Essays on the Economics of Higher Education

MISSING IMAGE

Material Information

Title:
Essays on the Economics of Higher Education
Physical Description:
1 online resource (174 p.)
Language:
english
Creator:
See, Edward Camaligan
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Economics
Committee Chair:
Kenny, Lawrence W
Committee Members:
Denslow, David A
Hamersma, Sarah E
Sindelar, Paul T

Subjects

Subjects / Keywords:
aa -- clast -- college -- economics -- education -- fcat
Economics -- Dissertations, Academic -- UF
Genre:
Economics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
Community colleges have played an increasingly important role in the nation's education and training system (Kane and Rouse, 1999), but little is known about the effects of education standards and college costs which shaped community college students' college-going decisions, educational attainment, and labor market success. I use rich administrative data on Florida students to analyze how education standards and college costs impact college enrollment and completion decisions, and to analyze the labor market returns to these decisions. First, I analyze how college education standards affect community college education. I determine how a college exit exam affects the number of semesters enrolled and the associates in arts degree completion. Using a regression discontinuity approach to compare the outcomes among students who narrowly pass and narrowly fail the College Level Academic Skills Test (CLAST), I find that failing the CLAST has no impact on the number of semesters enrolled but has small effects on associates in arts degree completion. Second, I (with Damon Clark) analyze how tougher standards affect student success. I particularly investigate how a tougher exit exam affects high school completion, post-secondary enrollment, and earnings. Using the difference-in-difference method, I find small negative effects of increasing the passing scores in Florida's Comprehensive Assessment Test (FCAT) on high school graduation rates, two-year college enrollment, and earnings. I cannot detect any positive effects of higher exit exam standards, although such effects may be too small to be picked up with my data. Third, I analyze how community college costs affect college-going decisions and earnings. Using the college cost data from the Integrated Postsecondary Education Data System (IPEDS), I find small negative effects of a higher fraction of the cost of attending community college on the number of semesters enrolled in community college and associates in arts degree completion. Using the fraction of cost as an instrumental variable for community college education, I find large returns to community college persistence and associates degree completion. These estimates suggest large returns among students whose community college-going decisions are shifted by the higher cost of attending community college.
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.
Statement of Responsibility:
by Edward Camaligan See.
Thesis:
Thesis (Ph.D.)--University of Florida, 2012.
Local:
Adviser: Kenny, Lawrence W.
Electronic Access:
RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2014-08-31

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2012
System ID:
UFE0044338:00001


This item is only available as the following downloads:


Full Text

PAGE 1

ESSAYSONTHEECONOMICSOFHIGHEREDUCATION By EDWARDC.SEE ADISSERTATIONPRESENTEDTOTHEGRADUATESCHOOL OFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENT OFTHEREQUIREMENTSFORTHEDEGREEOF DOCTOROFPHILOSOPHY UNIVERSITYOFFLORIDA 2012

PAGE 2

c 2012EdwardC.See 2

PAGE 3

ToBubbles,andtomyfamily... 3

PAGE 4

ACKNOWLEDGMENTS Thejourneytowardsthecompletionofmydissertationpassedthroughalong andwindingroad.Quitebumpyinthebeginning,butsmoothsailedtowardstheend. Ifitweretomyownefforts,successwouldbeimpossible.Fortunately,theoneabove blessedmewithpeoplewhohelpedandcontributedtoaccomplishthispieceofwork. Manyindividualsandorganizationshelpedtowardsthecompletionofthis dissertation.IwouldliketothankSarahHamersma,DavidDenslow,PaulSindelar, andRichardRomanoforusefulcommentsandforadvisingthisdissertationthroughout itscompletion.MysincerestthanksgotoDamonClarkandLarryKenny.Despite theirbusyschedules,theymanagedtogivecomments,suggestions,andideasthat shapedthisdissertation.Completingthisdissertationwouldbeimpossiblewithouttheir guidance.IamgratefultotheLockhartTravelEndowmentsforthetravelsupport.Iam thankfultotheFloridaDepartmentofEducationforprovidingtherichhighschooland communitycollegeeducationdata.Iamthankfultotheparticipantsatthe2011and 2012AssociationforEducationFinanceandPolicymeetinggsandthe2011Southern EconomicsAssociationmeetings.Iwouldliketothanktheseminarparticipantsat CharlesRiverAssociates,theERSGroup,andtheUniversityofFloridaforuseful commentsandsuggestions. Iacknowledgethesupportprovidedbymyfamilyandfriends,especiallytomy parentswhoarealwaysbeentheretosupportallmyendeavors,tomyUFPinoyfriends, andtoBubbleswhohasbeenmyinspirationincompletingthiswork. Lastly,IthankGodmycreatorandsavior,whoneverleftmeespeciallyduringthe timesoffailureandgreatneed.Iofferyou,myLord,thispieceofwork. 4

PAGE 5

TABLEOFCONTENTS page ACKNOWLEDGMENTS..................................4 LISTOFTABLES......................................7 LISTOFFIGURES.....................................9 ABSTRACT.........................................10 CHAPTER 1EFFECTSOFCOLLEGEEDUCATIONSTANDARDS..............12 1.1Introduction...................................12 1.2Background...................................15 1.2.1ReviewofRelatedLiterature......................15 1.2.2TheFloridaCLASTSystem......................17 1.3Data.......................................19 1.3.1DataConstruction...........................19 1.3.2CLASTMinimumScoreandCollegeOutcomeVariables......20 1.4EmpiricalStrategy...............................22 1.5Results.....................................25 1.5.1EstimatesoftheEffectofFailingtheCLASTontheNumberof SemestersEnrolled...........................25 1.5.2EstimatesoftheEffectofFailingtheCLASTonAssociatesinArts Degree..................................30 1.5.3RobustnessCheck...........................34 1.5.3.1Functionalform.......................34 1.5.3.2VaryingtheCLASTscorepointsincludedintheestimation sample............................35 1.6Discussion...................................35 1.6.1TheRolePlayedbytheNumberofPossibleTestAttempts.....36 1.6.2TheRolePlayedbytheAlternativeWaysofMeetingtheCLAST Requirements.............................38 1.6.3RelationshiptotheLiteratureonEffectsofHighSchoolExitExam39 1.7Conclusion...................................40 2THEIMPACTOFTOUGHEREDUCATIONSTANDARDS............69 2.1Introduction...................................69 2.2HighSchoolStandardsinFlorida.......................71 2.3TheoreticalModelandEmpiricalStrategy..................74 2.3.1EmpiricalStrategy...........................75 2.3.2SimpleModel..............................77 2.3.2.1Outcomesunderthelowstandard.............78 5

PAGE 6

2.3.2.2Theimpactsofahigherstandard..............79 2.3.3Discussion................................81 2.4Data.......................................81 2.4.1DataTypes...............................81 2.4.2AnalysisSample............................83 2.5Results.....................................83 2.5.1HighSchoolOutcomes.........................84 2.5.2PostsecondaryEducation.......................86 2.5.3Earnings.................................87 2.5.4RobustnessChecks..........................88 2.6DiscussionandInterpretation.........................89 2.7Conclusion...................................92 3EFFECTSOFCOMMUNITYCOLLEGEATTENDANCECOSTS........100 3.1Introduction...................................100 3.2Background...................................102 3.2.1ReviewofRelatedLiterature......................102 3.2.1.1Costofattendanceandcommunitycollegeeducation..102 3.2.1.2Returnstocommunitycollegeeducation.........105 3.2.2TheCostofAttendingaFloridaCommunityCollege........109 3.3Data.......................................112 3.3.1DataConstruction...........................112 3.3.2TheFractionofCostofAttendingtheNearestCommunityCollege andAnnualEarnings..........................115 3.4EmpiricalFramework..............................119 3.4.1ModelofSchoolingChoiceandEarnings..............120 3.4.2IdenticationConditions........................121 3.4.3TestsontheValidityoftheFractionofCollegeCost.........124 3.5Results.....................................129 3.5.1FirstStageResults...........................130 3.5.2InstrumentalVariablesEstimates...................133 3.6FalsicationTests................................138 3.7Discussion...................................141 3.8Conclusion...................................145 APPENDIX ATHEROLEPLAYEDBYTHENUMBEROFPOSSIBLETESTATTEMPTS..164 BPROPERTIESOFTHEGRAPHSINFIGURE2-1................166 REFERENCES.......................................169 BIOGRAPHICALSKETCH................................174 6

PAGE 7

LISTOFTABLES Table page 1-1NumberofCLASTTakersbyYearandSemester.................54 1-2CLASTPassingScores...............................55 1-3PercentageofStudentsWhoRetestedCLAST..................56 1-4DistributionofCLASTAttempts...........................57 1-5CLASTPassingRates................................58 1-6SummaryStatistics..................................59 1-7TestofRDAssumption...............................60 1-8CLASTEffectsbyCLASTAttempt.........................61 1-9CLASTEffectsbyGender..............................62 1-10CLASTEffectsbyRace...............................63 1-11CLASTEffectsbyYearofFirstCLASTAttempt..................64 1-12CLASTEffectsbyPercentofStudentsEverAttendedSUS...........65 1-13RDEstimatesontheNumberofSemestersEnrolledAftertheFirstCLAST Attempt........................................66 1-14RDEstimatesontheAssociatesinArtsDegreeCompletion...........67 1-15FractionofStudentswithAADegreebutdidnotPassCLASTEver.......68 2-1Descriptivestatistics.................................96 2-2Theimpactsoftougherstandardsonhighschooloutcomes...........97 2-3Theimpactsoftougherstandardsonpost-secondaryoutcomes.........98 2-4Falsicationtests...................................99 3-1EstimatesoftheCostofAttendingCommunityCollegesinFlorida.......148 3-2EstimatedTotalCost,AverageFinancialAid,andtheFractionofCost.....149 3-3SummaryStatistics..................................150 3-4NumberofSemestersEnrolledinCommunityCollege..............151 3-5AssociatesinArtsDegreeCompletionRates%.................152 3-6AverageAnnualTotalEarningsinYear2000$..................153 7

PAGE 8

3-7BalancingTest....................................154 3-8EstimatesoftheRegressionofFCATScoresonFractionofCost........155 3-9EstimatesoftheRegressionofCurrentYearFractionofCost onPriorYearCountyandCommunityCollegeCharacteristics.........156 3-10RegressionEstimatesofCollegeOutcomesonTestScores...........157 3-11FirstStageRegressionModelsReducedFormEstimates...........158 3-12FirstStageRegressionModelswithHeterogeneousEffects...........159 3-13InstrumentalVariablesEstimatesoftheReturnstoCommunityCollege EducationbasedontheFractionofCostofAttendingaNearbyCommunity College........................................160 3-14InstrumentalVariablesEstimateswithHeterogeneousEffects..........161 3-15RegressionEstimatesofSUSOutcomesontheFractionofCost........162 3-16InstrumentalVariablesEstimatesoftheReturnstoCommunityCollege EducationbasedontheInteractionoftheFractionofCost,LunchStatus,and Black.........................................163 8

PAGE 9

LISTOFFIGURES Figure page 1-1ProbabilityofPassingtheFirstCLASTTake....................42 1-2HistogramandMcCraryTestResultsforCLASTScore..............43 1-3McCraryTestResultsbyCLASTAttempt.....................44 1-4TestofRegressionDiscontinuityAssumption...................45 1-5RD:SemestersEnrolledAftertheFirstCLASTAttempt.............46 1-6RD:SemestersEnrolledAfter2ndto9thCLASTAttempt............47 1-7RD:SemestersEnrolledbytheExamYearofFirstCLASTAttempt.......48 1-8RD:SemestersEnrolledbyTypeofInstitution...................49 1-9RD:AssociatesinArtsDegreeAftertheFirstCLASTAttempt..........50 1-10RD:AssociatesinArtsDegreebyCLASTAttempt................51 1-11RD:AssociatesinArtsDegreebytheExamYearofFirstCLASTAttempt...52 1-12RD:AssociatesinArtsDegreebyTypeofInstitution...............53 2-1TheImpactofTougherStandards..........................95 2-2GroupDenitions...................................95 9

PAGE 10

AbstractofDissertationPresentedtotheGraduateSchool oftheUniversityofFloridainPartialFulllmentofthe RequirementsfortheDegreeofDoctorofPhilosophy ESSAYSONTHEECONOMICSOFHIGHEREDUCATION By EdwardC.See August2012 Chair:LawrenceW.Kenny Major:Economics Communitycollegeshaveplayedanincreasinglyimportantroleinthenations educationandtrainingsystemKaneandRouse,1999,butlittleisknownaboutthe effectsofeducationstandardsandcollegecostswhichshapedcommunitycollege studentscollege-goingdecisions,educationalattainment,andlabormarketsuccess. IuserichadministrativedataonFloridastudentstoanalyzehoweducation standardsandcollegecostsimpactcollegeenrollmentandcompletiondecisions, andtoanalyzethelabormarketreturnstothesedecisions.First,Ianalyzehowcollege educationstandardsaffectcommunitycollegeeducation.Ideterminehowacollege exitexamaffectsthenumberofsemestersenrolledandtheassociatesinartsdegree completion.Usingaregressiondiscontinuityapproachtocomparetheoutcomesamong studentswhonarrowlypassandnarrowlyfailtheCollegeLevelAcademicSkillsTest CLAST,IndthatfailingtheCLASThasnoimpactonthenumberofsemesters enrolledbuthassmalleffectsonassociatesinartsdegreecompletion. Second,IwithDamonClarkanalyzehowtougherstandardsaffectstudent success.Iparticularlyinvestigatehowatougherexitexamaffectshighschool completion,post-secondaryenrollment,andearnings.Usingthedifference-in-difference method,IndsmallnegativeeffectsofincreasingthepassingscoresinFloridas ComprehensiveAssessmentTestFCATonhighschoolgraduationrates,two-year 10

PAGE 11

collegeenrollment,andearnings.Icannotdetectanypositiveeffectsofhigherexitexam standards,althoughsucheffectsmaybetoosmalltobepickedupwithmydata. Third,Ianalyzehowcommunitycollegecostsaffectcollege-goingdecisionsand earnings.UsingthecollegecostdatafromtheIntegratedPostsecondaryEducation DataSystemIPEDS,Indsmallnegativeeffectsofahigherfractionofthecostof attendingcommunitycollegeonthenumberofsemestersenrolledincommunitycollege andassociatesinartsdegreecompletion.Usingthefractionofcostasaninstrumental variableforcommunitycollegeeducation,Indlargereturnstocommunitycollege persistenceandassociatesdegreecompletion.Theseestimatessuggestlargereturns amongstudentswhosecommunitycollege-goingdecisionsareshiftedbythehigher costofattendingcommunitycollege. 11

PAGE 12

CHAPTER1 EFFECTSOFCOLLEGEEDUCATIONSTANDARDS 1.1Introduction Asapartofaccountability-basedreform,theCommissiononFutureHigher Education,acommissionwithintheUSDepartmentofEducation,consideredimplementing exitexaminationsforcollegeanduniversitystudents. 1 Theexitexamswouldhave measuredproblemsolving,criticalthinking,andcommunicationskillsofgraduating collegestudents.Althoughneverimplementedatthenationallevel,severalstatessuch asCalifornia,Florida,andTennesseehaveadoptedcollegeexitexams. 2 Thesethree statesrequirestudentstotaketheexambeforegraduationbutonlyFloridarequires studentstopasstheexam.FailingtopassFlorida'sCollegeLevelAcademicSkillsTest CLASTdeniesastudenttoreceivetheassociatesinartsAAdegree. Adebatesurroundsexitexams.Ononesideofthedebate,proponentspush theideathatexitexamsareusefulmeasuresofastudent'sability.Anexitexamwill gaugeastudent'supper-divisionlevelcompetencyandhisreadinesstowork.Also, proponentsofexitexamsarguethatcredentialsaretiedtostandardsBetts,1998; Costrell,1994;incentivizingstudentswillmakethemworkharderresultinginhigher schoolperformance.Oneincentiveisdrivenbyhigherreturnstopassing.Ifwages arehigherforworkerswithadiploma,andifadiplomaisearneduponpassingthe exitexam,studentswouldhavetheincentivetopasstheexitexam.Ontheotherside ofthedebate,criticsbelievethatexitexamsjustaddtothelonglistofgraduation requirements.Thesecriticsarguethatanadditionalgraduationrequirementdecreases 1 SeeU.S.DepartmentofEducation2006CommissiononFutureHigherEducation Report. 2 CaliforniaimplementedtheCaliforniaCriticalThinkingandSkillsTestCCTST; TennesseeusedtheMeasureofAcademicProciencyandProgress;Florida administeredtheCollegeLevelAcademicSkillsTest. 12

PAGE 13

thechancethatsomestudents,particularlytheminorityandtheonesbelongingtothe lowsocio-economicclass,willgetadiploma.Inadditiontopreventingsomestudents fromgraduatingtheCan'tPassEffect,exitexamsmightdiscouragestudentsfrom attendingcollegeorcausethemtodropoutofcollegeearlierthantheywouldhadthey notfacedtheexamtheDiscouragementEffect. Theargumentsofthedebateechothosemadeaboutthestandardstiedtoother educationalcredentialssuchasthehighschoolexitexamsthatstudentsmustpassto receiveahighschooldiplomainsomestates.Yetwhilemuchisknownabouttheimpact ofhighschoolexitexamsDeeandJacob,2006;Martorell,2005;Papay,Murnane andWillett,2008;Reardonetal.,2009,littleisknownabouttheimpactofcollegeexit exams. IconsidertheeffectsoftheFloridacollegeexitexam:theCLAST.Therearetwo reasonswhyaCLASTanalysiscanshedlightontheissueofcollegeexitexams.First, asnotedabove,theCLASTistheonlyhigh-stakesexaminhighereducationtohave beenemployedintheUnitedStates.Second,theCLASTisinsomewaysdesignedto minimizeCan'tPasseffects.StudentscantaketheCLASTmultipletimesuntilthey pass.Thisgivesmetheopportunitytoinvestigatehowmultipletesttakingmitigatesany Can'tPasseffects. ToassesstheimpactsoftheCLASToncollegeoutcomesi.e.obtainingassociates inartsdegreeandwhetherstudentswerediscouragedbythetest,Iusearegression discontinuitydesign,comparingoutcomesamongstudentswhonarrowlyfailand narrowlypasstherstadministrationofthistest.Theregressiondiscontinuityapproach hasanumberofadvantages.First,sincethetwosetsofstudentsshouldbethesame onallobservableandunobservabledimensions,thiscomparisonshouldisolatethe causaleffectsofpassingtheCLAST.Second,Icanassesstowhatextenttheeffects ofpassingtheCLASTdependonthesubsequenttestingopportunitiesavailabletothe students.Forexample,Icancomparetheimpactoffailingthersttestontheeventual 13

PAGE 14

graduationwiththeimpactwewouldobserveifstudentswereonlyallowedtotake thetestonce,twice,thrice,andsoon.Thedrawbackoftheregressiondiscontinuity approachisthatIcannotassesseffectsonstudentsawayfromthepassingthreshold. MyestimatessuggestthatfailingtheCLASTontherstattempthasnoimpacton thenumberofsemestersstudentswereenrolledincommunitycollegeaftertherst CLASTattempti.e.,nodiscouragementeffectsandhasamoderatenegativeimpact ontheprobabilityofreceivinganAAdegreei.e.,can'tpasseffects.WhenIanalyzeby raceandgender,Indnodiscouragementeffectsforanygroup.Idohoweverndthat Can'tPasseffectsarelargeramongAfricanAmericans,Hispanicsandfemales.Ishow thatwhiletheCan'tpasseffectismoderatearoundvepercentagepoints-itwould besubstantiallylargerifstudentswerenotgivenmultiplechancestopassthetest. ThissuggeststhatthedesignoftheCLASTcangreatlyaffecttheoutcomes,especially amongAfricanAmericans,Hispanics,andfemales. Twocaveatsshouldbeappliedtothendingsinthispaper.First,Icannotaddress thebenetsofcollegeexitexamssincethedatausedaretakenfromanenvironment inwhichanexamwasinplace.Assuch,theanalysiscannotspeaktotherelative advantagesordisadvantagesoftheCLASTsystem.Second,theresultsinthispaper holdonlyforthesubsampleofthepopulationbeingstudied:namelythestudentswho narrowlypassandnarrowlyfailtheCLAST.Icannotassessimpactsonsomestudents whosetestscoresareawayfromtheCLASTpassingthreshold. Despitethesecaveats,theseresultssuggestthattheimpactsofcollegeaccountability couldgreatlydependonthedetailsoftheparticularaccountabilitysystem.Asthe accountabilitymovementgainsgroundinhighergrades,policy-makerswoulddowellto bearthisinmind. 14

PAGE 15

1.2Background 1.2.1ReviewofRelatedLiterature Collegeexitexamsbegmanyinterestingquestions.First,whatarethewelfare effectsofintroducingtheseexamsandmakingthemmoreorlessdifcult?Exitexams arebasedontheideathatcredentialsshouldbetiedtoeducationalstandards.Costrell 1994studiestherelationshipbetweeneducationalstandardsandwelfare.Costrell's modelpredictsthathigherstandardswilldenystudentsthecredentialssuchasa diploma,therebyincreasingincomeinequality.Costrellsuggeststhatloweringstandards willincreasegraduationrates.Betts1998buildsonCostrell'smodelbutassumes thatstudentsdifferinability.Whenstudentsdifferinability,thesocialplannerinBetts's worldprefershigherstandards.Bettsarguesthatincreasingthestandardswillraise theaverageabilityofthegroupthatfailedtomeetthenewstandards.Providedthat marketspaywagesaccordingtomarginalproductivity,earningsriseforthegroupwho failedtomeetthestandards.ContrarytoCostrell'sresults,Bettsconcludedthatraising educationalstandardsincreasestheearningsofstudentswhomeetandtheearnings ofstudentswhofailedtomeetthestandards.Ultimatelythetheorydoesnotgiveclear predictionsonthewelfareeffectsofeducationstandards. Second,whateffectsdoexitexamshaveoncollegedropoutrates?Althoughthere hasbeenlittleempiricalworkoncollegeexitexams,anumberofempiricalresearch papershaveassessedtheimpactofhighschoolexitexams.Onestrandofthiswork assesseswhathappenstotheaggregatewagesandtheaggregatedropoutrateswhen astateintroducestheexam.Thesestudiesusedifference-in-differencemethodsto identifytheimpactofhighschoolexitexamsDeeandJacob,2006;Warren,Jenkins andKulick,2006.Thesestudiesndthatonaverage,highschooldropoutrates increasebyonetotwopercentagepointsinthestatesthatimplementedhighschool exitexams.Usinggrade-leveldropoutdatafromMinnesota,DeeandJacob,2006nd thathighschoolexitexamsincreasedropoutratesamongstudentsintheurbanand 15

PAGE 16

high-povertyschooldistrictsandamongstudentsinschoolswithhighpopulationsof minoritystudents.Usingacross-sectionofstates,Warren,JenkinsandKulick2006 ndthatdifcultexitexamsareassociatedwithlowerpublichighschoolgraduationrates andwithhigherratesofGEDtesttaking.Likethetheoreticalliterature,theempirical literatureonthisquestiondoesnottopointtotheseexamshavinganyclear-cuteffects. Asecondstrandofthisworkaskswhether,withinstatesthathaveexitexams, failinganinitialadministrationoftheexamimpactsvariousstudentoutcomes.Studies inthisstrandofworkhaveusedregressiondiscontinuitymethodstoestimatethe causaleffectsoffailingtheinitialexitexamforthosestudentsonthemarginofpassing. Martorell2005,therstpaperinthiseld,useddatafromTexas.Hendsthatstudents donotdropoutearlyinhighschooluponfailingtheexamintherstadministrationin the10thorinthe11thgrade.Studentswhomeetallthegraduationrequirementsbut narrowlyfailthelastchanceexaminthesenioryearofhighschoolhaveagreater chanceofearningaGEDbutarelesslikelytoearnahighschooldiplomacompared withstudentswhonarrowlypassthelastchanceexam.UsingdatafromMassachusetts, Papay,MurnaneandWillett2008studytheimpactsonon-timegraduationand schoolpersistence.Theyndnegativeeffectsamonglow-incomeurbanstudentson themarginofpassingthemathematicsexamsbutnoeffectsonothergroups.While Martorell2005andPapay,MurnaneandWillett2008focusedontheeffectsof highschoolexitexamondropoutratesandschoolpersistence,Reardonetal.2009 lookedattheeffectsoffailingtheCaliforniaHighSchoolExitExamCAHSEEon studentachievement,graduationrates,andschoolpersistence.Theyndthatfailing theCAHSEEinthe10thgradedoesnotaffectstudentachievement,highschool persistence,andgraduationrates. Thispaperusesaregressiondiscontinuitydesignsimilartothatemployedby Martorell2005,Papay,MurnaneandWillett2008,andReardonetal.2009.While thesestudiesfocusedontheeffectsofhighschoolexitexam,thispaperistherstto 16

PAGE 17

examinetheeffectsofcollegeexitexamonassociatesinartsdegreecompletionand communitycollegepersistence. 1.2.2TheFloridaCLASTSystem OnAugust1,1984,theFloridaDepartmentofEducationimplementedtheCollege LevelAcademicSkillsCLASprogram.Inthisprogram,allassociatesinartsdegree seekingstudentsinFloridapublicpostsecondaryinstitutionsarerequiredtosatisfythe CollegeLevelAcademicSkillsrequirement.PriortoJanuary1,1996,studentsmetthe CLASrequirementbypassingtheCollegeLevelAcademicSkillsTestCLAST.Asa nalrequirementaftercompleting60credithoursofcoursework,acollegestudentis conferredanassociatesinartsAAdegreeafterpassingtheCLAST. The`2+2Program'inthestateofFloridaallowsAAdegreeholderstotransferto anyofthestateuniversitiesinFlorida.Fromtheperspectiveofacollegestudentwishing toadvancetothestateuniversitysystemSUS,theCLASTisahigh-stakestest.The needfortheCLASTscoressteadilyincreasedovertheyearswiththelargestnumber ofexamineesin1995.Table1-1showsthenumberofCLASTtakerssinceitsinception in1982.By1997,aFloridabillhasallowedsomealternativewaystomeettheCLAS requirements.StudentscansatisfytheCLASrequirementsbyearningagradepoint averageof2.5andhigherinasubjectsidentiedbytheStateBoardofEducationSBE orbypresentingpassingscoresfromnationalstandardizedtestssuchasSATorACT. TheFloridaDepartmentofEducationdiscontinuedtheadministrationoftheCLASTon July1,2009. Allstudentswhoaccumulatedatleast18credithoursofcollegecourseworkare eligibletotaketheCLAST.EligiblestudentscantaketheCLASTonlyonceevery semesterFebruary,June,October.NotallstudentstaketheCLAST.Studentswith speciclearningdisabilitiesareexempted.Sincetheycannotcompleteoneormore CLASTsubtests,studentswithlearningdisabilitiesapplyandobtainCLASTwaivers. 17

