Citation
Faculty Composition and Student Achievement in Charter Schools

Material Information

Title:
Faculty Composition and Student Achievement in Charter Schools
Creator:
Carruthers, Celeste
Place of Publication:
[Gainesville, Fla.]
Publisher:
University of Florida
Publication Date:
Language:
english
Physical Description:
1 online resource (131 p.)

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Economics
Committee Chair:
Figlio, David N.
Committee Members:
Kenny, Lawrence W.
Hamersma, Sarah E.
Sindelar, Paul T.
Graduation Date:
8/8/2009

Subjects

Subjects / Keywords:
Academic achievement ( jstor )
Charter schools ( jstor )
Classrooms ( jstor )
Mathematics ( jstor )
Mathematics teachers ( jstor )
Public schools ( jstor )
Schools ( jstor )
Students ( jstor )
Teachers ( jstor )
Test scores ( jstor )
Economics -- Dissertations, Academic -- UF
charter, economics, education, experience, mobility, public, quality, school, segregation, teachers
Genre:
Electronic Thesis or Dissertation
born-digital ( sobekcm )
Economics thesis, Ph.D.

Notes

Abstract:
Charter schools are independent public schools, free from many of the regulations facing mainstream public schools. They are competitive entrants in partially deregulated education markets. Charter schools offer students and their families more choice in publicly funded education, and they offer teachers more choice in their careers. Studies examining the effectiveness of charter schools per se are abundant, but less is known about the qualifications and instructional effectiveness of charter teachers. I utilize rich, unique data on North Carolina public schools, students, and teachers to develop an in-depth understanding of the state's charter teachers. First, I determine whether charter schools were drawing good teachers away from traditional public schools. I find nuanced patterns of teacher quality flowing to charter schools. Charters drew highly qualified, certified teachers, but low certification requirements attracted less qualified, uncertified teachers as well. Charters drew effective instructors from the pool of mobile teachers willing to change schools, but they did not skim the very best mainstream teachers. Inexperienced teachers have been cited as a root cause of low student achievement in charter schools. I show that new charter schools in North Carolina had exceptionally high rates of new teachers, but that inexperienced faculties were not responsible for sub-par student achievement. Rather, students had greater math achievement in new charter schools with higher rates of new teachers, relative to charter and mainstream students in schools with more experienced faculties. These findings suggest that inexperienced charter faculties were benign signals of development, not staffing failures. Charter schools can exacerbate racial segregation, which tends to widen racial achievement gaps, and more than half of North Carolina's charter schools were racially imbalanced. But there may have been a silver lining, in that predominantly nonwhite charter schools were largely staffed by nonwhite teachers. I show that high rates of nonwhite teachers marginally raised the math performance of black charter students. Nonetheless, these gains were insufficient to slow the growing black-white achievement gap in charter schools. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2009.
Local:
Adviser: Figlio, David N.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2010-02-28
Statement of Responsibility:
by Celeste Carruthers.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Carruthers, Celeste. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
2/28/2010
Resource Identifier:
489240957 ( OCLC )
Classification:
LD1780 2009 ( lcc )

Downloads

This item has the following downloads:


Full Text

PAGE 1

1

PAGE 2

2

PAGE 3

3

PAGE 4

Manyindividualsandorganizationsmadethisresearchpossible.MysincerestthanksgotoDavidFiglio,LarryKenny,SarahHamersma,andPaulSindelarforadvisingthisdissertationthroughoutitsprogress.IamgratefultotheLockhartandWalter-LanzillottiEndowmentsfortravelandresearchsupport.IamparticularlythankfultotheNorthCarolinaEducationResearchDataCenteratDukeUniversityformaintainingexceptionallyrichpubliceducationdataandfacilitatingprojectslikethese.Andnally,Iamgratefulforhelpfulcommentsandqueriesbyparticipantsatthe2008and2009AmericanEducationFinanceAssociationmeetings,the2008SouthernEconomicAssociationmeetings,the2008AssociationforPublicPolicyAnalysisandManagementmeetings,andseminarsattheUniversityofFlorida,theUniversityofTennessee-Knoxville,theUniversityofNorthCarolina-Greensboro,andtheUniversityofMississippi. 4

PAGE 5

page ACKNOWLEDGMENTS ................................. 4 LISTOFTABLES ..................................... 7 LISTOFFIGURES .................................... 8 ABSTRACT ........................................ 9 CHAPTER 1QUALIFICATIONSANDCLASSROOMPERFORMANCE ........... 11 1.1Introduction ................................... 11 1.2CharterSchoolsinNorthCarolina ....................... 16 1.2.1Background ............................... 16 1.2.2Data ................................... 21 1.3AnalyticMethodsandResults ......................... 23 1.3.1QualicationsofTeachersMovingtoCharterSchools ........ 23 1.3.2ClassroomPerformanceofTeachersMovingtoCharterSchools ... 32 1.3.3VariationinClassroomPerformance .................. 38 1.4Conclusions ................................... 42 2NEWTEACHERSANDSTUDENTACHIEVEMENT .............. 55 2.1Introduction ................................... 55 2.2NorthCarolinaCharterSchools,Students,andTeachers .......... 60 2.2.1NorthCarolinaCharterSchools .................... 60 2.2.2NorthCarolinaPublicSchoolData .................. 62 2.2.3NorthCarolinaCharterTeachers ................... 64 2.3AnalyticMethodsandResults ......................... 66 2.4Conclusions ................................... 77 3TEACHERS,STUDENTS,ANDRACIALIMBALANCE ............ 93 3.1Introduction ................................... 93 3.2RacialImbalanceinNorthCarolinaCharterSchools ............. 96 3.2.1Data ................................... 96 3.2.2DescriptiveFiguresandStatistics ................... 98 3.3AnalyticMethodsandResults ......................... 100 3.3.1Methods ................................. 101 3.3.2Results .................................. 103 3.3.3StudentFixedEectsAnalysis ..................... 108 3.4Conclusions ................................... 111 REFERENCES ....................................... 126 5

PAGE 6

................................ 131 6

PAGE 7

Table page 1-1NorthCarolinacharterschools,students,andteachers,1998-2007 ........ 44 1-2In-samplemobilitypatternsofcharterteachers .................. 46 1-3NorthCarolinapublicschoolteachers:Summarystatistics ............ 47 1-4Regressionresults:Qualicationsofteacherschangingschools .......... 50 1-5NorthCarolinapublicschoolteacherxedeects:Summarystatistics ...... 51 1-6Regressionresults:Classroomperformanceofteacherschangingschools ..... 52 1-7Variationinteacherquality ............................. 53 2-1StudenttransitionsinandoutofNorthCarolinacharterschools ......... 79 2-2NorthCarolinapublicschoolfaculties:Summarystatistics ............ 80 2-3NorthCarolinapublicschoolstudents:Summarystatistics ............ 84 2-4Fixedeectestimates:Theeectofcharterenrollment .............. 86 2-5Laggedachievementestimates:Theeectofcharterenrollment ......... 87 2-6Disaggregatedeectofcharterenrollmentonmathgains ............. 88 2-7Disaggregatedeectofcharterenrollmentonreadinggains ............ 90 3-1Blackfemalepublicschoolstudents:Summarystatistics ............. 113 3-2Blackmalepublicschoolstudents:Summarystatistics .............. 114 3-3Whitefemalepublicschoolstudents:Summarystatistics ............. 115 3-4Whitemalepublicschoolstudents:Summarystatistics .............. 116 3-5Theeectsofcharterenrollmentandracialimbalance:Blackfemales ...... 120 3-6Theeectsofcharterenrollmentandracialimbalance:Blackmales ....... 121 3-7Theeectsofcharterenrollmentandracialimbalance:Whitefemales ...... 122 3-8Theeectsofcharterenrollmentandracialimbalance:Whitemales ....... 123 3-9Studentmathxedeects,bycharter/mainstreamandracialimbalance ..... 124 3-10Studentreadingxedeects,bycharter/mainstreamandracialimbalance ... 125 7

PAGE 8

Figure page 1-12006Charterenrollmentandpenetration ...................... 45 1-2Densityestimates:Years'experienceofmobileteachers .............. 48 1-3Densityestimates:Meanlicensuretestscoresofmobileteachers ......... 49 1-4Densityestimates:Persistentteacherqualitydistributions ............ 54 2-1Averagerateofnewteachers,byageofschool ................... 81 2-2Averagefacultyturnoverrates,byageofschool .................. 82 2-3MeanreversioninEOGscores ............................ 83 2-4Marginaleectofnewcharterteachers,byageofschool ............. 92 3-1Densityestimates:Percentnonwhitestudentsinschools ............. 117 3-2Densityestimates:Percentnonwhiteteachersinschools .............. 118 3-3Localpolynomialestimates:Percentnonwhiteteachersandstudentsinschools 119 8

PAGE 9

Charterschoolsareindependentpublicschools,freefrommanyoftheregulationsfacingmainstreampublicschools.Theyarecompetitiveentrantsinpartiallyderegulatededucationmarkets.Charterschoolsoerstudentsandtheirfamiliesmorechoiceinpubliclyfundededucation,andtheyoerteachersmorechoiceintheircareers.Studiesexaminingtheeectivenessofcharterschoolsperseareabundant,butlessisknownaboutthequalicationsandinstructionaleectivenessofcharterteachers. Iutilizerich,uniquedataonNorthCarolinapublicschools,students,andteacherstodevelopanin-depthunderstandingofthestate'scharterteachers.First,Ideterminewhethercharterschoolsweredrawinggoodteachersawayfromtraditionalpublicschools.Indnuancedpatternsofteacherqualityowingtocharterschools.Chartersdrewhighlyqualied,certiedteachers,butlowcerticationrequirementsattractedlessqualied,uncertiedteachersaswell.Chartersdreweectiveinstructorsfromthepoolofmobileteacherswillingtochangeschools,buttheydidnotskimtheverybestmainstreamteachers. Inexperiencedteachershavebeencitedasarootcauseoflowstudentachievementincharterschools.IshowthatnewcharterschoolsinNorthCarolinahadexceptionallyhighratesofnewteachers,butthatinexperiencedfacultieswerenotresponsibleforsub-parstudentachievement.Rather,studentshadgreatermathachievementinnewcharterschoolswithhigherratesofnewteachers,relativetocharterandmainstreamstudentsin 9

PAGE 10

Charterschoolscanexacerbateracialsegregation,whichtendstowidenracialachievementgaps,andmorethanhalfofNorthCarolina'scharterschoolswereraciallyimbalanced.Buttheremayhavebeenasilverlining,inthatpredominantlynonwhitecharterschoolswerelargelystaedbynonwhiteteachers.Ishowthathighratesofnonwhiteteachersmarginallyraisedthemathperformanceofblackcharterstudents.Nonetheless,thesegainswereinsucienttoslowthegrowingblack-whiteachievementgapincharterschools. 10

PAGE 11

Docharterschoolsdrawgoodteachersfromtraditional,mainstreampublicschools?Usinga1997-2007panelofallNorthCarolinapublicschoolteachers,Indnuancedpatternsofteacherqualityowingintocharterschools.Highratesofinexperiencedanduncertiedteachersmovedtocharterschools,butamongcertiedteacherschangingschools,theon-paperqualicationsofchartermoverswerebetterornodierentthanthequalicationsofteachersmovingtocomparablemainstreamschools.Iestimatemeasuresofclassroomperformanceforasubsetofgrade3-5teachers,andshowthatchartermoversweremoreeectiveinmathandreadinginstruction,relativetoothermobileteachers.Chartermoverscomparedlessfavorably,however,tonon-mobileteachersandcolleagueswithintheirsendingschools.Thedistributionofclassroomperformanceamongfuturecharterteachers,adjustedforsamplingerror,wassignicantlylowerthanthedistributionforexclusivelymainstreamteachers. 11

PAGE 12

Friedman ( 1955 1997 )proposedvouchersasonewaytostokeschoolcompetition. Dee ( 1998 ), Hoxby ( 2003 ),andmostrecently, Booker,Gilpatric,Gronberg,andJansen ( 2008 )oerempiricalevidencethatmainstreamstudentperformanceimprovesinlightofcompetitionfromchoiceschools.Long-rungainsfromcompetitionwillrequirecharterstobeformidablecompetitors,however,andthejuryisstilloutastowhethertheyactuallyincreasestudentlearningrelativetomainstreamschools.Theemergingconsensusisthatnewchartershaveanegativeimpactonstudentachievementgrowth,andthispenaltyfadesasschoolsandstudentsgainexperience.2 BifulcoandLadd ( 2006 ), Booker,Gilpatric,Gronberg,andJansen ( 2007 ), Hanushek,Kain,Rivkin,andBranch ( 2007 ),and Sass ( 2006 ).3 ( 2005 )and Rocko ( 2004 )ndthataonestandarddeviationincreaseinteacherqualityincreasesstudentachievementbyabouttenpercentofastandarddeviation. 12

PAGE 13

Manuel ( 2007 )).Chartersarenotgenerallyboundbystatepayscales,theycanallocatebudgetsastheyseet,andfeasibly,theycanpayhigherteachersalaries.OneNewYorkCitycharterschoolfamouslyoersteachercompensationpackagesinexcessof$125,000( Gootman ( 2008 )).Nationwide,charterteachersalariesaremorecomparabletomainstreamsalaries,5butinsomestates,charterteachersearnsignicantlylessthanotherpublicschoolteacherswithsimilarqualications( MalloyandWohlstetter ( 2003 )).Evenifcharterschoolscannotoutbidmainstreamschoolsonsalaryalone,schoolleaderscaninuenceteachers'utilityinnon-pecuniaryways,byreducingtheirnon-instructionalduties,encouragingcollegialityamongfaculty,manipulatingclasssizeandcomposition,andgrantingteachersmorecreativelicenseandautonomythantheyareaordedinmainstreamschools.Earlyadvocatesofthechartermodelstressedtheprofessionalizationandempowermentof Clotfelter,Ladd,Vigdor,andDiaz ( 2004 ), Hanushek,Kain,O'Brien,andRivkin ( 2005 ), FalchandStrm ( 2005 ),and Scadi,Sjoquist,andStinebrickner ( 2007 ).5Nationally,charterteachershadanaveragesalaryof$37,000in2004,versus$44,500fortraditionalpublicschoolteachers.Thepaygapcoincidedwithasubstantialexperiencegap:43.4percentofcharterteachershadthreeorfeweryears'experience,comparedtojust17.1percentofmainstreamteachers( NationalCenterforEducationStatistics ( 2006 )). PodgurskyandBallou ( 2001 )and Hoxby ( 2002 )alsofoundcompetitiveteachersalariesincharterschools. 13

PAGE 14

Budde ( 1988 )and Kolderie ( 1990 )).Highteachersatisfactionratesincharterschoolstypicallystemmedfromgreaterautonomy(\freedomtoteachthewayIwant"),like-mindedcolleagues,andinnovativeteachingphilosophies.Teacherswhoweredissatisedincharterschoolscitedlowpay,lackofbenets,highworkload,andinsucientfacilities( MalloyandWohlstetter ( 2003 )). Inpractice,theintangiblebenetsofworkinginacharterschoolmaybetoolowtoosetlowpayandotherresourcelimitations.Commoncharternancemodelsallocateeachschoolaper-pupilrateroughlyequaltothesurroundingdistrict'saverageper-pupilcost,excludingthecostofbuildings.Ifadistrictenjoyssubstantialeconomiesofscaleinvariablecost,itsper-pupilexpenseswillbelessthanacharterschool'saveragecost.Charterswithcompetingusesforlimitedresourcesmaysacricesometeachingtalentinfavorofadministrativeandcapitalimprovementsifdoingsomaximizestheirobjectives(studentachievement,enrollment,andbudgetsizebeinglikelyobjectives).Furthermore,manystatesallowcharterstoemployahighrateofuncertiedteachers.Thispermitscharterstoattractteachersfromoutsidethetraditionalpipeline,butalsoincreasesthesupplyoflow-cost,low-skilledindividualseligibletoworkincharterschools,includinguncertiedmainstreamteachersnearingtheexpirationoftemporarylicenses.Recently,Wisconsinraisedsubject-basedcerticationrequirementsforitscharterteachers,promptingschoolleaderstoarguethattheycouldnotaordtohireteachersmeetingthenewstandard( Borsuk ( 2008 )).Charterlicensurerequirementsvaryacrossstates,andlittleisknownaboutthequalicationsofuncertiedteachersincharterschools,ortheimpactofrelaxedlicensurestandardsonstudentperformanceincharterschools. Muchofthedevelopingresearchoncharterteacherqualityexaminesthequalications,workload,andjobsatisfactionofthestockofcharterteachersnationwideorwithinparticularstates. PodgurskyandBallou ( 2001 )surveyedteachersinsevenstates,andfoundthatcharterteacherswerelesslikelytobecertied,morelikelytobeinexperienced, 14

PAGE 15

Hoxby ( 2002 ),usinga1998nationalsurveyofteachers,showedthatcharterteachershadtypicallytakenmoremathandsciencecoursesincollege,weremorelikelytohavegraduatedfromagoodcollege,andloggedmoreextracurricularhours.Interestingly,charterspaidapremiumforthesequalities,butnotforcerticationormaster'sdegrees. Taylor ( 2005 )alsofailedtondapremiumforadvanceddegreesinTexascharterschools,andshowedthatteachersrealizeda7.5percentpaycutuponmovingtoacharterschool. Whileapictureofteacherqualityincharterschoolsisemerging,littleisknownabouttheowofteachingtalentbetweenmainstreamandcharterschools,ortheclassroomperformanceofindividualcharterteachers.Here,IfullycharacterizetheresumequalicationsofallNorthCarolinapublicschoolteacherswhomovedtothechartersectorbetween1998and2007.Forasubsampleofelementaryteachers,Icharacterizetheirclassroomperformanceaswell.NorthCarolinaisararesettingwherepassivelycollectedadministrativedataincludelongitudinalschoolassignmentsforallcharterandmainstreampersonneloveraperiodexceedingtenyears.Furthermore,thedatalinksometeachersdirectlytotheirstudents,allowingmetoestimatemeasuresofinstructionaleectiveness.Byanalyzingtheowofteachersfromonesectortoanother,Ideterminewhethercharterschoolswere\creamskimming"goodteachersfrommainstreamschools.Ifhighlyqualiedandeectiveteacherswerevotingwiththeirfeetinfavorofcharterschools,theirmigrationisafavorablesignalofthedecentralizedmodel'sappeal,andmainstreamschoolsmayneedtoemulatecharterfeaturestoretainfaculty.Ifchartersweredrawinglessqualiedandlesseectiveteachers,whetherbecauseoflowpay,poororganization,orrelaxedlicensurestandards,thechartermodelisunlikelytofulllitspromiseasarevolutionaryvehiclefortheimprovementofpublicschools. Inthisstudy,IevaluatetheresumequalicationsofNorthCarolinachartermoversagainstthequalicationsofteachersmovingbetweenmainstreamschools,controllingforreceivingschoolproles.Chartermoverswerelessexperiencedthanothermovingteachers 15

PAGE 16

ThesendingsneitherarmnorrejecttheeectivenesswithwhichNorthCarolina'scurrentchartermodeldrawsgoodteachersfrommainstreamschools.Thesystemattractedhighlyqualied,certied,andeectiveteachers,butlowlicensurerequirementsattracteduncertied,lessqualiedteacherswhomayhavehadfewcareeroptionsinthemainstreamsector. 1.2.1Background 1.2.2 )havebeencollectedoveraneleven-yearperiodforallmainstreamandcharterteachersinthestate.Thesedataallowmetocharacterizetheon-paperqualicationsofeveryteachermovingtothestate'schartersector,andtoestimatethe 16

