhighends,lendinginsightintochangeinachievementdisparities.Giventhatspring2015wasthefirstyearofSmarterBalanced
DecentralizingFinance,OrganizationalChangeandAchievement–12
assessments,wegaugepossibleachievementeffectstwoschoolyearsaftertheinitialconcentrationgrantswereawardedtoeligibledistricts.
Figure3.Thedistributionofdistricts’unduplicatedpercentageofweightedstudentsstatewide
Staffassignmentandcoursedata.Tomeasureorganizationalchangewithinschools(aggregatedtothedistrict),weexploitdataonstaffingandcourse-takingcompiledbyCDE.Thesedatasetincludeeachcoursetakenbyhighschoolstudents,includingcoursesmeetingcollegeadmissionrequirementsoftheUniversityofCalifornia,knownastheA-to-Gcoursesequence,alongwithinformationoncompletionofAPcourses.StudentscompletingA-GorAPcoursesaregenerallyviewedastravelingamorerigorouscurriculumpath.
Theschoolstaffingandcourse-takingdataallowedustogeneratemeasuresof
organizationalfeaturesthatmayfosterhigherachievement.Thesemeasuresinclude(a)thenumberofnewlyhiredfirstyearteachersinthedistrict,(b)theaveragenumberofclassperiodsassignedtoteachers,onedimensionofworkingconditions,and(c)thepercentofcoursesmeetingA-Grequirements.ThestaffassignmentandcoursedatawerecollectedinOctober2014,whichfellbetweenthedistributionofLCFfunding(August2013)andtheadministrationofstateachievementtests(April2015).Table1showsdescriptivestatisticsfortheassignmentvariableandtheoutcomevariablesdiscussedabove.
Quasi-ExperimentalAnalysis–RegressionDiscontinuityDesign
WeemployanRDDanalysis,definedbyImbensandLemieux(2008),toestimatethemagnitudewithwhichconcentrationgrantsmayshapeschool-levelorganizationalchangeorpupilachievementwhenaggregatedtothedistrictlevel.Ourstrategytakesadvantageofthefactthatreceiptofaconcentrationgrantistriggeredwhendistrictenrollmentismadeupof
DecentralizingFinance,OrganizationalChangeandAchievement–13
55%weightedstudents,plusone.Figure4showshowconcentrationgrantsarrivewhentheunduplicatedpupilpercentage(UPP)exceeds55%,thenvarieswidelybasedontheunduplicatedcountofweightedstudents.
Table1.Descriptivestatisticsforassignmentandoutcomevariablesusedinregressiondiscontinuitydesign
Variable N Mean SD Min Max
Assignment variable
Unduplicated pupil percentage 949 0.60 0.25 0.01 1.00
Outcome variables 1. Social-organizational features of high schools
The number of newly hired first-year teachers in the district 414 11.7 24.76 0 414
The average number of class periods assigned to teacher 414 5.5 1.32 2.67 14.97
The percent of classes meeting A to G requirements 414 57.3 15.32 0.00 86.71
Outcome variables 2. Student achievement
The percent of students who passed an Advanced Placement (AP) Exam with a score of 3 or higher 400 57.4 19.11 0 100
The percent of 3-8 grade students who have scored “Standard Exceeded” in SBA ELA test 859 13.9 11.77 0 62
The percent of 3-8 grade students who have scored “Standard Not Met” in SBA ELA test 859 32.1 16.15 0 100
The percent of 3-8 grade students who have scored “Standard Exceeded” in SBA Math test 859 13.6 12.73 0 71
The percent of 3-8 grade students who have scored “Standard Not Met” in SBA Math test 859 34.8 16.81 0 100
Thepolicy’sstrictcutoffallowsustoutilizetheso-called“sharp”regressiondiscontinuitydesign.InsharpRDD,theprobabilityofreceivingthetreatmentequals1aboveagiventhresholdand0below,thatis! " = 1 % > %∗ = 1and! " = 1 % < %∗ = 0.Inoursettingtheassignmentvariable%isthepercentageofUPPandthethreshold%∗is55%.Thetreatment"iswhethertoreceivetheconcentrationgrants.
Weassumethatthetreatment-assignmentmechanismisdeterminedonlybythe
observable%,sothatthediscontinuityapproximatesarandomizedexperimentaroundthethreshold%∗.Schooldistrictsoneithersideofthethresholdandclosetoitareexpectedtobeverysimilar.Therefore,adifferenceinoutcomevariablescanbeattributedtothetreatment.
Theestimatorofaveragetreatmenteffect(ATE)is:
*"+ = + , % ∈ %∗, %∗ + 0 − + , % ∈ %∗, %∗ − 0 ,
where,isanoutcomevariable.SincethisATEisonlyidentifiedwhen0isaverysmallvalue(i.e.,for0 → 0),theATEisquantifiedinthedifferencebetweenregressionlinesrightatthe
DecentralizingFinance,OrganizationalChangeandAchievement–14
cutoffvalue%∗.Inotherwords,weestimatethejumpintheregressionlinerelatingtheunduplicatedpupilpercentagetooutcomevariablesrightattheUPPcutoffof55%atwhichthetreatmentassignmentswitches.ThisATEisinterpretedasanintent-to-treat(ITT)effectofassigningdistrictstothetreatmentgroup,thosereceivingconcentrationgrants.