PAGE 18

FoursubtestscomprisetheCLASTessay,reading,writing,andcomputational allofwhichmustbepassedtomeettheCLASTtestingrequirements.Theessaysection isa60minutecompositionwritingonagiventopic.The80minutereadingandwriting skillssectionsmeasureskillsinwordchoice,sentencestructure,grammarandspelling, andliteralandcriticalcomprehension.The90minutemathematicssectionteststhe knowledgeofarithmetic,geometry,algebra,statistics,andlogicalreasoning.TheFlorida DepartmentofEducationsetsthepassingscoresfortheseCLASTsubtests.These passingscoreshavechangedatleast4timesin1986,1989,1991,and1992where thelargestjumpinthepassingscoresoccurredin1989.Table1-2showstheCLAST passingscoresoftheCLASTsubtests.TheCLASTisdesignedinsuchawaythatthe difcultyofthequestionsiscomparablefromadministrationtoadministration. StudentswhofailedtheCLASTontherstattemptarepermittedtoretestasmany timesastheychoose.AstudentneedstoretaketheCLASTsectionsthatshefailed. Table1-3showsthepercentageofstudentswhoatleastretestedoneCLASTsubtest bysemester.Thepercentageofstudentswhoretestedsteadilyincreasedfromabout 32%inthefallofschoolyear1995-1996toabout45%inthesummerofschoolyear 2005-2006.Table1-4showsthedistributionofthenumberoftimestheCLASTwas taken.About53.8%ofthestudentstooktheCLASTatleastonce,23.3%atleasttwice, 10.5%atleastthrice,and12.4%fourormoretimes. Table1-5showstheCLASTpassingratesandthelikelihoodofpassingtheCLAST eventuallyafterfailingtheinitialexamacrossracialgroupsandgender.Theprobability ofpassingonthersttakeandthechanceofeventuallypassingtheCLASTafterfailing ontherstattemptareaboutthesameformalesandfemales.Table1-5revealsthe variationintheCLASTpassingratesacrossthemajorracialgroups.About51.7% ofWhitespasstheCLASTonrsttake,15.1%forBlacks,27.6%forHispanics.The chanceofpassingtheCLASTeventuallyforthosewhofailedtheCLASTthersttime is78.5%forWhites,47.5%forBlacks,and61.4%forHispanics.Theseobserved 18

PAGE 19

variationsintheinitialandeventualCLASTpassingratesacrossracialgroupspoint totheimpactoftheCLASTexampolicyondifferentracialgroups.Thesendings supportDavenportetal.2002notingthatuniformtestswithuniformpassingscores donotensureuniformapplicationofeducationalexpectationsforallstudentsp.13in Davenportetal.,2002. 1.3Data 1.3.1DataConstruction ThedatausedinthiswereobtainedfromFlorida'sK-20EducationalData WarehouseEDW.EDWintegratesinformationfrommanysourcesintoonedatabase, focusingonstudentswhoattendedFlorida'spubliceducationsystem.EDWfollows studentsfromschoolintopost-secondaryeducationandintothelabormarket.The EDWdatacombineseveraldatalesthatarelinkedatthestudentlevelusingaunique studentidentier.ThecombineddataincludeinformationonCLASTscores,student demographiccharacteristics,studentenrollmentrecordsincommunitycollegesand stateuniversities,andthedegreesawardedtothestudent. Thebasesampleconsistsof176,157studentswhoattendedFloridacommunity collegesandtooktheCLASTbetween1982and2006.Studentswhonevertookthe CLASTareexcludedfromthedata.ThesearethestudentswhoenrolledinFlorida communitycollegeswhomettheCLASrequirementswithouttakingtheCLASTe.g., becausetheyhavehighSATscores. Imakeanumberofsamplerestrictions.First,Iexcludestudentswhotookthe CLASTforthersttimebefore1995becausetheelectronicdatacollectionofcollege enrollmentrecordsonlystartedin1995.Second,Irestrictthesampletoincludeonly thosestudentswhotooktheCLASTwhileenrolledincommunitycollege.Inparticular, IdidnotconsiderthosestudentswhotooktheCLASTaftergainingadmissionstoany ofthestateuniversitiesinFlorida.Aboutlessthantenpercentofthestudentsinthe sampletooktheCLASTaftergainingadmissionstostateuniversities.Third,Irestrict 19

PAGE 20

thesampletoincludestudentswhotookallfourCLASTsubtestsatleastonce.Students whodidnottakeallfoursubtestscanbeconsiderednottohavetakentheCLASTat leastonce.About25percentofthesampledidnottakeallfoursubteststhersttime. Thenalsampleincludes52,111studentobservations.Summarystatisticsarereported inTable1-6.About68%ofthesampleareWhites,12%Blacks,and16%Hispanics. Femalestudentsaccountforabout67%ofthesample. 1.3.2CLASTMinimumScoreandCollegeOutcomeVariables Iusethesampledescribedabovetodenethefollowingvariables: CLASTScore .TheCLASTscoreistherunningvariableusedintheRDanalysis. TheCLASTconsistsoffoursubtestsmath,reading,writing,andessaywithdifferent passingscoresandscorerangesineachsubtest.ThesefeaturesoftheCLASTpresent atleastthreeproblemsfortheregressiondiscontinuityanalysis.First,anyofthe subtestscould,inprinciple,beusedastherunningvariablesincethereshouldbea discontinuityintherelationshipbetweenasinglesubtestscoreandwhetherornotthe wholeCLASTwaspassed.Second,thescorerangeinthemath,reading,andwriting subtestsaredifferentfromthescorerangeintheessaysubtest:math,reading,and writinghavemaximumscoresinthe400s,whileessayhasamaximumscoreoften. Third,althoughthescoresinmath,reading,andwritinghavethesamerange,the scoresinthesesubtestsareobservedatdifferentpointsintimeandhavedifferent passingscores.Toovercometheseproblems,Irstnormalizedmath,reading,and writingscorestohavezeroasthepassingthresholdscoringzeroandbelowmeans thestudentfailedthesubtest.Excludingtheessayscore,IthendenetheCLAST scorestobetheminimumoftheserescaledscores.Excludingtheessayscoredoes notposeabigproblemsinceonlyasmallfractionofstudentsfailtheessaysubtest. Figure1-1showstheregressiondiscontinuitygraphofpassingtheCLASTontherst attempt.TheRDgraphshowsthataroundeightpercentofthestudentsfailedtheessay subtestbutpasstherestofthesubtestsontherstattempt.TherstrowofTable1-6 20

PAGE 21

presentsthesummarystatisticsoftheCLASTscores.Thetablerevealsthevariationof theCLASTscoresacrossgenderandracialgroups.Whitestudentsscoredthehighest amongtheracialgroups.Whitesonaveragescoredabout1/4ofthestandarddeviation abovetheaverageCLASTscores,whileBlacksandHispanicsscore4/5and1/4ofthe standarddeviationbelowthemeanscore,respectively.Malestudentsscoredbetterthan theirfemalecounterpartswithmalesscoringslightlyabovetheaverageCLASTscores. NumberofSemestersEnrolledinCommunityCollegeaftertheFirstCLAST Attempt .ThereareatleasttworeasonswhytheCLASTcoulddecreasetheprobability thatastudentobtainsanassociatesinartsdegree.First,theexistenceoftheCLAST coulddiscourageastudent,therebycausinghertodropoutofcommunitycollegebefore theendofthecourseDiscouragementEffect. 3 Second,astudentcouldsatisfy allofthegraduationrequirementsexcepttheCLASTrequirement.TheCLASTwould therebypreventthisstudentfromobtaininganassociatesinartsdegreeCan'tPass Effect.Tomeasurethediscouragementeffects,Iusethenumberofsemestersenrolled incommunitycollegeaftertherstCLASTattempt.Theinformationonthenumberof semesterswereobtainedfromthemergedcommunitycollegeenrollmentrecordsand theCLASTscoresdata. 4 Thisvariableisdenedasthenumberofsemestersthe studentregisteredaftertherstCLASTattempt.ThefourthrowofTable1-6showsthe comparisonofmeansofthisvariableacrossgenderandracialgroups. EverAADegree .TomeasuretheeffectoftheCLASTontheprobabilityof receivinganAAdegree,IconstructedanEverAAdegreevariablefromthecommunity collegepost-secondaryawardsdata.Thedatacontaininformationonthedegrees awardedtothestudent,awarddegreedate,andtheawardinginstitution.Astudentis 3 Forexample,theCLASTwillcausehertodropoutearlybecausesheisworriedthat shewillneverobtainadegree. 4 ThedatesoftheCLASTexamsarematchedwiththeenrollmentdates. 21

PAGE 22

identiedtohaveeverobtainedanassociatesinartsdegreeifthestudentisawarded eitheranassociatesinartsoranassociatesinsciencedegree.ThesecondrowofTable 1-6revealstheAAdegreecompletionratesacrossgenderandracialgroups.Male communitycollegestudentshaveroughlythesamecompletionratesastheirfemale counterparts.Amongtheracialgroups,BlackandHispanicstudentshavethelowest AAdegreecompletionrateswhiletheirWhitecounterpartshavethehighestcompletion rates. 1.4EmpiricalStrategy Likeempiricalstudiesonhighschoolexitexams,Iusecomparisonsaroundthe CLASTpassingthresholdtomeasuretheeffectsoftheCLAST.Iusearegression discontinuityRDanalysistoseparatethecausaleffectsoftheCLASTfromthe confoundingeffectsofotherfactorsinuencingassociatesinartsdegreecompletion andthenumberofsemestersthestudentswereenrolledaftertherstCLASTattempt. LikeearlierstudiesthatusedtheRDstrategy,Iexploitthepassingthresholdseparating studentswhonarrowlypassandnarrowlyfailtheCLAST.IfpassingtheCLASThas aneffect,themeanoftheoutcomevariablewillshowadiscontinuityatthepassing threshold.Thediscontinuityestimateistheninterpretedasthecausaleffectofpassing theCLASTonstudentsonthemarginofpassing.Providedthattheunobservable factorsaffectingtheoutcomesofinterestdonotshowadiscontinuityatthepassing threshold,thediscontinuityestimateisinterpretedasthecausaleffectoftheCLAST. TheregressiondiscontinuityRDanalysisisbasedonthefollowingreducedform model: y i = f score + Pass i + i where y i representsthecollegeoutcomeofacommunitycollegestudent i score i representsthestudent'sCLASTscore, f isafunctionthatiscontinuousattheCLAST passingthresholdwithparameters e.g.ahigherorderpolynomial, Pass i isan indicatorwhetherastudenthaspassedtheCLAST,and i istheerrortermrepresenting 22

PAGE 23

theinuenceofotherfactors.Theparameter representsthecausaleffectofpassing theCLASToncollegeoutcome y i .Iftheoutcomebeingmeasuredisthereceiptof associatesinartsdegree,thenthiscausaleffectisjustthedifferenceintheprobabilityof receivingassociatesinartsdegreeoneithersideoftheCLASTpassingthreshold. ThekeyassumptionunderlyinganRDanalysisisthatunitsoneithersideofthe thresholdarecomparableoneverydimension.Thatis,unitsoneithersideofthe thresholdareasgoodasrandomlyassigned.Inthispaper,theRDassumptionis satisedifstudentsbelowandabovetheCLASTpassingthresholdarecomparablein allobservedandunobservedcharacteristics.InthecontextofEquation1,theRD assumptionis E [ i j P )]TJ/F25 11.9552 Tf 11.955 0 Td [(< score < P ] = E [ i j P < score < P + ] forsufcientlysmall ,where P isthepassingthreshold.Ifthisassumptionismet,even amerecomparisonofthemeanof y i oneithersideofCLASTpassingthresholdyieldsa consistentestimateoftheparameter Knowledgeofthetreatmentassignmentandcutoffrulesmayproduceunforeseen behavioralresponsesonthepartofstudentsImbensandLemieux,2008;Leeand Lemieux,2009;McCrary,2008.IntheCLASTsystem,studentshavetheoptionto taketheCLASTmultipletimesintheeventoffailinganyofthesubtest.TheCLAST retestingcouldresulttononrandomsortingaroundtheCLASTpassingthreshold, whichisaconcernforstudiesusingRDanalysisImbensandLemieux,2008;Leeand Lemieux,2009;McCrary,2008.Studentsinterestedinobtainingassociatesinarts degreewhoscoredbelowtheCLASTpassingthresholdarelikelytoretaketheCLAST. IfstudentswhofailtheCLASTinitiallyselecttotakethetestmanytimesuntiltheypass, theassumptionthatstudentsareasgoodasrandomlyassignedtoeithersideofthe CLASTpassingthresholdmayfailtohold.TheRDassumptionfailstoholdwhenthe unobservedfactorsjointlydeterminethecollegeoutcomesandthelikelihoodofpassing 23

PAGE 24

theCLASTafterretesting,subjectingtheresultstoselectionbias. 5 Withnonrandom sortingaroundtheCLASTpassingthreshold,onewouldexpecttoseemorestudents barelypasstheCLASTthanthosewhobarelyfail.Thisleadstoadiscontinuityinthe conditionaldensityoftheCLASTscoresatthepassingthreshold.Figure1-2shows thedistributionoftheCLASTscoresusingeachstudent'sentireCLASTscoreshistory. Graphicalinspectionshowstheprobleminusingtheentirehistoryofscores.Aspike occursatthepassingthresholdshowingmorestudentswhobarelypasstheCLAST thanstudentswhobarelyfail.TheMcCraryTestofManipulationprovidesaformaltest ofthepresenceofnonrandomendogenoussortingaroundthethresholdMcCrary, 2008.TheMcCraryTestresultsFigure1-2indicatethatthedensityofCLASTscores hasabreakattheCLASTpassingthresholdwithsignicancebelowtheconventional levels.Becauseoftheselectionbias,usingtheentirehistoryoftheCLASTscorescan compromisetheresultsoftheRDanalysisofcollegeoutcomes. Toaddresstheissueofnonrandomsortingaroundthepassingthreshold,Iusethe CLASTscoresonthe1stto9thattempt.TheRDanalysisofthecollegeoutcomesusing scoresontherstandsubsequentattemptswouldbelesslikelytosufferfromselection bias.Figure1-3showsthedistributionoftheCLASTscoresusingeachstudent'sscores onthe1stto9thattempt.Unlikethedistributionofscoresusingtheentirehistoryof CLASTscores,visualinspectionofthedistributionshowsnoapparentjumparound thepassingthreshold.Asaresult,theuseofCLASTscoresonthe1stto9thattempt separatelywouldlikelyidentifytheeffectsoftheCLASTintheRDanalysisofthe associatesinartsdegreecompletionandthesemestersenrolledaftertherstCLAST attempt. 5 Forexample,sincemotivatedstudentsaremorelikelytoobtainadegree,the unobservedmotivationfactormayconfoundtheeffectsoftheCLASTonobtainingthe degree. 24

PAGE 25

EvenwiththeabsenceofthenonrandomsortingaroundtheCLASTpassing threshold,resultsfromtheRDanalysiscanbecompromisedifotherfactorsaffecting collegeoutcomesalsochangeatthepassingthreshold.Figure1-4givesthevisualtest oftheRDassumption.GenderandracialcompositionsWhite,Black,andHispanic shownojumpsattheCLASTpassingthreshold.Table1-7presentsthetestof discontinuitiesingenderandracialgroups.Althoughtheteststatisticsarewithinthe 5percentlevel,thediscontinuityestimatesareverysmallindicatinglackofpractical signicance. 1.5Results InthissectionIpresenttheestimatesoftheeffectoffailingtheCLAST.Iassessthe impactoffailingtheCLASTonthenumberofsemestersenrolledaftertakingtheCLAST andtheprobabilityofreceivinganassociatesinartsAAdegree. 1.5.1EstimatesoftheEffectofFailingtheCLASTontheNumberofSemesters Enrolled Therstcollegeoutcomeofinterestisthenumberofsemestersenrolledafter takingtheCLAST.Lookingattherelationshipbetweenthenumberofsemesters enrolledandtheCLASTscoresprovidesinsightsabouttheDiscouragementeffects referredtoearlier. IstartwiththetopleftpanelofFigure1-5.Thisisbasedontheanalysissample discussedabove.Recallthatthissampleconsistsofstudentsthattookallfourpartsof theCLASTthersttime.OntheX-axisistheminimumofthemath,reading,andwriting rst-attemptscores,rescaledsothattheyaredenedinrelationtothepassingcutoff. Thismeansthatascoreabovezeroononesubtestmeansthatthesubtestwaspassed; aminimumgreaterthanzeromeansthatallthreesectionswerepassed.OntheY-axis 25

PAGE 26

arethesmoothedrunningmeansofthenumberofsemestersthesestudentsenrolled aftertherstattemptattheCLAST. 6 Itisnotclearinwhichdirectionwewouldexpectthegraphtoslope.Ontheone hand,wemightexpectstudentswithhigherrst-attemptscorestostayincommunity collegelonger:higher-abilitystudentswillbelesslikelytodropoutofcommunitycollege beforetheircourseiscompleted;theymaybeenrolledformoresemestersbecausethey areenrolledonlonger,perhapsmorerigorouscourses.Ontheotherhand,sincethe CLASTisagraduationrequirementandstudentshavemultiplechancestopass,we mightexpectstudentswithlowerscorestoremainincollegelongerinordertoretake andpasstheseexams.ThegraphonthetopleftpanelofFigure1-5surprisinglyshows nodenitepatternintheslope. Turningnexttotheextentofanyjumpinthisgraphatthepassingthreshold,it isagainnotclearwhetherwewouldexpectthisjumptobepositiveornegative.On theonehand,ifstudentsthatfailarediscouragedandquitcollegeearly,wewould expectthegraphtojumpupatthepassingthreshold.Alternatively,ifstudentsthat failtheexamhavetoremainincollegeinordertopassatasubsequentattempt,we wouldexpectthegraphtohaveanegativejump.TheRDgraphonthetopleftpanelof Figure1-5,however,showsnojumpattheCLASTpassingthreshold.Toplacestandard errorsontheseestimates,the1strowofthe3rdand4thcolumnsinTable1-8liststhe regressionversionoftheRDgraphonthetopleftpanelofFigure1-5.Theestimated discontinuitylistedis.057withastandarderrorof0.069. 7 Thisshowsnoevidenceof theDiscouragementeffectsontheCLASTrsttakewherestudentsonthemarginof 6 Ingeneratingthegraphofthesmoothedrunningmeans,abinwidthoftwoCLAST scorepointsisused. 7 Inthissection,allregressionestimatesaregeneratedusingawindowof100CLAST scorepointstotheleftandtotherightofthecutoff. 26

PAGE 27

failingtheCLASTdonotstaylongerinCCcomparedtothestudentsonthemarginof passingtheCLAST. StudentswhofailtheCLASTthersttimehaveanoptiontoretaketheCLAST severaltimesuntiltheyeventuallypass.TheRDgraphsfoundinFigure1-6arebased onthe2ndCLASTattemptsampleconsistingofstudentswhofailedtheCLASTthe 1sttime,3rdCLASTattemptsampleconsistingofstudentswhofailedtheCLASTthe 2ndtime,upto9thCLASTattemptsampleconsistingofstudentswhofailedtheCLAST the8thtime.Togivevisualdescription,Iagainusethesmoothed-runningmeansand connectedthepointsoneithersideoftheCLASTpassingthreshold.LookingattheRD graphsbasedonthe2ndto9thCLASTattemptsFigure1-6,weobservethesame downwardslopingpatternwithsomegraphsshowingsmalljumpsattheCLASTpassing threshold.FailingtheCLASTonthe2ndto9thattemptsappearstohaveaverysmall tonoeffectonthenumbersemestersenrolledaftertakingthatCLASTattempt.This showsthatfailingtheCLASTonsubsequentattemptsdoesnotmakestudentsonthe marginoffailingthatCLASTattempttostaylongerincollege.Columns3and4column inTable1-8alsoprovidetheregressionversionofFigure1-6.TheRDestimatesrange from.65to0.18,withthemajorityoftheseestimatesarestatisticallyinsignicantat theconventionallevels. TheestimatedeffectsonthetopleftpanelofFigure1-5andinTable1-8gaveus thepictureforthefullsample.Itispossiblethattheseestimatesmaskdifferencesin theeffectsformalesandfemales.Theremaybeeffectsforfemalesbutnotformalesif femalesaremorehighlymotivatedtogetanAAdegree.Themiddleandtoprightpanels ofFigure1-5plottheconnectedsmoothedrunningmeansofthenumberofsemesters enrolledaftertherstCLASTattemptagainsttheCLASTscoresontherstattemptfor maleandfemalesubsamples.Forbothmaleandfemalesubsamples,theRDgraphs shownoobviousdiscontinuityattheCLASTpassingthresholdindicatingnodifferences ineffectsacrossgender.Toplacestandarderrorsontheseestimates,columns3and4 27

PAGE 28

inTable1-9providetheregressionversionofthemiddleandtoprightpanelgraphsin Figure1-5.TheestimateddiscontinuitieslistedinTable1-9are0.005withastandard errorof0.127formales,0.109withastandarderrorof0.083forfemales. FailingontherstCLASTattemptappearstohavenoeffectonthenumberof semestersenrolledacrossracialgroups.ThesegroupsincludeWhites,Blacks, Hispanics,andothersrepresentingtheminoritiessuchasAsian,AmericanIndian, andmixedraces.IconsideredWhites,Blacks,andHispanicsintheanalysissince thesearethemajorracialgroupsrepresentedinFloridacommunitycollegesystem. 8 Iusethesmoothedrunningmeanstoplottherawdataforeachracialsubsampleto providevisualdescription.TheRDgraphsonthebottompartofFigure1-5showvery smalljumpsattheCLASTpassingthreshold.Tochecktheprecisionoftheseestimates, Table1-10providestheregressioncounterpartofthegraphsonthelastrowofFigure 1-5.Despitethesmalljumpsfound,theestimateddiscontinuitiesfoundincolumn3in Table1-10forthedifferentracialgroupsareallstatisticallynotdifferentfromzeroatthe conventionallevels.Thisshowsnodifferencesintheeffectsforanyofthemajorracial groups. ItmayalsobepossiblethatFloridacommunitycollegeschangetheretention andcourseremediationpoliciesovertheCLASTadministrationyears.Theremay beeffectsoftheCLASTonthenumberofsemestersenrolledintheyearswhenthe communitycollegesprovidemoreencouragementtostudentstocompletethedegree requirementsorhavetherulesrequiringstudentstoattendremedialclassesinthe eventoffailingtheCLAST.Forexample,theremaybegreatereffectsoffailingthe CLASTintheyearspriorto1997whentherulesgoverningtheCollegeLevelAcademic 8 Ididnotreporttheotherracescategoryintheanalysissincetheyonlyrepresent asmallpercentageinthedataandtheyarecomposedofdifferentraceswithdifferent experiencesandcharacteristics. 28

PAGE 29

SkillsCLASrequirementshavechanged.Thisgavestudentsmorewaystosatisfythe CLASrequirement,therebydecreasingthetimeoftheirstayinthecommunitycollege togetadegree.Figure1-7connectsthesmoothedrunningmeansofthenumberof semestersenrolledaftertherstCLASTattemptagainstthescoresontherstCLAST attemptforthedifferentCLASTyears.TheRDgraphsinFigure1-7showthatfailing theCLASTdoesnotaffectthenumberofsemestersenrolledaftertherstCLAST attemptforalltheCLASTexamyears.AlthoughtheRDgraphsforalltheCLAST yearsshownoobviousjumpsattheCLASTpassingthreshold,theaveragenumber ofsemestersenrolledassociatedwitheachCLASTscorehasdecreasedfrom1995 to2006.Toplacestandarderrorsontheseestimates,columns3and4inTable1-11 presenttheregressionversionwithestimateddiscontinuitiesandstandarderrors.With theexceptionoftheestimatefortheyear1995,theestimatedeffectsareverysmall rangingfrom.38to0.245andarestatisticallyinsignicantattheconventionallevels. Itispossiblethatsomecommunitycollegeshavedevelopedprogramsmakingsure thatstudentsretakeandpasstheCLAST.TheeffectsoftheCLASTmaybebiggerin communitycollegeswheretheadmissionstothestateuniversitiesinFloridaarenot important.ThesecommunitycollegeswouldcarelessabouttheCLASTfailuresand retakes.Toshedlightonthisconjecture,Icutthesampleintotwocategoriesbased onthepercentageofstudentswhoeverattendedthestateuniversitysystemSUS: abovemedianandbelowmedianpercentageofstudentswhoeverattendedanSUS. TheeffectsoffailingtheCLASTdonotvarywhenthesampleiscutintoaboveand belowmedianpercentgroups.Figure1-8connectsthesmoothedrunningmeansofthe numberofsemestersenrolledaftertherstCLASTattemptonthescoresontherst CLASTattempt.TheRDgraphsinFigure1-8shownoobviousjumpsattheCLAST passingthresholdforboththeaboveandbelowmedianpercentagegroups.Toplace standarderrorsaroundtheseestimates,theregressionversionfoundinthe3rdand 4thcolumnsofTable1-12liststheestimateddiscontinuitieswithestimatesof0.01with 29

PAGE 30

astandarderrorof0.087forabovemedianpercentgroupand0.027withastandard errorof0.116forbelowmedianpercentgroup.Theseresultsindicatetheabsenceofthe Discouragementeffectswhenthesampleiscategorizedaccordingtothepercentage ofstudentswhoeverattendedSUS. 1.5.2EstimatesoftheEffectofFailingtheCLASTonAssociatesinArtsDegree Exploringtherelationshipbetweentheprobabilityofreceivinganassociatesin artsdegreeandtheCLASTscoreswillprovideinsightsontheextentoftheCan't Passeffects.RecallthattheAAEveroutcomevariableisdenedasabinaryindicator takingavalueof1ifthestudenteverreceivedanAAdegreeand0ifnot.TheRD graphsaregeneratedusingthesamemethodasabove,plottingtheconnected smoothedrunningmeansoftheprobabilityofreceivinganAAdegreeagainstthe CLASTscores.Iusethesameanalysissampleofstudentswhotookallfourpartsofthe testthersttime. ThegraphwillshowadiscontinuityattheCLASTpassingthresholdintheevent thattheCLASThasaneffect.IfstudentswhofailtheCLASTdonotgettheAAdegree, thenwewouldexpectapositivejumpontherateofreceivinganAAdegreeatthe CLASTpassingthreshold.Infact,thegraphonthetopleftpanelofFigure1-9shows thattheprobabilityofreceivinganAAdegreejumpsupattheCLASTpassingthreshold indicatingthepresenceoftheCan'tPasseffect.Thegraphalsoslopesupward,an indicationthatstudentswhoscoredhighintheCLASTexamhavehigherratesof receivinganAAdegree.The1strowofcolumns5and6inTable1-8providesthe regressionversionofthegraphinthetopleftpanelofFigure1-9showinganestimated discontinuityof0.048withastandarderrorof0.006. SomestudentswhofailedtheCLASTontheirrstattemptretakeit.Withthis,it ispossibletolookathowtheprobabilityofreceivinganAAdegreevarieswitheach CLASTattempt.EffectsoffailingtheCLASTmaybehigherinsubsequentattempts sincestudentswhonarrowlyfailtheCLASThavehigherchancesofpassingtheCLAST 30

PAGE 31

thenexttimethanstudentswhoseCLASTscoresarefarbelowfromthepassing threshold.SimilartothesampleusedaboveforthedifferentCLASTattempts,Ilook atthesamplefromthepopulationofstudentswhofailedapreviousCLASTattempt i.e.,from2ndto9thattempt.Figure1-10showsthatfailingtheCLASTonthe2ndto 9thattemptalsoreducestheprobabilityofreceivinganAAdegree.TheRDgraphsin Figure1-10areplotsoftheconnectedsmoothedrunningmeansoftheAAEvervariable againsttheCLASTscoresfromdifferentCLASTattempts.Althoughtherearejumpsat theCLASTpassingthreshold,thenotableresultisthattheprobabilityofreceivingan AAdegreeisgreaterinthesubsequentCLASTattempts.ThismeansthattheCan't PasseffectsaregreateronsubsequentCLASTattempts.Theregressionversionof Figure1-10isfoundincolumns5and6inTable1-8.Theestimateddiscontinuitiesare statisticallysignicantwithlargerestimatesinthesubsequentCLASTattemptsexcept inthe8thattempt. TheestimatedCan'tPasseffectsonthetopleftpanelofFigure1-9andinTable 1-8potentiallymaskdifferencesintheeffectsbygender.Thedifferentialeffectsmay arisefromgenderspecicmotivationalfactorsinuencingthelikelihoodofcompleting anAAdegree.Ifyoungmalesareimpatient,failingacourseoranexamwouldleadto lowerratesofcompletinganAAdegree.Toprovidevisualdescriptionoftheeffect,the middleandtoprightpanelsofFigure1-9plotstheconnectedsmoothedrunningmeans ofAAEveragainstthescoresontherstCLASTattempt.TheRDgraphsshowthatthe rateofreceivinganAAdegreejumpsupattheCLASTpassingthresholdwiththejump formalesthatisstatisticallysignicantlygreaterthanthatforthefemales.Thismeans thattheCan'tPasseffectsarehigherinmales.Columns5and6inTable1-9showthe regressionversionofRDgraphsonthemiddleandtoprightpanelsinFigure1-9where 31