PAGE 17

TheNorthCarolinalegislatureauthorizedthestate'ssystemofcharterschoolsin1996.Thereareseveralstatedobjectivesofthesystem,includingincreasedlearningopportunitiesforstudentsandnewprofessionalopportunitiesforteachers.7Thestate'srstthirty-fourcharterschoolsopenedforthe1998schoolyear.Table 1-1 documentsthegrowthofNorthCarolina'schartersystemfromthatyearupto2007.8Charterstudentsaccountedfor2.0percentofstatewideenrollmentby2007,andcharterteachersaccountedfor2.1percentofpublicschoolteachers.Figure 1-1 illustratesthewidespreadgeographicrangeandconcentrationofcharterenrollmentin2006.Charterswereactivethroughoutthestate,thoughlesssointheruraleasterncounties.Fourofthestate'slargestcounties{Durham,Forsyth,Mecklenburg,andWake{hadthelargestcharterpresenceintermsof 17

PAGE 18

AdistinctivefeatureofNorthCarolina'scharterschoolsystemisapronouncedracialsegregation.12Justoverhalfofthestate'scharterschoolshadraciallyimbalancedstudentpopulations,meaningthatthepercentofstudentswhoarenonwhitewastwentypercentagepointsaboveorbelowthecounty-widenonwhiterepresentation.Bycomparison,justone-quarterofmainstreamschoolswereraciallyimbalanced.Charterschoolsarerequiredto\reasonablyreecttheracialandethniccompositionofthe ( 2006 )fullyexploredthesegregatingeectsofNorthCarolina'scharterprogram. 18

PAGE 19

Charterschoolsareheldtothesameaccountabilitymodelasmainstreampublicschools,withsomeexceptionsforchartersintheirrstyear.Eachyear,NorthCarolinaschoolsareawardedrecognitionlabelsaccordingtotheirstudents'performanceandgrowthonend-of-gradeandend-of-courseexams.14In2006,53percentofcharterschoolsweregivenoneofthelowestthreerecognitionlabels,comparedto48.1percentofpublicschoolsstatewide( Manuel ( 2007 )).Butcharterswerewell-representedatbothextremesoftheperformancedistribution.In2006,charterswere2.6timesaslikelyasmainstreamschoolstohaveperformancecomposites(reectingthepercentofstudentsatgradelevel)lowerthan50,15and2.3timesaslikelytohaveperformancecompositesabove90.Thewidevarianceinracialcompositionandstudentprociencyamongcharterschools NorthCarolinaStateBoardofEducation ( 2006 ,HSP-CSeries))15Schoolswithsub-50performancecompositesthatfailtomakeexpectedgrowthbenchmarksaregiven\Low-Performing"recognition.Schoolswithrepeatlow-performingstatus,includingcharterschools,mustcollaboratewithevaluationteamsassignedbytheStateBoardofEducationtodevelopcorrectiveactionplans.Charterswithsub-60 19

PAGE 20

Charterschoolsareallowedgreatexibilityintherecruitment,retention,andpayoftheirfaculties.Thestateimposesverylittleregulationonwhocanteachinacharterschool.Atleast75percentofcharterteachersinkindergartenthroughfthgradeclassroomsmustholdteachingcerticates.Thisnumberfallsto50percentforcharterteachersofgradessixthroughtwelve.16Uncertiedteachersaremuchlesscommoninmainstreamschools.Onlycertiedteachersareeligiblefortenureafterfourconsecutiveyearsofteachinginamainstreampublicschool.Tenuredmainstreamteacherswhowishtoteachinacharterschoolaregrantedoneyear'sleave,meaningthattheycanreturntotheiroriginalschoolafterayear,spacepermitting.Chartersarenotrequiredtooertenure,noraretheyrequiredtoparticipateinthestateretirementplan. Lowlicensurerequirementsforcharterfacultieswereputinplacetoattractnew,nontraditionalteachersfromfreshsources-non-teachingvocations,TeachforAmerica,andsoforth.Recentworkby Kane,Rocko,andStaiger ( 2008 )and Boydetal. ( 2006 )suggestthereislittledierencebetweenthequalitydistributionsofcertied,alternativelycertied,orevenuncertiedteachers.17Statesvaryintheirtreatmentofteacherlicensureincharterschools.Ofthefortystateswithactivechartersystemsin2008,fteenrequired performancecompositesforthreeconsecutiveyearsaredenied10-yearrenewal.( NorthCarolinaStateBoardofEducation ( 2006 ,EEO-USeries))16Uncertiedcharterteachersareexpectedtomeetthefederaldenitionof\HighlyQualied."Althoughthisrequirementdoesnotappeartobestrictlyenforced,uncertiedteacherscanmeetthestandardbymajoringorpassingaPraxisIIexamintheirsubjectarea.17Wewouldexpectthistobetruewithinschoolsifadministratorshireequivalentlyskilledcandidatesregardlessoflicensure.Itmaybethecase,however,thatlowlicensurestandardsputdownwardpressureontheacross-schooldistributionofteacherquality. 20

PAGE 21

CenterForEducationReform ( 2008 )) Muschkin,Bonneau,andDodge ( 2008 ). 21

PAGE 22

Table 1-2 describesteachermobilitypatternsbetweencharterandmainstreamschoolsforthe5,346teacherswhowereworkinginacharterschoolatsometimebetween1998and2007.Themajority,55.1percent,wereneverobservedteachingoutsideofthechartersector.Another33.6percenttaughtinamainstreampublicschoolbeforemovingtoacharter.Ofthese,two-thirdsmoveddirectlytoacharterschool,withoutleavingthepanelbetweenschools.Theremainingthirdtaughtinamainstreamschool,leftthepanelforoneormoreyears,andthenre-enteredinacharterschool.Ifocusondirectmovers,whowerecontinuouslyemployedovertheirtransitiontothechartersector.Icomparethequalicationsofdirectchartermoverstothequalicationsofotherteachersmakingdirectmovesbetweenmainstreamschools.19Byevaluatingchartermoversagainstothermobileteachers(asopposedtopublicschoolteachersmoregenerally),Iavoidselectionbiasesfromomittedvariablescontributingtomobilityperse,andIcanjudgewhatsortoftalentcharterschoolsweredrawingfromthepoolofteacherswillingtochangeschools.Aftercontrollingforreceivingschoolcharacteristics,Idetermineifcharterschoolswererecruitingmoreorlessofeachspecicteachercharacteristicthantheirmainstream 1.3.1 and 1.3.2 arequalitativelysimilar,butwithvaryingstatisticalprecision,ifIadoptmoreliberaldenitionsofteachermobility.Thesection 1.3.3 analysisapproximatesthedistributionofpersistentteacherqualityamongallpre-charterteachers,whetherornottheytookbreaksbetweenschools. 22

PAGE 23

1.3.1QualicationsofTeachersMovingtoCharterSchools 1-3 listssummarystatisticsforNorthCarolina'smainstreampublicschoolteachersfrom1997to2007.Teacherswereidentiedasschoolpersonnelwithteachingassignmentsinschoolactivityreports,excludingteachingassistants,facilitators,andDAREocers.Ideterminedthehighestdegreeattainedbyeachteacher:30.8percentofteachersheldapost-baccalaureatedegreeofsomekind.Ateacher'sdegree-grantinginstitutionwas\competitive"ifitwasclassiedassuch(or\competitiveplus,"\verycompetitive,"etc.)bythe1995editionofBarron'sProlesofAmericanColleges.20Justoverthree-quartersofNorthCarolinateachersgraduatedfromacompetitivecollegeoruniversity.NorthCarolinateacherstakeavarietyoflicensureexams,mostofwhichareinthePraxisfamily.Inordertoincludeallavailabletestinformation,Iscaledrawlicensuretestscorestohaveameanofzeroandstandarddeviationofonewithineachtestcodeandtestyear.Icalculatedthemeanstandardizedlicensuretestscoreforeachteacher,equaltotheaverageofallofheruniqueexamsrecords.21RegularlylicensedteachershadcompletedanapprovedteachereducationprogramandpassedthePraxisSeriesofexams,orattainedlicensingbyreciprocalorinterstateagreement.Thecomplementstoregularlylicensedteacherswereuncertiedteachersholdingtemporary,emergency,or 23

PAGE 24

1-3 ,wereearlierintheircareers,onaverage,andlesslikelytohaveagraduatedegreethanteacherswhowerenotchangingschools.24Mobileteachershadlowerlicensuretestscoresthannon-movers,by0.015standarddeviations.Mainstreamteachersmovingtocharterschools,summarizedinthefourthcolumnofTable 1-3 ,weretypicallylessqualiedthanothermovingteachers.Teacherstransitioningtothechartersystemwere4.5percentagepointslesslikelytohavegraduatedfromacompetitivecollegeoruniversity,and9.9percentagepointslesslikelytoberegularlylicensed.25NorthCarolina'spolicyofpermittingmoreuncertiedteachersincharterschoolsmayhavehadtheconsequenceofdrawinguntenuredmainstreamteachers Feller ( 2006 )).Indvirtuallynochangeinteachercerticationratesinthelateryearsofthepanel.23Thisaectedanon-trivialpercentofteacher-yearobservations(6.1),includingteacherswithrovingassignmentsandteacherswhoswitchedschoolsmid-year.24Amobileteacherisdenedthroughoutasoneobservedinschoolainyeartandschoola06=ainyeart+1.25Chartermoverswerealsomorelikelythanmainstreammoverstohavemissinglicensedata(8.1versus4.8percent)ThemainresultsareunaectedifIclassifytheseindividualsasunlicensed.Amoreproblematicdataqualityissueisthehighrateofmissinglicensuretestscoresamongchartermovers(11.8versus7.8percentformainstreammovers).Below,IdiscusswherethiscouldaectresultsandthesensitivitytestsIusedtoevaluatepotentialbiases. 24

PAGE 25

Figure 1-2 illustratescomparativekerneldensitiesfortheteachingexperienceofmobileteachers,bycharter/mainstreamdestination.Clearly,chartermoversweremorelikelytohavejustafewyearsofexperience,relativetomainstreammovers.Theywerealsomorelikelytohavearound30yearsofexperience.InthelowerpanelofFigure 1-2 ,limitedtoregularlylicensedmovers,thebimodalityofcharterteachers'experienceismorepronounced.Thedistributionofchartermovers'licensuretestscoresinFigure 1-3 alsohintsatanoisybimodality,butmoreimportantly,thedistributionoflicensedchartermovers'licensurescores(lowerpanel)appearstobetotherightofmainstreammovers'distribution.Chartermovers,particularlythosewithregularlicensure,weresomewhatmorelikelythantheirmainstreamcounterpartstohavehighlicensuretestscores,0.5-2.0standarddeviationsabovethemean.Thevisualdierenceisnotentirelyduetonoise.Wilcoxonrank-sumtestsindicatethatscoredistributionswerenotsignicantlydierentbetweencharterandmainstreammoversoverall,butthatlicensedchartermovers'scoresweresignicantlygreaterthanthoseoflicensedmainstreammovers. Thenonparametricevaluationofchartermovers'qualications,togetherwiththeparametricanalysestofollow,providearich,descriptivepictureofthevaluethatteacherstookwiththemwhentheymovedtothechartersector.Butthispictureisincompletewithoutanunderstandingoftheschoolstheseteacherweremovingbetween.Ifchartermoverswerehighlyqualiedrelativetoothermobileteachers,butleavinglow-performingschools,thiswouldhaveverydierentpolicyimplicationsthaniflessqualiedteachers 25

PAGE 26

Figures 1-2 and 1-3 oervisualdepictionsoftherangeofexperienceandachievementthatteachersbroughtwiththemwhentheymovedtothechartersystem,butcomparativekerneldensitiesdonotpermittheconclusionthatcharterswereattractingmoreorlessqualiedteachersthansimilarmainstreamschools.Towardthatend,IconductmoreparametricanalysesofcharterandmainstreammoversbyestimatingEquation 1{1 viaordinaryleastsquaresforeachNorthCarolinateacher(j)observedinyeart(1997-2007),schools,andcountyl: Equation 1{1 isareducedformexpressionforqualicationk,wherekindexestheon-paperqualicationssummarizedinTable 1-3 :graduatedegree,competitivecollegeeducation,meanlicensuretestscore,regularlicensure,andthreemeasuresofexperience.26 26Equation 1{1 wasestimatedseparatelyforeachqualicationk.Analternativewouldhavebeentoprojectteachers'mobilityontothespaceoftheirqualicationsandsendingschoolcharacteristicstogetasenseofthefactorsaectingthesupplyofcharterschoolteachers.Iemphasizethereduced-formempiricalstrategytounderscorethedescriptive,non-causalinferencegainedbyexaminingarelativelysmallsetofidiosyncraticlabordecisions.Asarobustnesscheck,Ialsoestimatedamultinomiallogitequationpredicting 26

PAGE 27

Moreexperiencedteachersmayseekgraduatedegreesoradditionalcerticationstoincreasetheirpay,soIcontrolledforteacherexperiencecategories(indicatorsforless thelikelihoodofdierenttypesofschoolchanges.Resultssuggestedthattherelativeriskofmovingtoacharterschoolsignicantlyincreasedforlessexperiencedandunlicensedteachers,andsignicantlydecreasedforteacherswithhigherlicensuretestscores.Thesendingsareinagreementwiththereduced-formresultstofollow.27ResultswerequalitativelysimilarwhenIcontrolledforsendingandreceivingschoolcharacteristics.Specicationswithreceivingschoolcharacteristicsalonearepreferred,astheybetterdepicttherelativeowofteacherqualicationstocharterandmainstreamschoolsofcomparablesizeandstudentcomposition. 27

PAGE 28

1{1 forlicensureandeducationvariables.Sincelicensedandunlicensedteachersmayhavedierentincentivestoconsidercharterschools,IlimitedEquation 1{1 toregularlylicensedmoversandproducedseparate\licensedmover"estimatesofmjtandcjtforallqualicationsexceptlicensureitself. Table 1-4 listsestimatesofcjtandmjtforeachresumequalication.28TherstcolumnofTable 1-4 listscoecientestimatesformjt,thetypicaldierenceinqualicationkbetweenteachersmovingtomainstreamschoolsandnon-movingteachers,controllingforreceivingschoolcharacteristics.Estimatesofmjtserveasthebaselinetowhichcjtestimatesarecompared.Moversweresignicantlydierentthannon-moverswithrespecttoeachqualications,withtheexceptionofgraduatedegrees.Moversweremuchlessexperienced,by3.69yearsonaverage,thantheirnon-movingcounterparts.Theywere13.2percentagepointsmorelikelytohavethreeyears'experienceorless,and10.2percentagepointslesslikelytohaveatleasttwenty-veyears'experience. ThesecondcolumnofTable 1-4 presentsestimatesofcjtfromEquation 1{1 .ColumnIIcoecientsanswerthequestion,\werechartermoversmoreorlessqualiedthanteachersmovingtocomparableschools?"Thequalierisimportant,giventheheterogeneityofcharterschoolworkingenvironments.Withrespecttograduateeducation,licensure,andyearsofexperience,chartermoversweresignicantlylessqualied.Theywere2.7percentagepointslesslikelytoholdagraduatedegree,relativetomainstream 28

PAGE 29

1-2 ColumnIandIIresultsweregeneratedfromanalysesofallpublicschoolteachers,regardlessoftheirlicensurestatus.Fullylicensedteachersmayhavehadmoreoptionsinthemainstreamsectorthanunlicensed,untenuredteachers.Accordingly,licensedteachers'mobilitydecisionsbetterrelectrevealedpreference.ColumnsIIIandIVlistresultsfromthesubsampleoflicensedteachers,whoaccountedfor89.0and79.1percentofmainstreamandchartermovers,respectively.Limitingthesamplehadlittleeectonresultsformainstreammovers;pointestimateswerenoteconomicallydierentbetweencolumnsIandIII.Butexcludingunlicensedteachersfromtheanalysissubstantiallyaectedconclusionsabouttherelativequalicationsofchartermovers.IncontrasttocolumnII,columnIVresultsshowthatlicensedchartermoverswerenotstatisticallydierentthanlicensedmainstreammoverswithrespecttograduatedegreesoryears'experience.Thedierencebetweenthefullandlimitedsampleisparticularlystarkforlicensuretestscores.Thecoecientforlicensedchartermoversinthetestscoreequationispositiveandsignicant.29Teachersmovingtocharterschoolstypicallyhadhighertestscoresthan 29

PAGE 30

Itissensiblethatcharterschoolswouldhavebeenabletoattractteacherswithhigherlicensuretestscores.Conditionalonlicensureitself,testscoresarenotrewardedinthestate'spayscale.Yetateacher'stestscoresaregoodindicatorsofhowwellhisorherstudentswilldoontheirowntests. Goldhaber ( 2007 )and Clotfelter,Ladd,andVigdor ( 2007 )haveshownthatNorthCarolinateacherswithhigherlicensuretestscoreswereassociatedwithhigherstudentachievementonend-of-grademathandreadingexams.Charterschools,perhapsrecognizingteachertestscoresasgoodsignalsofteacherquality,hadmoresuccessrecruitingindividualswithhighertestscoresthancomparablemainstreamschools. Thesendingsraisethepossibilitythatteachersviewedthechartersectorasalow-costjobchangeprecedingretirementorapermanentcareerchange.Sampleattritionwashighamongnewteachers,experiencedteachersnearingretirement,anduncertiedteachers,30andthesearethesamegroupsIobserveddisproportionatelyowinginto wouldbebiasedupwardsfromthetrueaverageofunderlyingknowledge.Ifchartermoverswithmissingtestdatacamefrommuchlowerintheunderlyingdistributionthanmainstreammoverswithmissingdata,theestimateddierencebetweencharterandmainstreammovers'testscoreswouldbebiasedinfavorofthechartermovers.Isimulatesituationslikethistogagethesensitivityofchartermovers'testscoreadvantagetovariouscounterfactualscoresforteacherswithmissingdata.Theresultthatlicensedchartermovershadsignicantlyhighertestscoreswasrobustuptoa0.5standarddeviationpenaltyforchartermoverswithmissingtestdata.Morethana2.0standarddeviationgapwasnecessarytoproducedtheresultthatlicensedchartermovershadsignicantlylowertestscoresthanmainstreammovers.30Therewere89,311uncensoredsampleexitsintheNorthCarolinateachers'panel.Ofthese,28.8percenthadthreeorfeweryears'experience,37.7percenthadatleasttwenty-ve,and30.6percentwereuncertiedorhadmissinglicensedata. 30

PAGE 31

Theresultsdiscussedinthissectionlendsomesupporttotheideathatcharterschoolshadarealizedadvantageinthelabormarketforpublicschoolteachers;amonglicensedteacherschangingschools,charterswerebetterabletoattracthighlyexperiencedteachers,andteacherswithhighlicensuretestscores.31Butalargeminorityofmainstreamteachersmovingtocharterschoolswerenotfullylicensedandattenuatedtheaveragequalicationsofchartermovers.Thiswaslikelyaconsequenceofthestate'slowlicensurerequirementsforcharterschools.Whatremainstobeseenisifthemigrationofuncertiedteacherstothechartersectorwaspredominantlydrivenbythecharters'demandforlow-costlabor,orbyuncertiedteachers'willingnesstosupplyit. Someresumeline-itemslikelicensuretestscoresarerobustsignalsofteacherquality,andcharterschoolswereeectivelycompetingforlicensedteacherswithrelativelyhightestscores.Butwerecharterschoolsdrawingteacherswithhistories,andnotjustsignals, 31

PAGE 32

Theproctorassociatedwitheachstudent'stestscorewasnotnecessarilyhisorherclassroomteacher.Tominimizethelikelihoodofbadteacher-studentmatches,Ifocusedonteacherswithself-containedclassroomsofstudentsingrades3-5.Self-containedclassroomsembodythetraditionalstructureofelementaryeducation,whereaclassofstudentsspendallorthemajorityofeachdaywithoneteacher.Iassembledgrade3-5studentEOGrecordsformorethan2.8millionstudent-yearsspanning1997to2007.Ateacher-studentmatchwasconsideredinvalidifanyofthefollowingfourconditionsweremet.Inparenthesesarethepercentofstudentsforwhicheachconditionwastrue. 32