Figure4.ThedistributionofCaliforniaschooldistricts’concentrationgrants,stemmingfrompercentageofunduplicatedweightedstudents
Estimationandbandwidthselection.Wefitthefollowingmodeltoestimatethetreatment
effectofdistrictreceiptofconcentrationgrants:
,3 = 4 %35 + 6"3 + 73
Inthismodel,,3 isthevalueofanoutcomeforthe8thschooldistrict,andtheassignmentvariable%35 istheunduplicatedpupilpercentageinthe8thdistrict,re-centeredsothatithasavalueofzeroatthecutoffof55%(%35 = %3 − %∗).Since"3 indicateswhetherthe8thdistrictreceivedaconcentrationgrant(1=receivedconcentrationgrant,0=otherwise),itsassociatedregressionparameter6representsthetreatmenteffectoftheconcentrationgrantsontheoutcome,estimatedattheunduplicatedpupilpercentageof55%.
4 %35 isagenericrepresentationofthefunctionalformofthehypothesizedrelationshipbetweentheoutcomeandthecenteredassignmentvariable%35.6Wechoosetousenonparametriclocallinearregressionforthe4 %35 ,whichprimarilyconsidersdatapointslocatedcloselyaroundthe55%cutoff.TheweightedlocallinearregressionmodelwiththeEpanechnikovkernelfunctionisfittedoneithersideofthecutoff,imposinghigherweightstoschooldistrictswhichareclosertothecutoff.Todeterminetheweightsofthelocallinearregressionatthecutoff,weusetheImbens–Kalyanaraman(Imbens&Kalyanaraman,2009)bandwidthselectionalgorithm(IKhereafter).TheIKalgorithmconsidersthesamplesizeaswellasthevariationandfunctionalformoftheoutcomevariablenearthediscontinuityingeneratinganidealbandwidthwhichminimizesthemeansquareerror.
DecentralizingFinance,OrganizationalChangeandAchievement–15
Figure5showstheestimatedkernelweightsforlocallinearregressionusingIKbandwidth
selectionwhentheoutcomeisthepercentof3-8gradestudentswhohavescoredStandardNotMetinthestateELAtest.SincetheestimatedIKbandwidthis7.98,datapointsoutsidetherangeof55 ± 7.98areexcluded,givingzeroweightstothosepoints.Fordatapointsreceivingnon-zerokernelweights,weseethatkernelweightsforthedistrictsclosertothe55%cutoffareestimatedhigher.
Wealsochecktherobustnessofourfindingsbyusingdifferentbandwidthselection
algorithmdevelopedbyCalonico,Cattaneo,andTitiunik(2015).Theirbandwidthselectionalgorithmattemptstocorrectforbiasduetoundersmoothing,whichmightemergebecausethefunctionalformoftheregressionlinesatthecutoffisnotwellapproximated.Inthefindingsection,wecalltheirestimatorasabias-correctedRDDestimator.Wealsopresentrobustestimateswhichcorrectstandarderrorsduetouncertaintyinthebiascorrection.SeeColonic,Cattaneo,andTitiunik(2015)formoredetail.
Figure5.TheestimatedkernelweightsforlocallinearregressionusingImbens–Kalyanaramanbandwidthselectionwhentheoutcomeisthepercentof3-8gradestudentswhohavescored“StandardNotMet”onELAtest.
Checkingassumptions.Identificationoftreatmenteffectsreliesontheassumptionthatthetreatment-assignmentmechanismbehavedasassumed.Forexample,theremaybeconcernsthatschooldistrictswithslightlylowerUPPscouldmanipulatetheirrecordstoreceiveconcentrationgrants.Inthatcase,thedistributionoftheassignmentvariablecouldbediscontinuousatthecutoff,withsurprisinglymanydistrictsjustbarelyeligibleforreceivingconcentrationgrantsandsurprisinglyfewfailingtoeligibility.Thesediscontinuitiesorbumpsinthedensityofassignmentvariablecanbevisuallyinspectedorformallytestedwithahypothesistest.WeperformtheMcCrarytest(McCrary,2008)toassesspotentialdiscontinuitiesatthecutoffoftheassignmentvariable(UPP).Figure6showstheresultsfromtheMcCrarytest,suggestingnosignificantdiscontinuityofdensityaroundthe55%cutoff.
DecentralizingFinance,OrganizationalChangeandAchievement–16
Figure6.TheMcCrarytestforcheckingdiscontinuitiesaroundthecutoffoftheassignmentvariable
AnotherassumptioncheckistoreplacetheactualoutcomeintheRDDanalysiswithnon-outcomecovariates,particularlythosecollectedpriortotreatmentimplementation.Ifwewouldobserveatreatmenteffectonapre-treatmentcovariate,doubtwouldbecastonthevalidityoftheRDDanalysis.Adesiredresultisthatnoeffectatthecutoffisfoundinanyofthepretreatmentcovariates.Wechosetwosetsofvariables—averagedailyattendanceineachgradespanandenrollmentbyethnicityandELstatus.Thesevariablescanclearlyberegardedaspre-treatmentcovariates,becausetheyaremeasuredandreportedbeforedistributingconcentrationgrants.
Findings
Wefirstdescribehowdistrictscomparewhenfallingoneithersideofthe55%cutpoint.