PAGE 32

theestimateddiscontinuityis0.074withastandarderrorof0.009formalesand0.037 withastandarderrorof0.006forfemales. 9 TheCan'tPasseffectestimatesmayalsovaryacrossmajorracialgroups. Differentialeffectsamongracialgroupscouldariseduetodifferencesinbackground, experience,andthesocio-economicstatusofthesegroupsthataffecttheprobabilityof receivinganAAdegree.ToprovidevisualdescriptiononthedifferencesintheCan't Passeffects,IplottheconnectedsmoothedrunningmeansoftheAAEverbinary variableagainstthescoresontherstCLASTattemptusingthesampleonWhites, Blacks,andHispanics.ThebottomgraphsinFigure1-9showthatthediscontinuities intheprobabilityofreceivinganAAdegreevaryacrossdifferentracialgroups.The RDgraphsshowthegreatestdiscontinuitiesforHispanicsandBlacksandtheleast discontinuityforWhites.Toplacestandarderrorsontheseestimates,columns5and6 inTable1-10listtheestimatedCan'tPasseffectof0.048withastandarderrorof0.006 forWhites,0.063withastandarderrorof0.021forBlacks,0.075withastandarderrorof 0.014forHispanics. ItisalsointerestingtoaskwhethertheCan'tPasseffectshavechangedoverthe historyoftheCLASTadministration.ThechangeintheCLASTrulesin1997allows studentstomeettheCollegeLevelAcademicSkillsrequirementsviaalternativeways. Onemayaskwhetherthesechangesintherulesmadeeasierforsomestudentsto getanAAdegree.Withthis,wemayexpectthattheeffectsoffailingtheCLASTare largerintheearlieryearsoftheCLASTadministration.Toshedlightonthis,Iplotthe connectedsmoothedrunningmeansofAAEvervariableagainstthescoresonthe rstCLASTattempt,disaggregatingthesampleintodifferentCLASTadministration years.TheRDgraphsfoundinFigure1-11showsobviousjumpsattheCLASTpassing 9 Theseestimatesarestatisticallydifferentbasedonthetestofthedifference betweenmeanswithalevelofsignicanceofatleast5percent. 32

PAGE 33

thresholdintheearlyyearsoftheCLASTadministration.NoticethattheCan'tPass effectsaregreaterintheearlyyearsoftheCLASTadministrationfrom1995to2000 whiletherestofyearsshowabsenceoftheCan'tPasseffects.Columns5and6in Table1-11providetheregressionversionofthegraphsinFigure1-11.Theestimated discontinuitieslistedare0.028withastandarderrorof0.01fortheyear1995,0.062with astandarderrorof0.012for1996,0.071withastandarderrorof0.023for1997,0.061 withastandarderrorof0.019for1998,0.07withastandarderrorof0.023for2000. 10 TheCan'tPasseffectsmayalsovaryacrosscommunitycollegesthathave programsensuringstudentstoretakeandpasstheCLAST.TheCan'tPasseffect maybelargerincommunitycollegeswherestudents'admissionprospectstoSUSare notabigdeal,sothatthesecollegescarelessabouttheCLASTfailuresaswellas encouragingstudentsmoretocompleteallthedegreerequirementsandreceiveanAA degree.Toshedlightonthisconjecture,theconnectedsmoothedrunningmeansofthe AAEvervariableareplottedagainstthescoresontherstCLASTattempt,cuttingthe sampleintocommunitycollegesbasedonthepercentageofstudentswhoeverattended anSUS.Aswiththeabovedenition,Icategorizethesampleasaboveandbelow medianpercentagesofstudentswhoeverattendedanSUS.TheCan'tPasseffectis higherininstitutionsthathistoricallyhavelowerpercentageofstudentswhoattended anSUSi.e.thebelowmedianpercentagegroup.Figure1-12showsthatthejumpat theCLASTpassingthresholdforthebelowmediangroupishigherthanthejumpforthe abovemediangroup.Columns5and6inTable1-12providetheregressionversionof Figure1-12.Thediscontinuityestimatesare0.047withastandarderrorof0.007forthe 10 Restrictingthesampletothe1995to1997CLASTadministrationyearsyieldan estimatedCan'tPasseffectof0.051withastandarderrorof0.007whiletheestimated Can'tPasseffectfromtheyears1998to2007is0.049withastandarderrorof0.008. TestofthedifferencebetweenmeansshowsthattheCan'tPasseffectishigherprior to1998,theyearsbeforetheCLASTruleshavechanged.Theestimateddifferenceis 0.011withastandarderrorof0.003 33

PAGE 34

abovemedianpercentagegroupand0.053withastandarderrorof0.008forthebelow medianpercentagegroup. 11 1.5.3RobustnessCheck Inordertochecktherobustnessofthediscontinuityestimates,Iconductedavariety ofrobustnesschecksinthissection. 1.5.3.1Functionalform Onewayofcheckingtherobustnessoftheestimatesistoverifywhetherthe discontinuityestimatesaresensitivetothefunctionalformofusedfortheregression discontinuityfunction.Theregressiondiscontinuityestimatesintheprecedingtables areestimatedusingaspecicfunctionalformfortheconditionalexpectationofthe probabilityofreceivinganAAdegreeandthenumberofsemestersenrolledafterthe rstCLASTattempt.Thatis,thelinearregressionofaversionofEquation1thatis linearin score i isused.Theregressionfunctiontakestheform y i = 1 + Pass i + 2 score i + 3 Pass i score i + i whereagain isthediscontinuityestimateand f Pass i score i ; = 1 + 2 score i + 3 Pass i score i .Tochecktherobustnessoftheseresults,Iintroduceaparametricmodel withhigherorderpolynomialsoftheform y i = 1 + Pass i + 2 score i + 3 Pass i score i + g Pass i score n i ; + i where n =2,3,4 .Tables1-13and1-14presentdiscontinuityestimatesfromthe regressionfunctionswithpolynomialsuptothe4thorderfortheoutcomesthe number ofsemestersenrolledafterthe1stCLASTattempt andthe probabilityofreceivingan AAdegree .Giventhestandarderrors,theestimatesfromeachtabledonotvarywith 11 Thesetwoestimatesarenotstatisticallydifferentbasedonthetestofthedifference betweenmeans. 34

PAGE 35

thepolynomialorderunderdifferentCLASTscorepointwindow.Broadlyspeaking,the estimatesarerobusttothepolynomialorderof score i 1.5.3.2VaryingtheCLASTscorepointsincludedintheestimationsample Anotherwayofcheckingtherobustnessoftheestimatesistoseewhetherthe discontinuityestimatesaresensitivetothechoiceoftheCLASTscorepointsincluded intheRDestimation.Tochecktherobustnessoftheseestimates,Iestimatethe regressionfunctionswithdifferentpolynomialordersvaryingtheestimationsampleby changingtheCLASTscorepointstotheleftandtotherightofthethresholdfrom )]TJ/F22 11.9552 Tf 9.298 0 Td [(112 to 112 12 Tables1-13and1-14showsthediscontinuityestimatesusingregressionswith polynomialsuptothe4thordervaryingtheCLASTscorepointstotheleftandtothe rightofthepassingthresholdfrom )]TJ/F22 11.9552 Tf 9.299 0 Td [(112 to 112 fortheoutcomes numberofsemesters enrolledafterthe1stCLASTattempt andthe probabilityofreceivinganAAdegree Fromthepolynomialoforderonerow,theestimatesfromthetwotablesdonotvarya lot.Also,theestimatesbecomepreciselyestimatedastheCLASTscorepointsincluded intheestimationincreases.Thesameconclusionholdsfortherestofthepolynomial orderspecicationsineachtable.Insum,thediscontinuityestimatesarerobusttothe choiceofCLASTscorepointsincludedintheRDestimation. 1.6Discussion InthispaperIconsidertheeffectsoffailingtheCollegeLevelAcademicSkillsTest CLASTonthenumberofsemestersenrolledandontheprobabilityofreceivingan associatesinartsdegree.Indamoderateeffectofaboutvepercentagepointsonthe probabilityofreceivinganAAdegreemoderateCan'tPasseffectandnoeffecton thenumberofsemestersenrollednoDiscouragementeffect.InthissectionIdiscuss whatmightexplaintheseresults. 12 Becauseoftherestrictionsappliedtothesample,theresultingCLASTscorepoints rangeisfrom-112to112. 35

PAGE 36

1.6.1TheRolePlayedbytheNumberofPossibleTestAttempts ThepossibilityofretakingtheCLASTmultipletimesmighthavedriventheresults outlinedabove.SincestudentscanretaketheCLAST,somestudentswhofailedthe rsttimeretakeitandsomepassiteventuallysothatthediscontinuityestimatesonthe probabilityofreceivinganAAdegreebasedonthescoresontherstCLASTattempt arereduced.Thus,itispossiblethattheCan'tPassandtheDiscouragementeffects wouldhavebeenlargerhadtherebeennoretakeopportunities. SupposetheobservedCan'tPasseffectsandnoDiscouragementeffectsare trulytheresultsofstudentretestingopportunities.Thequestioniswhydothestudents whohavefailedtheCLASTontherstattemptdecidetoretakeit?GiventhatCLAST canbetakenmultipletimes,whatdrivesthisdecision?Astudentwhonarrowlyfailed thersttestwouldlikelytoretakeitifheknowsthattheexpectedutilityofretakingis greaterthantheexpectedutilityofnotretakingthetest.Ifthemaximumnumberoftests istwo,thenthisexpectedutilitydependsonlyonthebenetsandcostsofretakingand theprobabilityofpassingthenexttest. 13 Thestudentonlyneedstogureoutwhether thegainsoutweighthecostsofretesting.Whenthemaximumnumberoftestsismore thantwo,thedecisiontoretestdependsnotonlyonthebenetsandthecostsoftaking thesecondtestbutalsowiththevalueoftakingthetestbeyondthesecondattemptin theeventthathefailsthistest.Eveniftheprobabilityofpassingthesecondtestislow, astudentmaystilldecidetotakethatsecondtestiftherearefuturegainsfromretaking becausethereisalwaysachancethathepassesthethirdtest,thefourthtest,etc. Thereisanoptionvalueintakingthesecondtestgiventhatthemaximumtestattempt 13 Theprobabilityofpassingthenexttestdependsatleastonthereliabilityofthetest andwhetherscoresimproveasthestudentsspendmoretimeintakingremediation classes 36

PAGE 37

ismorethantwo. 14 Theoptionvaluedependsontheprobabilityofpassingthesecond testandtheexpectedbenetstopassingthenexttest.Toillustrate,considerastudent whohasnarrowlyfailedthersttest.Supposethemaximumnumberoftestsistwoand therearebenets B andcosts C ofretaking. 15 Thenthestudenttakesthesecondtestif E [ U take 2 j fail 1, s 1 =0 ] = P pass 2 j fail 1, s 1 =0 B )]TJ/F39 11.9552 Tf 11.955 0 Td [(C > 0 where P pass 2 j fail 1, s 1 =0 istheprobabilityofpassingthesecondtestgiventhatthe studentfailedthersttesti.e. fail 1 ,where s 1 =0 meansthatthestudentnarrowly failedtheexamwith0asthepassingthreshold.Supposethestudenthavetomakethe samedecisionbutthistimethemaximumnumberoftestisthree.Thestudenttakesthe secondtestif E [ U take 2 j fail 1, s 1 =0 ] = Prob B + 1 )]TJ/F39 11.9552 Tf 11.956 0 Td [(Prob V )]TJ/F39 11.9552 Tf 11.955 0 Td [(C > 0, where Prob = P pass 2 j fail 1, s 1 =0 Clearly,theoptionvalue V determinesthedifferenceinexpectedutilityoftaking thesecondtestwhenthemaximumnumberoftestsistwoandwhenthemaximum numberoftestsisthree.If V ishighenough,thestudentwilltakethesecondtesteven iftheprobabilityofpassingthesecondtestislow.Iftheoptionvalueiszero,thenevery testretakedecisionwillbebasedonlythebenets,theconditionalprobabilityofpassing thesecondtest,andthecostsofretakingsothatintheeventofalowprobability 14 Theoptionvalueinthiscaseistheexpectedgainofpassingthetestbeyondthe secondtestintheeventthathefailedthatsecondtest 15 Thebenetsofretakingthetestinclude,butnotlimitedto,theincreasedlabor marketopportunitiesandhigherearningsintheeventthatthestudentpassthenext test.Thecostsofretakingthetestincludeatleasttheregistrationfeesandthetime costsdevotedinstudyingthematerialsforthenexttest. 37

PAGE 38

forecastofpassingthesecondtestthestudentmightdecidenottotakethe2ndtest. AppendixA Thepointisthatstudentswhonarrowlyfailedthersttestmightbeespecially likelytoretakethetestbecauseoftheoptionvalueassociatedwithbeingabletotake CLASTmultipletimes.Givenreasonablevaluesoftheoptionvalue,thesestudents mightretake,passthetest,andearnanAAdegreewhichwillreducethediscontinuity estimates.Although,therearenoexistingestimatesoftheoptionvalueofretestinga collegeexitexam,theinformationfoundintheliteratureaboutthereturnstocompleting anAAdegreeandthefeaturesofthecommunitycollegesysteminFloridawillprovide hintsthattheoptionvalueisfairlyhighforthecommunitycollegestudentsinFlorida. First,theestimatedreturnstocompletinganassociatesinartsdegreearequitelarge, about15to27percentoftheannualearningsKaneandRouse,1999.Second,the Florida2+2programmakestheAAdegreeevenmorevaluableBackgroundsection, page17.CompletinganAAdegreeguaranteesacommunitycollegestudentthe admissionsandhence,achancetoearnadegreefromanyofthestateuniversitiesin Florida.ThiswillincreasethelabormarketopportunitiesandfutureearningsofanAA degreeholderevenmore.Thesetwofactsmighthaveledsomestudentswhonarrowly failtheCLASTtoretestuntiltheypasstogettheAAdegree.Becausethestudentswill likelytoretaketheCLASTgivenalargeoptionvalue,thediscontinuityestimatesofthe AAdegreecompletionwillbereduced. 1.6.2TheRolePlayedbytheAlternativeWaysofMeetingtheCLASTRequirements Whenthe1997FloridabillrelaxedthemainuseoftheCLASTastheonly requirementtoobtainanAAdegreeinFloridacommunitycolleges,itpresentedstudents greateropportunitiestoearnanAAdegreethroughalternativeways.Uponpresentinga GPAof2.5orhigherinaCLASTequivalentcourseorupondemonstratingsatisfactory scoresinthenationalstandardizedtestssuchasSATandACT,astudentcanwaivethe 38

PAGE 39

CLASTrequirements.Withthis,itispossiblethatastudentwhofailedtheCLASTthe rsttimeormanytimesearnedanAAdegreebymeetingtheotherrequirements.With agreatnumberofstudentswhonarrowlyfailthersttestwhoearnedanAAdegree throughthealternativeways,theregressiondiscontinuityestimatesoftheprobabilityof receivinganAAdegreebasedontherstCLASTscoreswillbereduced. ThereisnoinformationinthedatathatastudentusedtheGPAorSATscoresto waivetheCLAST.Althoughthisinformationisnotavailable,wecanstillinferindirectly fromthedatabyobservingthestudentswhoearnedanAAdegreebutdidnotpass theCLASTever.Table1-15showsthepercentofstudentswhoearnedanAAdegree butdidnotpasstheCLASTever.ThepercentofstudentswhoearnedanAAdegree throughthealternativewayshaveincreasedfrom11.5percentin1995to30.2percent in2006.Also,takingandpassingtheCLASTtoearnanAAdegreebecameless popularamongstudents,leadingtoadeclineinthenumberofCLASTtakersstarting 1997.HadtherebeennootherwaysofearninganAAdegreeasidefrompassingthe CLAST,theeffectoffailingtheCLASTthersttimeontheprobabilityofearninganAA degreei.e.theCan'tPasseffectwouldhavebeenhigherassumingsomestudents failedtheCLASTcontinuouslyanddidnotpasstheCLASTever.Insum,thepossibility ofearninganAAdegreethroughthealternativewayshasreducedtheCan'tPass effects.BecausetherearealternativewaysofwaivingtheCLAST,somestudentswho wouldhavedroppedoutofacommunitycollegebecauseoffailingtheCLASTstayed duetomoreopportunitiesofearninganAAdegree. 1.6.3RelationshiptotheLiteratureonEffectsofHighSchoolExitExam Itisinterestingtoknowwhethertheresultsfoundinthispaperareconsistent withestimatesfoundintheliterature,atleastinthecontextofhighschoolexitexam. AlthoughIamunawareofpreviousanalysesoftheimpactofcollegeexitexam,there isatleastonestrandofrelatedresearchinthehighschoolexitexamliteratureto whichtheestimatesinthispapercanbecomparedto.Papay,MurnaneandWillett 39

PAGE 40

2008studytheimpactsoffailingahigh-stakestestonon-timegraduationandschool persistenceusingtheMassachusettsdata.Usingregressiondiscontinuityanalysis, theyndmoderatenegativeeffectsofabouteightpercentagepointsamonglow-income urbanstudentsonthemarginofpassingthemathematicsexamsbutnoeffectson othergroups.Sinceroughly89percentofthestudentswhofailedeventuallyretakethe test,theestimatedeffectofeightpercentagepointisquitelarge.Theyalsondthat nearlytwo-thirdsofthosewhofailedandneverpasstheexaminationstoptakingthe testanddropoutofschool.Theyattributethistopoortestperformancethatsomehow discouragedthestudents. ThendingsofPapay,MurnaneandWillett2008areconsistentwiththose presentedinthispaper.Thisisinterestingforseveralreasons.Thendingsofmoderate effectgivenmultipleretakeopportunitiesamongFloridacommunitycollegestudents isnotdrivenbythefocusonthewell-establishedCLASTprogramanditsrulesorby thesamplerestrictionsapplied.Onaverage,nearly80percentofcommunitycollege studentswhofailedtheCLASTinanygivenattemptcontinuetoretaketheCLAST. CoupledwiththealternativewaysofmeetingtheCLASTandearninganassociatesin artsdegree,theestimatedCan'tPasseffectofvepercentagepointsismoderateand quitesignicant. 1.7Conclusion InthispaperIestimatetheeffectofcollegeeducationalstandardsoncommunity collegeoutcomes.Usingaregressiondiscontinuityanalysis,Iestimatetheeffectof failingtheCLASTontherstattemptonthenumberofsemestersenrolledafterthe rstCLASTattemptandtheprobabilityofreceivinganassociatesinartsdegreefor thecommunitycollegestudentswhonarrowlypassandnarrowlyfailtheCLAST.Given thelargenumberofCLASTretestopportunities,Indamoderateeffectofaboutve percentagepointsontheprobabilityofreceivinganAAdegreei.e.,amoderateCan't 40

PAGE 41

PasseffectandnoeffectonthenumberofsemestersenrolledaftertherstCLAST attempti.e.,noDiscouragementeffect. Althoughtheseestimatesarelocaltothecommunitycollegestudentsonthe marginofreceivinganAAdegree,thesearethestudentswhoaregreatlyaffectedby collegeinterventionsinplacetoraisecollegeachievemente.g.remediationclasses. WhileIdonotestimatethevalueofacommunitycollegedegreeforthesestudents, theliteraturesuggestspositivereturnstocompletinganassociatesinartsdegree KaneandRouse,1999.Ifcommunitycollegedegreehasasignalingvalueinthelabor marketMartorellandClark,2010,thenpolicymakersmustfocusonprovidingeffective collegeinterventionstothedisadvantagedstudents. ThelargenumberofretestopportunitiesintheFloridaCLASTsystemandthe alternativewaysofmeetingtheCLASTrequirementssomehowdidnotdiscourage somestudentswhonarrowlyfailtheCLASTtocompleteacommunitycollegedegree. HadtherebeennoretakeopportunitiesortheotherwaysofmeetingtheCLAST,the effectsoffailingtheCLASTwouldhavebeengreater.Theevidenceprovidedbythe moderateCan'tPasseffectandnoDiscouragementeffectsuggeststhatthecollege accountabilitycouldgreatlydependonthedetailsoftheaccountabilitysystembeing used.Astheaccountabilitymovementgainsgroundinhighergradesandinhigher education,policymakerswoulddowelltobearinmindthedetailsoftheaccountability systemtobeimplemented. 41

PAGE 42

Figure1-1.ProbabilityofPassingtheFirstCLASTTake 42

PAGE 43

Figure1-2.HistogramandMcCraryTestResultsforCLASTScore 43

PAGE 44

Figure1-3.McCraryTestResultsbyCLASTAttempt 44

PAGE 45

Figure1-4.TestofRegressionDiscontinuityAssumption 45

PAGE 46

Figure1-5.RD:SemestersEnrolledAftertheFirstCLASTAttempt 46

PAGE 47

Figure1-6.RD:SemestersEnrolledAfter2ndto9thCLASTAttempt 47

PAGE 48

Figure1-7.RD:SemestersEnrolledbytheExamYearofFirstCLASTAttempt 48

PAGE 49

Figure1-8.RD:SemestersEnrolledbyTypeofInstitution 49

PAGE 50

Figure1-9.RD:AssociatesinArtsDegreeAftertheFirstCLASTAttempt 50

PAGE 51

Figure1-10.RD:AssociatesinArtsDegreebyCLASTAttempt 51

PAGE 52

Figure1-11.RD:AssociatesinArtsDegreebytheExamYearofFirstCLASTAttempt 52

PAGE 53

Figure1-12.RD:AssociatesinArtsDegreebyTypeofInstitution 53

PAGE 54

Table1-1.NumberofCLASTTakersbyYearandSemester SchoolYearFallSpringSummerTotal 1982-198332-5 1983-1984316120 1984-1985581023 1985-19861012931 1986-19871213732 1987-1988154445104 1988-1999698973231 1989-1990156394240790 1990-1991251366379996 1991-19924275975911,615 1992-19938081,2319863,025 1993-19941,9403,0922,9537,985 1994-19956,72321,01015,74443,477 1995-199621,54922,96912,30156,819 1996-199714,72715,87211,93642,535 1997-199812,14211,6018,15631,899 1998-19999,00510,1527,98427,141 1999-20008,85210,1557,22926,236 2000-20017,7108,6306,59422,934 2001-20027,71910,5558,14326,417 2002-20038,82510,3665,60724,798 2003-20048,83710,7107,43626,983 2004-20057,0908,0194,78919,898 2005-20065,4307,0515,13817,619 2006-20073,3673,367 54

PAGE 55

Table1-2.CLASTPassingScores DateEssayEnglishLanguageSkillsReadingMath August1,1984toJuly31,19864265260260 August1,1986toJuly31,19894270270275 August1,1989toSeptember30,19914295295285 October,11991toSeptember30,19925295295290 October,11992&thereafter6295295295 55

PAGE 56

Table1-3.PercentageofStudentsWhoRetestedCLAST SchoolYearFallSpringSummer 1995-199632.235.347.0 1996-199739.139.844.8 1997-199838.039.647.1 1998-199943.342.246.5 1999-200045.542.047.0 2000-200145.943.347.1 2001-200239.738.744.7 2002-200341.739.460.2 2003-200439.838.242.9 2004-200541.641.244.5 2005-200644.539.945.4 2006-200740.4 56

PAGE 57

Table1-4.DistributionofCLASTAttempts NumberofCLASTAttemptsFrequencyPercentCumulative 1219,49653.8453.8 295,08023.3277.2 342,86210.5187.7 422,2935.4793.1 511,9172.9296.1 66,6591.6397.7 73,7340.9298.6 82,1900.5499.2 91,3270.3399.5 108240.2099.7 11ormore13250.32100.0 57

PAGE 58

Table1-5.CLASTPassingRates PassFirstAttempt%FailFirstAttemptandPassEventually%Observations Gender Female70.540.987,675 Male70.442.646,848 Race Asian66.735.44,178 AfricanAmerican47.515.120,320 Hispanic61.427.620,479 AmericanIndian65.435.7535 Multiracial75.039.164 Non-residentalien60.821.51,555 White78.551.785,875 NotReported69.641.51,519 58

PAGE 59

Table1-6.SummaryStatistics FullSampleWhiteBlackHispanicOthersMaleFemale CLASTScore-3.082.19-20.54-11.37-6.98-1.28-3.98 AAEver%80.6084.2067.1075.5078.9081.6078.40 SemestersEnrolledAfterFirstAttempt6.576.597.297.427.486.667.23 after1stCLASTTake PassCLASTFirstTake%36.1049.4013.9024.9032.4039.5042.60 PassCLASTEver%65.2076.1043.5055.8061.6068.1068.80 PercentofObservations67.5011.8015.705.0066.6033.40 Observations52,11135,1916,1528,1822,58634,70317,407 Note:About4percentofthesamplebelongstotheotherracegroup Asian,AmericanIndian,mixedraces. 59

PAGE 60

Table1-7.TestofRDAssumption Estimatestd.errorp-value Gender Male0.0120.0070.067 Race White0.0130.0060.038 Black0.0090.0040.027 Hispanic-0.0200.0050.000 60

PAGE 61

Table1-8.CLASTEffectsbyCLASTAttempt ObservationsNumberofSemestersAAEver CLASTAttemptTakersFailedestimatestderrorestimatestderror 123456 1stAttempt52,11129,738-0.0570.0690.0480.006 2ndAttempt20,95810,733-0.6540.1140.0880.009 3rdAttempt8,8445,292-0.3100.1670.1270.015 4thAttempt4,3112,829-0.1610.2170.0730.022 5thAttempt2,2861,565-0.2470.2710.0840.033 6thAttempt1,246923-0.1430.2970.0960.046 7thAttempt707509-0.1380.3370.1240.063 8thAttempt4092870.1810.4450.0520.083 9thAttempt240190-0.2120.5030.0070.002 61

PAGE 62

Table1-9.CLASTEffectsbyGender NumberofSemestersAAEver GenderObsestimatestderrorestimatestderror 123456 Male17,4070.0050.1270.0740.009 Female34,7030.1090.0830.0370.006 62

PAGE 63

Table1-10.CLASTEffectsbyRace NumberofSemestersAAEver RaceObsestimatestderrorestimatestderror 123456 White35,191-0.1030.0800.0480.006 Black6,1520.5200.3100.0630.021 Hispanic8,182-0.2740.2130.0750.014 Others2,5860.5860.3330.0370.026 Note:Estimatesaregeneratedusingsubsamplesfromeachspecic racegroup. 63

PAGE 64

Table1-11.CLASTEffectsbyYearofFirstCLASTAttempt NumberofSemestersAAEver YearObsestimatestderrorestimatestderror 123456 199515,321-0.3600.1420.0280.010 199610,004-0.1080.1660.0620.012 19975,381-0.3790.2200.0970.016 19983,5920.1670.2510.0610.019 19993,430-0.0390.2450.0380.019 20002,8570.0380.2390.0680.021 20012,4460.2450.2520.0350.025 20022,789-0.2580.2070.0380.022 20032,293-0.3680.2010.0210.023 20042,369-0.0160.1550.0570.022 20059840.0580.1800.0730.037 2006645-0.1220.1130.0720.067 64