PAGE 33

Thestudent'sproctorwasunknownornotfoundintheassembledteacherpanel.33(19.5%) 2. Thestudent'sproctordidnothaveaself-containedclassroomassignment.(21.8%) 3. Thegrade-gstudent'sproctordidnothaveateachingassignmentwithstudentsingradeg.(<1:0%) 4. Thestudent'sexamgroupwaslargerthan30orsmallerthan5.(<1:0%) Theremaining58.3percentofstudentshadaproctorwhowasateacher,andwholedaself-containedclassroomwithstudentsinthesamegradeasareasonablenumberofEOGtest-takerslinkedtothatteacher.Theselimitationslendconsiderablevaliditytoeachallowedteacher-studentmatch.34Ofthe122,064EOGtest-takingclassroomswithaknownteacher,71.3percentwereconsideredvalidmatches.NorthCarolina'send-of-gradeexamsareinterval-scaled,meaningthataone-pointincrementreectsthesamedierenceinlearninganywhereonthescaleofrawscores.Scoresarecomparablewithinandacrossgradeseachyear,andtheminimumprocientscorerisesforeachgrade.Iscaledrawscorestohavemeanzeroandstandarddeviationequaltoonewithineachyear,grade,andschool.Thiscalibratedthedependantvariableofeacheducationproductionfunctiontoaccountforvarianceintherangeofrawscoresovertime,heteroscedasticityinrawscoresacrossgrades,anddistributionalshiftsinstudentperformanceacrossschoolsdueto Tiebout ( 1956 )sorting. 33

PAGE 34

VariablesinAictarestudentcharacteristics,includingrace,gender,parentaleducation,andlearningdisabilityindicators.Aictisavectorofaveragestudentcharacteristicsini'sclassroom(excludingstudenti),andTjctcontrolsfortwomeasuresofteacherinexperience.Xsltcontainsschool-levelvariables,includingquintileindicatorsforstudentbodysizeandcomposition,gradelevels,andadummyvariableequaltoonewhenstudentiisinanewschool.Thecoecientsjandltareteacherxedeectsandcounty-by-yeareects,respectively.35IestimatedEquation 1{2 andsavedestimatedteacherxedeects,^j.CoecientestimatesforEquation 1{2 wereunsurprising.Femalestudentshadlowermathbuthigherreadingscoresthanmales.Nonwhitestudentshadlowerscoresinbothsubjects,asdidstudentswithoutcollege-educatedparents.Learningdisabilitieswerestronglyassociatedwithlowerscores,moresofordisabilitiesdirectlyrelatedtothetestedsubject.Studentswithinexperiencedteachershadlowertestscoresinbothsubjects,especiallyiftheirteacherwasinherrstyearasopposedtohersecondorthird.Thepenaltyfromteacherinexperiencefellabout80percentfollowingateacher'srstyear. Inthissetting,teacherxedeectsareinterpretedaseachindividual'shistoryofclassroomperformancerelativetoexpectations,giventhecompositionofherstudents,intraschoolsorting,andtheteacher'sownexperience.Thisshouldbeimportanttoschoolslookingtohireteacherswitharecordofsuccessinraisingstudenttestscores,butdoesnotnecessarilypermittheinterpretationof^jasatransitiveindexofteachers'inherentqualityorvalueadded.Thelatterviewreliesontwostrongassumptions:(1)^j 34

PAGE 35

CameronandTrivedi ( 2005 )).Althoughteacherxedeectestimatesbenetfrommultiplestudent-levelsignalseachyear,niteclasssizeleadstoconsiderablesamplingerror.Teacherxedeectestimatesarenoisy,andtheirvarianceoverstatesthetruevarianceinteacherquality( Rocko ( 2004 )).Iaddressthisinthefollowingsubsectionbyisolatingthevarianceinpersistentteachervalueadded.Thesecondassumptionisinvalidifthereareunmeasuredstudentvariablesaectingtestscores,likemotivationorinherentintelligence,andifthesevariablessystematicallyaecttheteachertowhomastudentisassigned.Inthatcase,estimatesofjwillbeareectionofteacherqualityandstudentsorting.Positivematching,suchthatbetterstudentswereassignedtobetterteachers,wouldbias^jawayfromzeroandoverstateateacher'seectivenessorineectiveness.Negativematching,whichmaybethecaseifbetterstudentswereassignedtostrugglingteacherstoeasetheirburden,wouldbias^jtowardzero. Clotfelter,Ladd,andVigdor ( 2006 )foundevidenceofteacher-studentmatching,particularlypositivematching,inNorthCarolinaschools.Thebulkofstudentsortingwas Tiebout ( 1956 )sortingbetweenschools,whichIaddressedbycenteringtheEquation 1{2 dependantvariablebyyear,grade,andschool. Nonetheless,Equation 1{2 failstocontrolfornonrandomsortingofstudentswithinschools,suchaswouldbethecaseifparentsweresuccessfullylobbyingschooladministratorstoputtheirchildreninparticularclassrooms.36TwoadaptationstoEquation 1{2 -estimatingstudentgains,orincludinglaggedstudenttestscores-wouldaddresslikelypathwaysbywhichstudentswerenon-randomlymatchedtoteacherswithinschools.Eithermethodwouldeliminate3rdgradeteachersfromtheanalysishere,an Hui ( 2003 ), Crombie ( 2001 ),and Clotfelteretal. ( 2006 )) 35

PAGE 36

Rothstein ( 2008 )),norwillitcircumventtheinherentsamplingerrorofteacherxedeects.Itisimportanttoemphasizethatsortingbiaseswouldonlyhaveaectedtheanalysistothedegreethatchartermoversweredisproportionatelysubjecttonon-randomwithin-schoolstudentsortingpriortotheirmove.Estimatedteacherxedeectsprovideinsighttotherelativeperformanceofteachers'classrooms,whichwouldbeofinteresttopotentialreceivingschools. Iestimatedmorethan28,000teacherxedeectsforbothsubjects.Therewere13,752mobileteachersinthesample,257ofwhichweremovingtoacharterschool.Teacherxedeectsreectteachers'relativeperformancewithintheirschools.Thislimitsthescopeofinterpretationandunderstatesthevarianceinteacherqualityacrossschools,butadequatelyaddressesbetween-school Tiebout ( 1956 )sorting.38Table 1-5 summarizesteacherxedeectsestimates.Teachersmovingtoothermainstreamschoolshadlowerxedeectsthannon-movers,onaverage.Chartermovershadevenlowerxedeectsthanmainstreammovers,by3.7percentofastandarddeviationinmath.Forcontext,thecharter-mainstreammovergapinxedeectsrepresentedabout78.5percentofthe0.047standarddeviationgapbetweenmaleandfemalemathperformance,asestimatedby 1{2 withlaggedstudentachievement.Thisreducedthenumberofchartermoversforwhomxedeectscouldbeestimated,butnonetheless,thendingthatchartermovershadrelativelyhighwithin-schoolmathxedeectswasrobust.Resultsforreadingxedeects,however,werestatisticallyinsignicant.38Averagexedeectswere24-38percentlargerinabsolutevaluewhenstudents'rawscoreswerecalibratedtomeanstatewideperformanceratherthanmeanschool-wideperformance. 36

PAGE 37

1{2 .Socharterschoolswerenotdrawingmainstreamschools'bestteachers,asmeasuredbyteacherxedeects. SummarystatisticsfromTable 1-5 indicatethatteacherswhomovedtocharterschoolswererelativelylowinthedistributionofteacherqualitywithintheirsendingmainstreamschools.Butthesesimplemeansdonotcontrolforthetypeofschoolsteachersweremovingto,andcharterschoolsmayhaveattractedrelativelyhigh-performingteachers,comparedtotheowoflaborgoingtocomparablemainstreamschools.Iregressedteacherxedeectestimatesagainstmobilityindicators,receivingschoolcharacteristics,andreceivingcounty-by-yeareects: ^kj=mjt1(moving)+cjt1(tocharter)+Xrjsl(t+1)r+l(t+1)+"jslt(1{3) Subjects(mathandreading)areindexedbyk,teachersbyj,schoolsbys,countiesbyl,andyearsbyt.Table 1-6 presentsestimatesofmjtandcjt.ColumnIliststheestimateddierenceinxedeectsbetweenmainstreammoversandnon-movers(^mjtinEquation 1{3 ).Mobileteachersmovingbetweenmainstreamschoolstendedtohavelowermathandreadingxedeectsthannon-movingteachers,by1.8and1.4percentofastudent-levelstandarddeviation,respectively.ColumnIIliststheestimateddierenceinclassroomperformancebetweencharterandmainstreammovers.Chartermovers'xedeectswereestimatedtobe4.5percentofastandarddeviationhigherinmathand4.0percenthigherinreading,relativetothoseofteachersmovingtocomparable,albeitmainstreamschools.Equation 1{3 coecientsindicateareverseofthecharter-mainstreammoverperformancegapobservedinthedescriptivestatisticsofTable 1-5 .ColumnsIIIandIVlistanalogousresultsforthesubsampleofregularlylicensedteachers.Pointestimatesforcharterandmainstreammoverswerelargelyunchangedbyexcludingunlicensedteachers.39 39Therateofregularlicensurewasmuchhigheramongelementary-gradeteachers. 37

PAGE 38

38

PAGE 39

1{2 ,omittingteacherxedeects(j). Suppressingnotationforgradeandschool,theerrorsareeijct=j+"ijct.Therearetwocomponentstoeachstudentresidual,eijct:teacherj'spersistentvalue-added(j),andnon-persistentnoise("ijct)encompassingsamplingerrorandtransientshockstoaverageclassroomperformance.Theaveragestudentresidualforeachclasscanbeexpressedlikeso:^ejct=j+1 1{4 maybiasothercoecientsiftheyarecorrelatedwithj;this,inturn,willbiasestimatedresiduals,^eijct.CalibratingthedependantvariablebyschoolinEquation 1{4 limitsbiasesfrombetween-schoolsorting,butwithinschoolteacher-studentmatchingpatternsmaynonethelessaect2estimates.Solongasthecorrelationbetweenjand"jctisnotsystematicallydierentforsubsamplesofinterest,thecalculatedvarianceof 39

PAGE 40

CarrellandWest ( 2008 ),Iestimate2bycomputingtheaveragecovarianceofallclassroomresidualpairsbetweenteacherj'sclasscinyeartandc06=cinyeart06=t: ^2=24JXj=1CjXc=1^ejct^ejc0t035=N(1{5)Jisthenumberofteachers,Cjisthenumberofclassestaughtbyteacherj,andNisthenumberofpairs. Table 1-7 presentsestimatesoftotalandsignalstandarddeviations.40Allstandarderrors(inparenthesesbeloweachstandarddeviationestimate)wereestimatedbybootstrap,withanequalnumberofcharterparticipantsandnon-participantsselectedineachsampling.ThethirdandsixthcolumnsofTable 1-7 listthestandarddeviationofteacherxedeects,bygroup.Asexpected,estimatesofthevariationinpersistentteacherresiduals(signal)weremuchsmallerthanthevariationinteacherxedeects(SD(FE)).Thelattersuggeststhataone-standard-deviationincreaseinteacherqualityimprovedstudentmathperformanceby0.237standarddeviations,threetimesthebonusfromhavinganexperiencedteacherratherthananewone.Butjudgingbythepreferredmeasureofdispersion,aone-standard-deviationincreaseinteacherqualitywouldyieldastill-substantial0.179standarddeviationincreaseinstudentmathperformance,closertothedierencebetweenhavingacollege-educatedparentversusaparentwith\somecollege."Estimatesofsignalvarianceweresomewhatsmallerforreading;aone-standarddeviationincreaseinpersistentteacherqualitywaspredictedtoincrease 1-7 40

PAGE 41

1-7 alsolistssignalstandarddeviationsseparatelyforfuturecharterteachersandteacherswhowereneverobservedinacharterschool.Thevarianceofmathperformancewaswiderforfuturecharterteachers,butthevarianceofreadingperformancewasnarrower.Bothstatisticswerewithintwostandarderrorsofthecorrespondingsignalestimateforexclusivelymainstreamteachers. Following Kaneetal. ( 2008 ),IconstructedasimpleBayesianshrinkageestimatortoaccountforsamplingerrorinclassresidualsattributedtoteacherquality.IusedestimatesofsignalandnoisevariancelistedinTable 1-7 ,alongwiththenumberofclassesobservedforeachteacher(Cj)toscaleaverageclassresiduals(ej): ~j=ejCj Equation 1{6 shrinkseachteacher'saverageresidualtowardszeroaccordingtothetermsinparentheses.Residualsforteacherswithmoreclassesandgroupswithlargersignal-to-noiseratioswerescaledbyless,sincetheirresidualswereexpectedtobelessaectedbysamplingerror. Insection 1.3.1 ,Ishowedthatahighrateofchartermoverswereuncertiedandunder-qualied.Thiswouldnotbeproblematicforcharterschoolsifcertiedanduncertiedteacherswerepartofthesameunderlyingdistributionofteachingquality,solongaschartersdidnotdrawheavilyfromthelowerend.Inthatcase,lowlicensurerequirementsmaybeonewaytoretaineectiveteachersinpubliceducation.Inagreementwith Kaneetal. ( 2008 ),Indlittleobservabledierenceinteacherqualitydistributionsbetweenlicensuregroups,butafairlywide(albeitscaled)variancewithineachgroup.Forbothsubjects,Wilcoxonrank-sumtestsfailedtorejectthehypothesesthatteacherswithregularandtemporarylicensesweredrawnfromthesamedistribution 41

PAGE 42

1-4 plotscomparativedensitiesofteacherqualitybycharterparticipation.Thepersistentqualitydistributionforfuturecharterteacherswassignicantlylowerthanthatforexclusivelymainstreamteachers,especiallyformath.Figure 1-4 providesfurtherevidencethatteachersowingintothechartersectortypicallyhadlowerrelativeclassperformancewithintheirschoolsthanexclusivelymainstreamteachers. Goldhaber ( 2007 )and Kaneetal. ( 2008 )),althoughinNorthCarolinatheyareassociatedwithlowerstudentachievement( Clotfelteretal. ( 2007 )).Indnosignicantdierenceinpersistentteacherqualitybetweenbroadlicensuretypes,butawidevarianceofqualitywithineachtype.Ideally,charterschoolswouldrecruituncertiedteachers 42

PAGE 43

Foralimitedsampleofgrade3-5teachers,Iestimatemeasuresofteacherqualityusingclassroomperformanceonstandardizedend-of-gradeexams.Werecharterschoolscreamskimmingmoreeectiveteachers?Charterswerenotskimmingabove-averageteachersfromthestockofschoolfaculties,althoughaccordingtothemoreparametricclassroomperformanceanalysesofsection 1.3.2 ,chartersmayhavebeendrawinghigher-rankedteachersfromtheowofteacherschangingschools.Thisisnottosaythattheteacherswhomovedtocharterschoolsheldacommondeciencyinclassperformance.Ishowthatthedistributionoffuturecharterteacherquality,evenwhenitspersistentcomponentwasformallydissected,waswideandlargelyoverlappedthequalitydistributionofteacherswhonevertaughtinacharterschool. 43

PAGE 44

NorthCarolinacharterschools,students,andteachers,1998-2007 YearSchools(%)Students(%)Teachers(%) 199834(1.7)4,642(0.4)330(0.5)199957(2.7)8,555(0.7)601(0.8)200076(3.6)12,691(1.0)862(1.1)200190(4.1)15,523(1.2)1,086(1.4)200293(4.2)18,235(1.4)1,292(1.6)200393(4.2)20,420(1.5)1,390(1.7)200493(4.1)21,955(1.6)1,509(1.8)200597(4.3)25,248(1.8)1,669(1.9)200697(4.2)27,441(1.9)1,789(2.0)200792(3.9)27,700(2.0)1,894(2.1) Notes:Eachcountofcharterschools,students,andteachersrepresentstheindicatedpercent(%)ofallpublicschools,students,orteachersinthesample.Teacherandschoolcountsweretabulatedfromtheteachers'panel,describedfullyinsection 1.2.2 .StudentcountsweretabulatedfromNCESCommonCoredata. 44

PAGE 45

2006Charterenrollmentandpenetration.MajorpopulationcentersinCumberland(C),Durham(D),Forsyth(F),Guilford(G),Mecklenburg(M),andWake(W)counties. 45

PAGE 46

In-samplemobilitypatternsofcharterteachers TeacherMobilitypatternPercent Startedandendedinthechartersystem(rightcensored)21.5Startedandendedinthechartersystem(uncensored)33.6Mainstreamtocharter25.1Mainstreamtochartertomainstream8.5Chartertomainstream10.7Otherpatterns<1.0 Notes:n=5,346teachers.Thersttwomobilitypatternsapplytoteacherswhotaughtexclusivelyincharterschools.Rightcensoredcharterteachersenteredthesampleinthechartersystemandwerestillteachingtherein2007,thelastyearofthepanel.Uncensoredteachingspellsendedbefore2007.Thefollowingfourmobilitypatternsapplytoteacherswhotaughtincharterandmainstreamschools.Thepercentofallcharterparticipantswhofollowedeachpatternisindicatedatright. 46

PAGE 47

NorthCarolinapublicschoolteachers:Summarystatistics AllAllMainstreamCharterTeacherqualicationteachersmoversmoversmovers Holdsgraduatedegree(%)30.827.527.525.2(46.2)(44.6)(44.7)(43.4)Attendedcompetitivecollege(%)76.275.075.170.6(42.6)(43.3)(43.3)(45.6)Meanlicensuretestscore0.0300.0150.0150.029(0.857)(0.839)(0.838)(0.905)Regularlylicensed(%)89.588.889.079.1(30.7)(31.6)(31.3)(40.7)Teachingexperience(years)11.938.898.927.45(9.92)(8.72)(8.71)(8.97)Experience3years(%)26.136.336.147.5(43.9)(48.1)(48.0)(50.0)Experience25years(%)14.57.57.58.9(35.2)(26.3)(26.3)(28.5)Nonwhite(%)16.918.117.926.4(37.4)(38.5)(38.3)(44.1)Female(%)79.879.879.879.2(40.1)(40.2)(40.2)(40.6)n(teacher-years)886,34358,62957,4871,142 Notes:Standarddeviationsappearinparenthesesbeloweachmean.Dataformovingteachersareevaluatedintheyearimmediatelyprecedingaschoolchange. 47

PAGE 48

B Densityestimates:Years'experienceofallmobileteachers(A)andoflicensedmobileteachers(B).DensitieswereestimatedusingEpanechnikovkernelfunctionsandhalfwidthsof0.50years. 48

PAGE 49

B Densityestimates:Meanlicensuretestscoresofallmobileteachers(A)andoflicensedmobileteachers(B).DensitieswereestimatedusingEpanechnikovkernelfunctionsandhalfwidthsof0.05standarddeviations. 49

PAGE 50

Regressionresults:Qualicationsofteacherschangingschools,bylicensureandcharter/mainstreamdestination IIIIIIIVSampleAllAllLicensedLicensedDestinationMainstreamCharterMainstreamCharter(Equation 1{1 coecient)(^mjt)(^cjt)(^mjt)(^cjt) Graduatedegree0.003-0.0270.007-0.028(1.41)(2.03)(3.06)(1.70)Competitivecollege-0.011-0.016-0.0060.003(5.63)(1.20)(2.88)(0.21)Licensed-0.012-0.074(8.53)(6.15)Meanlicensuretestscore-0.0270.019-0.0230.094(7.00)(0.68)(5.77)(3.33)Experience(years)-3.69-1.07-3.66-0.44(81.75)(3.52)(76.29)(1.21)Experience3years0.1320.0700.1200.049(58.29)(4.58)(51.97)(2.90)Experience25years-0.1020.070-0.1000.059(65.22)(5.86)(60.91)(4.37) Notes:n=886,343teacher-years.ColumnIliststheestimateddierenceinqualicationkbetweenteachersmovingtomainstreamschoolsandnon-movers(mjtinEquation 1{1 ).CellsincolumnIIlisttheestimateddierenceinkbetweencharterandmainstreammovers(cjt).ColumnsIIIandIVpresentthesesamecoecientsforthesubsampleofregularlylicensedteachers.Controlvariablesincludedreceivingschoolcharacteristics(studentracialcomposition,performancecomposite,schoolsize,schoolage,graderangesserved),asetofdummyvariablesformissingdata,andcounty-by-yeareects.Theabsolutevaluesoft-statisticsarereportedinparenthesesbeloweachcoecient.Robuststandarderrorswereclusteredwithineachschoolandyear. 50