Table2presentsestimateddifferencesinthepre-treatmentcovariatesbetweentreatmentandcontrolgroups.Notethatpre-treatmentcovariatesweretransformedbytheirnaturallogarithmtomitigateagainsttheextremelyheavyrighttailsofthedistributionsofenrollmentvariables.Thefirstcolumngivesrawdifferences,whicharemeandifferencesbetweenalldistrictslocatedontheright-handsideofthecutoffandalldistrictsontheleft-handside.Itshowsthatdistrictsintreatmentgrouphavesignificantlymorekindergarten,elementary,middle,andhighschoolstudentsonaveragethandistrictsinthecontrolgroup,whilenodifferenceswerefoundin4-6gradestudents.ThedifferenceinsizebecomesstatisticallyinsignificantwhenwerestrictoursampletoobservationswithintheIKbandwidth.
ThesecondsetofestimatesusesImbensandKalyanaraman(2009)’sbandwidthselectionalgorithm,providingrawmeandifferencefordistrictswithinselectedrangeandweightedlocalconditionalmeandifferencewithintherange.ThelatterindicatesRDDestimates.Asshown,noneoftheRDDestimatesisstatisticallysignificant.Thissuggeststhatdistrictsfallingoneither
DecentralizingFinance,OrganizationalChangeandAchievement–17
sideofthe55%thresholdforreceivingconcentrationgrantsdonotdiffersignificantlybasedontheirstudentenrollmentforeachschoollevelorgradespan.Table2.Estimateddifferencesbetweentreatmentandcontrolgroupsin(log-transformed)pre-treatmentcovariates
Variable
Descriptive statistics Raw difference Imbens & Kalyanaraman (2009)’s
bandwidth selection algorithm
Mean (Std.Dev)
Total N
Mean diff.
Band width
Band range (non-zero weights)
N within band
Mean diff.
RDD Estimate (Std.Err)
Average Daily Attendance (ADA) in each grade span (log transformed)
Total ADA 7.15 (2.00) 949 0.02
(0.13) 5.18 [49.90, 60.10] 122 0.11 (0.34)
0.87 (0.68)
ADA K-3 5.63 (2.50) 949 0.30*
(0.17) 5.15 [49.87, 60.06] 122 -0.51 (0.44)
0.61 (0.79)
ADA 4-6 5.33 (2.49) 949 0.23
(0.17) 4.91 [50.20, 59.90] 120 -0.46 (0.44)
0.91 (0.90)
ADA 7-8 4.69 (2.65) 949 0.39**
(0.18) 4.81 [50.20, 59.60] 118 -0.35 (0.46)
0.51 (0.94)
ADA 9-12 5.63 (2.50) 949 0.30*
(0.17) 5.15 [49.87, 60.06] 122 -0.51 (0.44)
0.61 (0.79)
Enrollment by student subgroups (log transformed)
Total enrollment
7.18 (2.02) 937 0.02
(0.13) 5.23 [49.90, 60.20] 123 0.11 (0.34)
0.81 (0.67)
English learner 5.50 (2.26) 866 1.28***
(0.15) 5.91 [49.10, 60.90] 128 0.33 (0.42)
0.74 (0.73)
Latino/Hispanic 6.17 (2.33) 925 0.89***
(0.15) 5.66 [49.40, 60.60] 133 0.09 (0.38)
0.84 (0.71)
Black 3.63 (2.29) 767 0.10
(0.16) 6.00 [49.10, 60.90] 119 0.29 (0.42)
1.02 (0.84)
Note. ***p≤.01, **.01<p≤.05, *.05<p≤.10DistrictsinthetreatmentgrouptendtohavemoreLatinostudentsandEnglishLearners,
butthesedifferencesnolongerpersist,becominginsignificant,whenthesampleisrestrictedtodistrictswithintheIKbandandwhenwegivemoreweighttodistrictsclosertothe55%cutoff.Thus,wecanconcludethat,atleastonobservedpre-treatmentstudentcomposition,theconditionalexpectationsaresmoothatthediscontinuity.Thisboostsourconfidenceinthevalidityofanyestimatedtreatmenteffects.
Table3displaysasampleof10unifiedorhighschooldistrictslocatedclosetothe55%
cutoff.TheyvaryintermsofenrollmentsizeandtheshareofstudentsdesignatedasEnglishlearners.Yetotherwise,pupilcompositionslookquitesimilaracrosstheseillustrativedistricts.Notealsothattheamountofconcentrationgrantsreceivedvariesgreatly,whichlikelyconditionsthemagnitudeofpossibleeffects.Ouranalyticmethodtestsonlyforwhetherreceiptofaconcentrationgrant,nottheamount,ispredictiveoforganizationalchangeorachievementdifferences.