PAGE 65

Table1-12.CLASTEffectsbyPercentofStudentsEverAttendedSUS NumberofSemestersAAEver TypeObsestimatestderrorestimatestderror 123456 AboveMedian31,101-0.0990.0870.0470.007 BelowMedian21,0100.0270.1160.0530.008 65

PAGE 66

Table1-13.RDEstimatesontheNumberofSemestersEnrolledAftertheFirstCLASTAttempt CLASTScorePointWindow PolynomialOrder 2 3 5 10 15 20 50 75 100 105 110 112 Oneestimate0.0230.141-0.121-0.200-0.151-0.127-0.052-0.064-0.050-0.053-0.050-0.051 std.error0.3070.4570.2580.1470.1190.1030.0740.0700.0700.0700.0700.070 t-stat0.0800.3100.4701.3601.2601.2300.7100.9100.720-0.770-0.720-0.740 Twoestimate0.4420.066-1.205-0.127-0.133-0.149-0.009-0.009-0.051-0.042-0.051-0.047 std.error0.6780.3920.5510.2600.1900.1590.1050.0970.0930.0920.0920.092 t-stat0.6500.1702.1900.4900.7000.9400.0800.0900.550-0.460-0.560-0.510 Threeestimate0.0000.6973.277-0.439-0.355-0.221-0.257-0.124-0.024-0.040-0.029-0.038 std.error0.0000.8881.3190.4440.2960.2300.1400.1250.1150.1140.1130.113 t-stat-0.7802.4800.9901.2000.9601.8400.9900.210-0.350-0.260-0.340 Fourestimate0.0000.0000.480-0.488-0.150-0.230-0.137-0.211-0.210-0.156-0.158-0.144 std.error0.0000.0000.5790.7070.4470.3350.1770.1550.1390.1380.1380.138 t-stat--0.8300.6900.3400.6900.7701.3601.510-1.130-1.150-1.040 66

PAGE 67

Table1-14.RDEstimatesontheAssociatesinArtsDegreeCompletion CLASTScorePointWindow PolynomialOrder 2 3 5 10 15 20 50 75 100 105 110 112 Oneestimate0.0380.0650.0450.0530.0470.0400.0350.0430.0470.0480.0480.048 std.error0.0230.0330.0180.0100.0080.0070.0050.0050.0050.0050.0050.005 t-stat1.6701.9802.4805.2405.5705.4506.3908.3709.1709.2809.3809.380 Twoestimate0.0310.0180.0570.0500.0590.0560.0360.0250.0210.0390.0380.039 std.error0.0520.0290.0380.0180.0130.0110.0080.0070.0070.0070.0070.007 t-stat0.5900.6301.4802.7604.5105.0704.7103.6003.0505.7505.6405.710 Threeestimate0.0000.0700.0740.0540.0370.0570.0560.0480.0430.0440.0440.042 std.error0.0000.0680.0890.0310.0210.0160.0100.0090.0080.0080.0080.008 t-stat-1.0300.8401.7601.7803.5905.6705.4305.1005.3405.3305.080 Fourestimate0.0000.0000.0170.0420.0710.0420.0550.0590.0550.0520.0510.053 std.error0.0000.0000.0420.0490.0310.0230.0120.0110.0100.0100.0100.010 t-stat--0.4100.8702.2801.7704.4905.5005.4905.3005.2705.380 67

PAGE 68

Table1-15.FractionofStudentswithAADegreebutdidnotPassCLASTEver YearWithAAdegreebutdidnotPassCLASTEverTotalPercent 19951,76315,32111.5 19962,03110,00420.3 19971,1235,38120.9 19987083,59219.7 19997833,43022.8 20008042,85728.1 20015902,44624.1 20026162,78922.1 20035722,29324.9 20047392,36931.2 200529498429.9 200619564530.2 Total10,21852,11119.6 68

PAGE 69

CHAPTER2 THEIMPACTOFTOUGHEREDUCATIONSTANDARDS 2.1Introduction ThestandardsandaccountabilitymovementinUSeducationisoftencharacterized asasetofpoliciesdesignedtoimproveschooleffectiveness.Butseveralpolicies thatfallunderthestandardsandaccountabilityumbrellaaredesignedtoincentivize students.Thesepolicies,whichincludetest-basedpromotionandhighschoolexit exams,differindetail. 1 Thebasicideais,however,thesame:standardsareset,tests areusedtomeasurestudents'progressagainstthosestandardsandhigh-stakes consequencesaretiedtotheoutcomeofthosetests.Inthecaseofhighschoolexit exams,thefocusofthispaper,thestandardisdenedbyparticularthresholdsonmath andreadingtestsgiventostudentsintenthgrade.Studentscannotgraduatehigh schooliftheydonotpassthesetests. Thereareseveralissuessurroundingthepracticalimplementationofhighschool exitexams.Whichaspectsofthecurriculumshouldbecoveredbythesetestse.g., calculusoronlyalgebra?Whatformatshouldtheteststakee.g.,shortansweror multiplechoice?Shouldanystudentsbeexemptede.g.,limitedEnglishprocient students?Howmanychancesshouldstudentshavetoretakethetestiftheyfailrst timee.g.,ve,asinFlorida,ormore,asinsomeotherstates?Whereshouldthe standardi.e.,passingscorebeset? Inthispaperwefocusonthislastquestionandconsiderwhetherstudentsshould besubjectedtohigherstandards.Policy-makersareoftenperceivedassetting standardsthataretoolow.Moreover,economicmodelssuggestthisperceptionmaybe 1 GreeneandWinters2009discusstest-basedpromotioninFlorida. 69

PAGE 70

correct. 2 Yetthesemodelsarestatic,inthesensethattheyconsiderasetofstudents thatfaceaparticularstandard.Inpractice,studentscandropoutofeducationbefore theyhavetofacethestandard.Forexample,highschoolstudentscandropoutintenth oreleventhgrade,beforeanyexitexamstandardmustbemettheendoftwelfthgrade. Assuch,higherstandardsmightincreasedropoutrates,perhapsgeneratinganet humancapitalloss. Inthispaperweaimtoshedlightonthisquestionbyexaminingtheeffectsof thehighergraduationstandardsintroducedinFloridain2001.Theseincreased thescoresrequiredonboththemathandreadingportionsofthehighschoolexit examsuchthat19percentofthestudentsthatpassedundertheoldlowerstandard wouldhavefailedunderthenewhigherstandard.Weestimatetheseeffectsusing adifference-in-differenceapproachthatclassiesstudentsaccordingtowhether theyreachedtenthgradeintheyearbeforeortheyearafterthehigherstandardwas introducedlabeledpreandpostandaccordingtowhethertheirinitialexamscore wouldpassunderthelowerandthehigherstandard.Assumingthatthehigherstandard onlyaffectsbehavioraftertheinitialexamattemptintenthgradeandassumingthat thehigherstandarddidnotaffectstudentswhosescorewouldhavepassedunder bothregimes,pre-postcomparisonsofthestudentsthatwouldhavepassedunderthe 2 Inaformalmodelofstandardssetting,Costrell1994arguesthatpolicy-makers maysetstandardsbelowthosethatwouldbepreferredbyamedianvoter.Thisis becauseegalitarian-mindedpolicy-makerscaremoreaboutlower-abilitystudentsthan doesthemedianvoter.Betts1998arguesthatevenpolicy-makersthatcareabout low-abilitystudentsshouldsethigherstandards.Usingadifferenttheoreticalset-up, heshowsthathigherstandardsmakelower-andhigher-abilitystudentsbetteroff,with onlymiddle-abilitystudentslosingout.Theintuitionisthathigherstandardsreducethe fractionofstudentsthatpass,therebyimprovingthequalityofboththegroupthatfails andthegroupthatpassesthetest.Thelosersarethosethatwouldhavepassedunder thelowstandardbutdonotpassunderthehighstandard. 70

PAGE 71

higherstandardcanidentifytheeffectsofthehigherstandard. 3 Weexamineeffectson dropoutrates,graduationrates,postsecondaryenrollmentratesandearnings. Ourestimatespointtotwomainresults.First,thehigherstandardhadnoimpact ondropoutrates.Instead,thefractionofstudentsthatcompletedgradetwelveappears unaffected.Thisissurprising,sinceonemightexpectstudentsthatfailedtheinitial examtobecomediscouragedanddropoutofschool.Instead,itseemsthatthehigher standardcausedahigherfractionofstudentstocompletegradetwelvewithoutpassing theexam.Thenetresultoftheseeffectswasaverysmalldropingraduationrates. Second,thehigherstandardhadnoapparentimpactonpostsecondaryenrollment ratesandearnings.Thepostsecondaryenrollmenteffectsareunsurprisingconsidering thatstudentsdonotneedahighschooldiplomatoenroll.Theearningseffectsare consistentwithsmallstandardseffectsongraduationratesandsmallearningseffectsof graduationMartorellandClark,2010. 2.2HighSchoolStandardsinFlorida Floridawastherststatetomakehighschoolgraduationcontingentonpassing anexitexamin1978. 4 ThisexamwasknownastheHighSchoolCompetencyTest HSCT.In2000,bywhichtimeFloridahaddevelopedanaccountabilitysystembased onFCATtestsingradesthreethroughten,theHSCTwasreplacedbythegradeten FCAT. 3 Withouttheseadditionalassumptions,whichwejustifybelow,theimpactsofthe higherstandardwouldhavetobeidentiedusinganaggregatepre-postcomparison. Sincethistypeofestimatewouldbehighlysensitivetothespecicationofcohorttrends, wethinkthatourapproachismoreappealing. 4 ThissectiondrawsheavilyonseveralFloridaDepartmentofEducationpublications. TheseincludeFloridaDepartmentofEducation2005 a andFloridaDepartmentof Education2005 b 71

PAGE 72

ThegradetenFCATexamsareofferedinreadingandmathematics. 5 Ingeneral,all studentsenrolledingradetenshouldtaketheseexams. 6 Theexamscontainmultiple choiceitems,gridded-responseitemsandperformancetasks. 7 Rawscoresarescaled usingitemresponsetheorymethods.Thescaledscoresrangefrom100to500and covervepre-denedlevelsofachievement. 8 Topasstheexitexam,studentsneed toperformatthesecondlevelinbothexams.Forstudentsintenthgradein2000-2001 i.e.,tooktheexamforthersttimeinspring2001,thepassingscoreswere287for readingand295formath.Forstudentsintenthgradefrom2001-02onwardsi.e.,took theexamforthersttimefromspring2002onwards,thepassingscoreswere300in bothsubjects.Studentsandparentsreceiveafour-pagereportdescribingtheirexam performance.Thisliststhescoresobtained,whetherthestudentpassed,theassociated performancelevelsandthescoresrequiredtoobtainthevariousperformancelevelsand topass. 5 Thereadingexamisa160-minuteexamthatassessesstudents'reading comprehension.Thetestiscomposedofabout6-8readingpassagesinformational orliterarywithsetsof6-11questionsbasedoneachpassage.Themathexamis a160-minuteexamthatassessesperformanceonvestrands:NumberSense, ConceptsandOperations;Measurement;GeometryandSpatialSense; AlgebraicThinkingandDataAnalysisandProbability.Studentsareallowedtouse calculators. 6 LEPstudentscanbeexemptediftheyhavereceivedservicesinanLEPprogramfor oneyearorless.AnExceptionalStudentEducationESEstudentmaybeexemptedif hehasanIndividualEducationPlanIEPandmeetscertaincriteria. 7 Ingridded-responseitems,studentsanswerquestionsthatrequireanumerical responseandtheymarktheirnumericalanswersinresponsegrids. 8 ThesearelittlesuccesswiththecontentontheFCATlevel1,limitedsuccess withthecontentontheFCATlevel2,partialsuccesswiththecontentontheFCAT level3,successwiththecontentontheFCATbyansweringmostquestionscorrectly, exceptforthechallengingquestionslevel4,successwiththecontentoftheFCATby answeringmostquestionscorrectlylevel5. 72

PAGE 73

Studentsthatfailoneorbothoftheseexamshavenumerousopportunitiestoretake them.Studentsneedonlyretaketheexamstheyhavenotyetpassed.Initially,these retakestookthesameformatastheinitialexam.Startingfall2004,theretakeformat changedtoincludeonlymultiplechoiceandgriddedresponsequestions. 9 Studentsin tenthgradein2000-2001i.e.,tooktherstexaminspring2001hadvechancesto retaketheexam:thefallofeleventhgradeOctober2001,thespringofeleventhgrade February-March2002,thesummerofeleventhgradeJune2002,thefalloftwelfth gradeOctober2002andthespringoftwelfthgradeFebruary-March2003.Students intenthgradein2001-2002onwardshadanadditionalretakeopportunityinthesummer ofgrade10.Sincefewstudentsretaketheexaminthesummeroftenthandeleventh grades,studentstypicallyhavevechancestopassbeforetheendofgradetwelve. Schooldistrictsarerequiredtogiveintensiveremediationtoseniorsthathavenotyet receivedapassingscoreontheFCAT.Inpractice,schoolsofferremediationtostudents beforetheyreachtheirsenioryear.Studentsaregivenperformancereportsaftereach retake. Toobtainapublichighschooldiploma,studentsmustmeettheexamrequirement, maintaina2.0GPAandearncoursecreditsintherequirednumberanddistribution. Somestudentsthathavemettheothergraduationrequirementsbutnotpassedthe FCATcanbeexemptedfromtheFCATandcanreceiveadiploma.Exemptionsare providedforstudentswithdisabilitiesifthestudent'sIEPteamdetermineinthestudent's senioryearthattheFCATdoesnotaccuratelyreectthestudent'sabilities.StartingMay 2004,studentswithACT/SATscoresabovecertainthresholdsarealsoexemptfromthe 9 Thischangewasdesignedtoreducegradingcosts.TheDOEclaimedthatthenew formatwasalignedwiththeoldstandards.StudentswillreceivethisiftheirIEPteam determinesthattheFCATaccuratelyreectsthestudent'sabilities. 73

PAGE 74

FCATrequirement.Studentsthathavemettheothergraduationrequirementsbutnot passedtheFCATcanreceiveaCerticateofCompletion. 10 Studentsthathavemettheothergraduationrequirementsbuthavenotpassedthe FCATandarenotabletoreceiveanexemptioncanstillobtainadiplomaiftheypass anotheradministrationoftheFCATaftergradetwelve.StudentscanretaketheFCAT inthesummerofgradetwelveorinthefollowingacademicyear.Theycanpreparefor theFCATbytakinganadditionalsemesterorfullyearthirteenthyearofhighschool educationorbytakingremediationandFCATpreparationclassesavailableatadultand communitycolleges. Studentswishingtoenrollinstateuniversitiesmusthaveahighschooldiplomaand musthaveacquiredcoursecreditsinspecicamountsandtypes.Studentswishingto enrollincollegecreditcoursesatcommunitycollegesmusthaveahighschooldiploma ormusthaveacerticateofcompletionandpassacollegeplacementtest.Students thatdonotpassthecollegeplacementtestareplacedincollegepreparatorycourses. Studentscanenrollinadulteducationprogramswithoutadiplomaoracerticateof completion. 2.3TheoreticalModelandEmpiricalStrategy Inthissectionwediscusshowhigherstandardsmightaffectvariousoutcomesand wedescribetheempiricalstrategythatwewillusetoidentifythem.Webeginwiththe empiricalstrategy.Thiscanidentifytheeffectsofhigherstandardsunderassumptions thatwendplausible.Wethenpresentasimpletheoreticalmodel.Thisisuseful forxingideasaboutthemechanismslikelytogivetorisetoeffectsondropoutand graduationrates.Westressthatthismodelmakesseveralassumptionsthatarenot neededforourbasicidenticationstrategyandsoourresultsarerobusttoviolationsof 10 Specialeducationstudentsthathavemettheothergraduationrequirementsbutnot passedtheFCATcanreceiveaSpecialDiplomaforStudentswithDisabilities. 74

PAGE 75

theseassumptions.Theseassumptionsareusefulonlyinsofarastheyhelpsimplifythe theoryandtherebyilluminatepossiblemechanisms. 2.3.1EmpiricalStrategy TheoreticalmodelssuchasthosedevelopedbyCostrell1994andBetts 1998suggestthatahigherstandardcouldaffectstudentsatallpointsoftheability distribution.Thatisbecausethesearesignalingmodels,andbecauseahigherstandard couldchangethereturnsassociatedwitheachobservabletypeofstudente.g., studentsthatobtainadiplomaandstudentsthatdonot.Butifahigherstandardaffects allstudents,itisdifculttoconstructacounterfactualfortheeffectofhigherstandards onaparticularstudentand,byextension,theaggregateimpactofhigherstandards. Onepossiblestrategywouldbetoestimatetheaggregateimpactusinganinterrupted timeseriesdesign.Thatis,comparethecohortrstsubjecttothenewhigherstandards withthelastcohortsubjecttotheoldlowerstandards,adjustingforcohorttrends.This approachisnot,however,ideal:cohorttrendsaredifculttocontrolforandtheremaybe otherreasonswhyoutcomesdifferacrossadjacentcohorts. Withthisdifcultyinmind,weuseadifference-in-differenceDDapproachto identifytheimpactofthehigherstandard.Thiscomparesoutcomesamongtwogroups. Therstgrouparestudentswithinitialscoresthatwouldhavepassedunderthenew higherstandardirrespectiveofwhethertheytooktheexambeforeorafterthestandard wasraised.Welabelthesethepassbothcontrolgroup.Thesecondgroupare studentswithinitialscoresthatwouldhavepassedundertheoldstandardbutnotthe newstandard.Welabelthesethepassold,failnewtreatmentgroup.Wearguethat thisapproachwillbevalidprovidedtwoconditionsaremet.First,thehigherstandard musthavenoimpactontheoutcomesofthepass-bothtreatmentstudents.This ensuresthattheevolutionoftheiroutcomesthroughthestandardsincreaseservesas agoodcounterfactualfortheevolutionofoutcomesamongthepass-oldfail-newcontrol groupintheeventthatstandardshadnotincreased.Second,becausethesegroups 75

PAGE 76

aredenedintermsofinitialexamscores,thehigherstandardmustnothavechanged behaviorinadvanceoftheinitialexitexam.Thisensuresthattheeffectsofthehigher standardcanbemeasuredusingoutcomesthatoccuraftertheinitialexamistaken. Wethinkbothoftheseassumptionsareplausible.Whilethereturntopassing willnotbeindependentofthestandardinaformalmodelofsignaling,itmightbe plausibletosupposethatthisconditionholdsinpractice.Evenifitisnot,anincrease inthereturnstopassingcausedbythehigherstandardwillcauseusto over-estimate theimpactondropout. 11 Sincewendatworstsmalldropouteffects,thesearestill interestingifinterpretedasworst-casescenarios.Sincethehigherstandardwas announcedonlyafewmonthsbeforetheinitialexam,andsincestudentshavemany severalopportunitiestoretaketheexamiftheyfailrsttime,itseemsreasonableto supposethatthehigherstandardsdidnotchangebehaviorinadvanceoftheinitialexit exam.Thisisconsistentwithatleasttwoaspectsofthethedata.First,thedistribution ofinitialscoresissimilaramongthepre-andpost-changecohortsnotreported. Second,weestimatedifference-in-differencemodelsinwhichthedependentvariables includebackgroundcharacteristicssuchaseligibilityforafreeschoolmealandthe probabilitythatgrade10studentsareontime.Consistentwithnoanticipationeffects, weseenoimpactsonbackgroundcharacteristics. Wealsoextendthismodelbyconsideringasecondtreatmentgroup:studentswith initialscoresthatwouldhavefailedundertheoldlowerstandardsandthenewhigher standardswecallthisthefailbothgroup.Thisisaninterestinggrouptoconsider becausethehigherstandardsmeanttheyhadtomakelargerimprovementsinorder topass.Thiscould,forexample,havemadethemmorelikelytodropout.Thisgroup 11 Forexample,ifthehigherstandardshavenoimpactonthepass-oldfail-new group,butdecreasedrop-outratesamongthepass-bothgroupbecausethehigher standardsincreasethereturntocontinuingineducation,wewillconcludethatthe higherstandardscauseddropoutratestoincreaseamongthepass-oldfail-newgroup. 76

PAGE 77

presentsnonewconceptualproblems.Itdoeshoweverpresentapracticalproblem: howtodeneitwhenitcould,intheory,includeallstudentsthatfailedtheexam.We experimentwithseverallowerbounds.Thatis,wemakedifferentassumptionsabout howlowtheinitialscoremustbeforstudentsnottohavebeenaffectedbythehigher standards.Weshowthatthisdecisionhasnorealimpactonourmainresults. 2.3.2SimpleModel Tohelpxideasaboutthepossibleeffectsofhigherstandards,wenowsketcha simplemodel.Indoingso,wemakeanumberofassumptions.Theseincludethetwo assumptionsonwhichtheDDstrategyrestsandothersthathelpilluminatehowthe effectsofhigherstandardsmightoperate. Tobegin,weassumethatstudentstakeasingletestattheendofgrade10.We denotetheirscoreonthistestby t 1 ,where t 1 2 [ t t ] .Weassumethatthescore necessarytopassi.e.,thestandardisannouncedatthesametimeastheresultsof thistest,suchthatpre-testbehavioristhesameacrosscohortshencecanbeignored. Thismeansthatwecanthinkoftheinitialscoreasameasureofability.Weassumethe standardsforthepre-andpost-changecohortsare L and H ,where t < L < H < t Uponobservingtheirscoresandthepassingscore,weassumethatstudentsmake twodecisions.First,theydecidewhethertodropoutofschoolattheendofgrade10 orstayinschooluntiltheendofgrade12.Second,iftheydecidetostayinschool, theychoosehowmuchefforttoexert.Effortisdenoted e andthecostofefforti.e.,the costofraisingtestscoresisassumedtobe ce 2 2 .Iftheystayinschool,studentsretake thetestattheendofgrade12.Weassumetheirscoreonthistest,denotedby t 2 ,is determinedasfollows: t 2 = t 1 + e + where isanormallydistributedmean-zerorandomvariableand t 1 + e = E t 2 j t 1 .It followsthattheprobabilityofpassingandtherebygraduatingforastudentwithrst 77

PAGE 78

score t 1 thatexertseffort e is: P grd j t 1 e = P t 1 + e + > L = F [ e )]TJ/F22 11.9552 Tf 11.955 0 Td [( L )]TJ/F39 11.9552 Tf 11.955 0 Td [(t 1 ]2 Weassumethatthewagereturntoadiplomaisconstantacrossstudentsand independentofthepassingstandard.Itfollowsthattheutilityofstayinginschool andtheutilityofdroppingoutofschoolcanbewrittenas: U stay j t 1 e = U S 0 + RP grd j t 1 e )]TJ/F39 11.9552 Tf 13.151 8.087 Td [(ce 2 2 3 U dropout = U D 0 + where U S 0 isthereturntocompletingschoolwithoutgraduatingand U D 0 isthereturnto droppingoutofschoolattheendofgrade10.Bothareassumedindependentofability andeffort.Theterm R isthereturntograduatingconditionaloncompletinggrade12 i.e.,thereturntopassingtheexam.Theterm isanormallydistributederrorterm thatcanbethoughtofasaperson-specicutilitytermthatisindependentofability.We assumethisisalsonormallydistributedwithmeanzero. 2.3.2.1Outcomesunderthelowstandard ThemainresultscanbeseeninFigure2-1AppendixBcontainsformalderivations. Startingwiththebottompanel,thesolidlinecharacterizesoptimaleffortasafunction oftheinitialtestscoreandwhenfacingastandard L .Notethatoptimaleffortishighest whenthepassprobabilityis0.5.Thatisbecauseefforthasthegreatestmarginalreturn atthispoint.Itfollowsthatoptimaleffortforhigh-andlow-scoringstudentsislowerthan itisformedium-scoringstudentsi.e.,forthemedium-scoringgroup,effortmovesthe passprobabilityoverarangecloserto0.5.Forallstudentsthatfailedthersttesti.e., scoreslessthan L ,optimaleffortispositivesincethemarginalreturntoaunitofeffort exceedsthemarginalcost.Forallstudentsthatpassedthersttesti.e.,scoresgreater 78

PAGE 79

than L ,optimaleffortiszero:thesestudentshavealreadypassedhencecanobtaina diplomawithnoadditionaleffort. Movinguptothemiddlepanel,thesolidlinecharacterizestheimpliedpass probabilityasafunctionoftheinitialscoresgivenoptimaleffortunderstandard L .There aretworeasonswhy,comparedtolow-scoringstudents,medium-scoringstudentsare morelikelytopass.First,theyhaveahigherprobabilityofpassingconditionaloneffort. Second,theyexertmoreeffort.Incontrast,high-scoringstudentsexertlesseffortthan medium-scoringstudentsthebottompanel,but,asscoresincrease,effortdecreases bylessthanone-for-one,hencetheimpliedpassprobabilityrises.Movinguptothetop panel,thesolidlinecharacterizestheprobabilityofstayinginschoolunderstandard L Thisisincreasingininitialscoresbecausehigherinitialscoresimplyahigherequilibrium passprobability. Optimaleffortdepends,amongotherthings,on R c ,thediplomareturnrelativeto themarginalcostsofeffortAppendixB.Thelargeristhisratio,themoreeffortwill beexerted.Itfollowsthatthisparameterwillalsohelptodeterminetheequilibrium probabilityofpassingconditionalonstaying.Itwillalsohelptodeterminetheprobability ofstayinginschoolbeyondgradeten:thelargeristhisratio,themorelikelystudentsare tostay.Forexample,ifthisratioiszero,studentswillexertzeroeffort,willpasswithonly smallprobabilityandwillviewstayinginschoolasalessattractiveproposition. 2.3.2.2Theimpactsofahigherstandard Itiseasytoanalyzetheimpactsofahigherstandard H > L .Thatisbecause,under theassumptionsofthemodel,astudentwithrstscore t 1 facingstandard H > t 1 will makethesamedecisionsasastudentwithrstscore t 0 1 = L )]TJ/F22 11.9552 Tf 12.117 0 Td [( H )]TJ/F39 11.9552 Tf 12.116 0 Td [(t 1 facingstandard L .BehaviorunderthehigherstandardisdepictedbythedashedlinesinFigure2-1. Notethattherearetwotypesofstudentsthatfailtomeetthestandard H :thosethat wouldhavepassedundertheoldstandardthepassold,failnewgroupwithscores L t 1 < H andthosethatwouldnoti.e.,thefailbothgroupwithscores t 1 < L 79

PAGE 80

Forthepassold,failnewgroup,theeffectsofthehigherstandardareobvious. Underthelowstandard,theypassedrsttimehencereceivedadiplomawithzero additionaleffort.Underthehighstandard,optimaleffortispositiveandboththe equilibriumpassprobabilityandtheequilibriumprobabilityofstayinginschoolare increasinginthescore.Theextenttowhichthehigherstandardaffectsoutcomesfor thisgroupwillbeincreasingin c and R .Intuitively,foranygivengaintopassing,alarger c impliesthatitismoreexpensivetopasshencelessattractivetostayon.Forany givencostofpassing,alarger R impliesthatitismoreattractivetostayonunderboth standards,butlessattractiveinthehighstandardscasebecausetheprobabilitythatthe standardisnotpassedimposesanimplicittaxonthisextragain. Forthefailbothgroup,theefforteffectsofahighstandardareambiguous, althoughtheequilibriumpassprobabilityislower,asistheequilibriumstaying-on probability.Theintuitionfortheambiguousefforteffectisasfollows.Ifthestandardis onlyslightlyhigheri.e.,suchthat H )]TJ/F39 11.9552 Tf 12.008 0 Td [(L > L )]TJ/F39 11.9552 Tf 12.008 0 Td [(t M 1 ,thesestudentswillbeincentivizedto workhardandtrytomeetit.Ifitismuchhigher,theywillgiveuptryingtomeetit.The ambiguousefforteffectmeansitisnotclearhowthecostofeffortaffectstheimpactof higherstandardsontheprobabilityofstayingon.Againthough,theimpactonstayingon willbelargerthelargeris R .Theintuitionisasbefore:larger R increasesthereturnto stayingon,buthigherstandardsimposeahigherimplicittaxonthisgain. Wecansummarizethesepredictedimpactsasfollows.First,forthepassold,fail newgroup,higherstandardswillincreaseeffort,decreasetheprobabilitythatstudents passiftheystayinschoolanddecreasetheprobabilitythattheystayinschool.These effectsonstayinginschoolwillbeincreasinginboth R and c .Second,forthefail bothgroup,higherstandardswillhaveambiguouseffectsoneffort,willdecreasethe probabilitythatstudentspassiftheystayinschoolanddecreasetheprobabilitythatthey stayinschool.Theseeffectsonstayinginschoolwillbeincreasingin R 80