PAGE 51

NorthCarolinapublicschoolteacherxedeects:Summarystatistics AllAllMainstreamCharterFixedeectestimatesteachersmoversmoversmovers Math-0.016-0.036-0.035-0.072(0.258)(0.262)(0.263)(0.253)Reading-0.013-0.029-0.028-0.057(0.225)(0.227)(0.227)(0.242)n(teacher-years)167,24413,75213,495257 Notes:Cellsrepresentaveragexedeectestimates,bysubjectandmobilitystatus.Standarddeviationsappearinparenthesesbeloweachmean.Dataformovingteacherswereevaluatedintheyearimmediatelyprecedingaschoolchange. 51

PAGE 52

Regressionresults:Classroomperformanceofteacherschangingschools IIIIIIIVSampleAllAllLicensedLicensedDestinationMainstreamCharterMainstreamCharter(Equation 1{3 coecient)(^mjt)(^cjt)(^mjt)(^cjt) Math-0.0180.045-0.0170.044(7.20)(2.64)(6.76)(2.35)Reading-0.0140.040-0.0140.044(6.79)(2.41)(6.36)(2.43) Notes:n=167,244teacher-years.ColumnIliststheestimateddierenceinxedeectsbetweenmainstreammoversandnon-movers(mjtinEquation 1{2 ).ColumnIIliststheestimateddierenceinxedeectsbetweencharterandmainstreammovers(cjt).ColumnsIIIandIVpresentthesesamecoecientsforthesubsampleofregularlylicensedteachers.Unreportedcontrolvariablesincludereceivingschoolcharacteristics(studentracialcomposition,performancecomposite,schoolsize,schoolage,graderangesserved),asetofdummyvariablesformissingdata,andcounty-by-yeareects.Theabsolutevaluesoft-statisticsarereportedinparenthesesbeloweachcoecient.Robuststandarderrorswereclusteredwithineachschoolandyear. 52

PAGE 53

Variationinteacherquality SubjectMathMathMathReadingReadingReadingMeasureofvarianceTotalSignalSD(FE)TotalSignalSD(FE) Allteachers0.2910.1790.2370.2630.1370.205(0.009)(0.017)(0.013)(0.008)(0.015)0.012Neveracharterteacher0.2940.1560.2360.2600.1360.205(0.011)(0.019)(0.018)(0.008)(0.016)(0.011)Futurecharterteacher0.3070.1830.2550.2760.1320.232(0.009)(0.015)(0.008)(0.008)(0.014)(0.009) Notes:n=167,244teacher-years.Studentmathandreadingscoreswereregressedagainststudentcharacteristics,teacherexperienceindicators,peercharacteristics,andschool-by-yeareects(Equation 1{4 ).\Total"isthestandarddeviationofstudentresidualsfromEquation 1{4 estimates.\Signal,"calculatedbyEquation 1{5 ,isthestandarddeviationofteachers'persistentvalue-added,bygroup.\SD(FE)"isthestandarddeviationofteacherxedeects,estimatedbyEquation 1{2 .Standarderrors,inparenthesesbeloweachstandarddeviationestimate,wereestimatedbybootstrapwithanequalnumberofcharterparticipantsandnon-participantsselectedineachsampling. 53

PAGE 54

B Densityestimates:Persistentteacherqualityinmath(A)andreading(B)forfuturecharterteachersandexclusivelymainstreamteachers.DensitieswereestimatedusingEpanechnikovkernelfunctionsandhalfwidthsof0.025scaledstandarddeviations. 54

PAGE 55

Charterschools,particularlynewones,arestaedbyhighratesofnewteachers,andthismayhelptoexplainwhytheirstudentsstruggletomeetexpectations.Inexperiencedcharterfacultiesarewidelyacknowledged,buttodate,nostudyhaslinkedteacherexperiencetostudentperformanceincharterschools.IexaminewhetherinexperiencedfacultiesaectedstudentachievementinNorthCarolinacharters,usingatwelve-yearpanelofstudenttestdata.Theshareofnewteachersinthestate'scharterfacultieswasquitehighinnewschools,butfellsharplyastheschoolsaged.Consistentwithearlierresearch,Indsignicantreturnstocharterschoolage,butthismaturationcouldnotbeattributedtodecliningratesofnewteachers.Rather,charterstudentsbenetedfromhigherratesofnewteachersinrecently-newschools,moresoinmaththanreading.Thesendingssuggestthatinexperiencedfacultieswerenotarootcauseoflowstudentachievementinnewcharterschoolsandmayhavebeensymptomaticofdevelopment. NationalCenterforEducationStatistics ( 2004 )and Braun,Jenkins,andGrigg ( 2006 )).Thesestudiesusedcross-sectionalsnapshotsoffourth-gradestudentsin2003,anddespiteextensivecontrolsforobservablestudentcharacteristics,ndingsmayhavebeeninuencedbyselectionbias.Enrolling 55

PAGE 56

( 2007 )showedthatTexasstudentsincurredsignicantpenaltiesinnewcharterschools,butinschoolsolderthantwoyears,studentgainswerenotstatisticallydierentfromthoseofmainstreamstudents. Sass ( 2006 )foundsimilarlyoptimisticresultsforFlorida'scharterschools,whichimprovedtoparwithmainstreamschoolsbytheirfthyearofoperation.Bothstudiescontrolledforheterogeneityinstudentabilitywithlaggedachievementlevels.Analternativestrategyusesstudentxedeectstoparameterizeheterogeneous,unobservedability.Thismethodologywasemployedby BifulcoandLadd ( 2006 )and Bookeretal. ( 2007 )toexaminetheeectivenessofcharterschoolsinNorthCarolinaandTexas,respectively. BifulcoandLadd ( 2006 )showedthatdespiteevidenceofimprovementovertime,theeectofattendingaNorth ( 2008 )provideathoroughreviewoftheresearchonchartereectiveness. 56

PAGE 57

Bookeretal. ( 2007 )decomposedtheeectofcharterenrollmentinTexasbyschoolageandstudenttenure.StudentswhomovedtoacharterschoolinTexasexperiencedaninitialshockbutrecoveredwithinthreeyears.Charterschoolsarepromisingvehiclesforcompetitioninpubliceducationsystems,butlengthymaturationperiodsweakentheirappeal( Carnoy,Jacobson,Mishel,andRothstein ( 2006 )).Whydonewchartersstruggle,andwhatmighthastentheirimprovement? Facultydevelopmentmaybeasubstantialstart-upcostincharterschools.Therelativeinexperienceofcharterfacultiesiswidelyacknowledged.4Largesharesofyoung,inexperiencedteachersareviewedasasignofweaknessandimpracticalityinthechartermodel( Rimer ( 2003 )).Studieshaverepeatedlyshownthatnewteachersarelesseectiveatproducingstudentachievement.Thereturnstoexperiencearesteepandhighlynon-linear,withstatisticallysignicantgainscomingintherstthreetoveyearsofateacher'scareer.5Ifchartersareunabletoattractandretainexperiencedteachers,andifnewteachersfacethesamedicultiesincharterschoolsastheydointraditionalpublicschools,thenitshouldbenosurprisethatcharterstudentsdonotperformaswellasmainstreamstudents.Theconclusionthatnewteacherscontributetothestrugglesofjust-openedcharterswouldbeaneasyonetomake,despiteapaucityofquantitativeevidencelinkingfacultyinexperiencetostudentperformanceincharterschools. Hanusheketal. ( 2007 )foundthatinexperiencedfacultiesmayhavecontributedtotheestimatedpenaltyfromenrollinginanewTexascharterschool,althoughtheirstudentsbenettedfromsmallerclasssizes. Hoxby ( 2002 ), MalloyandWohlstetter ( 2003 ),and PodgurskyandBallou ( 2001 ).5Aselectionofstudiesreachingtheconclusionthatthereturnstoexperiencearesteepestoverateacher'srstfewyearsincludes Clotfelteretal. ( 2007 ), MurnaneandPhillips ( 1981 ),and Rocko ( 2004 ). 57

PAGE 58

Acharterschoolisawhollydierentenvironmentforstudentsandteachers,andweshouldnotbequicktoassumethatnewcharterteacherswillnegativelyaectstudentperformancetothesamedegreeobservedfornewteachersinmainstreamschools.Thereissomeexperimentalevidenceofnon-traditionalnoviceteachersoutperformingtheirtraditionalcounterparts( Glazerman,Mayer,andDecker ( 2006 )).Also,inapostsecondarysetting,studentsbenettedfromhavinglessexperienced,lesseducatedmathandscienceteachers,althoughtheyfailedtoretainthosebenetsinsubsequentfollow-oncourses( CarrellandWest ( 2008 )).Sonontraditionalnewteachersmaybemoreeectivethanthe 58

PAGE 59

Hiringyoungandinexperiencedteacherscouldbeanecientwayforcharterschoolleaderstoallocatetheirlimitedresources.Indoingso,theycandirectmoreoftheirbudgettowardrecruitingandoverheadexpenses.Butthisalternativepersonnelstrategycouldhavelargesocialcostsintheformoflowerstudentachievement.Ifso,charterschoolswithhighlyinexperiencedfacultieswouldnotbefulllingtheirrolesasviableoptionsforpublicschoolstudents.Thereisnoevidence,however,thatstudentsperformbetterorworseincharterschoolsthatrelyheavilyonnewteachers. Inthispaper,IlinkfacultyprolesofNorthCarolina'scharterandmainstreamschoolstoatwelve-yearpanelofstudentmathandreadingtestscores.Thisistherstquantitativeeorttodatethatfullyexploresrelationshipsbetweenfacultyinexperienceandstudentperformanceincharterschools.Idocumenthighratesofnewteachersandteacherturnoverincharterschools,relativetomainstreamschoolsofthesameage.Aschartersaged,however,therateofnewteachersdeclinedsteeply,andonaverage,convergedtothemainstreamlevelbythesixthorseventhyearofoperation.Iincorporatefacultystatistics{mostimportantly,theshareofnewteachers{intocommonmodelsofcharters'valueadded.Consistentwithearlierresearch,Indthatchartersimprovedwithage,butthismaturationwasnotduetodecliningratesofnewteachers.Duringtheirrstvetosixyearsofoperation,charterswithhigherratesofnewteachersproducedhigherstudentmathandreadinglevels,relativetoschoolswithmoreexperiencedfaculties.Thesendingssuggestthatyoungcharterschoolsbenettedfrom{oratleast,werenotharmedby{ratesoffacultychurnthatsignaltroubleinmainstreamschools.Whatislessclearisthemechanismbehindtheempiricalbenetsattributedtoinexperiencedcharterfaculties.WereNorthCarolina'scharterschoolshiringmoreeective,albeitlessexperienced,teachers,ordidlow-costfacultiesallowcharterstoredirectresourcesinwaysthatyieldedgreaterstudentachievement?Ineithercase,highratesofnewteachersincharterschools 59

PAGE 60

2.2.1NorthCarolinaCharterSchools NorthCarolinaisanidealsettingtostudytheeectivenessofcharterschoolsservingascomplementaryalternativestoexisting,fairlystablemainstreampublicschools.10Thestate'schartersystemistenyearsold,servesadiverserangeofstudentsandlocales,andbearsmanyfeaturesincommonwithotherstatechartersystems,particularlyregardingschoolnanceandfacultyrequirements.Eachcharterschoolreceivesaper-pupiltransfer NorthCarolinaStateBoardofEducation ( 2006 )).8Irefertoschoolyearsbytheyearoftheirtraditionalspringconclusion.Forinstance,2002referencesthe2001-2002schoolyear.9Althoughthecapisbinding,newchartershaveopenedeachyearsince2002toreplaceschoolsthatclosed.Also,statewidecharterenrollmentcontinuestogrowasexistingschoolsaddgradesandcampuses.10Charterschoolsassumethisroleinmanystates,butinsometroubledurbanschooldistricts,partoftheirpurposeistoreformchronicallyunderperforming,costlyschoolsystems.Theresultsdiscussedherewillbemoregeneralizabletothecomplementarytypeofchartersystem. 60

PAGE 61

2.2.2 )oncharterandmainstreamstudentperformancehasbeencollectedforsixgradesandtwelveyears,coveringtheentirehistoryofthestate'schartersystem.NorthCarolinahascontributedlargelytothenationaldebateoncharterschools,thanksinparttostudiesby BifulcoandLadd ( 2006 2007 ),whofoundstrongevidenceofsub-parstudentachievementgainsandracialsegregationinthestate'scharterschools.Irevisittheirndingsonstudentachievementinthispaper,withthebenetofveadditionalyearsofdata,andforthersttime,Ianalyzetheeectoffacultyexperienceoncharterstudentachievement. NorthCarolina'schartersystemisatacriticaljuncture.Littlehaschangedsince1996,butthe100-schoolcap,charterschoolfunding,andtheevaluationofindividualschoolsremainpolarizingissuesinstateandlocalpolitics.Oneresearchgrouphasrecommendedthestatemaintainthe100-schoolcap( Manuel ( 2007 )),citingpoorprogressinexistingcharters.Anothergroup,theBlueRibbonCharterSchoolCommission,adviseddoingawaywiththecapbutcrackingdownonlow-performingschoolsbyrevokingthecharterofanyschoolperformingbelowmedianstandardswithoutsignsofimprovement12( Owens ( 2007 )).Thispaperwillinformstateandnationaldebatesontheeectivenessof 61

PAGE 62

2006 2007 )evaluationsofNorthCarolina'scharterschools. Myprimarymeasuresofstudentperformanceareend-of-grade(EOG)testscorerecordsforstudentsingradesthreethrougheight.EOGtestscoverthestate'sStandardCourseofStudyinmathandreading.14Rawscoresareintendedtobeinterval-scaledacrossandwithingrades,meaningthataone-unitincreaserepresentsthesameknowledgegainatallpointsintherangeofscores,whichvariesovertime.Istandardizedrawscorestohavemeanzeroandstandarddeviationonewithineachgradeandyear.StandardizedEOGscores,then,representstudents'placeinthedistributionofscoresfortheircohort,andchangesinstudenttestscoresovertimerepresentmovementwithinthatdistribution.15Eachstudent'stestscoreislinkedtoananonymousidentierthat Muschkinetal. ( 2008 ).14Standardizedtestscoresareincompletemeasuresofstudentlearning,andfailtoassesnon-academicmeritsthatparentsvalue(discipline,forinstance).Thiscouldbeproblematicforthecurrentanalysisifcharterschoolsdierentiallyfocusonnon-academicmerits,andifparentswhosendtheirchildrentocharterschoolsplacelessemphasisontestscores.Nonetheless,EOGscoresandgainsarealargepartofNorthCarolina'sschoolaccountabilitymethodology,andcharterschoolsareheldtothesameaccountabilitystandardsastheirmainstreamcounterparts.15Theresultstofollowaresimilar{butattenuated{ifIstandardizebyyear,thereinpreservingtheverticalscaleacrossgrades.Educationinputsaectstudents'placeintheircohortdistributionsmoresothantheirplaceinthegrade3-8distribution,sopoint 62

PAGE 63

Iassembledapanelofstudentsingradesthreethrougheightbycohort,startingwiththecohortingradethreein1996,upthroughthecohortingradethreein2005.Iexcludedstudentswhorepeatedagradeorhadgapsintheirtimeseries.Changingschoolsisdisruptiveandtendstonegativelyaectastudent'stestscoregrowth( Hanushek,Kain,andRivkin ( 2004 )),soitisnecessarytocontrolforstudentmobilityintypicalstudentachievementproductionfunctions.Iidentiedvemutuallyexclusivetypesofschoolchanges:movementacrosscounties,movementwithinacounty,structuralmoves(usuallyindicatingthetransitionfromelementaryschooltomiddleschool),16movementfromamainstreamschooltoacharterschool,andviceversa. Thedataallowmetotrackmorethan1.7millionuniquestudents,ofwhom31,634attendedacharterschoolatsometime.Forcomputationalconvenience,Irandomlysampledcompletetestinghistoriesfortwentypercentofexclusivelymainstreamstudents,thosewhowereneverobservedinacharterschool.Table 2-1 categorizescharterstudentsbytheirmobility,orlackthereof,betweencharterandmainstreamschools.Nearlyforty-eightpercentofcharterstudentswereobservedinamainstreamschoolbeforemovingtoacharterschool{ofthose,28.2percenttransitionedbacktothemainstreamafteroneormoreyears.Justlessthansixteenpercentofcharterstudentsoriginatedinthechartersectorandthenmovedtoamainstreamschool.Inmodelsofchartervalue estimatestendtobelargerinabsolutevaluewhenEOGscoresarenormalizedbygradeandyearratherthanyearalone.Nonetheless,theconclusionthatinexperiencedcharterfacultieswereassociatedwithhigherstudentachievementwasrobusttoeitherdistributionofachievementlevels,aswellaschangesinthoselevels.16Followingtheliterature,Idenedastructuralmoveasanyschoolchangewhereatleastthirtypercentofthemover'sschool-gradecohortmadethesamemove( Hanusheketal. ( 2007 ), BifulcoandLadd ( 2006 ), Sass ( 2006 ),and Bookeretal. ( 2007 )). 63

PAGE 64

2.3 controlsforinherentstudentabilitywithlaggedtestscoresratherthanstudentxedeects,thereinpermittinganystudentwithatleasttwoyearsoftestrecordstocontributetoresults. MostEOGrecordsindicatestudents'examproctors,whomayhavebeenclassroomteachers.ThisisoneofthemanyvaluablefeaturesoftheNorthCarolinadata,becauseitallowsresearcherstolinkindividualteacherstotheirstudents.Unfortunately,charterteachersarenotwell-representedintheEOGles,soIcannotcondentlylinkindividualcharterteacherstotheirstudents.Charterteachersare,however,comprehensivelylistedinschoolactivityreports,allowingmetolinkfacultyprolestostudenttestrecords.Untilnow,nostudyhasexploitedthisuniquewindowonthetypicallyopaquedatawallsurroundingcharterfaculties. 64

PAGE 65

2-2 summarizescharacteristicsoftheschoolsandfacultieswithEOGtest-takers,bycharterdesignation,overtheyears1996-2007.Charterschoolsenrolledfewerstudentsonaverage,andtheyhadslightlysmallerstudent-teacherratios(15.0versus18.0).But13.6percentofcharterfacultieswerenewteachers,morethantwicetherateinmainstreamfaculties.Chartershadlowersharesoffemaleteachers,byelevenpercentagepoints,andsignicantlyhigherratesofblack,non-Hispanicteachers. Clearly,charterschoolsreliedonnewteacherstoamuchgreaterdegreethanmainstreamschools.Butsimpleaveragesmaskdynamictrendsinfacultyexperience.Figure 2-1 plotstherateofrst-yearteachersinschoolfaculties,byageofschoolandcharterdesignation.Onaverage,nearlythirtypercentofteachersinnewcharterschoolswereintheirrstyearofteaching.Thisisaboutthreetimestherateofnewteachersinjust-openedmainstreamschools.Buttherateofnewcharterteachersshrankdramaticallyastheirschoolsaged,reachingparwithmainstreamschoolsbythesixthtoeighthyearofoperation. Newteachersareatahighriskofleavingtheprofession.Stangaschoolwithalotofnewteachersmaybealow-costpersonnelstrategy,butaprecursortohighturnover. 65