DecentralizingFinance,OrganizationalChangeandAchievement–18
Table3.Tenschooldistrictsaroundthe55%threshold
District Enrollment %EL %Latino %Black UPP Concentration Grant (CG)
Per-pupil
CG
Per-pupil Total
Funding San Rafael City High 2,365 16.6 53.3 2.3 48.0 0 0 8,355
Hanford Joint Union High 3,845 8.6 61.2 5.8 53.3 0 0 9,262
Escalon Unified 2,695 20.5 47.7 0.4 53.7 0 0 8,855
Santa Barbara Unified 14,291 30.6 60.4 0.9 54.1 0 0 8,394
Imperial Unified 3,898 23.1 80.9 1.5 54.9 0 0 8,278
East Side Union High 23,685 14.8 46.5 2.9 55.2 196,879 15 9,640
Tracy Joint Unified 15,761 25.8 49.9 6.4 55.5 323,896 37 8,704
Elk Grove Unified 62,196 17.3 25.9 14.1 56.0 2,218,799 64 8,350
Grossmont Union High 17,580 12.8 34.1 5.9 56.2 845,789 86 9,522
Chaffey Joint Union High 24,598 10.1 63.5 8.1 57.7 2,828,169 199 9,614
EffectsonSocial-OrganizationalFeaturesofHighSchoolsTable4summarizesestimateddifferencesinoutcomevariablesbetweentreatmentandcontrolgroups,districtsfallingaboveorbelowthe55%cutoff,respectively.Thefirstsetofoutcomevariablespertaintosocial-organizationalfeaturesofhighschools,asreportedby412districtswithcompletedata.Rawdifferencesbetweenalltreatmentandcontroldistrictsarenotable.Weseealmostnodifferenceinthecountofnewteachershiredbetweendistrictsfallingbelowversusabovethe55%cutoff(1.4teachers).However,districtsabovethecutoffassigntheirhighschoolteachersaboutone-thirdmoreclassperiodseachday,onaverage,comparedwithdistrictfallingbelowthe55%marker.Theformerdistrictsofferalmost9%fewercoursesmeetingA-Gcriteriathanthelatterdistricts.
Butresultsarequitedifferentwhenestimatingthediscreteeffectofreceivingconcentrationgrantsafterpruningdistrictsfarfromthe55%cutoff.TheIKregression-discontinuityestimatesshowthatweighteddistrictsabovecutoffwereabletohire10additionalteachersbeyondtheaverageforweighteddistrictsfallingbelowthe55%cutoff.Thedistrictsreceivingconcentrationgrantsassignedteachersonefewerclassperiodseachdayonaverage,andofferedagreatershareofcoursesthatmetA-Grequirements(about10%higher),comparedwithweighteddistrictsnotreceivingconcentrationgrants.
Figure7vividlyillustratesthebumpexperiencedbydistrictswithhighschoolsthat
benefitedfromconcentrationgrants.Eachdotonthispairofplotsrepresentsaschooldistrict,placedinrelationtoitsUPPandtheoutcomeofinterest:newteachershiredorclassperiods
DecentralizingFinance,OrganizationalChangeandAchievement–19
Table4.Estimateddifferencesbetweentreatmentandcontrolgroupsfororganizationalfeaturesofschoolsandstudentachievement
Variable
Raw difference IK algorithm CCT algorithm
Total N
Mean diff.
(Std.err.)
Band widt
h
Band range (non-zero weights)
N within band
Mean diff.
within IK band (Std.err)
IK RDD
Estimate (Std.err.)
Bias-corrected
RDD Estimate (Std.err.)
Robust RDD
Estimate (Std.err.)
Outcome variables 1. Social-organizational features of high schools
The number of newly hired first-year teachers in the district 412 1.36
(2.12) 11.58 [43.90, 66.50] 122 2.21 (3.04)
9.87** (3.96)
13.62*** (4.43)
13.62** (5.39)
The average number of class periods assigned to teacher 412 0.30**
(0.12) 5.31 [50.20, 60.20] 56 -0.44 (0.30)
-1.12 (0.73)
-1.12** (0.50)
-1.12* (0.58)
The percent of classes meeting A to G requirements 412 -8.75***
(1.42) 10.24 [45.10, 64.70] 109 -1.39 (2.92)
10.31* (5.26)
12.34** (5.26)
12.34** (6.16)
Outcome variables 2. Student achievement The percent of students who passed an Advanced Placement (AP) Exam with a score of 3 or higher
400 -18.29*** (1.73) 9.36 [45.80, 64.40] 98 -3.01
(3.18) 17.21***
(6.08) 22.46***
(7.08) 22.46***
(8.14)
The percent of 3-8 grade students who have scored “Standard Exceeded” in SBA ELA test
859 -14.98*** (0.76) 7.61 [47.40, 62.60] 157 -1.30
(1.00) 3.61** (1.51)
3.31** (1.35)
3.31** (1.54)
The percent of 3-8 grade students who have scored “Standard Not Met” in SBA ELA test
859 20.26*** (0.86) 7.98 [47.20, 62.90] 160 1.33
(1.53) -6.11***
(1.99) -5.34***
(1.89) -5.34** (2.17)
The percent of 3-8 grade students who have scored “Standard Exceeded” in SBA Math test
859 -16.05*** (0.83) 8.04 [47.20, 63.00] 161 -2.13*
(1.13) 3.01* (1.75)
2.67* (1.50)
2.67 (1.74)
The percent of 3-8 grade students who have scored “Standard Not Met” in SBA Math test
859 21.94*** (0.88) 8.96 [46.20, 63.80] 175 1.71
(1.59) -4.49** (2.21)
-4.61** (2.03)
-4.61** (2.34)
Note. ***p≤.01, **.01<p≤.05, *.05<p≤.10; IK=Imbens&Kalyanaraman(2009),CCT=Calonico,Cattaneo,andTitiunik(2015)bandwidthselectionalgorithm.
assignedtoteachers(leftandright-handpanels).Thedarklinestracetheestimatedassociationbetweenthetwovariables(aregressionline).Thelightgraylinesrepresenttheconfidenceintervalwithwhichtheregressionestimatesarecalculated.Thesharpbreaksatthe55%cutoffillustratesthejumpinnewteachershired(leftpanel)andfewerclassperiodassigned(rightpanel)indicatetheeffectsofconcentrationgrantsondistricts’organizationalbehaviors.