PAGE 81

2.3.3Discussion Asalreadynoted,thismodelmakesmanyassumptions.Mostofthesewedonot needforouridenticationstrategytohold;theymerelysimplifythetheoreticalanalysis. Threesetsofassumptionsstandoutasbeingespeciallyunrealistic.First,themodel treatsthelabormarketinaverysupercialway.Inparticular,itassumesthatthereis areturntoobtainingadiplomabutotherwisenoadditionalreturnstoeffort.Amore generalmodelwouldallowforreturnstoeffortinadditiontopossiblesheepskineffects associatedwithgraduatinghighschool.Second,themodelignorespostsecondary education.Again,thisisdonetosimplifytheanalysis.Third,onlyonecomponent ofeffortismodeled-thatrequiredtopassthehighschoolexitexam.Otherrelevant componentsofeffortincludethoserequiredtostayinschooli.e.,persistence,as opposedtolearningandthatwhichimprovesproductivityinthelabormarket.Again, thegoalistodrawattentiontoaparameterthatweexpectwillhavealargeimpacton theeffectsofhigherstandards:thecostofimprovingtestscores. 2.4Data ThedatausedinthispaperwereprovidedbytheFloridaEducationandTraining PlacementInformationProgramFETPIP,aunitwithintheFloridaDepartmentof Education.FETPIPfollowsstudentsfromschoolintopost-secondaryeducationand intothelabormarket.TheFETPIPdatacombineseveraldatalesthatarelinkedat thestudentlevelusingidentifyinginformationsuchasthestudent'sname,dateofbirth andsocialsecuritynumber.Sincethedatawerelinkedbeforetheywereprovidedtous, matchratesareunknown. 2.4.1DataTypes TheFloridadataconsistsofabaseenrollmentrecordmatchedtoseveralother typesofdata.Thebaseenrollmentrecordreferstotheacademicyearinwhicha studentwasrstenrolledingradeteninaFloridapublicschooldenedtoinclude charterschoolsbutnotprivateschools.EveryindividualintheFloridadataistherefore 81

PAGE 82

associatedwithauniquebaseenrollmentrecordandweusethistodenethecohort towhichtheindividualbelongs.Inparticular,wedenethepreandposttreatment cohortstoincludeindividualsingradetenin2000-01and2001-02respectively. Severaltypesofdataarematchedtothisbaseenrollmentrecord: 1. SubsequentenrollmentsinFloridapublicschools :Theseenrollment recordsandthebaseenrollmentrecordincludeschoolandgradedetailsand time-varyingstudentdetailssuchasfreelunchstatusandspecialeducation classication. 2. Demographicdata :Theseincludeinformationonsex,monthandyearofbirth andrace.RaceisclassiedasWhite,Black,Hispanic,Asian,andother. 3. Grade10FCATdata :Inprinciple,theseincludedetailsofallgradetenFCAT attempts,includingdatesandscoresobtainedonboththemathandreading sections.Fortherstandsecondcohorts,gradetenFCATdataareonlyavailable forspringexamsandretakesi.e.,notforsummerandfallretakes.Sinceboth therstandlast-chanceexamsareadministeredinspring,thisisnotanimportant constraint. 4. Awardsdata :Theseincludedetailsofallcerticatesanddiplomasawarded tostudents.Thesedataincludethetypeofdiplomaawardede.g.,highschool diplomaandfurtherdetailsoftheroutebywhichitwasobtainede.g.,met standardrequirements,exemptedfromFCATrequirement. 5. Postsecondaryenrollmentandawardsdata :Theseareavailableforallstudents thatattendstatecommunitycollegesCCsandstateuniversitiesSUsinFlorida. Thedataincludeenrollmentandawardslessimilartothoseavailableatthehigh schoollevel.Weusethepostsecondaryinformationtodenevariablesincluding semestersinCCsandsemestersinSUS. 6. Earningsdata :ThesecomefromtheUnemploymentInsuranceUItaxreports submittedtotheFloridaDepartmentofRevenuebyemployerscoveredbythe state'sUIlaw.Coveredemployersarerequiredtoreport,onaquarterlybasis, thewagespaidtoeachemployeeinordertodeterminetherm'staxliability. WageswillbereportedfornearlyallindividualsworkingforearningsinFlorida. Themajorcategoriesnotcoveredwillbethoseworkingintheinformalsector,the self-employedandthoseworkinginthemilitary.Foreachindividualinourdata wehave,foreachquarter,earningsinformationasprovidedbyeachemployer. Wesumthistoobtaintotalearningsinthequarteranddeateitto$2000using theCPI-Useries.Notethatearningscouldbeobservedtobezerobecause thesampleshavezeroearnings,becausetheyworkintheuncoveredsectoror becausetheyhaveleftthestate.Hencewhilewedonothaveanyattritionfromour 82

PAGE 83

data,wemayhavesomeobservationswithfalsezeroearnings.Martorelland Clark2010considerthispossibilityandarguethatitwillnothavearst-order impactontheestimatesreportedthere.Similarconclusionslikelyimplyhere. 2.4.2AnalysisSample Theanalysissampleincludesstudentsingrade10in2000-01and2001-02. Wedenethreegroupsofstudentsbasedontheirinitialtestscores.Thegroupsare depictedinFigure2-2.Therstgroupfailedbothscoredatlevelsthatwouldhave failedunderbothstandardsstudentsfailwhenatleastonescoreisbelowthepassing cutoff.Thesecondgrouppassedold,failednewscoredatlevelsthatwouldhave passedundertheoldstandardbutfailedunderthenew.Thethirdgrouppassedboth scoredatlevelsthatwouldhavepassedunderboth.Wecouldhaveincludedmore studentsinthefailbothandpassbothgroupsbutchosenotto.Thecommontrends assumptionunderlyingthedifference-in-differencestrategyseemmoreplausiblewhen thegroupsaredenedusingtightertestscoreranges. Table2-1presentssomedescriptivestatisticsforthesegroups.Amajorityofthese studentsaregirls,alargefractionaroundonequarterareeligibleforafreeorreduced pricelunchFRPLandalmostonequarterdidnotreachgrade10ontime. 12 Not surprisingly,studentswithinitialscoresthatfailunderbothstandardsfailbothare moredisadvantagedthanstudentsthatpassundertheoldlowerstandardpassold, failneworpassunderboththeoldlowerstandardandthenewhigherstandardpass both. 2.5Results Inthissectionwereportdifference-in-differencesestimatesoftheimpactsofthe tougherstandards.Asdiscussedintheempiricalstrategysection,theseestimates arethecoefcientsontheinteractionofthetwogroupdummiespasstheoldexam, 12 Wegeneratethisvariableusingdataonyearandmonthofbirth. 83

PAGE 84

failthenewexamandfailbothexamsandthepostcohortdummyinregressions ofoutcomesonvariouscontrolvariables,thetwogroupdummies,apostdummythat indicatesthecohortfacesthenewhigherstandardandtheinteractionsofthegroup dummiesandthepostdummy.Theexcludedgroupisthepassbothgroup-students whoseinitialscoreswerehighenoughtopasstheexamunderboththeoldlowerand newhigherstandards. Webeginbyestimatingthesemodelsonthesubsampleofstudentswithmathand readingscoresinthe280-310range.Thecommontrendsassumptionunderlying thedifference-in-differenceanalysisismoreplausibleoverthisnarrowerrange.The rstcolumnpresentsestimatesfrommodelswithoutcontrols.Inthesecondcolumn weaddbaselinecontrolsreduced/freelunch,white,and10thgradeontime.Inthe thirdcolumnweaddschoolxedeffects.Inthefourththroughsixthcolumns,wereport estimatesgeneratedfromthesamemodelsbutestimatedoveranevennarrowerrange ofdata-305.Theseprovideacheckonthecommontrendsassumption,since violationsofthisassumptionshouldberevealedbydifferencesbetweentherstthree andnextthreeestimates.Inthesevenththroughninthcolumnswereportplacebo treatmentestimatesgeneratedusingasetofstudentsthatwouldallhavepassed underboththeoldlowerandnewhigherstandardi.e.,scores300-310,someof whichareincorrectlycodedashavingfailedtomeetactitioushigherstandardof305. Sincetheseplacebotreatmenteffectestimatesshouldbezero,theyprovideauseful specicationtest.Inotherwords,non-zeroplacebotreatmenteffectestimatessuggest violationsofthecommontrendsassumptionandsuggestthatweshouldbeskepticalof theestimatesreportedincolumnsonethroughsix. 2.5.1HighSchoolOutcomes Wefocusrstonhighschooloutcomesandwebeginwithhighestgradecompleted i.e.,theinverseofhighschooldropout.Recallthatwearguedthattougherstandards couldreducehighestgradecompletedifthehigherstandardsdiscouragestudents.We 84

PAGE 85

assumethatthehigherstandardsdonotdetertheyoungerofthesecohortsfromtaking theinitialexamandwearguedthatthiswasconsistentwiththeempiricaldistributions ofinitialscores.Instead,wefocusonhighestgradecompletedconditionalontakingthe initialexam. TheestimatesincolumnsthroughofTable2-2suggestthatthehigher standardshadlittleimpactonthisoutcome.Thesignsswitchacrossspecications andallestimatesaresmallandstatisticallyindistinguishablefromzero.Theplacebo estimatesarealsosmallandstatisticallyinsignicant,atleastoncethebaselinecontrols havebeenadded.Evenintheworstcasescenariofortougherstandardscolumn, theestimatessuggesteffectsonhighestgradecompletedof-0.04. Theseestimatessuggestthatthetougherstandardshadnoimpactondropout. Sincethetougherstandardsincreasedthefractionofstudentsthatfailedtheinitial attempt,wemightexpecttondalargerfractionofstudentstakingthelastchance exam-theretakewhichtakesplaceattheendofgradetwelve,astudent'slastchance topassbeforescheduledgraduation. 13 AsseeninthesecondrowofTable2-2,thisis exactlywhatwend.Inparticular,theseestimatessuggestthatthetougherstandards increasedtheprobabilityoftakingthelast-chanceexambyaroundfourpercentage pointsforthepassold,failnewgroupandbyaroundthreepercentagepointsforthe failbothgroup.Theseestimatesarereasonablyrobusttothenarrowobservation window.Wecannotimplementtheplacebospecicationbecausestudentsthatpassed underbothstandardsareneverobservedtoretakeattheendofgrade12. Becausewendnodropouteffectsandbecausethetougherstandardincreases thefractionofstudentsretakingattheendofgrade12,wemightexpectthehigher 13 Ifwefoundnoeffectsonthisoutcomeitwouldimplythatstudentsthatfailedatthe newhigherstandardpassedoneoftheretakesadministeredbeforethelast-chance exam. 85

PAGE 86

standardtohavedecreasedthegraduationratei.e.,thefractionofstudentsthatpassed theexamhencemetallofthegraduationrequirementsandobtainedahighschool diploma.Specically,becausethehigherstandardincreasedtheprobabilityoftaking thelast-chanceexambyfourpercentagepoints,andbecausethepassrateonthis examisaroundonethird,wemightexpectthetougherstandardstodecreasethe graduationratebyoneortwopercentagepoints.Thisisroughlywhatweseeinthethird panelofTable2-2.Forthepassold,failnewgroup,thereisaroughly1.5percentage pointdecreaseinthegraduationrate.Amongthefailbothgroup,thedecreaseis smallerandnotstatisticallysignicant.Theplacebotestsarecomfortablysatised. InFlorida,butnotnecessarilyinotherstates,studentsthatcompletetwelvegrades butdonotreceiveadiplomaareeligibletoreceiveacerticateofcompletion.Wemight expectthegraduationratereductioncausedbythetougherstandardstobemirroredby anincreasethefractionofstudentsreceivingsuchacerticate.Asseeninthebottom panelofTable2-2,thisisroughlywhatwend.TheCerticaterateincreaseislower thanthegraduationratedecrease,perhapsbecausestudentschoosenottoobtaina Certicatebecausetheyplantoretaketheexamafterthetwelfthgradeandtrytoearna diploma. Tosummarize,wendthatthetougherstandardsdidnotcausestudentstodropout earlierthantheywouldhavedoneotherwise.Thisisanimportantresult,whichwereturn tobelow.Instead,wendthatbytheendoftwelfthgrade,ahigherfractionofstudents havestillnotpassedtheexam.Asaresult,thetougherstandardsreducethegraduation rate,oneconsequenceofwhichistoincreasethefractionofstudentsthatobtaina CerticateofCompletion. 2.5.2PostsecondaryEducation Wenotedabovethatstudentsdonotneedahighschooldiplomatoenrollin college.Assuch,itisnotclearwhywewouldexpecttondanyeffectsoftougher standardsonpostsecondaryoutcomes.Consistentwiththislineofreasoning,thetop 86

PAGE 87

twopanelsofTable2-3donotrevealanyclearimpactsonthenumberofsemestersthat studentswereenrolledintwo-yearcollege.Thereappeartobesignicantandpositive i.e.,wrong-signedimpactsonthenumberofsemestersthatstudentswereenrolledin four-yearcollege,butonlyforthefail-bothgroup,onlyatinyfractionofwhomenrollin afour-yearcollege.Theplacebotestsunderlinethefragilityofthisresult:theseshowa negativeandstatisticallysignicantimpactofthefalseplacebotreatmentonthenumber ofsemestersenrolledinfour-yearcollege. 14 2.5.3Earnings Ourestimatesoftheeffectsoftougherstandardsonhighschooloutcomeshave ambiguousimplicationsfortheirimpactsonearnings.Ontheonehand,sincetougher standardsreducegraduationrates,theycouldreduceearnings.Thisassumesthereis asignalingvaluetoahighschooldiploma,suchthatworkersthatdonotgraduatesuffer anearningslossrelativetoworkersthatdo.Ontheotherhand,byforcingstudentsto retaketheexamsthattheyfailed,thetougherstandardscouldincreasestudenteffort andincreasetheamountoftimespentonbasicskillsandtheamountofattention receivedbyteachers.AlloftheseeffectscouldservetoincreaseearningsTyler, 2004. 15 Eveninthebest-andworst-casescenariosassociatedwiththesehypotheses, itisdifculttoimaginethetougherstandardshavinglargeimpactsonearnings.For example,assumingthetougherstandardscausedathreepercentagepointdropinthe graduationrateatthetopendofourestimatesandassumingasignalingvalueof30 14 Theseoutcomesexcludeenrollmentinprivatecollegesorincollegesoutsideof Florida.Sincestudentscanenrollintwo-yearcollegeswithoutahighschooldiploma, itisnotobviouswhythetougherstandardsshouldimpactthefractionofstudentsthat enrollinthesetypesofcolleges. 15 Ifstudentsaredemotivatedbyfailingexamse.g.,sufferalossofcondence,then thiscouldcounteracttheseefforteffects. 87

PAGE 88

percent-attheverytopendoftherangeofestimatesfoundintheliteratureMartorell andClark,2010,wewouldexpecttoseenegativeearningseffectsofaroundone percentagepoint-around$400whentheoutcomeiscumulativeearningsovertherst veyearsaverageof$37,000inTable2-3.Assumingthetougherstandardsincreased effortbroadlyconstruedby10%,andassumingarateofreturntoayearofeducation of10percentinlinewiththeestimatesreportedbyCard1999,wewouldexpect toseepositiveearningseffectsofaroundonepercentagepoint.Iftheresomeofboth effectsoperate,wewouldexpectnetearningseffectstobesomewherewithin$400of zero. Weconsidertwoearningsoutcomes:earningsintherstyearafterhighschool i.e.,therstfourquartersafterthescheduledhighschoolgraduationandearnings intherstveyearsafterhighschooli.e.,therst20quarters.Theestimatesare reportedinthebottomtwopanelsofTable2-3.Theearenearlyallinthepredicted range.Theyarenot,however,preciseenoughtorevealwhichtypesofeffectsdominate. Indeed,incolumnsand,weseeamixtureofpositiveandnegativeestimates. Theonlyreasonableconclusionisthatanyearningseffectsaretoosmalltobedetected inthesedata.Notethataddingdataonearningsatolderagesisunlikelytosolvethis problem.Thatisbecausewemightexpecttheeffectsoftougherstandardstobeweaker atolderages.Wewouldcertainlyexpectsignalingeffectstoweakenovertimeasrms acquiredmoreproductivityinformation.Efforteffectsmightalsoweaken,although thedynamiceffectsofincreasedeffortarenotobvious.Eitherway,itisplausibleto supposethataddingearningsatlaterageswoulddiluteanyearningseffectsoftougher standards. 2.5.4RobustnessChecks Finally,wecheckrobustnessbyestimatingeffectsonpre-determinedcharacteristics Table2-4.Becausethesecharacteristicsarepre-determined,thehigherstandard cannotpossiblyaffectthem.Asaresult,theseestimatesgiveusasenseofthelikely 88

PAGE 89

robustnessofourndings.Inparticular,signicanteffectsonthesecharacteristicswould suggestthatthetreatmentiscorrelatedwithobservableseveninthedifference-in-difference framework,hencemayalsobecorrelatedwithunobservables.Infactmostofthese estimatesarestatisticallyindistinguishablefromzeroasareothersnotreported. Theonlyoutcomeforwhichonemightsuspectthatthereareeffectsisgrade10on time.Sincetheeffectsarenegativeacrossallspecications,includingtheplacebo specication,itmaybethattheabilityproleofretentioninearliergradeswasdifferent inthepreandthepostcohorti.e.,higher-abilitystudentsweremorelikelytoberetained inthepostcohort.Fortworeasons,wedonotthinkthisismajorconcern.First,nearly allofourestimatesarerobusttoincludingcontrolsforthisvariablecomparecolumns andandandinTables2-2and2-3.Second,nearlyallofourplacebo estimatessuggestnoeffects,eventhoughthereisaplaceboeffectonthisvariable albeitnotstatisticallysignicantTable2-4. 2.6DiscussionandInterpretation Ouranalysissuggeststhatthetougherstandardshadnoimpactonhighschool dropoutrates,butledtosmallreductionsinhighschoolgraduationrates.Theseare associatedwithincreasesinthefractionofstudentstakingthelast-chanceexam, increasesinthefractionofstudentsthatobtainaCerticateofCompletion,butno impactonpostsecondaryenrollmentorearnings. Toputthesegraduationeffectsintoperspective,weusedourestimatestoderivean estimateoftheimpactofthetougherstandardsontheoverallgraduationratei.e.,not justtheimpactonthefailbothandfailold,passnewgroups.Todoso,wecalculatedthe pre-reformgraduationrateforeveryscoreintheinitialmathandreadingdistributions. Wethenusedthesepre-reformgraduationratesandourestimatesofthegraduation rateimpactsofthereformtocalculatepost-reformgraduationratesateachscorelevel. Fromthereitiseasytoaggregateupandcalculatethedifferenceingraduationrates inthepreandpostreformperiods.Incalculatingthepost-reformgraduationrate,we 89

PAGE 90

assumednoeffectonthepassbothgroup,weimposedtheestimatedeffectonthepass old,failnewgroupandweimposedtheestimatedeffectonthefailbothgroup. Sinceitisdifculttoknowhowtodenethefailbothgroupi.e.,atwhatpointdoes thisbecomesobigthatthedifference-in-differenceassumptionsnolongerhold?we repeatedthisprocedureformanypossibledenitionsofthisgroup.Usingthedenition employeduntilnowi.e.,onlyincludingstudentswithscoresgreaterthan280inFigure 2-2yieldsanoveralleffectof0.2percentagepoints.Changingthe280thresholdto200 increasesthistoonepercentagepoint.Fromthereitisrobusttothresholdsbasedon evenlowerscores.Althoughthisrangeofestimatesisawideone,thisexercisedoes atleastruleoutanoverallgraduationrateeffectofbiggerthanonepercentagepoint. Interestingly,aFloridareportOfceoftheProgramPolicyAnalysisandGovernment Accountability,2007showsthatthefractionofstudentsreceivingastandarddiploma decreasedby0.4percentagepointsbetween2003and2004,whenmostofthetenth gradestudentsinthepreandpostcohortswouldhavegraduated. Itisinterestingtoconsiderwhethertheseresultsareconsistentwithestimates foundinthepreviousliterature.Althoughweareunawareofpreviousanalysesofthe impactoftougheningstandards,thereareatleasttwostrandsofrelatedresearch. Onestrandlooksattheimpactsofintroducingthesetypesofexams.DeeandJacob 2007isthelatestandperhapsthemostcomprehensivepaperinthislineofresearch. Theyuseanacross-statedifference-in-differencestrategythatexploitsthefactthat differentstatesadoptthesepoliciesindifferentyears.Likethepreviousliterature,they ndnoclearevidencethathighschoolexitexamsreducedhighschoolcompletionrates i.e.,completionofgrade12.Theyndsmallstatisticallysignicanteffectswhenthey splitbysexandrace,althoughtherace-andgender-speciceffectsareunrelatedto academicperformance,atleastasproxiedbythebaselinecompletionrate.Theynd 90

PAGE 91

noeffectsoncollegeenrollmentandnoclearevidenceonlabormarketoutcomes. 16 A secondstrandofliteratureusesregressiondiscontinuitydesignstolookattheimpacts offailingtheinitialexamonoutcomesincludinghighschooldropoutandgraduation rates.UsingdatafromTexas,Martorell2010ndsthatfailinganinitialexamdoes notdiscouragestudents.Hendshowever,thatstudentsthatfailaninitialexamare lesslikelytograduate,becausetheycannotpasstheexambytheendoftwelfthgrade. UsingdatafromMassachusetts,Papay,MurnaneandWillett2008andOu2010nd similarresults. 17 Bothsetsofndingsarebroadlyconsistentwiththosepresentedhere.Thisis interestingfortworeasons.First,itsuggeststhatourndingsofsmalleffectsisnot drivenbyourfocusonawell-establishedprogramorbyourdecisiontoconditionon gradetenscores.Second,itsuggeststhattheregressiondiscontinuityestimates,which identifyeffectsonlyforstudentsatthepass-failmargin,generalizetostudentsfurther awayfromthecutoff,suchasthoseinthefailold,passnewandfailbothgroupsthat weconsider. Itisalsointerestingtoconsiderwhytougherstandardsappeartohavesuch smalleffects.Onepossibilityisthatstudentswerenotawareofthepolicy.Thatis,the studentsinthepassold,failnewgroupdidnotrealizetheywouldhavetopassthe examatalaterattemptinordertograduate;thestudentsinthefailbothgroupdidnot realizetheywouldhavetomeetahigherstandardinordertograduate.Thisdoesnot seemplausible,especiallywhenappliedtotherstofthesetwogroups:thosestudents 16 InaseparateanalysisoftheimpactondropoutratesusingdatafromMinnesota, theyndthattheintroductionofanexitexamledtoaslightincreaseingraduationrates. Whentheyfocusonlow-incomedistrictshowever,theyndthattheexamdecreased graduationratesbytwoorthreepercentagepoints. 17 Papay,MurnaneandWillett2008andOu2010dohoweverndalargeeffectof failingthemathematicsexamforlow-incomeurbanstudents. 91

PAGE 92

musthavebeentoldtheywouldhavetoretaketheexam.Asecondpossibilityisthat thereturntograduation R inthetheoreticalmodelislow.Intheextremecase,if R was zero,suchthattherewasnosignalingvaluetothehighschooldiploma,rationalstudents wouldactasiftherewasnothresholdatwhichtheexamwaspassedandwouldinstead chooseeffortlevelsinawaythatoptimallytradedoffthecurrentcostsofeffortagainst thefuturelabormarketreturnstoeffort.EvidencepresentedbyMartorellandClark 2010suggeststhatthesignalingvalueofadiplomais,indeed,small,althoughthis doesnotimplythatstudentsareawareofthisinformationandactonit. Athirdpossibilityisthatstudentsareovercondent,andunder-estimatethe probabilitythattheywillbeunabletopasstheexam.Combinedwithatendencyto followthedefaultrouteofstayinginschoolunlessforcedtodootherwise,thiswouldbe apowerfulforceservingtobluntanyeffectsofthetougherstandards.Withthedataat hand,itisdifculttodeterminetheimportanceofover-condence.Nevertheless,there isampleevidencetosuggestitislikelyafactor.Forexample,Fischhoffetal.2000use datafromNLSY1997toshowthatonly7%ofteensexpecttohavenotcompletedhigh schoolbyage20,comparedwiththe16%thattheyreportfromthen-currentdata.Other datasuggestover-condenceisnotconnedtolower-abilityoryoungerteenagers.In astudyoftheimpactofnancialincentivesonthecompletionoftherstyearofan economicscourseattheUniversityofAmsterdam,Leuven,Oosterbeekandvander Klaauw2010reportevidencefromabaselinesurveywhichsuggeststhatalmosttwice asmanystudentsexpecttocompletetherstyearasactuallycompletetherstyear. 2.7Conclusion OuranalysissuggeststhatatougheningofFloridahighschoolgraduation standardsledtoasmalldecreaseinhighschoolgraduationrateswithnoadverse impactsonotheroutcomes.Whilesomemightconcludethatstandardsshouldbemade eventougher,itisnotclearthatthetougherstandardsyieldedanybenets.Wecan bereasonablysurethattheyresultedinsomestudentsspendingmoretimestudying 92

PAGE 93

basicmathandreadingsincetheyhadtoretaketheexamsandsinceweknowthat schoolstypicallyenrollretakersinremedialmathandreadingclasses,yetthisimplies thatlesstimeisspentonothercoursesoractivities.Itisnotclearwhetherthisimproves earnings.Unfortunately,eventhelargesamplesthatweemployaretoosmalltodetect anysuchearningseffects.Moreover,itisnotclearthatamoredrastictougheningof standardswouldhavehadthesameeffect.Itmaybethatstandardscanrisewithout consequenceuntilatippingpointisreached,atwhichpointthehigherstandardcauses largenumbersofstudentstodropoutofschool.Again,giventhepolicybeingstudied andthedatatohand,wecannotshedlightonthispossibility. Othersmightdrawtheoppositeconclusion,thatweakerstandardsdonoharmand hencethattheseexamscouldprotablybeabolished.Again,suchaconclusioncannot reasonablybedrawnfromtheevidencepresentedhere.Forexample,theabolitionof theseexamscouldresultinalessintensefocusonmathandreadingskills,andthis mayharmstudents'labormarketprospects. Inourview,therearelikelydiminishingreturnstoadditionalevaluationofexitexam policies-whetherbasedoncross-statedifference-in-differencemethods,within-state regressiondiscontinuitymethodsorthewithin-statedifference-in-differencemethod usedhere.Theresearchbasedoneachoftheseapproachespointstobroadlythe sameconclusion,anditishardtoimaginefurtherevaluationscomingtoverydifferent conclusions. 18 Amorefruitfulapproachmaybetofocusonthemechanismsthatwill ultimatelydeterminetheeffectsofthesepolicies.Forexample,whilethetheorybehind exitexamsassumesthatdiplomascarryasignalingvaluee.g.,Costrell1994and 18 Wheretheresultsdifferacrossthesestudiese.g.,theeffectsonparticular subgroupsorinparticularschools,thesedifferencesmayreectspecicfeaturesof thesystemsbeingstudied.Sincethesesystemsdiffersomuchacrossstateshowever, onedoubtsthatthesedifferencescanbeusedtoidentifytheeffectsofspecicfeatures ofthesesystems. 93