PAGE 66

2-2 ,notethatteacherturnoverwashighinNorthCarolina{19percent,onaverage,meaningthataboutoneinveteachersinaparticularschoolwerenottherethefollowingyear.Butturnoverincharterschoolswasevenhigher,33.6percent,meaningoneinthreecharterteachersdidnotreturn.Butagain,thesimpleaverageismisleading,becauseteacherturnoverfellsharplyascharterschoolsaged.Figure 2-2 illustratesturnoverincharterandmainstreamschoolsintheirrstthroughninthyear.Muchliketherateofnewteachers,thetypicalrateofteacherturnoverincharterschoolswasveryhighinitially(almostfortypercent)butdeclinedsharplyastheschoolsaged.Bytheseventhyearofoperation,averageturnoverincharterschoolswasstatisticallyindistinguishablefromturnoverinmainstreamschools. Theevaluationofcharterschooleectivenesshasrepeatedlypointedtoapatternofdicultearlyyearsfollowedbygradualimprovement. Bookeretal. ( 2007 )attributedpartoftheestimatedpenaltytothefactthatnewcharterschoolsnecessarilyenrollhighratesofschool-changers,whotendtomakelowerachievementgainsuponmovingtoanewschool.Additionalvolatilitymaycomefromhighratesofinexperiencedandexitingteachers.Canweattributethematurationofcharterschoolstofallingratesofnewteachers?Section 2.3 suggeststhatnewteacherswerenotresponsibleforlowstudentachievementinNorthCarolina'scharterschools. 66

PAGE 67

InthemodeldescribedbyEquation 2{1 ,theeectsofinputsfrompriorgradesandyearsdecayaccordingtokgt.Also,theeectsofinputsareallowedtovaryacrossgrades{forinstance,itmaybethecasethatmakinganintra-district,non-structuralmoveismoredisruptiveinhighergrades.Similarly,astudentmayhaveaninherentlygoodmatchwiththefourth-gradecontent,makingi4largerthanherotherxedeects.Unfortunately,thisexibilityrendersthemodelinestimable.Thenumberofparametersequalsthenumberofobservations,anditisthetaskoftheeconometriciantoimposereasonablerestrictionsthatpermitidentifyingvariationinthevariablesofinterest.First,Iassumethatstudentxedeectswereconstantacrossgrades:iG=i8g.Similarly,IassumethattheeectsofinputsabbreviatedbyCigt,Aigt,Migt,andFigtwereconstantacrossgrades.Irestricttheeectofpriorinputstobezeroforthehistoryofgrade-years:gt=08g
PAGE 68

2{2 isamodelofstudentachievementwhereastudent'sinherentpropensitytoproduceachievementisrepresentedbyanunobserved,time-invariantxedeect(ki).Students'gender,race/ethnicity,parentalincome,andsubsidizedlunchstatuswerelargelytime-invariant,soAiGTcanbeomittedsinceitsvariablesarehighlycollinearwithki.Ialsoestimateamodelwithlaggedachievement,insteadofstudentxedeects: Notetheadditionofcountyeects(C)in 2{3 .Studentxedeects,inconjunctionwithcross-countymoveindicators,negatetheneedtocontrolforcountyeectsinEquation 2{2 .Butintheabsenceofstudentxedeects,Cindicatorsarenecessarytocontrolfordierencesintypicalstudentperformanceacrosscounties.Equations 2{2 and 2{3 requireatleasttwocontinuousyearsoftestingdatatoidentifytherelativeeectofcharterenrollment;accordingly,Ilimitthesampletostudentswithatleastoneprioryearofmathandreadingtestscores.Todate,studiesonthequalityofcharterschoolshavetypicallyusedeitheraxedeectsorlaggedachievementmethodology,butnotboth.IpresentresultsfromEquations 2{2 and 2{3 side-by-side,andshowthattheeectivenessofcharterschoolsappearsquitedierentacrossmodels.Thedierencecanbereconciledby(1)thesourceofidentifyingvariationeachmodelrelieson,and(2)theassumptionsthatledtotheirderivationfromEquation 2{1 Equation 2{3 hasconceptualandoperationaladvantagesoverEquation 2{2 .ThexedeectsspecicationrepresentedbyEquation 2{2 makesthestrongassumptionthatprioryears'inputsdonotdecay,whereasEquation 2{3 ismoreexible,allowingktobeestimatedratherthanassumed.Equation 2{3 ,whichcontrolsforinitialprociency,isalsolesssubjecttomeanreversionthanEquation 2{2 .Meanreversion,inthissetting,isproblematicwhenstudentswithbelow-averageinitialscorestendtomakeabove-averageadvancementswithintheircohortdistribution.NorthCarolina'sEOGtestscoregainsaregreatlyaectedbymeanreversion,particularlyinmath.Figure 2-3 plotslocalpolynomial 68

PAGE 69

Additionally,thegainsmodelwithlaggedachievementoersawiderscopeforinference,albeitalesscausalone.Fixedeectmodelsevaluatetheeectivenessofachartereducationbymeasuringastudent'scharterperformanceagainsthermainstreamperformance.Astudent'smainstreamachievementisanintuitivecounterfactualforherperformanceincharterschools.Butthisidenticationstrategyhingesonsector-switchers,whomaybesystematicallydierentthanstudentswhospentfewerthantwoyearsinmainstreamschools.Equation 2{3 isunder-parameterizedbycomparison,andidentifyingvariationincharterenrollmentispredominantlycross-sectional,ratherthanlongitudinal.Accordingly,inferenceregardingtheeectofcharterschoolsorfacultycharacteristicsshouldbeattributedtocross-sectionalvariationinthoseinputs.CoecientestimatesfromEquation 2{3 willreectglobalrelationshipsbetweeninputsandstudentachievement,whereasestimatesfromEquation 2{2 withxedeectswillshowhowstudentachievementreactedtochangesineachinput. TheadvantagesofthelaggedachievementmodelrepresentedbyEquation 2{3 comeatconsiderablecosttotherigoroustreatmentofselection.WithZkiG1T1inthelistofinputs,studentxedeects(i)andtime-invariantstudentcharacteristicslikegender,race/ethnicity,parentaleducation,andexceptionalities(variablesrepresentedbyAiGTinEquation 2{1 )wouldbeendogenous.Thexedeectsmodel,however,willcontrolforanydierentialselectionintocharterschoolsstemmingfromstudents'inherentrateofachievementgrowth.BothmodelsdealwithheterogeneousabilityinawaythatnegatestheneedtocontrolforobservablestudentcharacteristicsinAiGT.Beforemovingon,Iwill 69

PAGE 70

Table 2-3 listssummarystatisticsforindividualstudentcharacteristicscontainedinAiGT,bycharter/mainstreamenrollment.19Charterstudentshadlowermathscoresandgainsthanmainstreamstudents,buthigherreadingscores.Charterstudentswereslightlymorelikelytobewhite,althoughsummarystatisticsforraceandethnicityfailtodemonstratetheextentofracialpolarizationinNorthCarolina'scharterschools.Charterparentsweremoreeducatedonaverage.Amongcharterstudents,32.7percenthadaparentwithafour-yearcollegedegree,comparedtojust21.6percentofmainstreamstudents.Charterparentswerealsomorelikelytoholdpostgraduatedegrees.Thelikelihoodofvariouslearningdisabilitieswassimilarforstudentsinbothsectors,althoughmainstreamstudentswerethreetimesaslikelytobeacademicallygifted.Charterstudentswerealsolesslikelytohavefreeorreduced-pricelunch,by16.9percentagepoints.Judgingbythesesummarystatistics,atypicalcharterstudentcamefromamoreadvantagedbackgroundthanhismainstreampeers,wasrelativelystronginreading,butrelativelyweakinmath. IestimateEquation 2{2 usingthe\within"xedeectsestimator,andIestimateEquation 2{3 usingordinaryleastsquares(OLS).Inbothmodels,Iallowforrobuststandarderrorstobeclusteredbystudentidentifers.SelectedcoecientestimatesfromEquations 2{2 and 2{3 arepresentedinTables 2-4 and 2-5 ,respectively. Turningrsttothexedeectsspecicationswithoutfacultyvariables(Table 2-4 ,columnsIandIII),theeectofachartereducationappearstobenegativeformathandreading.Charterstudentssawadropof0.169standarddeviations(sd)inmathanda 2-3 summarystatisticsarelimitedtostudentswithatleastoneprioryearoftestdata,thesamesampleforwhichEquations 2{2 and 2{3 areestimated.SamplemeansarequalitativelysimilarforbroadergroupsofEOGtest-takers. 70

PAGE 71

ColumnsIIandIVofTable 2-4 listcoecientestimateswhenfacultycharacteristicsareincludedinthexedeectsmodel.Addingfacultycharacteristicschangestheestimatedeectivenessofcharterschools,butnotsignicantly.CondenceintervalsoverlapforcolumnIIandIVpointestimates.Ofparticularinterestforthispaperistheeectofnewteachersonstudentachievement.Atenpercentage-pointincreaseintherateofnewteachersatastudent'sschooldecreasedhisachievementgainsby0.017sdinmathand0.013sdinreading.Giventhis,andtheprevalenceofnewteachersincharterschools,wemightconcludethatcharterschoolswouldhavebeenmoreeective,butfortheirrelianceonnewteachers.Butthisinferenceassumesthatnewteachershavethesameeectinbothsectors,anassumptionIrelaxlateron. AddingfacultycharacteristicstothexedeectsspecicationdoesnotchangetheconclusionthatNorthCarolinawereineectiveatraisingstudentachievement,butswitchingtothelaggedachievementspecicationdoes.CoecientestimatesforthelaggedachievementmodeloftestgainsarelistedforbothsubjectsinTable 2-5 .The 2-4 issimilarto BifulcoandLadd ( 2006 )preferredxedeectsmodelintheirstudyofNorthCarolinacharterschools,althoughtheirdependentvariablewasthechangestandardizedlevelscoresratherthanthelevelitself.Theyfoundthatcharterenrollmentreducedstudentgainsby0.160sdinmathand0.095inreading. 71

PAGE 72

Despitedierentmodesofinference,estimatedeectsoffacultycharacteristicsonstudentachievementweresimilarinxedeectandlaggedachievementmodels.Atenpercentage-pointincreaseintherateofnewteacherssignicantlydecreasedmathandreadinggainsinbothmodels,althoughtheestimatedpenaltytoreadinggainsattributedtonewteachersgrewsignicantlyfrom-0.013sdinthexedeectsmodelto-0.020sdinthelaggedachievementmodel.Fortheremainderofthissection,Iusethelagged 72

PAGE 73

ColumnIofTables 2-6 (math)and 2-7 (reading)replicatesthecolumnsIIandIVspecicationsfromTable 2-5 ,respectively,butbreaksthesingledummyvariableforcharterenrollmentintotenindicatorsrepresentingrstthroughtenth-yearcharters.Resultsshedlightondynamictrendsinchartereectiveness.Themathperformanceofstudentsinrst-yearcharterswas0.081sdlowerthanthatofmainstreamstudents.Studentsinfth-yearandolderchartershadmathscoresthatwerestatisticallyindistinguishableorgreaterthanmainstreamstudentscoresscores.Thepatternofimprovementwasmoredramaticforreadingachievement.First-yearcharterschoolshadanegativeimpactonreadingachievement,second-andthird-yearcharterswereonparwithmainstreamschools,andbeginningwithfourth-yearcharters,readingscoresweresignicantlyhigherincharterschools,byasmuchas0.042sd(about5.7percentoftheblack-whitegapinreading). Whatroledidfacultydevelopmentandexperienceplayintheimprovementofnewcharterschools?Recallthatratesofnewteachersandteacherturnoverwereveryhighfornewcharterschools,butfellquitesharplyovertheyearswhereimprovementinstudenttestscoreswasgreatest.ColumnIIspecicationsinTables 2-6 and 2-7 addtheinteraction,\New(%)xcharter"tothelistofinputs.Thecoecientfor\Newteachers(%)"suggeststhatforevery10-percentage-pointincreaseintherateofnewteachers,mainstreamstudentsrealizedabouta0.024sddropintheirmathscoresanda0.022sddropinreadingscores.Coecientson\New(%)xcharter"measurehowtheeectofnewteacherswasdierentincharterschools,relativetothebaselineeectofnewteachersinmainstreamschools.Iftheshareofnewteachershadasimilarormagniedeectoncharterstudents'testscores(thatis,ifthecoecientontheinteractionisstatisticallyinsignicantornegative),thenfacultydevelopmentandretentionwouldexplainmuchofsub-parachievementinnewcharterschools,andthesubsequent 73

PAGE 74

ColumnIIIspecicationsinTables 2-6 and 2-7 interactcharteragewiththepercentoffacultywhowerenewtounderstandtherelationshipbetweennewteachersandstudentperformanceincharterschoolsofvariousages.22Coecientsforthecharterageindicatorsrepresenttherelationshipbetweencharterenrollment(byageofschool)andstudenttestscores,asidefromwhatwasstatisticallyattributedtonewteachers.Coecientsoncharterenrollmentinteractedwiththepercentofteacherswhowerenewrepresenttheadditionalpenalty(orreductioninthatpenalty)fromacharterschoolhavingatenpercentage-pointincreaseinthepercentofnewteachers,relativetothebaselineeectofnewteachersonmainstreamstudentperformanceestimated. 74

PAGE 75

2-4 ,usingthepointestimateslistedincolumnIIIofTables 2-6 and 2-7 .Thesolidlineineachpanelrepresentsthebaselinepenaltytomainstreamstudenttestscoresfromatenpercentage-pointincreaseinnewteachers,estimatedtobe0.022sdformathand0.024sdforreading.Scatteredpointsarethesumofbaselineeectsandtherelativeeectoftheinteraction,\New(%)xyeartcharter."InthetoppanelofFigure 2-4 ,vemarginaleectsaregreaterthanzero,meaningthatincharterschoolsofparticularvintages,lessexperiencedfacultiesincreasedmathachievement,relativetomoreexperiencedcharterandmainstreamfaculties.Asimilar,butattenuated,patternwasfoundforreadingtestscores.Theseresultsimplythatfresh,inexperiencedfacultieswerenotresponsibleforlowachievementinrecently-newcharterschools.Butstudentsinolderchartersweretypicallybetterowithmoreexperiencedfaculties. 75

PAGE 76

2-6 and 2-7 controlforstudent-teacherratiosinadditiontothelistofinputsincolumnIII.Aswiththeotherschool-levelinputs,Iallowstudent-teacherratiostohaveadistinctimpactinthechartersector.Morestudentsperteacherslightlyincreasedstudentachievementinmainstreamschools,by0.003sdinmathand0.005sdinreading.Thissmallbutcounter-intuitiverelationshipisnotunheardofinnon-experimentalsettings.24Largerclassesincharterschoolsoset(ormorethanoset,inthecaseofmath)baselineeects.Moreimportantly,controllingforclasssizehadlittletonoeectontheestimatedbenetsthatcharterstudentsrealizedfromattendingschoolswithlessexperienced RiceandSchwartz ( 2008 )forareviewofresearchontheeectivenessofclasssizereductions. 76

PAGE 77

2{1 :levelswithstudentxedeects(Equation 2{2 )andlevelswithlaggedachievement(Equation 2{3 ).Someconsistentpatternsemergedacrossmodels.Enrollinginarst-yearcharternegativelyaectedstudents'testscorelevels.Studentswhomovedfromamainstreamschooltoacharterschoolofanyagesawlowergainsimmediatelyfollowingtheirmove.Higherratesofnewteachersdecreasedmainstreamstudents'mathandreadinglevels.Theseconclusionswerereachedwithvaryingstatisticalandeconomicsignicanceinbothmodels,despitedierenttreatmentsofheterogeneous,unobservedstudentability.Theestimatedimpactofattendingacharterschoolonstudentachievementwasconsiderablymoresanguineinthelaggedachievementmodel,highlightingtheimportanceofstatisticalassumptionsandidentifyingvariationinpolicy-relevantanalyses. Idocumentedaclearpatternofimprovementincharterstudentachievementfollowingcharters'rstyearofoperation,whentherateofnewteachersincharterfacultieswasfallingsharply.Buttheweakperformanceofstudentsinnewcharterschoolscouldnotbeattributedtoinexperiencedfaculties.Priortocharterschools'sixthyear,highersharesofnewteacherswereassociatedwithslightlyhigherstudentachievement,particularlyformath.Thisrelationshipwasambiguousbeyondyearsix,suggestingthatinolder,moreestablishedcharters,theeectoffacultyexperiencewasnodierentthanitwasintraditional,mainstreampublicschools. 77

PAGE 78

78

PAGE 79

StudenttransitionsinandoutofNorthCarolinacharterschools PercentofstudentsTypeoftransition(s)observedincharterschools onlyobservedincharterschools35.6movedfrommainstreamtocharter34.3movedfromchartertomainstream15.5movedfrommainstreamtochartertomainstream13.5otherpatternswithatleasttwotransitions1.1 Notes:n=31,634totalindividualsobservedincharterschools. 79

PAGE 80

NorthCarolinapublicschoolfaculties:Summarystatisticsforschoolswithgrade3-8EOGtest-takers. TypeofschoolMainstreamCharterStatisticMeanMeanDierence numberofstudents536.5236.5300.0(237.7)(183.9)numberofteachers31.615.815.8(14.0)(10.7)student-teacherratio18.015.03.0(15.9)(8.0)percentnewteachers6.413.6-7.2(6.5)(16.2)teacherturnover(percentleaving)19.033.6-14.6(14.5)(25.6)percentfemaleteachers88.977.911.0(11.2)(16.8)percentwhiteteachers83.067.815.2(19.0)(34.8)percentblack,non-Hispanicteachers15.027.5-12.5(17.6)(33.7)percentHispanicteachers0.71.7-1.0(2.0)(5.1)percentother,non-Hispanicrace/ethnicity1.33.0-1.7(5.9)(10.3)n(school-years)18,948701 Notes:Standarddeviationsareinparenthesesbeloweachmean.indicatesasignicantdierenceat95%condenceormore. 80

PAGE 81

Averagerateofnewteachers,byageofschoolandcharter/mainstreamdesignation.Thinnerlinesrepresentcondenceintervals. 81

PAGE 82

Averagefacultyturnoverrates,byageofschoolandcharter/mainstreamdesignation.Thinnerlinesrepresentcondenceintervals 82

PAGE 83

B Meanreversioninmath(A)andreading(B)EOGtestscores.Testgainswereestimatedasrst-degreelocalpolynomialfunctionsofinitialscores,with0.50standarddeviationhalfwidths.InitialscoredensitieswereestimatedusingEpanechnikovkernelfunctionsand0.50standarddeviationhalfwidths. 83

PAGE 84

NorthCarolinapublicschoolstudents:Summarystatisticsforgrade3-8EOGtest-takers TypeofschoolMainstreamCharterStatisticMeanMeanDierence Mathscore0.1100.0320.078(0.967)(0.988)Readingscore0.1050.142-0.036(0.951)(0.966)Female0.5090.513-0.005(0.500)(0.500)White,non-Hispanic0.6180.633-0.015(0.486)(0.482)Black,non-Hispanic0.2830.292-0.008(0.451)(0.454)Hispanic0.0460.0220.023(0.209)(0.148)other,non-Hispanic0.0530.0530.001(0.224)(0.223)Missingrace/ethnicity0.0010.0020.000(0.035)(0.042)Highschoolorless0.4310.2330.198(0.495)(0.423)Lessthan4yearsofcollege0.1960.1920.004(0.397)(0.394)Four-yearcollegegraduate0.2180.337-0.119(0.413)(0.473)Postgraduatedegree0.0490.075-0.025(0.217)(0.263)Missingparentaleducation0.1060.163-0.057(0.307)(0.369)Learningdisabled-math0.0190.023-0.004(0.137)(0.150)Learningdisabled-reading0.0410.0410.000(0.198)(0.198)Learningdisabled-writing0.0420.044-0.001(0.202)(0.205)Learningdisabled-language0.0070.009-0.002(0.086)(0.095)Otherlearningdisability0.0020.002-0.001(0.040)(0.048)n(student-years)741,73657,480 Notes:Standarddeviationsareinparenthesesbeloweachmean.denotessignicantdierencesatthe5%condencelevel. 84

PAGE 85

TypeofschoolMainstreamCharterStatisticMeanMeanDierence Academicallygifted0.1740.0540.120(0.379)(0.227)Subsidizedlunch0.3320.1680.164(0.471)(0.374)Missingsubsidizedlunchstatus0.1630.182-0.019(0.370)(0.386)n(student-years)741,73657,480 Notes:Standarddeviationsareinparenthesesbeloweachmean.denotessignicantdierencesatthe5%condencelevel. 85