EffectsonStudentAchievementTable4alsoshowsdifferencesinstudentachievementlevelsfordistrictaboveorbelowthe55%cutoff.WeobtainedCalifornia’sstandardizedtestscoresinmathandELAforstudentsenrolledingrades3-8,withn=859districtsreportingcompletedata.Inaddition,wereportontheperformanceofsecondaryschoolpupilsonAPexams,scoring3orhigher,withn=412districtsoperatinghighschoolsandprovidingcompletedata.Weagaindisplayrawdifferencesbetweenthetwosetsofdistrict,thenestimatedifferencesbetweenweighteddistrictslayingwiththebandwidtharoundthe55%cutoff.
Let’sfirstlookatachievementdifferencesinELAamongpupils,grades3-8.Weseethatover20%(20.3)morestudentsfailtomeettheproficiencystandard(“StandardNotMet”)indistrictsabovethe55%cutoff,comparedwithpeersattendingschoolsindistrictsbelow.Thisisnotsurprisinggiventhattheformergroupofstudentsismorelikelytocomepoorornon-Englishspeakinghomes.Thiscanbeclearlyseenintheupper-rightpanelofFigure8.TheupwardlyslopingregressionlineindicatesthatthepercentageofstudentsfailingtomeetproficiencyclimbsrapidlyasUPPsharesrise.
Butnotethedropintheregressionlinesatthe55%cutoff:detectinganinterruptioninthe
expectedassociationbetweenpupilbackgroundandachievement.BackonTable2weseethattheshareofpupilsfailingtomeettheELAstandardwasabout6%(6.1)lesswhencomparedwithweighteddistrictsbelowthe55%cutoff.Similarly,theshareofstudentsmeetingtheELAstandardwasabout4%(3.6)higherindistrictsabovethecutoffandthusreceivingconcentrationgrants(seenintheupper-leftpanel,Figure8).TheseresultsforELAareconsequentialandstatisticallysignificant.
Differencesbetweendistrictsbelowandabovethe55%arelessremarkablewhenitcomes
tomath.Thelower-leftpanelinFigure8showstheshareofpupilsmeetingthemathstandard,grades3-8,plottedagainstUPPenrollmentshares.Asexpected,thepercentageofstudentsexceedingthemathstandardfallsasUPPsharesrise.Butnearthe55%cutoffweseeajumpupintheestimatedpercentageofstudentsclearingthemathproficiencybar.Thisadvantageequals3%.Butthedifferenceisonlymarginallysignificant(atp<.10)andnotsignificantforoneofthetestsinmeandifferences(Table4).Ontheotherhand,the4.5%fewerstudentswhofailtomeetthemathstandardisstatisticallysignificant.Overall,mathachievementeffectsstemmingfromconcentrationgrantsarelessconsistentthantheencouragingeffectsonpupils’ELAperformanceindistrictsabovethe55%cutoff.
DecentralizingFinance,OrganizationalChangeandAchievement–21
Number of first-year teachers
Percent of courses meeting A-G requirements
Figure7.Positiveorganizationaloutcomesfordistrictsclosetothe55%cutoff.Linesrepresentlocallinear
regressionsusinganEpanechnikovkernelandIKbandwidth.Dotsrepresentschooldistricts.
Finally,weseethattheshareofhighschoolersscoringa3orhigheronAPexamswasconsiderablegreaterfortheaverageweighteddistrictabovethe55%cutoff.Thepercentageofpupilsscoring3orhigherequaled17%percent(17.2)greaterthantheaverageweighteddistrictabovetheline,comparedwiththeaveragedistrictbelow.Thisdifferenceishighlysignificantandconsequentialinpragmaticterms.
Wealsotestedforheterogeneoustreatmenteffectsbyestimatingseparatemodelsfor
specificracialorethnicgroups.Table5showsthatmostsignificanteffects,stemmingfromreceiptofconcentrationgrants,arefeltbyLatinostudentsandnotbyothergroups.ThisistrueforthepercentofhighschoolstudentspassingAPcourseswithascoreof3orhigher,aswellastestscoreresultsforpupilsingrades3through8.
Thisspecificityofstudentachievementeffectsmaynotbesurprising,giventhatdistrictswithmorethan55%UPParedominatedbyLatinopupils.Ontheotherhand,thelackofeffectsforBlackstudentsisworrisome.ThelackofpositiveeffectsonWhitestudents–adistinctminorityindistrictswinningconcentrationgrants–doessuggestthatthetreatmentoverallmaynarrowachievementgapsbetweenLatinoandWhitestudents.Moreworkisrequiredtounderstandhowgap-closingworksamongschoolswithindistricts.
DecentralizingFinance,OrganizationalChangeandAchievement–22
ELA Test Scores – Standard Exceeded ELA Test Scores – Standard Not Met
Math Test Scores – Standard Exceeded Math Test Scores – Standard Not Met
Figure8.EachCaliforniadistrictplacedinrelationtopercentageofstudentsmeetingornotmeetingproficiency
standardandUPP.LinesrepresentlocallinearregressionsusinganEpanechnikovkernelandIKbandwidth.
DiscussionandPolicyImplications
California’swageronrestoringbudgetauthoritytolocalboards–thenawardingmanywith$41billioninprogressivelydistributedfundinginrecentyears–offersahugeexperimentindecentralizedpublicfinance.TheLocalControlFundinginitiativestemmedfromgrowingfrustrationwithdivergentstreamsofcategoricalaid,fragmentedprogramsthatwerecentrallydesignedandtightlyregulated.Collateralfaithinsite-basedmanagement,freeingprincipalstoselecttheirownstaff,managetheirbudgets,andperhapsfocusoninstructionalgains,alsoshiftedthepolicydiscourseinCaliforniaoverthepastgeneration.