PAGE 94

Betts1998,MartorellandClark2010ndthesignalingvalueofadiplomatobe small.Furtherresearchonsignalingcaninformthediscussionofexitexams.Research intostudentexpectationsandmotivationinthefaceofhigh-stakesexamscanalso informthediscussionandhelppolicy-makersmakemoreinformeddecisionsinrelation totheexistenceanddesignofexitexamsandrelatedpoliciesdesignedtoincentivize studentsinhighschool. 94

PAGE 95

Figure2-1.TheImpactofTougherStandards. Figure2-2.GroupDenitions 95

PAGE 96

Table2-1.Descriptivestatistics FullsampleGroup1Group2Group3 FailbothPassold,failnewPassboth 20002001200020012000200120002001 Male45.8245.6642.9044.1347.8146.4046.1846.02 White54.0252.9048.4845.9053.4152.2558.4358.02 Black22.2121.5927.5727.5822.3321.9718.3517.36 Hispanic19.5020.9719.9922.2419.9521.3018.7719.85 Freeorreducedpricelunch27.9126.4830.3930.0328.3826.9825.7623.73 Limitedenglishprocient17.2118.7418.6821.2517.8319.4615.6416.49 Specialeducation5.996.457.858.875.856.444.804.87 Gifted0.660.800.250.310.520.731.071.18 Grade10ontime77.4977.0874.1272.8977.6677.0079.7179.89 Observations31,75935,2598,7299,16110,60212,06312,42814,035 Note:Firsttwocolumnspresentstatisticsforstudentsinallthreegroups.Other columnsareforeachgroup. 96

PAGE 97

Table2-2.Theimpactsoftougherstandardsonhighschooloutcomes. Usingscores280-310Usingscores285-305Usingscores300-310 XschoolFEXschoolFEXschoolFE HighestgradecompletedFailboth0.0020.0030.005-0.009-0.0050.000 std.error0.0130.0120.0120.0180.0180.017 Passold,failnew-0.008-0.007-0.006-0.014-0.011-0.009 std.error0.0110.0110.0110.0130.0130.013 Score 305-0.027-0.010-0.006 std.error0.0130.0130.013 Last-chancesampleFailboth4.0433.9974.012.9822.9342.945 std.error0.3800.3790.3820.5080.5060.512 Passold,failnew4.2034.2084.2114.2034.2094.214 std.error0.1830.1830.1870.1830.1830.189 Score 305N/AN/AN/A HighschooldiplomaFailboth-1.209-1.138-1.224-0.613-0.286-0.082 std.error0.7900.7770.7731.1191.1001.095 Passold,failnew-2.101-2.066-2.093-1.658-1.539-1.408 std.error0.6900.6820.6780.8190.8080.806 Score 305-0.7630.2670.422 std.error0.7770.8140.812 CerticateofcompletionFailboth0.7850.7760.7680.2790.2720.262 std.error0.2020.2030.2040.2690.2690.272 Passold,failnew0.7940.7940.8040.7080.7070.742 std.error0.1070.1070.1090.1140.1140.115 Score 305-0.029-0.024-0.007 std.error0.0830.0910.092 Observations67,01867,01867,01845,53045,53045,53026,46326,46326,463 Note:Columns-reportdifference-in-differenceestimatesof theimpactoftougherstandardsi.e.,thepostperioddummyonthe groupsfailbothandpassold,failnew.Columns-usethe mainsamplediscussedinthetext.Columns-useasubset ofthissample.Columns-estimateplacebotestsusinga subsampleofthedataandahypotheticaltreatmentscorelessthan 305onbothtests. 97

PAGE 98

Table2-3.Theimpactsoftougherstandardsonpost-secondaryoutcomes. Usingscores280-310Usingscores285-305Usingscores300-310 XschoolFEXschoolFEXschoolFE Semestersin2-yrcollegeFailboth0.0160.0380.037-0.0030.0280.029 std.error0.0600.0590.0590.0840.0830.082 Passold,failnew0.0460.0450.0400.0350.0360.038 std.error0.0570.0560.0560.0680.0670.066 Score 305 -0.039-0.052-0.040 std.error 0.0610.0650.065 Semestersin4-yrcollegeFailboth0.1370.1410.1380.1080.1290.130 std.error0.0430.0420.0420.0570.0560.056 Passold,failnew0.0750.0710.0630.0350.0380.034 std.error0.0460.0450.0450.0540.0530.053 Score 305 -0.118-0.102-0.093 std.error 0.0480.0520.052 Earningsin1styrpostHSFailboth64.21084.10074.930-16.330-1.390-13.000 std.error109.420108.710108.820131.400130.560132.130 Passold,failnew-81.490-59.020-60.610-128.620-136.530-146.920 std.error109.940109.270108.980158.740157.830158.550 Score 305 -243.680-70.210-86.144 std.error 123.700128.890133.800 Earningswithin5yearsFailboth276.160408.470407.290-266.970-131.780-102.710 std.error662.300660.140659.360784.560781.430787.410 Passold,failnew-147.590-44.970-123.090-308.590-340.690-394.000 std.error683.750681.820680.560964.140961.080966.440 Score 305 -1459.160-138.300-236.780 std.error 753.480788.420820.030 Observations67,01867,01867,01845,53045,53045,53026,46326,46326,463 Note:seenotestoTable2-2. 98

PAGE 99

Table2-4.Falsicationtests. Usingscores280-310Usingscores285-305Usingscores300-310 XschoolFEXschoolFEXschoolFE MaleFailboth0.0140.029 std.error0.0100.013 Passold,failnew-0.013-0.015 std.error0.0090.011 Score 305-0.058 std.error0.010 EligibleforfreelunchFailboth1.6751.378 std.error0.8691.220 Passold,failnew0.6401.214 std.error0.7990.950 Score 305-1.671 std.error0.874 Ontimeingrade10Failboth-1.415-2.044 std.error0.8251.170 Passold,failnew-0.839-1.453 std.error0.7450.888 Score 305-1.098 std.error0.830 Observations67,01867,01867,01845,53045,53045,53026,46326,46326,463 Note:seenotestoTable2-2. 99

PAGE 100

CHAPTER3 EFFECTSOFCOMMUNITYCOLLEGEATTENDANCECOSTS 3.1Introduction ManyresearchstudieshaveshownthehighreturnstocollegeeducationKane andRouse,1999.Withtherisingpremium,manyhighschoolgraduatesaspireto acquirecollegeeducationinthehopeofexpandingtheirlabormarketopportunities.The estimatesofthereturnstocompletinganassociatesinartsdegreerangefrom15to 27percentofannualearnings,whileattendingcollegecouldraisetheannualearnings byabout10to12percentKaneandRouse,1999.Despitetheexistingpremiumto collegeeducation,notallhighschoolgraduatesgotocollege. Muchefforthasbeendevotedtoanalyzecollegeaccessaswellasitsdeterminants, mostespeciallyamongdisadvantagedstudents.Agreatdealofattentionhasfocused ontherapidlyrisingcostofcollegeeducationanditsimplicationonthecollege-going decisionsofstudents.Thishighcostofcollegeeducationpreventsalargefractionof highschoolgraduatesfromattendingcollege.Between1999-2000and2009-2010, costsoftuition,roomandboardrose37percentatpublicinstitutions,25percentat privateinstitutionsDigestofEducationStatisticsAnnualReport,NationalCenter forEducationStatistics,2010.Theimpactoftherisingcollegecostsismorelikely feltbycommunitycollegestudentswhoaresensitivetothechangesincollegecost KaneandRouse,1999.Between1999-2000and2009-2010,about66percentof thehighschoolgraduatesenrolledincollegeinthefallimmediatelyafterhighschool graduation.Amongthehighschoolgraduateswhochoosetogotocollege,only37 percentmatriculatedatcommunitycollegesConditionofEducationAnnual Report,NationalCenterforEducationStatistics,2011.Asthehighlabormarket premiumfavoringcollegeeducationpersistsandthatacollegedegreebecomesa prerequisitetojointheranksofthemiddleclass,thehighcostofcollegeeducation appearstobeoneofthebarrierstostudentsuccess.Tomakecollegeeducation 100

PAGE 101

affordable,somearguedtheneedformoreaggressivenancialsupportpoliciesto assiststudentsinmeetingtheirgoalsofacquiringhighereducation.Providinghigh qualityandaffordablecollegeeducationremainsapublicpolicyconcernPernaandLi, 2006. InthispaperIconsidertheeffectsofthecostofattendingacommunitycollege.I analyzetheeffectsoftheratioofthenetcosttothetotalcostofattendingthenearest communitycollegeoncommunitycollegeoutcomesofFloridastudents:semesters enrolledincollegeandassociatesinartsdegreecompletion. 1 Iexploitthenatureof thecostofattendingaFloridacommunitycollegeasanexogenoussourceofvariation incommunitycollegeeducation.ThisallowsmetoemployinstrumentalvariablesIV analysistogetmorecredibleestimatesoftheeffectsofcollegeeducationonearnings. Indthatincreasingthefractionofthecollegecostdecreasesthenumberof semestersstudentswereenrolledbyaboutathirdofasemesteranddecreasesthe associatesinartsdegreecompletionratebyabouttwopercentagepoints.These estimatesaremoderatelylargewhencomparedtotheaveragesemestersenrolledin communitycollegeandtheaverageAAdegreecompletionrates.Cuttingthesample intodifferentdemographicgroups,Indthatthefractionofthecollegecosthasalarger negativeeffectonboththenumberofsemestersenrolledincollegeandtheAAdegree completionrateinthesubsampleoffemales,Hispanics,studentswhoparticipatedin free/reducedlunchprogram,andthestudentswhoreached10thgradeontime.TheIV estimatesbasedonthefractionofcollegecostsindicatethatasemesterofenrolmentin communitycollegeisassociatedwithabouta20percentincreaseinannualearnings. TheIVestimateofthereturnstocompletingassociatesinartsdegreearequitelarge, withawiderangeofestimatefromaboutoneto650percentoftheaverageannual 1 Thenettototalcostratioreectsthefractionofthetotalcostofattendancethata familywouldhavetocontributetopayforthecollegeeducation. 101

PAGE 102

earnings.TheseIVestimatesareconcentratedamongwhitestudents,studentswhodid notparticipateinthefree/reducedlunchprogram,andstudentswhoreached10thgrade ontime.TheIVestimatesprovidesuggestiveevidenceofthelargereturnstocommunity collegeeducation. 3.2Background 3.2.1ReviewofRelatedLiterature 3.2.1.1Costofattendanceandcommunitycollegeeducation Severalinterestingquestionssurroundthecostofattendingacollege.First,how docostsaffectthedecisionstoattendacollegeandothercollege-goingbehavior?One strandofresearchexaminedtherelationshipbetweentuitioncostsandthecollege-going behaviorofstudents.LeslieandBrinkman1988,Rouse1994,andKane1995 studytherelationshipbetweentuitionandcollegeenrollment.Manski1989,St.John 1990,St.JohnandStarkey1995,Dynarski2008studytheeffectsoftuitioncostson collegecompletionandcollegepersistence.Thesecondstrandofresearchaskedhow theeffectsofcollegetuitioncostsvaryacrosssocio-demographicgroups.Kane1994, 1995,1999,EllwoodandKane2000,andCarneiroandHeckman2002examined thesensitivityofvariousincomegroupstotuitioncosts. Thestandingunresolvedquestioniswhetherandhowcollegecostsaffectthe collegeattendanceandtheothercollege-goingbehavioramonghighschoolgraduates e.g.persistence.Althoughtherehasbeenlittleempiricalworkontheeffectsofcollege costs,anumberofresearchpapershaveassessedtheeffectsoftuitioncostsand nancialaidonfour-yearcollegeenrollment.Intheircomprehensivereviewofthehigher educationcosts,LeslieandBrinkman1988haveestimatedthata$1,000increase intuitioncostsisassociatedwiththreetofourpercentagepointdecreaseincollege enrollments.St.John,1990studiedhowstudentsrespondtocollegeprices.Using thedatafromtheHighSchoolandBeyondClassof1982survey,St.Johnanalyzed 102

PAGE 103

theeffectsoftuitionandaidonstudentenrollment.St.Johnconcludedthatallformsof nancialaidgrants,work,andloanswereeffectiveinpromotingenrollment. Someskepticscallintoquestiontheresultsofthesestudies.Theestimatesfrom thesestudieswerecommonlygeneratedfromtimeseriesandcross-sectionalmethods, exploitingthevariationsintheaveragestatepublictuitionlevelstoidentifytheeffects oftuitioncostsonenrollment.Rouse1994,Kane1995,CameronandHeckman 1999,andLong2004pointedoutthatestimatesbasedonstatevariationsintuition costsaretobetreatedwithcautioniftuitioncostsarecorrelatedwiththeunobserved factorsrelatedtocollegeenrollment.Somestudiesexploitedthequasi-experimental natureofthedatatoexaminetheeffectsoftuitioncostsonenrollment.Dynarski 2003usedtheshiftinnancialpolicytheeliminationofSocialSecurityStudent BenetProgramin1982thataffectedonlysomestudentsasthesourceofthe exogenousvariationtoidentifytheeffectsofnancialaidoncollegeattendance.Using adifference-in-differencemethodology,shendsthattheeliminationoftheSocial Securitystudentbenetprogramreducedthelikelihoodofcollegeattendanceand collegecompletion.Dynarskisuggestedthata$1,000increaseinaidwouldincreasethe probabilityofcollegeattendanceby3.6percentagepoints. Lessisknownonhowcollegecostsaffectcollegecompletionandcollege persistence.Therelatedstudiesinthisareausedtheinformationonthetuitioncosts andnancialaidtoexplaincollegecompletionandpersistence.Manski1989argued thattheneteffectofreducingtuitioncostoncollegecompletionandpersistenceis ambiguous.IntheframeworkofManski,studentswhoareinducedtoattendcollege duetolowertuitioncostswillsoondiscoverwhetherornottheyarecollegematerial.By loweringtuitioncostsforthelesspreparedstudents,bothenrollmentandthedropout behaviorcouldincrease.Ontheotherhand,loweringtuitioncostscouldattractmore preparedbutcreditconstrainedstudentswhoaremorelikelytocompletecollege. Manskiconcludedthattheneteffectofreducingtuitioncostsisambiguous.St.Johnand 103

PAGE 104

Andrieu1995assessedtheinuenceofstudentaidonthewithin-yearpersistenceby traditionalcollegestudents.Usingthe1987NationalPost-secondaryStudentAidStudy, theyconcludedthattuitionchargeshadanegativeeffectonpersistence.DesJardins, AhlburgandMcCall2002,onasimilarnote,examinedtheeffectsofscholarships onretention.Theyshowedthatchangingloanstoscholarshiphasalargeeffecton persistence.Dynarski2008examinedthecausallinkbetweencollegecostsand degreecompletioninstateswithscholarshipprograms.Usingatreatment-comparison groupresearchdesign,Dynarskiconcludedthatcollegecompletionincreasedinstates withscholarshipprograms. Littleisknownonhowtheeffectsofthecollegecostsvaryacrosssocio-economic groups.Economistshavesuggestedthatthedifferentialeffectsoftuitioncostsarisedue toindividualcreditconstraints.Comparedtotheafuentindividualswhotypicallyhave higherlevelsofschooling,lowincomeindividualsattainloweryearsofpost-secondary schoolingduetotheirinabilitytoborrowagainstfutureincome.Ifcreditconstraints reducecollegeattendanceintheselowincomeindividuals,thentheseindividualsare expectedtobemoresensitivetotuitioncostsandnancialaid.Kane1994,1995, 1999andEllwoodandKane2000ndlow-incomeyouthtobemoresensitivetothe tuitioncosts.CameronandHeckman1998,1999andCarneiroandHeckman2002 showedthatconditionalonability,theeffectsoftuitionareuniformacrossfamilyincome categories.Theevidenceinthisareaofempiricalresearchismixed. Whilethereareempiricalresearchpapersthatexaminedthelinkbetweencollege tuitioncosts,attendance,andcompletion,theseresearchpapersareconductedwith thefour-yearcollegestudentsinmind.Thereisalittleconsensuswhetherandhow thedirectandindirectcollegecostsaffectattendanceanddegreecompletionamong two-yearcollegestudents.Communitycollegesoftenservestudentsonthemargin ofattendingcollege,whichincludeimmigrantsandtheeconomicallydisadvantaged. Thesemarginalstudentsaretheoneswhoaresensitivetothechangesincollege 104

PAGE 105

costs.Kane1995andMarcotteetal.2005arguedthatcommunitycollegestudents aretypicallymoresensitivetothecollegecoststhanthetraditionalfour-yearcollege studentsbecausetheformertendtoseekcredentialsandtypicallycombineschooland work.Sincecommunitycollegedegreeismostofthetimetheterminaldegreeamong thesetwo-yearcollegestudents,collegecompletionisparticularlyimportantbecause notallstudentswhoattendtwo-yearcollegesplantoenrollinfour-yearcollegesRouse, 1995. Thispapercontributestotheliteratureontheeffectsofcollegecostsoncommunity collegeeducationoutcomescommunitycollegeenrollmentandassociatesinarts degreecompletion.Thispapermakesuseofthebetweencommunitycollegecost variationsamongthe28Floridacommunitycollegesinidentifyingtheeffectsofthe collegecostsonthenumberofsemestersenrolledincollegeandontheassociatesin artsdegreecompletion.Thispaperisthersttouseameasureofthefractionofthe collegecosts,the fractionofcostofattendingcommunitycollege .Also,Iprovideseveral testsandrobustnesscheckstowarrantthecredibilityoftheestimatesbasedonthe fractionofthecollegecosts,whicharenotcommonlyseeninmanyempiricalresearch papersinthiseld. 3.2.1.2Returnstocommunitycollegeeducation Findingtheextenttowhichcollegeeducationimproveslabormarketsuccessof studentshadbeenthecentralresearchquestionofmanystudiesinthepast.Although thereareempiricalresearchpapersthatdocumentedtheeffectsofcollegeeducation onthelabormarketsuccessandearnings,onlyafewexaminedtheimpactofthe communitycollegeeducation.Becauseoftheavailabilityofthedata,themajorityofthe empiricalresearchpapershadfocusedonthereturnstothefour-yeareducationCard, 2001. Understandingtheeconomicreturnstocommunitycollegeeducationisimportant atleastfor3reasons.First,communitycollegesprovideopportunitiestodisadvantaged 105

PAGE 106

studentswhoaretypicallyleftbehindduetopooracademicpreparationCohenand Brawer,2003.Second,communitycollegesplayamajorroleintrainingtoenhance theskillsofthoseindividualswhoarealreadyintheworkforceMarcotteetal.,2005. Third,sincecommunitycollegesenrollalargeshareofcollegestudentsandare supportedbythefederalandstategovernments,theseinstitutionsareamenableto policyinterventionsRouse,1994. Despitethebenetsfromthecommunitycollegeeducation,thevalueofthe communitycollegeeducationinthelabormarketremainedunderstudied.Theearly researchpapersshowedthatcommunitycollegeeducationhavenoorlittleeffecton earnings.Dougherty1987madeacomprehensivestudyoftheresearchpapersonthe returnstocommunitycollegeeducation.Basedonthesynthesizedndings,Dougherty concludedthatbaccalaureatedegreeseekingstudentswhoenteredcommunitycolleges attainlesseducationallyandeconomically.The1990sshowsomepositivestrideswith moreempiricalresearchpapersdocumentedthepositivereturnstothecommunity collegeeducation.UsingthedatafromtheNationalLongitudinalSurveyClassof 1972NLS-72,Grubb1993comparedtheearningsofthestudentswhocompleted associatesinartsdegreeandthestudentswhonishedhighschoolonly.Grubb ndsthatthestudentswhocompletedassociatesinartsdegreeearnedmore.Ona similarnote,Grubb1997,2002ndsapositiveearningseffectofsub-baccalaureate credentials.UsingthedatafromtheSurveyofIncomeandProgramParticipation,Grubb showedthatthestudentswhoearnedcollegecredentialshavehigherearningsthan thosestudentswhofailedtoearncredentials.Intheirinuentialpaper,KaneandRouse 1995usedtheNationalLongitudinalSurveyClassof1972and79NLS-72and79 andtheNationalLongitudinalSurveyofLaborMarketExperienceofYouthNLSY-79 tocomparethereturnstothetwo-yearandfour-yearcollegedegrees.Onenotable conclusionoftheirpapersuggeststhattheearningsofthecommunitycollegedegree 106

PAGE 107

holdersarehigherbyroughly15to25percentcomparedtothestudentswhocompleted highschoolonly. Theimportanceofunderstandingtheeffectsofcollegepersistenceonearnings cannotbediscountedfortworeasons.First,theadditionalyearsofschoolingalways have,ifnotlarge,somepositiveeffectsonearnings.Second,somestudentswho enteredcommunitycollegedropoutearlyandfailtonishadegree,probablyduetothe costsassociatedwithcollegeeducation,skillsandacademicpreparation,andtosome psychologicalreasons.Findingsthatpointtopositivereturnstocollegepersistencecall forpoliciesthatmakestudentsstayincollege.KaneandRouse1995andLeighand Gill1997investigatedhowcourseworknotleadingtoadegreeaffectsearnings.These researchstudiesndthatayearofcourseworkincreasesearningsbyaboutvetoeight percent.Similarly,Jacobson,LaLondeandSullivan2005usedtheadministrativedata ofdisplacedworkersinWashingtonStateandndthatanacademicyearequivalentof creditsearnedincreasesearningsbyninepercentformenand13percentforwomen. Usingthe2000Follow-UpoftheNationalEducationLongitudinalSurvey,Marcotteetal. 2005ndpositiveearningseffectsamongstudentswhocompletedandwhofailedto earncredentials. Theseearlystudiesthatattemptedtomeasurethereturnstocommunitycollege educationhaveseveralissues.First,thedatasetsusedintheseearlyresearchpapers aremostlybasedonasurveydata.Surveydataonearningsandotherinformationare commonlymismeasured.Asaresult,estimatesofthereturnstocommunitycollege educationbasedonsuchdatasetarepotentiallybiased.Moreover,theresultsgenerated fromsomeoutdateddatasetsprovidelittlevaluetothecurrentpolicymakerssincethe structureofthecommunitycollegesandthelabormarkethavechangeddramatically since1980Marcotteetal.,2005.Theearlyresearchpapersalsohavesuffered frommethodologicalissuesthatcastdoubtsonthecredibilityoftheirestimates.The estimatesfromthesepapersarepotentiallybiasedsincethepredominantestimation 107

PAGE 108

methodusedismostlytheOLSregressionanalysiswithonlythefamilybackground variablesandtestscoresincludedtocontrolatleasttheabilitybias.Aconvincing estimateofthereturnstocommunitycollegeeducationneedsanexogenousvariationin thecommunitycollegeeducationoutcomes.Tocontrolforthebiasfromunobserved factorssuchasability,theexogenousvariationintheeducationoutcomescan beexploitedunderquasi-experimentalresearchdesigns.KaneandRouse1995 attemptedtousetheinstitutionalfeaturesofthecommunitycollegesintheiranalysisas thesourceoftheexogenousvariationinthecommunitycollegeeducation.Theyused tuitioncosts,transcriptofrecords,andthedistancetothenearestcommunitycollegeas theinstrumentsforcollege-goingbehavioranddegreecompletion.AlthoughKaneand Rousewereabletoproducemorecredibleestimatesofthereturnstocommunity collegeeducation,theywerenotabletoestimateaneffectduetothelackofpowerin theirtests. Thispaperattemptstouseaninstrumentalvariablesframeworkintheanalysis ofthereturnstocommunitycollegeeducation.Iofferseveralimprovements.First,I userichadministrativedatathatincludecollegecostsandnancialaidinformation fromFlorida.Theadministrativedatahasinformationonearnings,collegeenrollment, collegedegreecompletion,anddemographicvariables.Thisadministrativedata alsoincludeimportantpre-collegeinformationsuchastestscoresandinformation whetherastudentisontracktonishhighschool.Thesepre-collegeinformationare valuablesincetheyrepresentskills,whichmaycontrolforunobservedability.The useofadministrativedataisalsolesspronetothemeasurementerror,especiallythe earningsinformation.Second,similartotheapproachofKaneandRouse1995, Iwillmakeuseofaninstitutionalfeatureofthecommunitycollegesthecostof attendingthenearestcommunitycollegeasthesourceofexogenousvariationinthe communitycollegeeducation.Inparticular,Iwillusethenettototalcollegecostratio ofattendingcommunitycollege,whichmeasuresthefractionofthecollegecosts,as 108

PAGE 109

aninstrumentalvariable.Third,Iwillshowhowtheidenticationconditionsarisingfrom themodelundertheinstrumentalvariablesframeworkcanbetested.Inthisway,Ican establishthatmyestimatesaremorecredibleandthattheinstrumentalvariableIuseis legitimate. 3.2.2TheCostofAttendingaFloridaCommunityCollege Acredibleestimateofthereturnstocommunitycollegeeducationrequiresan exogenoussourceofvariationincommunitycollegeeducation.Onepotentialsourceof thisexogenousvariationisthedifferencesinthecostofattendingcommunitycolleges inFlorida.Ahighcostofcollegeattendanceisexpectedtodiscourageafractionof studentsfromattendingcollege.InthissectionIdescribethenatureandthedifferences inthecostofattendingcommunitycollegesinFlorida. Prospectivestudentsareexpectedtopayfortheeducationalandlivingexpenses whileattendinganyofthe28Floridacommunitycolleges.Typically,astudent'scollege expendituresincludeeducationalandlivingexpenses.Educationalexpensesconsist oftuitionandfeesandbookexpenses,whilelivingexpensesincluderoomandboard, transportation,andotherpersonaleffects.Intheschoolyear2001-2002,studentson averagespentmostonroomandboard,whichaccountsforabout56percentofthecost ofattendingcommunitycollegeinFlorida.Inthatsameschoolyear,about18percentof thecostofattendanceisattributedtotuitionandfees,13percenttothetransportation expense,and12percenttopersonaleffectsReportNo.03-33OfceoftheProgram PolicyAnalysisandGovernmentAccountability,2003. Overtheyears,tuitionandfeesinthecommunitycollegesaresetbytheFlorida legislature.TheStateBoardofEducationSBEinFloridaprescribesascheduleofthe tuitionandfeesthateachcommunitycollegecanchargetotheircurrentandincoming students.Thelegislaturealsogiveseachcommunitycollegesomeexibilitytocharge additionalfees.Ontopoftheprescribedtuitionandfees,thelegislatureallowsthe BoardofTrusteesofeachcommunitycollegetochargeadditionalfeeswiththeprovision 109