PAGE 86

Fixedeectestimates:Theeectofcharterenrollmentontestscoregains IIIIIIIVSubjectMathMathReadingReading Charter-0.169-0.156-0.085-0.081(31.20)(28.56)(15.93)(14.98)Movedtocharter-0.013-4.86e-3-2.71e-31.79e-3(2.58)(0.96)(0.51)(0.33)Movedfromcharter-0.039-0.0413.53e-32.49e-3(7.34)(7.58)(0.61)(0.43)Structuralchange4.59e-37.43e-32.23e-31.77e-3(3.67)(5.49)(1.64)(1.20)Schoolchange-0.015-0.014-8.79e-3-8.51e-3(7.72)(7.13)(4.11)(3.96)Countychange-0.016-0.017-5.37e-3-5.44e-3(4.69)(4.87)(1.41)(1.42)Newteachers(%)-0.017-0.013(16.86)(12.60)Nonwhiteteachers(%)-0.011-2.60e-4(16.85)(0.38)Femaleteachers(%)-1.72e-3-2.05e-3(2.84)(3.36) Notes:n=799,216student-years.Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Robuststandarderrorsallowforclusteringbystudentidentiers. 86

PAGE 87

Laggedachievementestimates:Theeectofcharterenrollmentontestscoregains IIIIIIIVSubjectMathMathReadingReading Charter-0.006775.45e-30.0150.027(2.92)(2.34)(6.00)(10.75)Movedtocharter-0.171-0.153-0.104-0.087(32.78)(29.29)(18.60)(15.57)Movedfromcharter0.1150.1210.0620.06819.0220.01(9.68)(10.64)Structuralchange5.90e-45.67e-31.72e-35.06e-3(0.35)(3.29)(0.95)(2.72)Schoolchange-0.057-0.051-0.057-0.051(26.96)(23.67)(24.61)(21.80)Countychange-0.052-0.051-0.043-0.042(14.71)(14.42)(11.19)(10.91)Newteachers(%)-0.020-0.020(21.01)(19.40)Nonwhiteteachers(%)-0.021-0.021(52.14)(47.81)Femaleteachers(%)-2.25e-3-4.07e-3(4.66)(7.93) Notes:n=799,216student-years.Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Robuststandarderrorsallowforclusteringbystudentidentiers. 87

PAGE 88

Theeectofcharterenrollmentonmathachievementgains,disaggregatedbyschoolage SpecicationIIIIIIIV Year1charter-0.081-0.139-0.184-0.119(6.17)(9.88)(8.78)(4.94)Year2charter-2.22e-3-0.038-0.0190.041(0.26)(4.16)(1.66)(2.67)Year3charter-3.78e-3-0.031-0.0480.013(0.52)(4.04)(4.48)(0.86)Year4charter-0.019-0.045-0.0574.65e-3(2.85)(6.43)(5.52)(0.29)Year5charter0.0282.41e-3-0.0290.031(4.50)(0.36)(3.24)(2.22)Year6charter-2.23e-3-0.025-0.0170.042(0.36)(3.83)(1.79)(2.81)Year7charter0.0277.57e-30.0360.095(4.30)(1.16)(3.96)(6.59)Year8charter-0.012-0.032-0.0110.048(1.65)(4.36)(1.05)(3.25)Year9charter-7.27e-3-0.0247.00e-40.060(0.88)(2.80)(0.07)(4.02)Year10charter0.0164.80e-4-0.0170.039(1.28)(0.04)(0.95)(1.89)Newteachers(%)-0.018-0.024-0.024-0.024(17.04)(20.57)(20.58)(20.00)Newxcharter0.026(10.72)Newxyear1charter0.0410.040(7.06)(6.87)Newxyear2charter0.0140.014(3.07)(2.97)Newxyear3charter0.0400.038(6.13)(5.93)Newxyear4charter0.0360.033(5.19)(4.80)Newxyear5charter0.0540.052(9.16)(8.81)Newxyear6charter0.0180.017(2.45)(2.28) Notes:n=799,216student-years.CoecientswereestimatedbyEquation 2{3 .Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Robuststandarderrorsallowforclusteringbystudentidentiers. 88

PAGE 89

SpecicationIIIIV New(%)xyear7charter-9.39e-3-0.011(1.10)(1.30)Newxyear8charter9.70e-4-3.90e-4(0.10)(0.04)Newxyear9charter-0.013-0.016(1.32)(1.62)Newxyear10charter0.0490.045(2.37)(2.17)Student-teacherratio3.18e-3(9.30)Student-ratioxcharter-3.66e-3(5.33) Notes:n=799,216student-years.CoecientswereestimatedbyEquation 2{3 .Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Robuststandarderrorsallowforclusteringbystudentidentiers. 89

PAGE 90

Theeectofcharterenrollmentonreadingachievementgains,disaggregatedbyschoolage SpecicationIIIIIIIV Year1charter-0.033-0.059-0.081-0.016(2.46)(4.08)(3.90)(0.67)Year2charter0.014-0.002490.0230.087(1.56)(0.26)(1.96)(5.28)Year3charter-1.34e-3-0.014-0.0300.033(0.17)(1.68)(2.54)(1.97)Year4charter0.0260.014-0.0050.056(3.51)(1.75)(0.46)(3.17)Year5charter0.0240.012-0.0120.050(3.36)(1.62)(1.22)(3.21)Year6charter0.0330.0230.0220.080(4.95)(3.25)(2.14)(4.85)Year7charter0.0390.0300.0610.120(5.66)(4.22)(6.06)(7.45)Year8charter0.0330.0240.0380.098(4.40)(3.12)(3.52)(6.19)Year9charter0.0220.0140.0110.072(2.38)(1.55)(0.91)(4.37)Year10charter0.0420.0350.0470.103(3.08)(2.55)(2.30)(4.33)Newteachers(%)-0.019-0.022-0.022-0.021(16.67)(16.96)(16.92)(16.20)Newxcharter0.012(4.49)Newxyear1charter0.0200.019(3.40)(3.21)Newxyear2charter-3.65e-3-5.30e-3(0.74)(1.06)Newxyear3charter0.0250.023(3.26)(3.03)Newxyear4charter0.0280.026(3.51)(3.21)Newxyear5charter0.0340.030(4.84)(4.35)Newxyear6charter0.0130.014(1.66)(1.72) Notes:n=799,216student-years.CoecientswereestimatedbyEquation 2{3 .Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Robuststandarderrorsallowforclusteringbystudentidentiers. 90

PAGE 91

SpecicationIIIIV New(%)xyear7charter-0.027-0.028(2.75)(2.90)Newxyear8charter-5.10e-3-8.70e-3(0.50)(0.86)Newxyear9charter0.0180.014(1.63)(1.30)Newxyear10charter-6.62e-3-0.010(0.29)(0.45)Student-teacherratio4.72e-3(12.75)Student-ratioxcharter-3.69e-3(4.90) Notes:n=799,216student-years.CoecientswereestimatedbyEquation 2{3 .Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Robuststandarderrorsallowforclusteringbystudentidentiers. 91

PAGE 92

B Marginaleectsofaten-percentage-pointincreaseinnewteachersoncharterstudents'math(A)andreading(B)levels,byageofschool 92

PAGE 93

Anunintendedconsequenceofdecentralizedcharterschoolsystemsisthattheycanexacerbateracialsegregationacrossschools.Racialisolationincharterschoolshasbeenshowntoreducetheachievementgainsofblackstudentsandwidentheblack-whiteachievementgap.Buttheremaybeasilverlining:NorthCarolinacharterswithpredominantlynonwhitestudentpopulationswerestaedbyrelativelyhighratesofnonwhiteteachers,whomayhavebeenmoreeectiveinstructorsofnonwhitestudents.Ishowthatdisproportionatelynonwhitefacultiesmarginallyraisedthemathperformanceofblackstudentsincharterschools.Butthesegainswereinsucienttoslowthegrowinggapbetweenstudentsinpredominantlywhiteandpredominantlynonwhitecharterschools. Colemanetal. ( 1966 ))haveshownthatstudentachievementissignicantlylowerinschoolswithhigherratesofnonwhitestudents.Thiscorrelationisdriveninpartbycausalpeerexposureeects,andinpartbythewayfamiliesandteacherssortacrossneighborhoodsandschools.Familypreferencesforschoolandneighborhoodqualityleadtoself-segregationalongthosedimensions( Bayer,Ferreira,andMcMillan ( 2007 )),andconsequently,heterogeneousdemographicprolesinneighborhoodschools.Avarietyofstudieshavetriedtocircumventtheeectofsortingtounderstandthecausaleectofsegregationonstudentoutcomes. Guryan ( 2004 )usedintertemporalandgeographicvarianceinthetimingofcourt-ordereddesegregationplanstoshowthatlargegainsinwhiteexposureindices(thefractionofstudentswhowerewhiteinatypicalblackstudent'sschool)yieldedmodestreductionsinblackdropoutrates. CutlerandGlaeser ( 1997 )reliedonaggregatecity-leveldata(presumablylessaectedbysortingthanschool-orstudent-leveldata)toshowthatmoresegregatedcitieshadsignicantlylarger 93

PAGE 94

CardandRothstein ( 2007 )extendedthisconsiderably,showingthattheblack-whitegapinSATscoreswaspositivelycorrelatedwithcities'degreeofsegregation,andthatneighborhoodsegregationappearedtobemoreofafactorthanwithin-schoolsegregation. Despiteintegrationeortsacrossthecountry(court-orderedandvoluntary),racial,ethnic,andeconomicsegregationpersistsinmanypublicschooldistricts.Thesocioeconomictopologyofurbanandsuburbanareasandthewidevarianceofeconomicprosperitywithindemographicgroupshindereventhemostambitiousschoolassignmentmechanismsfrommaintainingracialoreconomicbalanceacrossschools.1Somehavehopedthatschoolchoicewouldamelioratesegregationpatternsbygivingdisadvantaged,urbanfamiliestheoptiontosendtheirchildrentomoreauent,outlyingschools(see,e.g., Rabinovitz ( 1997 )and Schneider,Teske,andMarschall ( 2000 )).Butinpractice,manyelementsofschoolchoice{includingcharterschools,privateschoolvouchers,andopenenrollment{haveincreasedracialisolationindices.2 BifulcoandLadd ( 2007 )exploredtheeectsofracialimbalanceinNorthCarolinacharterschoolsonstudenttestoutcomesandtheblack-whiteachievementgap.Blackstudentswhomadeasegregatingmovetoacharterschooltendedtoseelowermathgainsasa Silberman ( 2006 )).InSanFrancisco'spublicschoolsystem,economicintegrationeortsmayhaveactuallyintensiedracialandethnicsegregation( GlaterandFinder ( 2007 )).2See,forexample, BifulcoandLadd ( 2007 ), Frankenberg,Lee,andOreld ( 2003 ), InstituteonRaceandPoverty ( 2008 ),and Levin ( 1998 ). Hoxby ( 2003 )notedthatnegativepeereectsarisingfromschoolchoiceandstudentsortingcouldbemorethanosetbyproductivitygainsinderegulatededucationmarkets. 94

PAGE 95

Nonetheless,theremaybeanintrinsic{ifnotcompletelyosetting{benettocharterschools'polarizedstudentcompositions.Raciallyimbalancedstudentpopulationsinchoiceschoolsaregenerallycoincidentwithraciallyimbalancedfaculties.IshowthatthiswasthecaseforthersttenyearsofNorthCarolina'schartersystem{charterschoolsweremuchmorelikelythanmainstreamschoolstohavedisproportionatelyhighratesofnonwhiteteachers.Researchfromnon-choiceandexperimentalsettingshaveshownthatassignmenttoademographicallysimilarteachercanbebenecial.Soevenifblackstudentachievementwassub-parinlargelyblackschools,studentsmayhavederivedsomebenetfromexposuretomoreblackteachers. Ehrenbergetal. ( 1995 ),usingthe1988and1990wavesoftheNationalEducationalLongitudinalStudy(NELS),showedthatassignmenttoanown-raceorown-genderteachertypicallyhadnoeectonsubjecttestgains,butdidyieldhighersubjectiveevaluationsbyteachers.Awealthofstudiesineducation,sociologic,andmulti-disciplinaryliteraturesprovideevidencethatstudentsbenetfrombeingpairedwithateacherofthesamerace{however,themagnitudeofthisbenetandthecausalmechanismsresponsibleforitareunclear( Ferguson ( 1998 )). Themostcompellingevidenceonstudent-teachermatchingbyracecomesfrom Dee ( 2004 2005 ).Intheearlierstudy,DeeexploitedtherandomassignmentofstudentstoclassroomsinTennessee'sProjectSTARexperimenttoshowthatassignmenttoanown-raceteachersignicantlyincreasedmathandreadingscoresby3-5percentagepoints(10-18percentofastandarddeviation). Dee ( 2005 )revisitedtheNELSdata,employedamatched-pairsmethodologytocontrolforstudentheterogeneity,andexaminedhowtwodierentsubjectteachersevaluatedthesamestudentatthesametime.Resultssuggested 95

PAGE 96

Theliteratureonstudent-teacherracialpairinghasyettoexaminethegainorlossfromdemographicallysimilarcharterteachers,likelybecauseofdataavailabilityandendogeneousstudentandteachersortingintotheseschools.Butifstudentsandteachersaresortingintocharterschoolsonthebasisofrace,eectivelyre-segregatingafractionofpublicschools,policystakeholdersshoulddevelopacloseanalysisofthebenetsandcoststherein.Idocumentconsiderableracialimbalanceincharterfaculties(whichappearstohavebeenmoresevereinschoolswithveryhighsharesofnonwhitestudents),anddeterminewhetherstudentsbenetted{empirically,atleast{fromattendingacharterschoolwithmoreteacherswhosharedtheirraceorgender.Iincorporatefacultyracialprolesintocommonmodelsofchartereectiveness,estimateanaverageeectofcharterenrollmentonachievementgainsforfourrace/gendercategories,andexaminehowthateectvariedwithchangesinfacultyraceandgenderproles.Indthatblackstudentshadmarginallygreatermathperformanceincharterswithdisproportionateratesofnonwhiteteachers,relativetoblackstudentsinmorerepresentativeschools.Butthiseectwassmallcomparedtotheeectofracialimbalanceincharterstudentpopulations,whichwidenedtheperformancegapbetweenlargelywhiteandlargelynonwhiteschools. 3.2.1Data Muschkinetal. ( 2008 ). 96

PAGE 97

Iconstructedcampus-by-yearfacultyprolesfromaggregatedschoolactivityreports.Activityreportsdocumenteachschool-relatedactivitywherepersonnelhavedirectcontactwithstudents{classroominstruction,lunch,eldtrips,etc.Iidentiedteacherstobeindividualswithteachingassignmentslistedintheactivityreports,excludingteachingassistants,facilitators,andDAREocers.Thereportshaveverylittleinformationaboutteacherqualications,buttheydolistteachers'race/ethnicity,gender,andstatusasarst-yearteacher.Iusedthesedatatoconstructfacultyprolesforcharterandmainstreamschoolsoveryears1996-2007.Iaggregatedrace,gender,andexperienceindicatorstotheschoollevel,andcalculatedthepercentofteacherswhowerewhite,nonwhite(black,Hispanic,orothernon-Caucasion),female,andnewtopubliceducation.Theseproleswerethenlinkedtostudenttestrecords.Ilimitedtheanalysistoblackandwhitepublicschoolstudents.Longitudinalmicrodataarenotavailableforprivateschoolstudents.Hispanicandothernon-black,non-Caucasianstudentsaccountedfor 97

PAGE 98

3-1 through 3-4 andillustratedinFigures 3-1 3-2 ,and 3-3 Tables 3-1 through 3-4 presentcomparativesummarystatisticsforcharterandmainstreamstudents,byraceandgender.Blackcharterstudentshadsignicantlylowermathandreadinglevelscoresthanblackmainstreamstudents,andsignicantlylowergainsinmath.Whitecharterstudentshadlowermathlevelscoresandgainsthantheirwhitemainstreamcounterparts,buthigherreadinglevelsalongsidelowerreadinggains.Malestypicallyperformedbetterinmath,andfemalesperformedbetterinreading.Theblack-whiteachievementgapwasabout0.098standarddeviationswiderinthechartersectorforfemales,and0.115standarddeviationswiderformales. Tables 3-1 through 3-4 alsoidentifythelikelihoodofbeingenrolledinaraciallyimbalancedschool.Throughoutthepaper,Idenearaciallyimbalancedstudentpopulationtobeonewhoseshareofnonwhitestudentswastwentypercentagepoints 98

PAGE 99

BifulcoandLadd ( 2007 ).Forty-eightpercentofblackcharterstudentswereenrolledinaschoolwitharaciallyimbalanced(i.e.,disproportionatelylarge)nonwhitestudentpopulation,comparedtoabout29percentofblackmainstreamstudents.Blackcharterstudentswerealsomorelikelytobeenrolledinadisproportionatelywhiteschool.Whitecharterstudentswere3.5timesmorelikelythanwhitemainstreamstudentstobeinawhite-imbalancedschool,andwererarelyfoundinanonwhite-imbalancedschool.Raciallyimbalancedfacultiesweremorecommonincharterschools,particularlytheschoolsblackstudentsattended.Charterstudentsofbothracesweremorelikelytobeinaschoolwithanimbalancedstudentpopulationandandanimbalancedfaculty,thoughthiswasmuchmoreprevalentforblackcharterstudents.Fortypercentofblackcharterstudentswereindoublyimbalancedschoolslikethis,morethanthreetimestherateofblackmainstreamstudents.7ThesesamplemeanssuggestalackofstudentdiversityNorthCarolina'scharterschools,andalackoffacultydiversityincharterswithrelativelylargesharesofblackstudents.Figures 3-1 3-2 ,and 3-3 conrmandelaboratethesepatterns. 99

PAGE 100

3-1 plotskerneldensityestimatesofthecampus-levelpercentofstudentswhowerenonwhite,bycharter/mainstreamdesignationandweightedbyenrollment.Thedistributionforcharterschoolswasstarklybimodal;chartersweremuchmorelikelythanmainstreamschoolstohavefewerthan20percentormorethan90percentnonwhitestudents.Facultycompositionmaybeafactorinhowstudentsandtheirfamiliessortintocharterschools.Figure 3-2 plotsweightedkerneldensityestimatesofthecampus-levelpercentofteacherswhowerenonwhite,bycharter/mainstreamdesignation.Chartersweresomewhatlesslikelythanmainstreamschoolstohaveverylowratesofnonwhiteteachers,butmorelikelytohaveveryhighrates,intherangeof60-100percent.Figure 3-3 illustratesthenonparametricrelationshipbetweenstudentandfacultycompositionsincharterandmainstreamschools.Theshareofnonwhiteteachersinschoolstendedtoincreasewiththeshareofnonwhitestudents,buttherelationshipwasmuchsteeperforcharterswhereatleast60percentofstudentswerenonwhite. BifulcoandLadd ( 2007 )showedthatthesegregatingeectsofthestate'scharterschoolsincreasedracialisolationindicesandwidenedtheblack-whiteachievementgap.However,thereisevidencefromothersettingswherestudentsperformedbetter,orwereperceivedbetter,byteachersoflikerace.Wemightexpecttoseesignicantgainsresultfromstudent-facultyracialdynamicsinthechartersector,wherestudentandteachersortingisnotdrivenbycentralizedassignment.Itseemsreasonablethatstudentswhoenrolledinacharterschoolwithadisproportionateshareofown-raceteachersarestudentswhoseparentsbelievedtheywouldbenetfromthematch.Thatistosay,ifthereisasignicantgain(tosomestudents)fromraciallyimbalancedfaculties,weshouldseeevidenceofthisincharter 100