DecentralizingFinance,OrganizationalChangeandAchievement–23
Table5.Heterogeneouseffectsofconcentrationgrantsonstudentachievementbyethnicitysubgroups
Variable Estimate Combined White Latino Black
The percent of students who passed an Advanced Placement (AP) Exam
with a score of 3 or higher
N (Bandwidth)
400 (9.36)
380 (10.80)
388 (8.92)
241 (9.59)
IK RDD Estimate (Std.err.)
17.21*** (6.08)
6.04 (3.77)
20.13*** (7.72)
11.37 (16.97)
The percent of 3-8 grade students who have scored “Standard Exceeded” in SBA ELA test
N (Bandwidth)
859 (7.61)
690 (8.75)
681 (6.47)
293 (8.68)
IK RDD Estimate (Std.err.)
3.61** (1.51)
1.10 (1.84)
2.14** (1.01)
2.19 (2.52)
The percent of 3-8 grade students who have scored “Standard Not
Met” in SBA ELA test
N (Bandwidth)
859 (7.98)
690 (8.27)
681 (8.77)
293 (10.33)
IK RDD Estimate (Std.err.)
-6.11*** (1.99)
-2.92 (2.21)
-3.67* (2.20)
-4.31 (4.28)
The percent of 3-8 grade students who have scored “Standard
Exceeded” in SBA Math test
N (Bandwidth)
859 (8.04)
690 (8.77)
680 (6.59)
293 (7.24)
IK RDD Estimate (Std.err.)
3.01* (1.75)
0.52 (2.20)
0.14 (1.17)
-1.89 (2.90)
The percent of 3-8 grade students who have scored “Standard Not
Met” in SBA Math test
N (Bandwidth)
859 (8.96)
690 (8.90)
680 (8.60)
293 (10.45)
IK RDD Estimate (Std.err.)
-4.49** (2.21)
-2.79 (2.62)
-3.81** (1.90)
-3.50 (6.91)
Note. ***p≤.01, **.01<p≤.05, *.05<p≤.10; IK=Imbens&Kalyanaraman(2009)bandwidthselectionalgorithm.
TheGoldenStatedevisedahybridfinancestructure:devolvingbudgetauthoritywithina
frameworkofeightstatepolicygoals,progressivelydistributingdollarsamongdistricts,andmandatingaseeminglydemocraticprocessfordevisingbudgetslocally.Manyadvocatesarguethattwopiecesaremissing:holdingdistrictsaccountableforwhethertheychanneldollarstothekidsthatgeneratenewrevenuethroughsupplementalandconcentrationgrants;andanexplicitstrategytoevaluatetheschool-leveleffectsofthismassivepublicinvestment.PositiveOrganizationalandAchievementEffects
Wedofindencouragingresultsfromtheconcentration-grantelementoftheLCFreform.Districtswithmorethan55%oftheirenrollmentsintheweighted-pupilcategories–thuswinningconcentrationgrantsfromSacramento–aremorelikelytoengageinorganizational
DecentralizingFinance,OrganizationalChangeandAchievement–24
changesthatmaymediatelearninggains.Additionalanalysiswilltelluswhetherimprovedworkingconditionsforteachers(e.g.,byteachingfewerclassperiods)oramorerigorouscurriculum(offeringahighershareofA-Gcourses)helpstoexplaintheachievementgainsthatwedetected.Thepresentanalysisatleastestablishesthatconcentrationgrantsofferedfreshresourcesthatpromptedorganizationalchangeattheschoollevel.
Followingalongthishypothesizedcausalpathway,weestimatedhigherachievementinELA
forstudentsenrolledindistrictsthatreceivedconcentrationgrants.Weseegainsatboththelowandhighendsofpupilperformance.Thatis,theshareofstudents,grades3-8,whofailedtomeetthestate’sproficiencystandardeasedamongdistrictsabovethe55%cutpoint.Andthesharethatdidmeetstandardrangedhigher,comparedwith(weighted)districtsfallingbelow55%,failingtowinconcentrationgrants.TheseachievementbenefitswerefeltbyLatinostudents.Pupilsinotherethnicgroupsdidnotshowcommensurategainsintestscores.
LimitationsandPersistingQuestions
OuranalysisislimitedtothefirsttwoyearsofLCFimplementation.Theinitiativeisabouttoenteritsfifthyearofimplementation.Asdatabecomeavailable,wecantestforwhethertheseorganizationalandpupil-achievementgainspersistovertime.Wehopetoidentifyadditionalorganizationalfeaturesofschools–includingassignmentofnewteachers–whichmayhelptoexplainchangesinpupilachievement.Wecanfurtherdisaggregatehowstudentsubgroupsmaydifferentiallybenefitfrominfusionsofconcentrationgrants.
ThefactthatgainswerefeltbyLatinostudentsisencouraging,suggestingthatachievement
gapsarenarrowingvis-à-visWhitepeers.Moreworkisrequired,however,tounderstandwhetherconcentrationgrantsorparallelelementsofLCFareclosingdisparitiesamongstudentgroupswithindistricts,andthroughwhatorganizationalchanges.OurinabilitytodetectgainsforBlackstudentsisworrisome.