PAGE 110

thatthesefeesshouldnotexceedvepercentoftheprescribedtuitionandfeesand thatthesumoftheseadditionalfeesshouldbeexpendedonlyforthelocalsafetyand security.Thetuitionandfeesforagivenschoolyearmaybethoughtofasanoutgrowth ofthehistoricalcostsofoperationandtheyeartoyearexibilityofeachcommunity collegetoaddsomeadditionalfees. Eachcommunitycollegeprovidesestimatesofthelivingcostsineveryschool year.Variouscostestimatesforseveralhousingarrangementsaretypicallypublished ineachcommunitycollegewebpage.Theestimatedcostoftheon-campushousing, off-campushousingnotwithfamily,andoff-campushousingwithfamilyareamongthe gurespublishedinthecolleges'webpages.Transportationcostsaretypicallybased onthestudents'dailycommutingdistancetoandfromthecampus.Sinceeachcollege geographicsituationisunique,theestimatesoftransportationcostsvarywidelyacross thecommunitycolleges.Whilesomecommunitycollegeshaveconservativeestimates ofthetransportationcosts,somecommunitycollegeschoosetoincludeparkingfees, gas,automobileinsurance,andmaintenanceintheirestimatesofthetransportation costs.ThecommunitycollegesinFloridadonotfollowaspecicformulaincalculating thelivingexpensesforanaveragestudent.Instead,thesecommunitycollegesfollow theirownguidelinesinestimatingthiscostitemresultingtolargedifferencesinthe estimatesoflivingcost.Sincetheseestimatescountintothecomputationofnancialaid awards,thelargedifferencesinthelivingcostestimatesbecomesignicantlyimportant tocurrentandprospectivestudentsReportNo.03-33OfceoftheProgramPolicy AnalysisandGovernmentAccountability,2003. Table3-1presentstheestimatesoftheeducationalandlivingcostsamong Floridacommunitycollegesfrom2003to2006.Thecostestimatesvariedwidely. Forinstance,thelargestdifferenceinthecostestimatescanbeseeninthehousing andtransportationcosts,withtheestimatesrangingfrom$3,629atSouthFlorida CommunityCollegeto$13,591atFloridaKeysCommunityCollegeduringtheschool 110

PAGE 111

year2003.Inthatsameschoolyear,thereisasmallvariationintheestimatedtuition andfeesforregistering30credithours,withthesmallestandthehighestcostestimates ofaround$1,320TallahasseeCommunityCollegeand$1,950PalmBeachState CommunityCollege. ThestateofFloridaachievedbigstridesinmakingcollegestudentsbenetfromthe state'scollegeeducationsystem.Thestateprovidesgenerousnancialaidtodefray thecostsoftuitionandfeesforin-statestudents.TheresidenttuitioninFloridaisset ataround25percentoftheinstructionalcostsinagivenschoolyear.Thisineffect makeseveryresidentstudenttoreceiveatuitionwaiverofalmost75percentofthe publishedtuitionandfees.Tocovertherestofthecost,avarietyofneedandmerit basednancialsupportsintheformofgrants,scholarships,andwork-studyprograms weremadeavailabletothecurrentandprospectivecollegestudents.Tonameafew, popularprogramssuchasBrightFuturesScholarship,federalPellgrants,andother institutionalstudentaidsprovidenancialsupporttoqualiedstudentsthroughouttheir collegeyears.Forthesequaliedstudents,attendingcollegeinFloridaisrelatively inexpensive.Studentswhodonotqualifyforthesegrantsandscholarshipprograms oftenapplyforstudentloans. Ahighschoolstudentwhowishestogetanynancialsupportforcollegeeducation isadvisedtogetestimatesoftheeligibilitytoreceivefederalgrantsearlyinhissenior year.Onthatsameyear,thisseniorhighschoolstudentshopsforcollegesandapplies forcollegeadmission.Bythespringofthestudent'ssenioryear,heneedstoaccomplish theFreeApplicationforFederalStudentAidFAFSAifhewishestoreceivenancial support.TheFAFSAassessmentwillprovidetheStudentAidReportSARwhich includestheamountoftheExpectedFamilyContributionEFCbasedonthestudent's familyresources.TheEFCindicatestheamountofthestudent'sfamilyresources thatareavailabletohelppayforthecostofcollegeeducation.Thenalstepinthe applicationforthenancialsupportrequiresthestudenttocompletetheStateofFlorida 111

PAGE 112

FinancialAidApplicationpriortohighschoolgraduation.Bythetimethestudent registersforcollegeclasses,heisawardedthenancialsupport. Eligiblestudentswhoseekadmissionstodifferentcommunitycollegesdonot necessarilygetthesameamountofnancialaid.Inthebasicformula,theamountof nancialsupportthataneligiblestudentwillreceiveisequaltothedifferencebetween theestimatedtotalcostofattendanceandtheExpectedFamilyContributionEFC. IftheestimatedtotalcostexceedstheEFC,thedifferenceistheamountofnancial supportthatisawardedtothestudent.Thelargevariationsinthenancialaidawards amongcommunitycollegesareprimarilyduetothewidevariationintheestimated costofattendanceineachcommunitycollegeandthevariationintheexpectedfamily contribution.Table3-2showstheestimatedcostofattendanceandtheamountsofthe totalnancialaidinthe28Floridacommunitycolleges. 3.3Data 3.3.1DataConstruction Toexaminetheeffectsofthecostofattendingcommunitycollegeoncollege-going behaviorandlabormarketreturns,Iuserichadministrativestudent-leveldata.Several dataleswereprovidedbyFloridaEducationandTrainingPlacementInformation ProgramFETPIP,aunitwithinFloridaDepartmentofEducation.Thedataleswere mergedusingauniquestudentidentier.TheFETPIPdataisthenmergedwiththe communitycollegeleveldataoncostsobtainedfromtheIntegratedPostSecondary EducationDataSystemIPEDS.Thecombineddataincludeinformationonstudent enrollmentinFloridapublicschools,demographiccharacteristicssex,race,yearand monthofbirth,gradetenFCATscoresmathematicsandreading,post-secondary awardscommunitycollegeanduniversityawards,andearnings.Theearningsdata comefromtheUnemploymentInsuranceUItaxreportssubmittedbytheemployers coveredbytheFlorida'sUnemploymentInsurancelawtotheFloridaDepartmentof Revenue.Theenrollmentrecordsarebasedontheacademicyearinwhichastudent 112

PAGE 113

wasrstenrolledingradeteninaFloridapublicschool,includingcharterschoolsbut notprivateschools.Withthis,eachstudentinthedatahasauniqueenrollmentrecord whichisusedtodenethecohortafliationofeachstudent.Iusefourstudentcohorts thatincludestudentsingradeteninschoolyears2000-2001,2001-2002,2002-2003, and2003-2004.Thebasesamplehasabout712,142studentobservations.Table3-3 showsthedistributionofobservationsbycohort. ThedataobtainedfromtheIPEDSincludeinformationontheestimatedrelevant costsofattendingcommunitycollegesinFloridaaswellasrecordsoftheaverage nancialsupportthatthestudentsreceivedinaschoolyear.Thiscostdatainclude directinstructionalcostssuchastuitionandfees,bookexpenses,andlivingcosts includingonandoffcampushousing,transportation,andotherpersonalexpenses. Summingupthesecostsyieldstheestimatedcostofattendanceforeachofthe28 Floridacommunitycolleges.Thenancialaidrecordscontainaverageamountsofthe federalaid,institutionalaid,andstudentaid.Thesumyieldstheaveragetotalnancial aidthatstudentsineachcommunitycollegereceive.Forthetotalcostofattendance andthetotalnancialaiddata,therelevantschoolyearsincludedare2003-2004, 2004-2005,2005-2006,and2006-2007.Theuseoftheserelevantyearsisbasedon theassumptionthathighschoolstudentsonaveragestarttoshopforcollegestwo yearsaftertheytookthegradetenFCAT.Forexample,gradetenstudentsincohort oneduringtheschoolyear2000-2001areexpectedtonishgrade12highschool inschoolyear2002-2003andwillshopforcollegestoattendforthecomingschool year2003-2004.Thus,studentsincohortoneonaveragewillusetherelevantcostof attendanceinformationavailableintheschoolyear2003-2004.Also,gradetenstudents incohortfour,theyoungestamongthecohortgroups,whotookthegradetenFCAT duringtheschoolyear2003-2004willusetherelevantcostofattendanceinformation availableintheschoolyear2006-2007. 113

PAGE 114

Igeneratethecostofattendingthenearestcommunitycollegefromthelocation ofthehighschoolattendedbyeachstudentintheFloridadata. 2 Theideaisthat thestudentswilllikelyattendthenearestandthecheapestcollegefromtheirhigh school.Sincethereisonlyonecommunitycollegepresentinastudent'scountyof residencewherethehighschoolislocated,astudentwouldhighlyconsiderattending thiscommunitycollegeandwouldfacethiscollege'spublishedtotalcostofattendance intheeventhedecidestomatriculateinthiscollege.Thestudentenrollmentrecords containinformationonthenameandthephysicallocationofthehighschoolinstitution thestudentsattendedwhileingradeten.Theinformationonthephysicallocation madeitpossibletocollectthedataonthecoordinateslatitudeandlongitudeofeach highschoollocationfromthepubliclyavailableschoolinformationdatabasefoundin theFloridaDepartmentofEducation.Theotherrequirementinndingthenearest communitycollegeisthephysicallocationandthecoordinatesofallthe28Florida communitycolleges.FloridaDepartmentofEducationpublishesinitswebsitethe physicaladdressesofallthecommunitycollegecampusesinthestateofFlorida. Usingeachhighschoolasthebaselocation,Indthenearestcommunitycollegeby calculatinggeodeticdistancesinmiles. 3 Afterndingthenearestcommunitycollege,I mappedthecommunitycollegelevelcostofattendanceandnancialaiddatatoeach studentintheFloridadatausingtheidentiednearestcommunitycollegeandcohort ID. 4 2 StudentswhoattendedprivatehighschoolsinFloridaareexcludedinthesample. 3 Geodeticdistanceisthelengthoftheshortestcurvebetweentwopointsalongthe surfaceofthemathematicalmodeloftheearth.Thedrivingdistancesofeachhigh schoollocationtothenearestcommunitycollegeareapproximatedbythecalculated geodeticdistance.Thisassumesthattheearthisatwithnorivers,swamps,mountains andthattravellingfrompointAtopointBcanbeachievedinalinearfashion. 4 ThecohortIDisbasedonthecohortgroupyearafliationofthestudent. 114

PAGE 115

Imakeanumberofsamplerestrictions.First,Iexcludestudentswhodidnottake theFCATwhenitwasinitiallyofferedandthosestudentswhonevertooktheFCAT test.Second,Iexcludethosestudentswhoattendedhighschoolprogramsthathave missingphysicaladdressesintheFloridaDepartmentofEducationSchoolInformation database.Thenalsampleincludes584,300studentobservations.Summarystatistics arereportedinTable3-3.About48%ofthesamplearemales.Inthesample,54 percentarewhites,21percentareblacks,20percentareHispanics.Theaverageage ofthestudentsinthesampleis16.About77percentofthestudentsinthesample reachedgradetenontime.Thereareonlyaboutaquarterofstudentsinthesample whoparticipatedinthefree/reducedlunchprogram. 3.3.2TheFractionofCostofAttendingtheNearestCommunityCollegeand AnnualEarnings Iusethesampledescribedabovetodenethefollowingvariables: TheFractionoftheCostofAttendingCommunityCollege .Iconsiderusing ameasureofthecostofcollegeattendancebasedonthetotalcostratherthanusing thetuitionandfeesgurescommonlyusedinmanystudies.Collegeboundstudents willlikelycomparethetotalcostofattendancewiththepotentialreturnsofacquiring collegeeducation.Theywilllikelyviewthecostofattendanceasaninvestmentthatwill generatereturnsaftertheyhavecompletedorreceivedacollegedegree. ThetotalcostofattendingthenearestcommunityCollegeisoneofthetwo componentsneededtocalculatethefractionofthecostofattendingthenearest communitycollege.Incalculatingtheestimatedcostofattendingcommunitycollege,I madesomecomputationalassumptionsonsomeofthecollegecostcomponents.First, Iusedthepublishedin-stateguresfortuitionandfeessincemostofthestudentsare Floridaresidents.Second,Iusedtheaverageoftheon-campusandtheoff-campus roomandboardcostsforasinglestudentnotlivingwiththefamilyincalculatingthe housingcostcomponent.Third,theguresIusedtocalculateotherlivingcostsarethe 115

PAGE 116

averageoftheon-campusandtheoff-campusotherlivingexpensesforasinglestudent notlivingwiththefamily.Thetotalofthesecostcomponentstuitionandfees,book expenses,roomandboard,transportation,andotherlivingcostsconstitutethetotal costofattendingFloridacommunitycolleges.Ontheaverage,studentspay$12,985.86 s.d.$3,596.6inannualtotalcostoflivingandattendingacommunitycollegeinFlorida intheperiod2003to2006Table3-3. Thenetcostofattendingacollegeiscalculatedbysubtractingeachcollege's averagetotalnancialaidfromtheestimatedtotalcostofattendance. 5 Theaverage nancialaidisbasicallythesumofthefederalaid,institutionalaid,andstudentaid.The averagenetcostofattendanceisaround$7,680.35s.d.$3,840.9intheperiod2003to 2006Table3-3. Finally,Icalculatethefractionofthecostofattendingcommunitycollegeasthe ratiooftheestimatednetcosti.e.,totalcostlessnancialaidandtheestimatedtotal costofattendingcommunitycollege.Afractionofcollegecostequaltoonemeansthat astudentwouldhavetoshoulderallofthetotalcostofattendingcommunitycollege whileafractionofcollegecostequaltozeromeansthatthestudentwillshouldernone ofthetotalcost. Usingthefractionofthecollegecosthasanumberofadvantages.First,unlike thetotalcostofattendance,thefractionofthecollegecostwillreectthefractionof thetotalcostthatstudentswouldhavetopayforthecollegeeducation.Second,unlike thenetcostofattendance,thefractionofthecollegecostisaunitfreemeasure.This isespeciallyimportantwhenIcomparetheeffectsofthecostsforstudentswhoface similarnetcostsbutdifferenttotalcostsofattendance.Table3-3showsthatstudents 5 TheIPEDScostdataalsohaveinformationontheaveragenancialaidofferedin eachofthe28Floridacommunitycolleges. 116

PAGE 117

onaverageshoulderaround56percentofthetotalcostofcollegeattendance.Also,the averagefractionofthecollegecostappearstobestableacrosscohorts. NumberofSemestersEnrolledinCommunityCollege .Thereareatleasttwo reasonswhythefractionofthecostofattendingcommunitycollegewouldhavean impactonastudent'scollege-goingdecisionandpersistenceincollege.First,ahigher fractionofthecollegecostwilldiscouragestudents,mostespeciallythosewhoare nanciallychallengedbythehighcostofpursuingacollegedegree.Second,those whoarealreadyincollegemightconsiderdroppingoutearlyduetothehighcost ofcompletingadegree.Tomeasuretheeffectofthefractionofthecollegecoston theattendanceandcollegepersistenceamongcommunitycollegestudents,Iuse thenumberofsemestersenrolledincommunitycollege.Thenumberofsemesters enrolledisderivedfromthepostsecondaryenrollmentrecordsintheFloridadata.I denethisvariableasthenumberofsemestersthestudentsregisteredforclassesin Floridacommunitycolleges.Iexpectthatahigherfractionofthecollegecostwillcause somecostsensitivestudentstodropout,decreasingthenumberofsemestersthat thesestudentswereenrolled.Table3-4showsthecomparisonofthemeansofthis variableacrossgender,race,andotherdemographicgroups.Table3-4revealsthat ontheaveragefemalestudents,whitestudents,studentswhodidnotparticipateinthe free/reducedlunchprogram,studentswithoutlimitedEnglishprociency,non-special students,andstudentswhoreachedgradetenontimeenrolledmoresemesters whileattendingacommunitycollegeinFlorida.Thispatternisstableacrosscohorts. Moreover,theoldercohortsi.e.,2000cohortregisteredmoresemestersincollege. AssociatesinArtsDegreeCompletion .Tomeasuretheimpactofthefraction ofthecollegecostsonthecommunitycollegedegreecompletion,Iconstructthe associatesinartsdegreevariableusingthepost-secondaryawardsdataintheFlorida data.Thepost-secondaryawardsdatahaveinformationonthedegreesawardedtothe student,theawarddate,andtheawardinginstitution.Thestudentisidentiedtohave 117

PAGE 118

everobtainedanAAdegreeifthestudentisawardedeitheranassociatesinartsor anassociatesinsciencedegree.Iexpectthatahigherfractionofthecollegecostwill decreasethestudent'slikelihoodofreceivinganAAdegree.Table3-5showstheAA degreecompletionratesbygender,majorracialgroups,andotherdemographicgroups. Ontheaverage,femalestudents,whitestudents,studentswhodidnotparticipateinthe free/reducedlunchprogram,studentswithoutlimitedEnglishprociency,non-special students,andthestudentswhoreachedgradetenontimehavehigherassociatesin artsdegreecompletionrates.Thepatternisoncemorestableacrosscohortswiththe oldestcohorti.e.,2000cohorthavethehighestAAdegreecompletionrates. AverageAnnualTotalEarnings .Measuringthereturnstocollegeeducation needsagoodmeasureofthelabormarketoutcomes.Moststudieshaveusedhourly wagestomeasuretheeffectofanextrayearofschoolingonwages.Inthispaper, however,Iusetheaverageannualtotalearningstomeasurethereturnstocommunity collegeeducation.Thisisnotamajordisadvantage.Firmsmaybehiringbasedon education-basedsignalse.g.,AAdegreewheretheeffectsmanifestinthetotal,butnot hourly,earningsMartorellandClark,2010.Inconstructingtheaverageannualtotal earnings,Irstaggregatethequarterlyearningsforeachyearreportedfornearlyall individualsworkingforearningsinFlorida.Theaggregatedearningsforeachindividual isdeatedby2000$basedontheCPI-Useriestogettheearningsinconstantdollars. Togettheaverageannualtotalearnings,Iaveraged5yearsofearningsdata,startingin the2ndyearafterthestudentgraduatedfromhighschool.Onaverage,astudentwho nisheswithanAAdegreewouldhavespentatleast2yearsinacommunitycollege andcouldhavestartedworkingatleastinthe2ndyearafterhighschoolgraduation. Thelackofyearsusedtoevaluatetheeffectsofcommunitycollegeeducationon earningswouldposesomelimitations.Although5yearsafterreceivingadegree wouldprobablybenotenoughtomeasurelongtermeffectsofthecommunitycollege educationonearnings,theAAdegreeholdersorthosewhostayedlongerincollege 118

PAGE 119

wouldhavesomehowexperiencedtheeffectsofthecommunitycollegeeducation. Table3-6showsthecomparisonoftheaverageannualtotalearningsamongdifferent demographicgroupsacrossallcohorts.Malesonaverageearnslightlyhigherthan females.Also,whitestudents,studentswhodidnotparticipateinthefree/reduced lunchprogram,studentswithoutlimitedEnglishprociency,non-specialstudents,and studentswhoreachedgradetenontimeareonaveragehavehigheraverageannual totalearnings. 3.4EmpiricalFramework Thereisaconsensusthatbettereducatedworkersearnhigherearnings.Many researchstudieshaveestimatedthegapinearningsbetweenmoreandlesseducated individuals.Someofthesestudiesmadesignicantprogressinprovidingcredible estimatesofthereturnstoeducation.Acredibleestimateofthereturnstoeducation requiresanexogenoussourceofvariationintheschoolingoutcomessinceitisa wellknownfactthateducationlevelsarenotrandomlyassignedinthepopulation. Collegechoiceisnoexceptionwhereindividualsmakecollege-goingdecisions. Instudiesinvolvingcollegechoiceanditseffectonearnings,researchersmade signicantprogressbyusingthefeaturesoftheeducationalinstitutionsasthesource oftheexogenousvariationCard,2001.Inthisstudy,Iproposethatthevariationin thefractionofthecostofattendingcommunitycollegeisapotentialsourceofthe exogenousvariationincommunitycollegeeducation.Ifindeedthefractionofthe collegecostisanexogenousdeterminantofthecommunitycollegeeducation,then wecancanuseittogeneratemorecredibleestimatesofthereturnstocommunity collegeeducation.Inthefollowingsection,Idiscussthebasicmodelofcollegechoice andearningsaswellastheeconometricissuesinidentifyingthecausaleffectofthe communitycollegechoiceonearnings. 119

PAGE 120

3.4.1ModelofSchoolingChoiceandEarnings ThefollowingmodelisinspiredbyCard2001.Therepresentativeindividual maximizeslifetimeutilityoverconsumptionandschoolingundertheconstraintthathis lifetimeconsumptionequalsearningswhileinschoolandearningsafterschool.This representativeindividualwillchoosetogetmoreschooling S untilthemarginalbenetof anextraunitofschoolingequalsthemarginalcostofthatlastunitofschooling.Therst orderconditionexpressesthisrelationas MB S = MC S .Letthemarginalbenetand marginalcostofschoolingfunctionstakethefollowinglinearform: MB S = b i )]TJ/F39 11.9552 Tf 11.955 0 Td [(k 1 S i MC S = c i + k 2 S i Theterms b i and c i areindividualspeciccomponentsofthemarginalbenetandthe marginalcostfunctionswithmean b and c ,respectively.Thepresenceofsubscriptin theparameter b i allowsforheterogeneityinthemarginalbenetofschoolingacross individuals.Setting MB S = MC S yieldstheoptimalschoolingchoice: S i = b i )]TJ/F39 11.9552 Tf 11.955 0 Td [(c i = k where k = k 1 + k 2 Giventhelevelofschooling,thisindividualearnsaowofearningsduringandafter attendingaschool.Letthestructuralearningsequationthatrepresentstherelation betweenanindividual'schoiceofschoolingandearningstaketheform W i = i + b i S i + u i Withthedataonearningsandthelevelofschooling,theordinaryleastsquaresOLS proceduremaygiveaninconsistentestimateoftheindividualspecicreturnsto schooling b i duetothepresenceofunobservedfactorsin u i e.g.,abilitythatwill potentiallyaffectboththelevelofearningsandtheschoolingchoice.Aconsistent estimateof b i canbeobtainedifwecanndsomeobservablefactorsthataffect 120

PAGE 121

schoolingchoicebutnotearnings.Forexample,ifschoolingwererandomlyassigned amongindividualsthenwecanexploittherandomizationprocesstoestimate b i using theinstrumentalvariablesIVapproach.Ifpurerandomizationisabsent,wecanstill consistentlyestimate b i onlyifwecanndacausaldeterminantofschoolingthatcanbe excludedinthestructuralearningsequationinEquation3Card,1995.Manystudies inthisareahaveusedtheinstitutionalfeaturesofschoolsandcollegestoidentify thefactorsthatdriveschoolingdecisions.Inthispaper,Iwillexploitaninstitutional featureofthecommunitycollegesinFloridaasthesourceofexogenousvariationin communitycollegeeducation.Inthenextsection,Idiscusstheidenticationconditions tobesatisedattheminimumtoobtainmorecredibleestimatesofthereturnstothe communitycollegeeducation. 3.4.2IdenticationConditions Letthemarginalcostofschoolingcomponent c i inEquation3takestheform c i = Z i + i where Z i representsobservablefactorsthataffectthemarginalcostofschooling. Thevectoroftheobservablefactors Z i containstheinstrumentalvariable.Inserting Equation3intotheoptimalschoolingchoiceequationinEquation3andaddingand subtracting b ,wegetthereducedformoftheschoolingchoiceequation S i = Z i + i where Z i = b )]TJ/F39 11.9552 Tf 12.692 0 Td [(Z i and i = )]TJ/F39 11.9552 Tf 5.48 -9.684 Td [(b i )]TJETq1 0 0 1 271.585 215.326 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 271.585 205.35 Td [(b )]TJ/F25 11.9552 Tf 11.955 0 Td [( i = k Thegoalistoestimatetheaverage returnstoschooling b .Wecanestimatethisparameterusingthestructuralearnings equationwithsomeidenticationconditions.Addingandsubtracting b ,wecanrewrite theearningsequationinEquation3as W i = a 0 + a i + bS i + )]TJ/F39 11.9552 Tf 5.48 -9.683 Td [(b i )]TJETq1 0 0 1 304.98 107.74 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 304.98 97.764 Td [(b S i + u i with a i = i )]TJ/F39 11.9552 Tf 11.955 0 Td [(a 0 121

PAGE 122

Theparameter a i representstheindividualspeciclevelofability.Replacingthe schoolingvariable S i inEquation3byEquation3,wegetthereducedform earningsfunction W i = a 0 + a i + b Z i + i + )]TJ/F39 11.9552 Tf 5.48 -9.684 Td [(b i )]TJETq1 0 0 1 351.442 634.342 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 351.442 624.366 Td [(b Z i + i + u i Takingtheexpectationof3conditionalon Z i yields E W i j Z i = E a 0 + a i + b Z i + i + )]TJ/F57 10.9091 Tf 5 -8.837 Td [(b i )]TJETq1 0 0 1 288.885 565.707 cm[]0 d 0 J 0.436 w 0 0 m 5.896 0 l SQBT/F57 10.9091 Tf 288.885 556.604 Td [(b Z i + i + u i j Z i = E a 0 j Z i + E a i j Z i + E b Z i + i j Z i + E )]TJ/F57 10.9091 Tf 9.545 -8.836 Td [(b i )]TJETq1 0 0 1 387.014 540.295 cm[]0 d 0 J 0.436 w 0 0 m 5.896 0 l SQBT/F57 10.9091 Tf 387.014 531.192 Td [(b Z i + i j Z i + E u i j Z i = a 0 + bZ i + E )]TJETq1 0 0 1 207.998 514.882 cm[]0 d 0 J 0.436 w 0 0 m 5.896 0 l SQBT/F57 10.9091 Tf 207.998 505.779 Td [(b i j Z i + E a i j Z i + E )]TJ/F57 10.9091 Tf 9.546 -8.836 Td [(b i )]TJETq1 0 0 1 349.231 514.882 cm[]0 d 0 J 0.436 w 0 0 m 5.896 0 l SQBT/F57 10.9091 Tf 349.231 505.779 Td [(b Z i j Z i + E )]TJ/F57 10.9091 Tf 9.545 -8.836 Td [(b i )]TJETq1 0 0 1 452.551 514.882 cm[]0 d 0 J 0.436 w 0 0 m 5.896 0 l SQBT/F57 10.9091 Tf 452.551 505.779 Td [(b i j Z i + E u i j Z i = a 0 + bZ i + E a i j Z i + E )]TJ/F57 10.9091 Tf 9.545 -8.837 Td [(b i )]TJETq1 0 0 1 288.605 489.469 cm[]0 d 0 J 0.436 w 0 0 m 5.896 0 l SQBT/F57 10.9091 Tf 288.605 480.366 Td [(b i j Z i + E u i j Z i wheretheterms E )]TJETq1 0 0 1 181.011 455.475 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 181.011 445.5 Td [(b i j Z i =0 and E )]TJ/F39 11.9552 Tf 10.461 -9.683 Td [(b i )]TJETq1 0 0 1 314.283 455.475 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 314.283 445.5 Td [(b Z i j Z i =0. Toobtainconsistent estimatesoftheaveragereturntocommunitycollegeeducation,weneedtosatisfythe followingidenticationconditions: i E a i j Z i =0 ii E u i j Z i =0 iii E )]TJ/F39 11.9552 Tf 10.46 -9.684 Td [(b i )]TJETq1 0 0 1 133.129 335.936 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 133.129 325.96 Td [(b i j Z i =0 Conditioniimpliestheabsenceoftheabilitybiaswhichrequirestheindependence ofabilityandtheinstrumentalvariable. 6 Conditioniirequirestheindependence oftheinstrumentalvariableandanyunaccountedfactorintheearningsequation thataffectsbothearningsandtheschoolingchoice.Forexample,thefactorsand policiesthatimprovelabormarketconditionsleadingtohighemploymentandearnings potentialsmaysimultaneouslyaffectearningsandtheschoolingchoice,whichwill induceindividualstoworkinsteadofattendingacollege.Conditioniiirequiresthe 6 Itispossiblethathighabilityindividualsacquiremoreeducationandearnmoreat theworkplaceregardlessoftheirchoiceofeducation. 122