PAGE 101

Inputsincludeanindicatorforenrollmentinacharterschool(CiT),facultycharacteristics(FsT,someofwhichareallowedtohaveadistinctimpactincharterschoolsviaCiTfsT),andstudentmobilityindicators(MiT).Observableendowmentssuchasparentaleducationandincome,aswellasunobservableendowmentslikestudents'inherentability,areabsorbedinthelaggedlevelscore,ZiG1T1.ThecoecientkCTrepresentscounty-yearxedeects,andkGrepresentscommongrade-leveleects.Therateofproportionaldepreciationintheeectofprioryearinputsisgivenbyk,andisassumedtobeconstantacrossgradesandtime.IestimateEquation 3{1 byordinaryleastsquares(OLS).Robuststandarderrorsallowforclusteringbystudentidentiers.Clusteringwillmodifystandarderrorsforanystudent-speciccorrelationinresiduals,butwillnotparameterizethatcorrelationlikexedeects.Ialsoestimatehowtheseinputsaectastudent'sgaininstandardizedmathandreadingscores: ZkiGT=(1k)ZkiG1T1+CiTkC+FsTkF+(CiTfsT)kf+MiTkM+kG+kCT+"kiGT(3{2) Following Hanusheketal. ( 2007 )and Sass ( 2006 ),IinstrumentforZiG1T1inEquation 3{2 usingtwicelaggedlevelscores,ZiG2T2.AsinEquation 3{1 ,Iallowforrobust 101

PAGE 102

3{1 and 3{2 separatelyforblackfemales,blackmales,whitefemales,andwhitemales.Thelevelsspecicationrequiresatleasttwoyearsofcontinuoustestdata,whereasthegainsspecicationrequiresthree.Toensurethatresultsfrombothmodelsreecttheachievementofthesamesetofstudents,Ilimitthelevelsanalysistostudentswithatleasttwoprioryearsoftestdata.Consequently,theresultstofollowdescribetheeectofcharterenrollmentandracialcompositionsontheachievementofgradevethrougheightstudents. ThevectorFsTcontainsschool-widefacultystatisticsattimeT:thepercentofteacherswhowerenew,thepercentwhosharedi'sgender,andfourindicatorsofracialimbalanceinstudentandfacultycompositions.Changingschoolstendstohaveanegativeeectonstudentgains( Hanusheketal. ( 2004 )),particularlyifastudentismovingtoacharterschoolfromthemainstreamsector(see,e.g., BifulcoandLadd ( 2006 ), Bookeretal. ( 2007 ), Hanusheketal. ( 2007 ),and Sass ( 2006 )).Accordingly,MiTisavectorofmutuallyexclusivemobilityindicatorsforstudentswhomovedwithintheircounty,acrosscounties,toacharterschool,fromacharterschool,andforstudentsmakingstructuralschoolchanges.Followingtheliterature,Idenedastructuralchangeasanymovesharedbyatleastthirtypercentofastudent'scohort.8 102

PAGE 103

Thesemodelsdrawfromseveralrecentstudiesontheeectivenessofcharterschools. BifulcoandLadd ( 2006 )identiedtherelativeeectivenessofcharterenrollmentforNorthCarolinastudentswithexperienceincharterandmainstreamschools,andshowedthatthesesector-switchersperformedsignicantlyworseintheformer. Bookeretal. ( 2007 )brokedowntheeectofachartereducationbyschoolageandstudenttenure,demonstratingsignicantreturnstoboth.Bothstudiesusedstudentxedeectstoparameterizeheterogeneityinunobservedability.Bycontrast, Sass ( 2006 )and Hanusheketal. ( 2007 )usedlaggedachievementmodelssimilartothelevelsandgainsspecicationsemployedheretoestimatetheaverageeectivenessofcharterschoolsbyschoolage.Inallofthesestudies,themoveryearexperienceforastudenttransitioningfromamainstreamschooltoacharterschoolwassignicantlynegative,intermsoftestperformance.Studentsmovingintheotherdirection,however,madeupformuchofthislossintheirrstmainstreamyear. 3-5 through 3-8 presentcoecientestimatesfromthelevelsandgainsequations,whichwereestimatedseparatelyforblackfemales(Table 3-5 ),blackmales(Table 3-6 ),whitefemales(Table 3-7 ),andwhitemales(Table 3-8 ).Iproceedthroughtheseresultsbyrstoutliningsomecommonthemesacrossrace/gendercells(regardingtheeectofmovingtoorfromacharterschoolandtheeectofmoremaleteachersin 103

PAGE 104

Consistentwiththeliteratureontheeectivenessofcharterschools,Indthatmovingtoacharterschoolresultedinsignicantpenaltiestomathandreadingachievement.Thesepenaltieswerelargecomparedtotheeectofothertypesof(unreported)schoolchanges.Themajorcostofstratifyingthesampleinthiswayislossofprecision,butdegreesoffreedomappeartobelargeenoughtodrawinference,eveninthesmallestrace/gendercells.TheprimarybenetofstraticationisthatIcanallowcoecientstovarybyraceandgender.Forinstance,whitestudentsincurredalargeraveragepenaltyuponmovingtoacharterschoolthanblackstudents,althoughcondenceintervalsoverlapped.Femalecharterstudentsofbothracessawsignicantlylargerlevelsandgainsinbothsubjects,relativetomainstreamfemales,whereastheeectofcharterenrollmentonmales'testoutcomeswaseithernegativeorinsignicant.Achartereducationwaspredictedtoincreasethemathlevelscoresofblackfemalesby0.221standarddeviations(sd),relativetotoblackmainstreamfemales.Thisaccountsforabout30percentoftheblack-whitegapinmathlevels. NorthCarolina'scharterschoolshadhigherratesofmaleteachers:22percent,versus11percentinmainstreamschools.Malesarehistoricallyunder-representedinpublicteaching,and Dee ( 2007 )showedthatmalestudentsintheNELSsurveyperformedsignicantlybetterwhenassignedtoamaleteacher.IndthatinNorthCarolina,therewasgenerallyanegativecorrelationbetweenstudentperformanceandthefractionoffacultiesthatweremale,exceptforwhitemalestudentsincharterschools.Coecientestimatesfor\%maleteachers"inTable 3-8 reporttheeectofatenpercentage-pointincreaseinmalefacultyonmainstreamtestoutcomesforwhitemales.Theeectwasnegativeandsignicantformathandreadingoutcomes(byasmuchas0.015sdforreadinglevels),meaningthatmainstreamwhitemalesperformedworseinschoolswithmoremaleteachers.Coecientsontheinteraction,\charterx%maleteachers"represent 104

PAGE 105

3-8 ,andlargerinabsolutevaluethanbaselineeects.Sothemarginaleectofmoremalecharterteacherswaspositive:higherratesofmaleteachersincharterschoolswereassociatedwithgreatermathandreadingperformanceforwhitemalecharter.Forblackmalesandfemalesofbothraces,theeectofhighermalerepresentationinschoolfacultiestendedtobenegative,butlesssoincharterschools. Tables 3-5 through 3-8 includeestimatesoftheimpactofstudentandfacultyracialimbalanceincharterandmainstreamschools.Schoolswereidentiedashavingimbalancednonwhite(white)studentpopulationsiftheshareofnonwhitestudentswastwentypercentagepointsabove(below)thecounty-wideshare.Racialimbalanceinfacultiesweresimilarlydened,butforblackstudents,Iincludedtwodegreesofnonwhitefacultyimbalance.Inmoderatelyimbalancednonwhitefaculties,theshareofnonwhiteteacherswas20-40percentagepointsabovethecountymean,andinseverelyimbalancednonwhitefaculties,thesharewasmorethan40percentagepointsabovethecountymean.Baselineimbalancecoecientsrepresenttheestimatedeectofaparticularracialimbalanceonstudentachievement,relativetostudentachievementinbalancedschools.Interactioncoecientsshowhowtheeectofthatimbalancewasdierentincharterschools.InTable 3-5 ,forexample,coecientestimatesfor\nonwhitestudentimbalance"showthatblackfemalestendedtosee0.014sdlowermathlevelsand0.016sdlowerreadinglevelsinschoolswithdisproportionatelylargesharesofnonwhitestudents,relativetotheircounterpartsinbalancedschools(theomittedcategory).Butthepenaltyattributedtoimbalancedratesofnonwhitestudentswasevenlargerincharterschools,by0.038-0.043sd.Sointotal,ablackfemaleinalargelynonwhitecharterschoolwasexpectedtoscore0.054-0.057sdlowerthanablackfemaleinaraciallybalancedmainstreamschool.Blackfemalesinschoolswithdisproportionatelylargewhitestudent 105

PAGE 106

ThelastfourrowsofTable 3-5 describehowracialimbalanceamongblackfemales'teachersaectedtestperformance.Blackfemalessawsignicantlylowermathlevelsandgainsinschoolswithmoderateorseverenonwhitefacultyimbalances.Iftherewereosettingeectsofnonwhitefacultyincharterschoolsonmathlevels,theywerenotstatisticallysignicantatconventionallevels.Mathgains,however,increasedby0.40sdinschoolswithseverenonwhitefacultyimbalances,relativetoblackfemalecharterstudentswithbalancedfaculties.Relativetoallstudents,charterorotherwise,enrollinginacharterschoolwithaseverelyimbalancedandlargelynonwhitefacultyincreasedblackfemales'mathgainsby0.017sd(-.023+0.040).Thisaccountsformorethan40percentoftheblack-whitegapinmathgainsformainstreamfemales.9Butblackfemalesbenettedmuchmorebyenrollingincharterswithwhite-imbalancedfacultiesinallmeasuresofachievementexceptreadinggains.InTable 3-6 ,weseethatblackmalesalsobenettedfromwhite-imbalancedcharterfaculties,byasmuchas0.199sdinmathlevels(about28percentoftheblack-whitegapamongmainstreammales).Blackcharterstudentswereaboutsixpercentagepointsmorelikelythanblackmainstreamstudentstobeinaschoolwithanimbalancedshareofwhiteteachers,relativetothesurroundingcountry.Thereweremorethan350blackstudentsincharterslikethis(withmorethan1100student-yearsoftestingdata),andofthese,81.2percentwereobservedinmainstreamschoolsbeforemovingtothechartersystem.SoIcanobservethetypeofschoolstheyoptedoutof.Evenifthechartersystemasawholewasraciallypolarized,nontrivialnumbersofminoritystudentsmayhavebeentakingadvantage 3-1 and 3-3 ,averagemathgainswere0.286sdforblackfemales,and0.327sdforwhitefemales.The0.017sdbonusfromseverelyimbalancedfacultiesrepresents41.5percentofthe0.041sdgap. 106

PAGE 107

InTable 3-6 (reportingresultsforblackmales),coecientestimatesfor\charterxmoderatenonwhitefacultyimbalance"showthattherewasnostatisticallysignicantbonusfrommoderatelyimbalancedfaculties.Butseverelyimbalancednonwhitecharterfacultiesincreasedthemathlevelsandgainsofblackmalesby0.047and0.043sd,respectively,relativetoblackmalesincharterswithbalancedfaculties.Theseeectswerelargerinabsolutevaluethanthebaselineeectsofseverelynonwhite-imbalancedfaculties(-0.027sdandastatisticallyinsignicant-0.013sd,respectively).Judgingbycoecientestimatesalone,themarginaleectofseverelyimbalancedsharesofnonwhitecharterteacherswassmallbutpositiveforblackmales'mathperformance,increasingmathlevelsbyanestimated0.020sd.Thisisaneconomicallymodestboost,representingjust2.8percentoftheblack-whitegapinmathlevelsformales.Also,notethatamongblackmalesincharterschools,nonwhitefacultyimbalanceswerehighlycoincidentwithnonwhitestudentimbalances.Soanybenetfromnonwhitefacultywouldlikelyhavebeenpairedwithnegativepeereectsarisingfromlargesharesofnonwhitestudents,whichareestimatedtohavedecreasedblackmales'mathlevelsby0.017sd. Tables 3-7 and 3-8 reportresultsforwhitefemalesandwhitemales,respectively.Whitefemaleshadsignicantlyhighermathlevelsandgainsinschoolswithwhite-imbalancedstudentpopulations.Thebonusfromthisown-raceimbalancewasjust0.016sdformathlevelsand0.009sdformathgains,butincharterschoolsitwasconsiderablylarger,by0.052and0.050sd,respectively.Soadisproportionateshareofwhitepeerswasassociatedwithatotalincreaseincharterfemales'mathlevelsby0.068sd,andan 107

PAGE 108

Theresultsoutlinedabovesuggestthatracialimbalanceincharterstudentpopulationsincreasedperformancegapsbetweenwhite-imbalancedandnonwhite-imbalancedschools,particularlywithrespecttomathachievement.Blackstudentsachievedsignicantlylowermathandreadinglevelsinschoolswithdisproportionateratesofnonwhitestudents,andforblackfemales,thispenaltywasevenlargerincharterschoolswithnonwhitestudentimbalances.Whitestudentsrealizedgreatermathlevelsandgainsincharterswithimbalancedratesofwhitestudents,relativetowhitesinraciallybalancedmainstreamandcharterschools.Together,thesepatternssuggestthatwhilelargelywhitecharterschoolswerepullingaway,largelyblackcharterswerefallingbehind.Iinvestigatedwhethertherewasasilverliningtothisnormativelyundesirablepattern,inthatblackcharterstudentsmayhavederivedsomebenetfrompredominantlynonwhitefaculties.Blackstudentsdidinfacthavesomewhathighermathachievementincharterswithhighratesofnonwhiteteachers,relativetotheircounterpartsinschoolswithraciallybalancedfaculties.Insection 3.3.3 ,Idiscusshowthispremiummayhavebeenaectedbythewaystudentssortedintoraciallyimbalancedschools. 108

PAGE 109

3{2 : ZkiGT=MiTkM+FsTkF+GT+ki+"kiGT(3{3) AsinEquation 3{2 ,Equation 3{3 includescontrolsforstudentmobility(MiT)andfacultycharacteristics(FsT).Students'inherentgrowthisrepresentedbyaxedeect(ki),ratherthanlaggedachievement.Notethatinorderforfuturecharterstudentstohavevalidxedeectsestimates,theymusthavehadatleastfourcontinuousyearsoftestingdata(veyearsisthemaximumlengthofastudent'stimeseries),includingatleastthreeyearsinmainstreamschools.Igeneratedxedeectsestimatesfor10,160futurecharterstudents,representing24.9percentofcharterstudentsfromsection 3.3.2 .Thevastmajorityofthesestudents(94percent)ultimatelysortedintobalancedornonwhite-imbalancedcharterschools. Tables 3-9 and 3-10 summarizeaveragestudentxedeectsonmathandreadingexams,respectively,forstudentsinschoolswithraciallybalancedfaculties(^kbal),andaveragexedeectforstudentswithmoderatelyorseverelynonwhite-imbalancedfaculties(^kmod;^ksev).ThethirdcolumnofTable 3-9 liststhecharter-mainstreamdierenceinaveragemathxedeectsforeachfacultycompositioncategory.Amongallstudentsinseverelyraciallyimbalancedschools,charterstudentshad0.026sdlowermathxed 109

PAGE 110

3-10 ,amongallstudentsinschoolswithraciallybalancedfaculties,charterstudentshad0.031sdlowerreadingxedeectsthanmainstreamstudents.Otherwise,charterstudentswerenotstatisticallydierentfrommainstreamstudentsinsimilarlystaedschools.ThefourthrowofTables 3-9 and 3-10 liststhebalanced-moderateimbalancedierenceinaveragexedeects(^kbal^kmod),foreachpubliceducationsector.Studentsinbalancedschoolsoutperformedstudentsinmoderatelyimbalancedschoolsinmath,butnotreading.Thefthrowofeachtableliststhebalanced-severeimbalancedierence(^kbal^ksev),bysector.Again,studentsinbalancedschoolshadhighermathxedeects,byasmuchas0.049sdinthechartersector,butsignicantlylowerreadingxedeects.Sostudentswhosortedintobalancedcharterschoolstendedtohavehigherinherentmathgains,butlowerreadinggains,thanstudentswhosortedintoimbalancedcharters.Butthesamecouldbesaidformainstreamstudents;mainstreamstudents'mathxedeectswerehigherinschoolswithraciallybalancedfaculties,butreadingxedeectswerehigherinschoolswithnonwhite-imbalancedfaculties. Charterstudents'electivesortingwouldnothavebiasedresultsagainstnonwhitecharterfacultiesunlessthebalanced-imbalancedgapwaswiderinthechartersectorthaninthemainstreamsector.Thatis,theestimatedbenettoblackcharterstudentachievementthatIattributedtolargelynonwhitefacultieswouldhaveunderstatedthetruebenetifstudentswhosortedintotheseschoolshadlowerrelativegrowthratesthanmainstreamstudentswithnonwhite-imbalancedfaculties.TheintersectionsofdierencerowsandcolumnsinTables 3-9 and 3-10 showthecross-sectordierenceinbalanced-imbalancedgaps,calculatedaccordingtothefollowingexpression: [^kbal^kj]charter[^kbal^kj]mainstreamj=mod;sev(3{4) Themathachievementgapbetweenbalancedandseverelynonwhite-imbalancedschoolswas0.020sdwiderinthechartersector,andthisdierencewasmarginallysignicant 110

PAGE 111

3-5 through 3-8 )mayhaveoverstatedthetrueeect. Thisrobustnesscheckwaslimitedbythefactthat75percentofcharterstudentshadinsucientexperienceinmainstreamschoolstoassesstheirinherentgrowthpriortoenrollinginacharterschool.Nonetheless,theanalysisprovidesevidencethatthecorrelationbetweendisproportionateratesofnonwhiteteachersandhighermathperformanceamongblackmalesmayhavebeenunder-estimated. 111

PAGE 112

Racialsegregation(orre-segregation)isanincreasinglycommonoutcomeofpublicschoolchoiceplans.Thebestavailableevidencesuggeststhatsegregationattributabletoschoolchoiceincreasestheachievementgapsbetweenwhiteandblackstudents,orbetweenhigh-SESandlow-SESstudents(see,e.g., BifulcoandLadd ( 2007 )and Hastingsetal. ( 2008 )),andmyndingsreiteratethispoint.Charterschoolshaveconsiderableexibilitytorecruitnontraditionalteachers.NorthCarolina'schartersystem,likemanystatesystems,hasrelaxedlicensureandpaystandardsforcharterfaculties.Thisexibilitymayhaveallowedthestate'scharterschoolstorecruitmoremalesandnonwhiteteachers,twounder-representedgroupsintraditional,mainstreampublicschools.Ishowthatwhilesomeblackstudentsbenettedfromaninuxofnonwhiteteachersincharterschools,theirgainswereinsucienttoslowthegrowinggapbetweenstudentsinpredominantlywhiteversuspredominantlynonwhiteschools. 112

PAGE 113

Blackfemalepublicschoolstudents:Achievementandracialimbalancesummarystatistics CharterMainstreamDierence Mathlevelscore-0.143-0.008-0.135(0.805)(0.748)Mathgain0.2540.286-0.032(0.470)(0.446)Readinglevelscore0.0300.069-0.039(0.786)(0.749)Readinggain0.2940.301-0.007(0.511)(0.499)Attendsschoolwithnonwhitestudentimbalance0.4680.2810.187(0.499)(0.449)Attendsschoolwithwhitestudentimbalance0.0700.0480.022(0.255)(0.213)Attendsschoolwithnonwhitefacultyimbalance0.4520.1800.272(0.498)(0.384)Attendsschoolwithwhitefacultyimbalance0.0910.0250.066(0.288)(0.156)Attendsschoolwithnonwhitestudentand0.3950.1300.265facultyimbalances(0.489)(0.336)Attendsschoolwithwhitestudentand0.0240.0110.013facultyimbalances(0.153)(0.105)n(studentyears)6,83583,473 Notes:Standarddeviationsareinparenthesesbeloweachmean.denotesastatisticallysignicantdierenceat95%condence.Forcomputationalconvenience,Isampled20%ofstudentswhowereneverobservedinacharterschool.nincludesallsampledstudents(including100%ofallstudentswhowereeverobservedinacharterschool)withatleastthreeyearsofmathandreadingtestdata,byrace/gender. 113