Staffingdatacanbefurtherexploitedtobetterunderstandpossibleeffectsonthekindsof
teachingorstudentsupportpositionsaddedtoschoolpayrolls,alongwithrelatedorganization-levelchange.OurworkinLosAngelesrevealsthatnewLCFfundinghasledtoshrinkingsharesofdollarsgoingforteachingpostsinhighschools,asdollarsgoforcounselors,assistantprincipals,andpupilsupportstaff(UnitedWay,2017).
Theeffectsweestimatestemsolelyfromtheconcentration-grantportionofLCF,which
equaled$3.3billionstatewidein2016-17,oraboutone-sixthofthegovernor’sprogram.Wecouldestimatehowsupplementalgrantscontributetotheseeffects–perhapsinteractingwithlevelsofconcentrationgrants–althoughsuchestimatescouldbecontaminatedbyconfoundersproxiedbythelevelofsupplementalgrants.Inaddition,perpupilspendingtiedtoconcentrationgrantsisnotasdiscontinuousaswhethergrantsaregivenornot(discussedaboveandseetheAppendix).Furtheranalysisisrequiredtolearnhowspendinggainsshapeschool-levelorganizationalchangeandachievement.
DecentralizingFinance,OrganizationalChangeandAchievement–25
Thispaperoffersonemethodforunderstandingoneelementofthismassiveexperimentin
progressivepublicfinance.Weemphasizehoworganizationalchangeorshiftsinteacherqualitieslogicallyprecedesgainsinstudentengagementandlearning.Simplyassociatingnewfundingwithpupiloutcomesdoeslittletobuildtheoryoverhowdecentralizingfinancemaysetinmotionpracticesonthegroundthatalterschoolsandpupilmotivation,ornot.Nordogoodintentionsorhigh-soundingrhetoricabouthelpingpoorchildrensuffice.Whenthepoliticalstarsdoaligntoprogressivelyfundschools,wemustrigorouslydigintowhetherthesewell-meaninginitiativestrulyadvancefairnessand,ifso,throughwhatpracticesinsideschools?
Appendix
Thedollarincreaseperpupilreceivedbydistrictsthatreceiveconcentrationgrants,seenbelowfor2013-14,isnotasdiscontinuousatthe55%cutpointrelativetotheoverallintent-to-treatdiscontinuity.Futureworkmightuse“kinkedregression”techniquestoassociatespendingchangeswithorganizationalpracticesanddifferencesinstudentachievement.
References
Augenblick,J.,Myers,J.,Anderson,A.(1997).Equityandadequacyinschoolfunding.FutureofChildren,7,63-78.Baker,B.(2003).ReviewofweightedstudentfundingforCalifornia.Boulder:NationalEducationPolicyCenter.Berne,R.,&Stiefel,L.(1979).Taxpayerequityinschoolfinancereform:Theschoolfinanceandpublicfinance
perspectives.JournalofEducationFinance,5,36-54.Brown,J.(2013).GovernorBrownsignshistoricschoolfundinglegislation.Sacramento:OfficeoftheGovernor.Access
at:https://www.gov.ca.gov/news.php?id=18123.Bersin,A.,Kirst,M.,&Liu,G.(2008).Gettingbeyondthefacts:ReformingCaliforniaschoolfinance.Stanford,CA:Getting
DowntoFactsProject,StanfordUniversity.Bryk,A.,Sebring,P.,Allensworth,E.,Luppescu,S.,&Easton,J.(2010).Organizingschoolsforimprovement:Lessonsfrom
Chicago.Chicago:UniversityofChicagoPress.
DecentralizingFinance,OrganizationalChangeandAchievement–26
CaliforniaOfficeoftheLegislativeAnalyst(2013).AnoverviewoftheLocalControlFundingFormula(updated).
Sacramento.CaliforniaOfficeoftheLegislativeAnalyst(2016).EdBudgetTables,July.Accessat:
http://www.lao.ca.gov/Publications/Report/3491.Calonico,S.,Cattaneo,M.D.,&Titiunik,R.(2015b).Rd-robust:Anrpackageforrobustnonparametricinferencein
regression-discontinuitydesigns.RJournal,7,38–51.Chambers,J.,Levin,J.,&Shambaugh,L.(2010).Exploringweightedstudentformulaasapolicyforimprovingequityfor
distributingresources:AcasestudyoftwoCaliforniaschooldistricts.EconomicsofEducationReview,29,283-300.
Chambers,J.,Shambaugh,L.,Levin,J.,Muraki,M.,&Poland,L.(2008).Ataleoftwodistricts:Acomparativestudyof
student-basedfundingandschool-baseddecisionmakinginSanFranciscoandOaklandunifiedschooldistricts.MenloPark,CA:AmericanInstitutesofResearch.
Fensterwald,J.(2015).Torlaksonreinterpretsdepartment'sstanceonteacherraises.EdSource,June15.Accessat:
https://edsource.org/2015/torlakson-reinterprets-departments-stance-on-teacher-raises/81528.Finn,C.,Manno,B.,&Wright,B.(2016).Charterschoolsatthecrossroads:Predicaments,paradoxes,possibilities.