PAGE 114

Blackmalepublicschoolstudents:Achievementandracialimbalancesummarystatistics CharterMainstreamDierence Mathlevelscore-0.203-0.070-0.133(0.776)(0.774)Mathgain0.2470.271-0.025(0.486)(0.468)Readinglevelscore-0.146-0.080-0.066(0.803)(0.798)Readinggain0.2980.301-0.003(0.556)(0.546)Attendsschoolwithnonwhitestudentimbalance0.4960.2910.205(0.500)(0.454)Attendsschoolwithwhitestudentimbalance0.0670.0500.017(0.250)(0.219)Attendsschoolwithnonwhitefacultyimbalance0.4860.1890.297(0.500)(0.391)Attendsschoolwithwhitefacultyimbalance0.0840.0250.058(0.277)(0.157)Attendsschoolwithnonwhitestudentand0.4280.1370.291facultyimbalances(0.495)(0.344)Attendsschoolwithwhitestudentand0.0210.0120.009facultyimbalances(0.143)(0.108)n(studentyears)6,01773,096 Notes:Standarddeviationsareinparenthesesbeloweachmean.denotesastatisticallysignicantdierenceat95%condence.Forcomputationalconvenience,Isampled20%ofstudentswhowereneverobservedinacharterschool.nincludesallsampledstudents(including100%ofallstudentswhowereeverobservedinacharterschool)withatleastthreeyearsofmathandreadingtestdata,byrace/gender. 114

PAGE 115

Whitemalepublicschoolstudents:Achievementandracialimbalancesummarystatistics CharterMainstreamDierence Mathlevelscore0.5740.611-0.037(0.857)(0.843)Mathgain0.2840.327-0.043(0.461)(0.437)Readinglevelscore0.7100.6510.059(0.751)(0.759)Readinggain0.2750.296-0.022(0.479)(0.474)Attendsschoolwithnonwhitestudentimbalance0.0160.051-0.035(0.126)(0.220)Attendsschoolwithwhitestudentimbalance0.4480.1210.327(0.497)(0.326)Attendsschoolwithnonwhitefacultyimbalance0.0180.022-0.004(0.134)(0.146)Attendsschoolwithwhitefacultyimbalance0.1220.0240.098(0.328)(0.154)Attendsschoolwithnonwhitestudentand0.0060.009-0.003facultyimbalances(0.078)(0.094)Attendsschoolwithwhitestudentand0.1160.0180.098facultyimbalances(0.320)(0.133)n(studentyears)14,303172,325 Notes:Standarddeviationsareinparenthesesbeloweachmean.denotesastatisticallysignicantdierenceat95%condence.Forcomputationalconvenience,Isampled20%ofstudentswhowereneverobservedinacharterschool.nincludesallsampledstudents(including100%ofallstudentswhowereeverobservedinacharterschool)withatleastthreeyearsofmathandreadingtestdata,byrace/gender. 115

PAGE 116

Whitefemalepublicschoolstudents:Achievementandracialimbalancesummarystatistics CharterMainstreamDierence Mathlevelscore0.6230.641-0.017(0.877)(0.877)Mathgain0.2830.323-0.040(0.472)(0.456)Readinglevelscore0.6070.5600.047(0.782)(0.796)Readinggain0.2720.298-0.026(0.509)(0.502)Attendsschoolwithnonwhitestudentimbalance0.0220.054-0.032(0.146)(0.225)Attendsschoolwithwhitestudentimbalance0.4190.1230.296(0.493)(0.329)Attendsschoolwithnonwhitefacultyimbalance0.0220.024-0.002(0.147)(0.153)Attendsschoolwithwhitefacultyimbalance0.1120.0250.087(0.316)(0.158)Attendsschoolwithnonwhitestudentand0.0070.009-0.003facultyimbalances(0.082)(0.097)Attendsschoolwithwhitestudentand0.1060.0190.086facultyimbalances(0.307)(0.138)n(studentyears)13,585171,626 Notes:Standarddeviationsareinparenthesesbeloweachmean.denotesastatisticallysignicantdierenceat95%condence.Forcomputationalconvenience,Isampled20%ofstudentswhowereneverobservedinacharterschool.nincludesallsampledstudents(including100%ofallstudentswhowereeverobservedinacharterschool)withatleastthreeyearsofmathandreadingtestdata,byrace/gender. 116

PAGE 117

Densityestimates:Percentofschools'studentswhowerenonwhite,bycharter/mainstreamdesignation 117

PAGE 118

Percentofschools'teacherswhowerenonwhite,bycharter/mainstreamdesignation 118

PAGE 119

Localpolynomial(degreezero,meansmoothing)estimates:Nonparametricrelationshipbetweennonwhiteteachersandnonwhitestudentsinschools. 119

PAGE 120

Theeectsofcharterenrollmentandracialimbalanceonblackfemaleachievement EconometricmodelLevelsLevelsGainsGainsSubjectMathReadingMathReading Enrolledincharter0.2220.1090.1850.034(6.42)(3.06)(5.34)(0.93)Movedtocharter-0.116-0.055-0.124-0.050(9.81)(4.31)(10.04)(3.61)Movedfromcharter0.0470.0200.0930.055(4.14)(1.72)(7.91)(4.31)Femaleteachers(%)0.0270.0190.0250.016(13.23)(8.82)(12.01)(7.42)Charterx%femaleteachers-0.029-0.011-0.023-2.4e-3(6.72)(2.36)(5.32)(0.52)Nonwhitestudentimbal.-0.014-0.016-1.8e-32.3e-3(3.50)(3.71)(0.46)(0.54)Charterxnonwhitestudentimbal.-0.043-0.038-0.0173.5e-4(2.64)(2.20)(1.04)(0.02)Whitestudentimbal.0.0310.0190.0194.3e-3(4.30)(2.62)(2.75)(0.61)Charterxwhitestudentimbal.-0.0210.013-0.0270.011(0.95)(0.53)(1.25)(0.46)Moderatenonwhitefacultyimbal.-0.020-0.011-0.017-3.8e-3(3.98)(1.91)(3.29)(0.68)Charterxmoderatenonwhitefacultyimbal.0.0120.0170.0200.020(0.60)(0.77)(0.93)(0.85)Severenonwhitefacultyimbal.-0.031-7.1e-3-0.0231.2e-4(3.83)(0.79)(2.78)(0.01)Charterxseverenonwhitefacultyimbal.0.035-5.5e-30.0408.9e-3(1.87)(0.28)(2.17)(0.44)Whitefacultyimbal.-0.012-2.4e-3-0.0084.6e-3(1.18)(0.23)(0.80)(0.41)Charterxwhitefacultyimbal.0.1850.1030.1280.049(7.45)(3.97)(5.37)(1.84) Notes:n=90,308student-years.CoecientsestimatesforEquation 3{1 (Levels)andEquation 3{2 (Gains),limitedtoblackfemales.Continuousfacultycharacteristicswerescaledsothatestimatedcoecientsaretheeectofa10percentage-pointincrease.Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Cluster-robuststandarderrorsallowforstudent-levelcorrelationinresiduals.Inschoolswitha\moderatenonwhitefacultyimbalance,"thepercentoffacultywhowerenonwhitewas20-40percentagepointsabovethecountymean.A\severenonwhitefacultyimbalance"indicatesthesharewasmorethan40percentagepointsabovethecountymean. 120

PAGE 121

Theeectsofcharterenrollmentandracialimbalanceonblackmaleachievement EconometricmodelLevelsLevelsGainsGainsSubjectMathReadingMathReading Enrolledincharter-0.0440.022-0.0250.025(3.22)(1.38)(1.83)(1.57)Movedtocharter-0.102-0.059-0.112-0.068(7.76)(4.14)(8.09)(4.35)Movedfromcharter0.0320.0380.0800.083(2.55)(2.84)(6.14)(5.62)Maleteachers(%)-0.024-0.021-0.021-0.017(10.71)(8.44)(9.42)(6.88)Charterx%maleteachers0.0148.9e-30.0105.6e-3(3.07)(1.72)(2.10)(1.05)Nonwhitestudentimbal.-0.017-0.013-7.4e-34.3e-3(3.96)(2.63)(1.76)(0.89)Charterxnonwhitestudentimbal.-0.027-0.0246.1e-30.027(1.47)(1.16)(0.34)(1.30)Whitestudentimbal.0.0150.0154.0e-36.5e-3(1.86)(1.81)(0.54)(0.80)Charterxwhitestudentimbal.-0.013-0.065-0.011-0.068(0.51)(2.35)(0.44)(2.52)Moderatenonwhitefacultyimbal.-0.034-0.026-0.028-0.020(5.85)(4.12)(4.89)(3.02)Charterxmoderatenonwhitefacultyimbal.0.0290.0180.0230.018(1.30)(0.68)(1.00)(0.66)Severenonwhitefacultyimbal.-0.027-0.013-0.0134.1e-4(3.03)(1.35)(1.47)(0.04)Charterxseverenonwhitefacultyimbal.0.047-0.0230.043-0.031(2.27)(0.98)(2.03)(1.33)Whitefacultyimbal.5.3e-3-1.1e-38.0e-31.5e-3(0.48)(0.09)(0.74)(0.12)Charterxwhitefacultyimbal.0.1990.1560.1330.102(6.41)(4.88)(4.43)(3.26) Notes:n=79,113student-years.CoecientsestimatesforEquation 3{1 (Levels)andEquation 3{2 (Gains),limitedtoblackmales.Continuousfacultycharacteristicswerescaledsothatestimatedcoecientsaretheeectofa10percentage-pointincrease.Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Cluster-robuststandarderrorsallowforstudent-levelcorrelationinresiduals.Inschoolswitha\moderatenonwhitefacultyimbalance,"thepercentoffacultywhowerenonwhitewas20-40percentagepointsabovethecountymean.A\severenonwhitefacultyimbalance"indicatesthesharewasmorethan40percentagepointsabovethecountymean. 121

PAGE 122

Theeectsofcharterenrollmentandracialimbalanceonwhitefemaleachievement EconometricmodelLevelsLevelsGainsGainsSubjectMathReadingMathReading Enrolledincharter0.0750.1900.0530.117(2.99)(7.13)(2.14)(4.59)Movedtocharter-0.135-0.082-0.149-0.091(14.82)(8.69)(15.96)(8.96)Movedfromcharter0.1340.0760.1730.106(11.46)(6.79)(14.72)(8.66)Femaleteachers(%)0.0160.0150.0150.015(11.96)(10.76)(11.26)(10.07)Charterx%femaleteachers-0.015-0.023-0.009-0.013(4.70)(6.93)(2.99)(4.07)Nonwhitestudentimbal.-0.020-0.011-0.0090.002(4.31)(2.37)(2.26)(0.34)Charterxnonwhitestudentimbal.-0.0430.025-0.0330.037(1.42)(0.76)(1.04)(1.09)Whitestudentimbal.0.016-0.0010.009-0.004(4.57)(0.37)(2.73)(1.30)Charterxwhitestudentimbal.0.0520.0070.0500.010(6.51)(0.89)(6.61)(1.26)Nonwhitefacultyimbal.-0.049-0.032-0.028-0.003(6.94)(4.14)(4.01)(0.37)Charterxnonwhitefacultyimbal.0.027-0.0630.038-0.059(0.91)(1.97)(1.25)(1.72)Whitefacultyimbal.0.032-0.0010.0310.001(4.08)(0.15)(3.97)(0.16)Charterxwhitefacultyimbal.-0.0230.014-0.032-0.001(1.55)(0.94)(2.15)(0.04) Notes:n=186,628student-years.CoecientsestimatesforEquation 3{1 (Levels)andEquation 3{2 (Gains),limitedtowhitefemales.Continuousfacultycharacteristicswerescaledsothatestimatedcoecientsaretheeectofa10percentage-pointincrease.Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Cluster-robuststandarderrorsallowforstudent-levelcorrelationinresiduals.Inschoolswitha\moderatenonwhitefacultyimbalance,"thepercentoffacultywhowerenonwhitewas20-40percentagepointsabovethecountymean.A\severenonwhitefacultyimbalance"indicatesthesharewasmorethan40percentagepointsabovethecountymean. 122

PAGE 123

Theeectsofcharterenrollmentandracialimbalanceonwhitemaleachievement EconometricmodelLevelsLevelsGainsGainsSubjectMathReadingMathReading Enrolledincharter-0.075-0.047-0.038-0.017(8.33)(4.83)(4.39)(1.77)Movedtocharter-0.137-0.066-0.151-0.077(14.53)(6.78)(15.56)(7.19)Movedfromcharter0.1410.0990.1770.134(11.90)(8.43)(14.67)(10.53)Maleteachers(%)-0.014-0.015-0.012-0.014(9.59)(10.03)(8.74)(8.83)Charterx%maleteachers0.0210.0240.0140.016(6.21)(6.80)(4.23)(4.56)Nonwhitestudentimbal.-0.017-0.002-0.0080.008(3.68)(0.43)(1.77)(1.75)Charterxnonwhitestudentimbal.-0.011-0.005-0.005-0.003(0.39)(0.18)(0.18)(0.08)Whitestudentimbal.0.0120.0040.006-0.001(3.23)(1.04)(1.84)(0.17)Charterxwhitestudentimbal.0.044-0.0150.037-0.018(5.17)(1.77)(4.58)(2.18)Nonwhitefacultyimbal.-0.053-0.029-0.034-0.003(7.15)(3.58)(4.79)(0.41)Charterxnonwhitefacultyimbal.-0.020-0.0140.0050.016(0.74)(0.48)(0.18)(0.56)Whitefacultyimbal.0.0440.0100.0440.015(5.47)(1.15)(5.40)(1.67)Charterxwhitefacultyimbal.-0.0030.019-0.0100.001(0.17)(1.17)(0.68)(0.08) Notes:n=185,211student-years.CoecientsestimatesforEquation 3{1 (Levels)andEquation 3{2 (Gains),limitedtowhitemales.Continuousfacultycharacteristicswerescaledsothatestimatedcoecientsaretheeectofa10percentage-pointincrease.Absolutevaluesoft-statisticsareinparenthesesbeloweachcoecient.Cluster-robuststandarderrorsallowforstudent-levelcorrelationinresiduals.Inschoolswitha\moderatenonwhitefacultyimbalance,"thepercentoffacultywhowerenonwhitewas20-40percentagepointsabovethecountymean.A\severenonwhitefacultyimbalance"indicatesthesharewasmorethan40percentagepointsabovethecountymean. 123

PAGE 124

Averagestudentxedeectsonmathexams,bycharter/mainstreamdesignationandracialimbalance TypeofschoolCharterMainstreamDierence Raciallybalancedfaulty9.1e-30.014-5.2e-3(0.384)(0.219)(0.004)Moderatenonwhite-imbalancedfaculty0.014-0.0190.033(0.383)(0.229)(0.017)Severenonwhite-imbalancedfaculty-0.040-0.014-0.026(0.337)(0.231)(0.010)Dierence-5.3e-30.033-0.038(balanced-moderate)(0.017)(0.001)(0.017)[0.31][23.70][2.22]Dierence0.0490.0290.020(balanced-severe)(0.011)(0.003)(0.011)[4.60][11.01][1.86]n(student-years)10,160500,520 Notes:Studentxedeectswereestimatedformathandreadinggains,describedbyEquation 2{2 .Thetablegivestheaveragestudentmathxedeectsforthreetypesofschoolsineachsector:schoolswithraciallybalancedfaculties,schoolswithmoderatenonwhite-imbalancedfaculties(wherethepercentofnonwhiteteacherswas20-40percentagepointsabovethecountyaverage),andschoolswithseverenonwhite-imbalancedfaculties(wherethepercentofnonwhiteteacherswasmorethan40percentagepointsabovethecountyaverage).Charterstudentxedeectswereproxiedbythexedeectsestimatedforfuturecharterstudents,whowereobservedinmainstreamschoolsbeforemovingtothechartersector.Standarddeviationsareinparenthesesbeloweachmean,standarderrorsareinparenthesesbeloweachdierenceinmeans,andt-statisticsareinbracketsbeloweachdierenceinmeans. 124

PAGE 125

Averagestudentxedeectsonreadingexams,bycharter/mainstreamdesignationandracialimbalance TypeofschoolCharterMainstreamDierence Raciallybalancedfaulty-0.034-3.4e-3-0.031(0.459)(0.231)(0.005)Moderatenonwhite-imbalancedfaculty0.0164.9e-30.012(0.463)(0.239)(0.020)Severenonwhite-imbalancedfaculty0.0310.0150.015(0.435)(0.258)(0.013)Dierence(balanced-moderate)-0.051-8.3e-3-0.042(0.014)(0.003)(0.015)[3.52][3.07][2.89]Dierence(balanced-severe)-0.065-0.019-0.046(0.014)(0.003)(0.014)[4.76][6.46][3.31]n(student-years)10,160500,520 Notes:Studentxedeectswereestimatedformathandreadinggains,describedbyEquation 2{2 .Thetablegivestheaveragestudentreadingxedeectsforthreetypesofschoolsineachsector:schoolswithraciallybalancedfaculties,schoolswithmoderatenonwhite-imbalancedfaculties(wherethepercentofnonwhiteteacherswas20-40percentagepointsabovethecountyaverage),andschoolswithseverenonwhite-imbalancedfaculties(wherethepercentofnonwhiteteacherswasmorethan40percentagepointsabovethecountyaverage).Charterstudentxedeectswereproxiedbythexedeectsestimatedforfuturecharterstudents,whowereobservedinmainstreamschoolsbeforemovingtothechartersector.Standarddeviationsareinparenthesesbeloweachmean,standarderrorsareinparenthesesbeloweachdierenceinmeans,andt-statisticsareinbracketsbeloweachdierenceinmeans. 125

PAGE 126

|||(2007):\SchoolChoice,RacialSegregation,andTest-ScoreGaps:EvidencefromNorthCarolina'sCharterSchoolProgram,"JournalofPolicyAnalysisandManagement,26,31{56. |||(2008):\TheEectsofCharterSchoolsonTraditionalPublicSchoolStudentsinTexas:AreStudentsWhoStayBehindLeftBehind?"JournalofUrbanEconomics,64,123{145. http://www.jsonline.com/news/education/29423774.html 126

PAGE 127

|||(2007):\TeacherCredentialsandStudentAchievement:LongitudinalAnalysiswithStudentFixedEetcs,"EconomicsofEducationReview,26,673{682. |||(2004):\Teachers,Race,andStudentAchievement:EvidencefromaRandomizedExperiment,"TheReviewofEconomicsandStatistics,86,195{210. |||(2005):\ATeacherLikeMe:DoesRace,Ethnicity,orGenderMatter?"AmericanEconomicReview,95,158{165. |||(2007):\TeachersandtheGenerGapsinStudentAchievement,"JournalofHumanResources,42,528{554. 127

PAGE 128

|||(1997):\PublicSchools:MakeThemPrivate,"EducationEconomics,5,341{345. http://aida.econ.yale.edu/~jh529/HKS Combined 200806.pdf 128

PAGE 129

http://pubpol.duke.edu/centers/child/ep/nceddatacenter/ |||(2006):SchoolsandStangSurvey,2003-2004,Washington,DC:U.S.DepartmentofEducation,OceofEducationalResearch. http://sbepolicy.dpi.state.nc.us http://www.wral.com/news/local/story/2042877 129

PAGE 130

http://econweb.tamu.edu/workshops/PERC%20Applied%20Microeconomics/ Lori%20L.%20Taylor.pdf 130

PAGE 131

CelesteK.CarrutherswasborninGarner,NorthCarolina.Herparents,BarbaraandBillCarruthers,spentmostoftheircareersaseducatorsandspecialistsintheWakeCountyPublicSchoolSystem.CelestegraduatedmagnacumlaudefromAppalachianStateUniversityin2004,withmajorsinaccountingandeconomics.In2005,sheearnedaMasterofArtsdegree(economics)fromtheUniversityofNewHampshireandbeganfurthergraduatestudiesattheUniversityofFlorida.Herresearchinterestsincludepubliceconomics(withemphasisontheeconomicsofeducation),antitrusteconomics,andregulatoryeconomics.InAugust2009,shewilljointheDepartmentofEconomicsfacultyattheUniversityofTennessee,Knoxville. 131