Cambridge,MA:HarvardEducationPress.Fuller,B.(2015).Organizinglocally:Howthenewdecentralistsimproveeducation,health,andtrade.Chicago:University
ofChicagoPress.Fuller,B.,Loeb,S.,Arshan,N.,Chen,A.,&Yi,S.(2007).Californiaprincipals’resources:Acquisition,deploymentand
barriers.Stanford,CA:GettingDowntoFactsProject,StanfordUniversity.Fuller,B.,Marsh,J.,Stetcher,B.,&Timar,T.(2011).DeregulatingschoolaidinCalifornia:How10districtsrespondedto
fiscalflexibility,2009-2010.SantaMonica:RAND.Fuller,B.,&Tobben,L.(2014).LocalControlFundingFormulainCalifornia:Howtomonitorprogressandlearnfroma
grandexperiment.Berkeley:SchoolofLaw.Hanushek,E.,&Woessmann,L.(2017).Schoolresourcesandstudentachievement:Areviewofcross-countryeconomic
research.Pp.149-170inCognitiveabilitiesandeducationoutcomes,editedbyM.Rosen.NewYork:Springer.HawleyMiles,K.&Roza,M.(2006)Understandingstudent-weightedallocationasameanstogreaterschoolresource
equity.PeabodyJournalofEducation,81,39-62,Imbens,G.,&Kalyanaraman,K.(2012).Optimalbandwidthchoicefortheregressiondiscontinuityestimator.TheReview
ofEconomicStudies,79,933–959.Imbens,G.,&Lemieux,T.(2008).Regressiondiscontinuitydesigns:Aguidetopractice.JournalofEconometrics,142,
615–635.Jackson,K.,Johnson,R.,&Persico,C.(2015).Theeffectsofschoolspendingoneducationalandeconomicoutcomes:
Evidencefromschoolfinancereforms.Cambridge,MA:NationalBureauofEconomicResearch.Kholi,S.(2016).HowdidLAUSDspend$450million?Notonthehigh-needsstudentsthemoneywasfor,stateofficials
say.LosAngelesTimes,August9.Accessat:http://www.latimes.com/local/education/la-me-edu-lausd-state-funding-decision-20160807-snap-story.html.
DecentralizingFinance,OrganizationalChangeandAchievement–27
King,R.(1983).EqualizationinNewMexicoschoolfinance.JournalofEducationFinance,9,63-78.Lafortune,J.,Rothstein,J.,&Schanzenbach,D.(2016).Schoolfinancereformandthedistributionofstudent
achievement.Berkeley:DepartmentofEconomics.Leppert,&Routh,(1980).Weightedpupileducationfinancesystemsinthreestates:Florida,Utah,andNewMexico.
Washington,DC:NationalInstituteofEducation.Malen,B.,Dayhoff,J.,Egan,L.,&Croninger,R.(2015).Thechallengesofadvancingfiscalequityinaresource-strained
context.EducationalPolicy,29,1-28.Odden,A.,&Picus,L.(2014).Schoolfinance:Apolicyperspective.NewYork:McGraw-Hill.PartnershipforLosAngelesSchools(2017).MakingequitythefoundationforL.A.Unifiedbudgeting.LosAngeles.Schwartz,A.,Rubenstein,R.,&Stiefel,L.(2008).Whydosomeschoolsgetmoreandothersless?Anexaminationof
school-levelfundinginNewYorkCity.NewYork:InstituteforEducationandSocialPolicy.Shipps,D.(2006).SchoolreformChicagostyle,1880-2000.Lawrence:UniversityPressofKansas.UnitedWayandCLASSCoalition(2017).Incrementalprogresstowardequity:Trackingthedistributionof$3.8billion
LCFFdollarsintheLosAngelesUnifiedSchoolDistrictsince2013.LosAngeles:UnitedWayofGreaterLosAngeles.
Wirt,F.,&Kirst,M.(1972).ThepoliticaldynamicsofAmericaneducation.Richmond,CA:McCutchan.Wolf,R.,&Sands,J.(2016).ApreliminaryanalysisofCalifornia’snewLocalControlFundingFormula.EducationPolicy
AnalysisArchives,24.Accessat:http://files.eric.ed.gov/fulltext/EJ1100156.pdf.
Endnotes
1ThanksgotoBillLuciaoftheEdVoiceorganizationinSacramentoforthisvividanalogy.2Thestatecountsweightedstudentsjustonce,so-called“unduplicatedpupils”,whentheyfallintomorethanonecategoryofdisadvantage.3Thestate’scentrallyarticulatedgoalsofferamixofintentions,whichemphasizeimplementationofCommonCoreStateStandards;improvingparentparticipation,schoolclimate,andbasicservices;widening“courseaccess”anddeepeningstudentengagement,alongwithraisingstudentachievementand“otherstudentoutcomes”(California,2013).4Themedium-termassessmentofGov.Schwarzenegger’sconsolidationofcategoricalaidprogramsdidfindawarmreceptionfromdistrictofficials,manyreportingthattheywelcomedthenewfiscaldiscretion.Atthesametime,districtsdisplayedlimitedcapacityoverallinassessingschool-levelchangeorachievementeffectsstemmingfromneworrecastprogramsaimedatliftingstudents(Fuller,Marsh,Stetcher,&Timar,2011).5Thatis,thepercentageofadistrict’senrollmentthatincludesstudentsfromlow-incomefamilies,Englishlearners,orthoseinfostercare.6Weincludedadditionalcovariatessuchasthenaturallogoftotalenrollmentandthetypeofschooldistrict(unified,elementary,orhighschool)inthemodelspecification.Thisisnotnecessaryfortreatment-effectidentification.Still,itcanbeusefulinimprovingprecision.Theresultswithorwithoutincludingcovariatesdidnotdiffersignificantly.