PAGE 123

absenceofcomparativeadvantagebiasindividualswhohavehigherreturnsto schoolingtendtoacquiremoreeducation.Satisfyingtheseconditionsattheminimum arecrucialinobtainingconsistentestimatesoftheaveragereturnstocommunitycollege education. Theidenticationconditionsabovewouldapplyundertheassumptionofheterogeneous returnstoeducationin3.Ifheterogeneityisabsenti.e., b = b i ,theconditional expectationofearningscollapsesto E W i j Z i = a 0 + bZ i + E a i j Z i + E u i j Z i withtheterm E )]TJ/F39 11.9552 Tf 10.461 -9.684 Td [(b i )]TJETq1 0 0 1 189.358 502.848 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 189.358 492.872 Td [(b i j Z i =0. Theidenticationconditionsinthiscaseare i' E a i j Z i =0 ii' E u i j Z i =0. Withouttheheterogeneousreturns,theidenticationconditionsini'andii'only requiretheindependenceoftheinstrumentalvariable,ability,andthethefactorsthat shiftlabormarketconditions. Theidenticationconditionsareoftenviolatedinmanystudiesthattrytoisolate thecausalimpactofschoolingonearningsduetothepotentialendogeneityinthe schoolingchoice.Iusethefractionofthecostofattendingthecommunitycollegeasthe instrumentalvariabletoaddresstheendogeneityinthecommunitycollegeeducation. Providedthatthefractionofthecollegecostismeanindependentatleastwiththe unobservedability,labormarketconditions,andtheunobservedcomparativeadvantage ofindividualswithhighreturnstoeducation,itwouldbepossibletoobtainmorecredible estimatesoftheeffectofthecommunitycollegeeducationonearnings.Itistherefore imperativetoestablishthevalidityofthefractionofthecostofattendingcommunity collegeasaninstrumentalvariable. Sincetheassumptionsaboveareuntestable,Iemploythefollowingindirecttests: 123

PAGE 124

1.Balancingtestsontheobservablecovariateswhichlooksattheuncorrelatedness betweentheinstrumentalvariableandtheobservedcovariates. 2.Testoftheuncorrelatednessbetweentheinstrumentalvariableandthetest scores,withthetestscoresusedastheproxyforability.Thisindirectlytests conditioni. 3.Testofuncorrelatednessbetweentheinstrumentalvariable Z i andsomekey indicatorsofthelabormarketconditionsaswellasthecommunitycollege indicatorsthatarepotentiallyrelatedtotheinstrumentalvariable.Thisisthe indirecttestforconditionii. 4.Testforthepresenceofthesystematiccorrelationbetweentheinstrumental variable Z i andtheresidualearningscomponent )]TJ/F39 11.9552 Tf 5.479 -9.684 Td [(b i )]TJETq1 0 0 1 390.016 537.701 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 390.016 527.725 Td [(b i whichwilltest conditioniii.Thefeasiblewayoftestingthissystematiccorrelationisby comparingtheeffectsofthetestscoresonthecommunitycollegeoutcomes underdifferentvaluesofthefractionofthecollegecost. 7 3.4.3TestsontheValidityoftheFractionofCollegeCost Therstordertestthattheinstrumentalvariablemustpassisthebalancingtest. Theideaisthattheinstrumentalvariableofinterestmustbebalancedinwhenthe instrumentisbinaryoruncorrelatedwithwhentheinstrumentiscontinuousthe observedandunobservedfactors.Thisisparticularlyimportantwhentheunobserved factorse.g.,abilityandcomparativeadvantageovereducationarebelievedtovary systematicallyacrossobservedcharacteristicssuchasgenderorracialafliation. Sincetheobservedfactorsarealreadyavailableinthedata,thebalancingtestscanbe easilyappliedbylookingatthecorrelationanditsstatisticalsignicancebetweenthe instrumentalvariableandtheobservedfactors.Asfortheunobservedfactors,itisthe creativityoftheresearchercombinedwithluckinndinggoodproxyvariablesthatwill potentiallymakeasensibledevicetotestoutthebalancednessamongtheunobserved factors.Becausesomeoftheunobservedfactorscouldpotentiallyberelatedtothe observedfactors,carryingoutthebalancingtestsontheobservedfactorsisanindirect 7 Forexample,wecancomparetheeffectsoftestscoresforstudentswhoattended communitycollegeswithbelowandabovemedianfractionofthecollegecost. 124

PAGE 125

wayoftestingwhetherornottheunobservedfactorsvarywiththeinstrumentalvariable. Forexample,ifabilityvarieswithracewithwhiteshavingthehighestabilityamongthe racialgroups,thenndingthattherearemorewhiteindividualslivingnearcolleges thatofferlowercostofattendancewouldindirectlyimplythattheseindividualswiththe highestabilityliveclosertothesecollegesthatofferlowercostofattendance.Withthis, thereisareasontobelievethattheestimatesfromaregressionofcommunitycollege outcomeontheinstrumentalvariableispotentiallybiasedevenaftercontrollingforan indicatorofbeingawhiteindividual. Table3-7showstheresultsfromtheregressionoftheobservedcovariatesincluding age,gender,majorracialgroups,free/reducedlunchstatus,indicatorofbeinglimited Englishprocientandbeingaspecialchildontheinstrumentalvariable.Therst columninTable3-7containscoefcientestimatesfromthesimpleregressionwithout controllingfortheothercovariates.Indthatsomecovariatesincludingthefraction ofmales,fractionofwhites,andfractionofblacksarestatisticallysignicant.What dowemakeoftheseestimates?Thenegativeandstatisticallysignicantestimate intheregressionwiththefractionofmaleswouldmeanthattherearemoremale studentsinplaceswherethehighschooltheyattendedareclosertothecommunity collegesthatofferlowercostofattendance.Ifabilitynorcomparativeadvantage overattainingmoreeducationdoesnotvaryacrossgender,however,thisestimate wouldnotmeananythingexceptthattherearebasicallymoremalesthanfemaleson average.ThesecondcolumninTable3-7presentsthecoefcientestimateswhen othercovariatesareaddedincludingcohortdummies,communitycollegedummies, demographiccharacteristics,andtestscores.Withtheinclusionofthesecovariates,all thecoefcientestimatesbecomestatisticallynotdifferentfromzero,whichrendersthe covariatestobebalancedonobservables.Withthis,Icanmakeassumptionsusing selection-on-observables,whichisaweakerconditionrelativetotheconditionsfound ini,ii,andiii. 125

PAGE 126

Perhapsacloserbutanindirectwayoftestingwhetherornottheinstrumental variableiscorrelatedwithabilityisbylookingatthecorrelationbetweentheinstrument andtestscores.TheFloridadataincludesinformationonFCATscoresthatarerst takenwhenthestudentsareingradeten.TheFCATisdesignedtotestthereadiness ofFloridahighschoolstudentstoengageinhighereducationandtowork.Thismakes theFCATscoresagoodmeasureofstudents'abilitytogetintocommunitycolleges andtoattainahighereducationdegrees.Intestingthecorrelation,Iestimateseparate regressionsforthemathematicsandreadingFCATscores.Theregressionsinclude yeardummies,communitycollegedummies,anddemographiccharacteristics.The coefcientestimatesoftheinstrumentarepresentedinTable3-8.Thecoefcient estimatesarepositivebutstatisticallynotdifferentfromzero.Thissupportstheabsence ofcorrelationbetweentheinstrumentalvariableandtheFCATtestscores. Toindirectlytestconditionii,Isetupapaneldataofthekeyindicatorsofthe labormarketconditionsandcommunitycolleges.Themainideaistotestwhether ornotthesekeyindicatorscausetheinstrumenttochange.Forexample,anding thatthecommunitycollege'scostofoperationsaffectsthepublishedtotalcostof attendanceandeducationalattainmentwhicheventuallyaffectsearningspotentialwill confoundtheeffectsoftheinstrumentalvariableonthecommunitycollegeeducation andearnings.Table3-9presentsthecoefcientestimatesofthevariouslabormarket andcommunitycollegekeyindicatorsbasedontheregressionwiththecurrentvalues oftheinstrumentalvariableasthedependentvariable.Intheseregressions,Iuse thepreviousyear'svaluesforthekeyindicatorsbecauseofthetimingelementinthe publicationofthecostofthecollegeattendanceforaparticularschoolyear.Typically, thecostofattendanceforecastsaremadelessthanayearaheadofthecomingschool yearanditmakessensetoseeifthepreviousyear'skeycollegeandlabormarket indicatorsaffectthecostestimatesorthenancialaidoffers.Thecoefcientestimates arestatisticallynotdifferentfromzero,implyingtheabsenceofcorrelationbetween 126

PAGE 127

thefractionofthecollegecostandthekeyindicators.Withregardstothecostof operations,theresultisnotsurprisingsincethecostofoperationsandthecommunity collegeattendancecostsareindependentlyset,withFloridacommunitycolleges calculatingtheestimatesofthecostofoperationsandwiththestategovernment settingthetuitionandfeesscheduletobeadoptedbyeachcommunitycollege.Since mostofthecollegeattendancecostcomponentsaresetbyFloridalegislature,itisnot surprisingtoseethatthekeylabormarketconditionindicatorssuchasaveragecounty unemploymentrateandaverageannualcountyearningsdonotaffectthefractionofthe collegecost. Finally,Itestthesystematiccorrelationbetweentheinstrumentalvariableand theresidualearningscomponent )]TJ/F39 11.9552 Tf 5.479 -9.683 Td [(b i )]TJETq1 0 0 1 278.497 478.942 cm[]0 d 0 J 0.478 w 0 0 m 6.461 0 l SQBT/F39 11.9552 Tf 278.497 468.967 Td [(b i bycomparingtheeffectsoftestscores oncollegeoutcomesforstudentswhofacelowandhighfractionsofthecollegecost. Evenifabilityasproxiedbytestscoresisuncorrelatedwiththefractionofthecollege costconditionalontheobservedcovariates,thecorrelationbetweeneducationand abilityasmeasuredbytestscoresmaybedifferentforindividualswhoattended collegeswithlowandhighfractionsofthecollegecost.Thesedifferencesintheeffects ofthetestscorescouldbearesultofthechangingmappingbetweentheindividual abilityandeducationoutcomes.Thiswouldarisewhenindividualswithhigherreturns tocollegeeducationwhopresumablyalsohavehigherabilitywillalwayswanttoget moreeducationandthus,willliveclosertocommunitycollegesthatofferalowfraction ofthecostofattendance.Hence,ifthereisasignicantdifferenceonhowability affectscollegeoutcomesinindividualswhoattendedcollegesthatofferlowandhigh fractionsofthecollegecost,thatwouldprovidesubstantialevidenceofthecorrelation betweentheresidualearningscomponentandabilitywhichwouldviolateconditioniii. Iproceedtestingthiscorrelationintwoways:rst,bycomparingtheeffectsbetween thosewhofacebelowandabovemedianfractionsofthecollegecostofattendance; andsecond,bylookingatthecoefcientestimatesoftheinteractiontermbetweenthe 127

PAGE 128

fractionofthecollegecostandthetestscores.Theabsenceofthecorrelationamounts tohavingstatisticallyidenticalcoefcientestimatesintherstapproachandhaving astatisticallyinsignicantestimateoftheinteractionterminthesecondapproach. Consistentwiththeideathatindividualswhofacealowerfractionofthecollegecostare morelikelytoattendcollegeandthattheindividualswhoattendedcollegeonaverage willearnmore,Icannotdetectasignicantdifferenceintheeffectsofthetestscoreson thecommunitycollegeoutcomesforindividualswhoattendedbelowandabovemedian fractionsofthecollegecost.Table3-10conrmsthis.Theestimatesoftheeffectofthe mathematicstestscoresandReadingforthebelowandabovemedianfractionsofthe collegecost,ifnotstatisticallydifferentfromzero,areverysimilarinmagnitudeinthe regressionswithCCTermsortheAAdegreeasthedependentvariable.Turninginto theregressionversionthatincludestheinteractiontermbetweentheratiovariableand testscores,thecoefcientestimatesoftheinteractiontermarestatisticallynotdifferent fromzeroinregressionswithCCTermsorAAdegreeasthedependentvariable.This impliesthattherearenovariationsonhowthetestscoresaffectcommunitycollege outcomesamongstudentswhopaysmallorlargerfractionsofthecollegecost.These resultssomehoweasethedoubtthattheunobservedcorrelationbetweenabilityandthe residualearningscomponentwillbiastheestimatesofthereturnstocommunitycollege education. Thebatteryofindirecttestsaboveshowthattheinstrumentalvariableatthe minimumsatisestheidenticationconditions 8 .Therearesomequalitativeevidence tosupporttheclaimthatthefractionofthecollegecostcanbetreatedasexogenous. TheOfceofProgramPolicyAnalysisandGovernmentAccountability,anarmofFlorida 8 Totestthemeritsofotherpotentialinstruments,thesebatteryoftestswerealso carriedoutforthenetcost,totalcost,anddistancetocommunitycollege.Thesethree alternativeinstrumentsfailedatleasttwoofthetestsabove. 128

PAGE 129

legislature,providedseveralreportsontheintriguingissueabouttheextensivevariation inFlorida'sstateuniversityandcommunitycollegesystemReportNo.03-33Ofce oftheProgramPolicyAnalysisandGovernmentAccountability,2003.Inthisreport, tuitionandfeesandtheexpensesforroomandboardreceivedthegreatestmentionof beingthetwolargestcomponentsofthecostofattendance,withthetuitionandfees contributingtothelargestchangeovertheyears.TuitionandfeesaresetbytheFlorida legislaturethroughtheStateBoardofEducation.Thetuitionandfeesscheduleto beadoptedinagivenschoolyearwasbasedonnotlessthan15yearsofhistorical costdata.Forexample,thetuitionandfeesscheduleadoptedintheschoolyear 2000-2001wasbasedonthehistoricalcostanalysisthatusesthecostsintheschool year1999-2000,whichwasbasedonthecostanalysisofthepastyears.Because,the tuitionandfeescomponentofthetotalcostofattendanceisbasedonthehistorical values,itisnothardtosee,forexample,whythecostofattendanceisindependentof onetoaboutthreeyearsworthoflabormarketconditionsinformationoranyinformation aboutthecharacteristicsofFloridacommunitycollegese.g.,enrollmentheadcounts. TheOPPAGAreportconcludedthatthewidevariationinthetotalcostofattendanceis primarilyduetothedifferencesinthewaythecostcomponentsarecalculated.With this,itisafactthatthetotalcostofattendingacollegeinFloridaisbasedonFlorida legislatureandtheuniquefeaturesofthe28Floridacommunitycolleges,anevidenceto supportthatthefractionofcollegecostcanbetreatedasexogenous. 3.5Results InthissectionIpresenttheestimatesoftheeffectofthefractionofthecostof attendingcommunitycollege.Irstassesstheimpactofthefractionofthecostontwo communitycollegeoutcomes:theNumberofSemestersEnrolledandtheLikelihood 129

PAGE 130

ofReceivinganAssociatesinArtsdegreefromacommunitycollegeinFlorida. 9 I alsopresenttheinstrumentalvariablesestimatesofthereturnstocommunitycollege outcomesbasedonthefractionofthecostofattendingcommunitycollege. 3.5.1FirstStageResults Thecommunitycollegeoutcomesofinterestsinthispaperarethenumberof semestersenrolledincommunitycollegeandtheassociatesinartsdegreecompletion. Lookingattherelationshipbetweenthenumberofsemestersenrolledandthefraction ofthecollegecostprovideswillshedlightontheextenttowhichthecostaffectsthe attendanceandpersistenceincommunitycollege.Ontheotherhand,therelationship betweenthelikelihoodofreceivingassociatesinartsdegreeandthefractionofthe collegecostwillshedlightonhowthecollegecostaffectthecollegedegreecompletion. Table3-11presentstherststageOLSregressionestimatesoftheeffectsof thefractionofthecostofattendingcommunitycollegeonthesecollegeoutcomes controllingforthestandardizedmathematicsandreadingtestscores,indicatorsof gender,race,free/reducedlunchstatus,limitedEnglishprociency,specialstudent, reachedgrade10ontime,yeareffects,communitycollegedummies,andinteraction terms.Inalltheregressionspecications,Iusetherobuststandarderrorsthatare clusteredbyyearandcommunitycollegeafliationofindividualsinthesample. Thecoefcientestimatesofthefractionofthecollegecostarenegativeand preciselyestimatedintheregressionmodelswiththenumberofsemestersenrolled incollegeandtheassociatesinartsdegreecompletionasthedependentvariables. Achangeinthefractionofcollegecostfromzerotoonewilldecreasethecollege enrollmentbyabout0.4ofasemestert-stat=-4.86andwilldecreasetheassociates 9 Theregressionoftheassociatesinartsdegreecompletionorthenumberof semestersenrolledonthefractionofthecostisbasicallytherststageregressionunder theinstrumentalvariablesprocedure. 130

PAGE 131

inartsdegreecompletionratebyabouttwopercentagepointst-stat=3.21. 10 The estimatesofthet-statisticsaregreaterthanthreeimplyingthatthefractionofthecollege costsatisestherequirementofagoodinstrument.Wecandrawoneconclusion fromTable3-11.Thefractionofthecollegecostappearstohaveanegativeimpact onboththenumberofsemestersenrolledincollegeandtheassociatesinartsdegree completion.Thisimpliesthatahighercontributioninpayingforthecollegeeducationwill makethesestudentstoenrolllessincollegeandwillmakecollegedegreecompletion lesslikely. 11 Theaboveestimatesmaymaskdifferencesintheeffectsacrossthemajorracial groups.Theestimatesmightbedifferentacrossracegroupsduetothedifferencesin familybackground,culture,andresourcesthathavedirectimpactonthecollege-going behaviorandmotivationtocompleteacollegedegree.Forexample,studentswithmore resourcesareprobablylesssensitivetothehighercollegecosts,andasaresult,the effectsofthecollegecostswouldbesmalleronthem.Table3-12showsthecoefcient estimatesfromtheregressionofthecommunitycollegeoutcomesonthefractionof thecollegecostforthemajorracialgroups.Theregressionmodelsincolumns andusethesamecontrolvariablesasintheregressionspecicationinTable3-11. Standarderrorsarerobustandclusteredbyyearandcommunitycollegeafliationofthe student.Theestimatesincolumnindicatethatanincreaseinthefractionofcollege 10 Increasingthefractionofthecostfromzerotooneimpliesthatthecollegecost shoulderedbystudentincreasesfromzerotothefullcostofcollegeattendance. 11 Whenroomandboardexpensesareexcludedinthecalculationofthecostof attendance,theestimatedeffectsarereducedbyasmuchas50percent.Theeffect ontheassociatesinartsdegreecompletionisreducedtoonlyonepercentagepoint statisticallysignicantat0.0llevelwhiletheeffectonsemestersenrolledisreduced to0.15ofasemesterstatisticallysignicantat0.05level.Theevidencesupportthat therearesomepeoplewhodonotliveathome.Themainresultsareconsistentwith studentswhodonotliveathomeandhavespentonroomandboardwhileincommunity college. 131

PAGE 132

costfromzerotoonewilldecreasethenumberofsemestersenrolledby0.33ofa semesterforwhitest-stat=-3.4,0.25forblackst-stat=-2.76,and0.53forhispanics t-stat=-3.73.Thefractionofthecollegecostalsohasdifferentialnegativeeffectson theassociatesinartsdegreecompletion.Theregressionestimatesincolumnshow thattheassociatesinartsdegreecompletionratewithdecreaseby1.8percentage pointsforwhitest-stat=-2.3,0.9percentagepointsforblackst-stat=-2.14,and 2.8percentagepointsforhispanicst-stat=-2.76.Theseestimatespointoutthat thefractionofthecollegecosthasnegativeeffectsontheassociatesinartsdegree completionrateswiththelargesteffectonhispanics. Theaboveestimatesmayalsomaskdifferencesacrossgender.Thedifferences intheeffectsmayariseduetothegenderdifferencesinmotivationandpersistence towardsacquiringcollegecredentials. 12 Theestimatesincolumnshowsthatthe estimateddecreaseinthenumberofsemestersenrolledis0.27t-stat=-3.19ofa semesterformalesand0.46t-stat=-5.9ofsemesterforfemaleswhiletheestimated decreaseintheassociatesinartsdegreecompletionrateincolumnisabout1.5 t-stat=-2.4percentagepointsformalesand2.5percentagepointst-stat=-3.46for females.Theseestimatesindicatethatthefemales'college-goingbehaviorandthe desiretocompleteacollegedegreeareaffectedmorebyahigherfractionofthecollege cost. Theeffectsofthefractionofthecollegecostmayalsobedifferentacrossstudents whoandwhodidnotparticipateinthefree/reducedlunchprogram.Thedifferential effectsmayariseduetodifferencesinfamilyresources,whichisanimportantinputtoa student'ssuccessincollege.Also,lookingatthedifferencesintheeffectsacrossthese groupsofstudentsisespeciallyimportantsincetheprogramrulesforthefree/reduced 12 Leppel2002ndsthatcollegepersistenceishigherinfemales,especiallyamong African-Americanfemales. 132

PAGE 133

lunchprogramcanbeinuencedbypolicies.Thecoefcientestimatesincolumn showthattheestimateddecreaseinthenumberofsemestersenrolledis0.44 ofasemestert-stat=-3.11forstudentswhoparticipatedinthefree/reducedlunch programand0.35ofasemestert-stat=-4.45forstudentswhodidnotparticipatein thefree/reducedlunchprogram.Also,theestimateddecreaseintheassociatesinarts degreecompletionincolumnis1.5percentagepointst-stat=-2.5forstudentswho participatedinthefree/reducedlunchprogramand2.3percentagepointst-stat=-3.28 forstudentswhodidnotparticipateinthefree/reducedlunchprogram. Theeffectsofthefractionofthecollegecostmayalsobedifferentamongstudents whoreachedanddidnotreachgradetenontime.Becausethesestudentsreached gradetenontime,thesestudentspassedthebatteryoftestsadministeredbythe FloridaDepartmentofEducationstartingwhenthestudentwasin3rdgrade.Itislikely thatthesesetofstudentswhoareontrackhavehigherabilityandpossiblyhavebetter educationalresourcescomparedtothosestudentswhofailedtoreachedgradeten ontime.Thisdifferenceinthetimingofreachinggrade10mayhaveimplicationson persistenceanddegreecompletionincollege.Theestimatesincolumnsandof Table3-12showevidenceofthisdifference.Onlytheestimatesforthestudentswho reachgrade10ontimeappeartobestatisticallynotdifferentfromzero.Incolumn, theestimateddecreaseinthenumberofsemestersenrolledis0.44t-stat=-5.98 ofasemesterforstudentswhoreached10thgradeontimeandaboutaneightt-stat =-0.97ofasemesterforstudentswhodidnotreach10thgradeontime.Also,the estimateddecreaseintheassociatesinartsdegreecompletionincolumnis2.4 t-stat=-3.43percentagepointsforstudentswhoreached10thgradeontimeand0.6 t-stat=-1.2percentagepointsforstudentswhodidnotreach10thgradeontime. 3.5.2InstrumentalVariablesEstimates InthissectionIpresenttheinstrumentalvariablesestimates.Iestimatethereturns tothecommunitycollegeoutcomesthenumberofsemestersenrolledincommunity 133

PAGE 134

collegeandtheassociatesinartsdegreecompletionusingtheinstrumentalvariables IVprocedurebasedonthefractionofthecollegecost.Thestandardresultsinthe studiesthatestimatetheeffectofschoolingonearningsshowthattheindividualswith betterschoolingoutcomeshavehigherwages/earnings.Inthispaper,Iexpectthat thecommunitycollegestudentswhoshowpersistenceandenrolledmoresemestersin thecommunitycollege,ontheaverage,havehigheraverageannualtotalearningsand thecommunitycollegestudentswhoreceivedanassociatesinartsdegreearealso, ontheaveragehavehigheraverageannualtotalearnings.Irstassesstheeffectsof thesecommunitycollegeoutcomesusingtheOLSregressionmodel.ThenIcompare theIVestimatestothebenchmarkOLSestimates. AlthoughtheOLSestimateshadbeenproventobebiasedinmanyempirical studies,theseestimatesarestillimportantsincetheycanserveascomparisonsto theotherestimatesgeneratedbyanalternativeestimationstrategy,withthehopeof identifyingthepotentialestimationbias.Forexample,thestandardexplanationofhaving arelativelyhigherinstrumentalvariablesestimatesisthepresenceofabiasattributed tothemeasurementerrororthepresenceofheterogeneouseffects.Ontheotherhand, whentheOLSestimatesarelargerthantheIVestimatesthestandardexplanationisthe presenceofabiasduetotheomittedvariables.WhiletheIVandOLSestimatescannot becompareddirectly,thedifferencesintheseestimateswouldindicatethepresence, thedirection,andthepotentialsourceofthebias. Table3-13showstheestimatesofthereturnstocommunitycollegeoutcomes basedontwoestimationmodels.First,theOLSestimatesofthereturnstocommunity collegeoutcomesincolumnaregeneratedfromtheregressionoftheaverage annualtotalearningsonthecommunitycollegeoutcomes.Second,theinstrumental variablesestimatesofthereturnstocommunitycollegeoutcomesincolumn2are generatedfromthestructuralmodelsofearningswiththefractionofthecollegecost astheinstrumentalvariableforthecommunitycollegeoutcomes.BoththeOLSandIV 134

PAGE 135

regressionscontrolforthemathematicsandreadingtestscores,indicatorsofgender, race,free/reducedlunchstatus,limitedEnglishprociency,specialstudent,reached gradetenontime,yeareffects,nearestcommunitycollegeindicators,andinteraction terms.Inalltheregressionspecications,Iusetherobuststandarderrorsthatare clusteredbyyearandcommunitycollegeafliationofindividualsinthesample. ColumnshowtheOLSestimatesofthereturnstothenumberofsemesters enrolledincommunitycollegeandthereturnstocompletinganassociatesinarts degree.TheOLSestimatesindicatethattheaverageannualtotalearningswillincrease byabout$323.62about6.7percentofthemeanaverageannualtotalearningsfor eachsemesterenrolledincollegeandarehigherby$1,070.80about22percentof themeanaverageannualtotalearningsforstudentswhocompletedassociatesinarts degree.TheinclusionofthecontrolvariablesintheOLSregression,however,does notguaranteethevalidityoftheOLSestimatesascasualestimatesofthereturnsto communitycollegeoutcomes.DespitethepotentialbiaspresentintheOLSestimates, wecandrawaconclusionfromtheseresults:thestudentswhopersistedandenrolled moresemestersincommunitycollegeandthestudentswhocompletedanassociatesin artsdegreetendtohavehigherearnings. TheOLSresultsshouldbeinterpretedwithacaveat.Asdiscussedabove,these OLSestimatesarepotentiallybiasedsincestudentscanself-selecttoattendmore semestersincollegeandcompleteadegreeespeciallywithstudentswithhigher abilityandwithhigherpotentialgainsfromattendingcollege.Iusethefractionofthe collegecostasthesourceoftheexogenousvariationincollegeoutcomestoaddress theproblem.Columnpresentstheinstrumentalvariablesestimatesofthereturns tothecommunitycollegeoutcomesbasedonthefractionofthecollegecostasthe instrumentalvariableforcollegeoutcomes.RelativetotheOLSestimates,theIV estimatesarebigger.Theestimatedreturnstothenumberofsemestersenrolledin collegeareabout3timeslargerwhiletheestimateofthereturnstocompletingan 135

PAGE 136

associatesinartsdegreeareabout19timeslargerthantheOLS.TheIVregression estimatesaremarginallysignicant.Theestimatesincolumnshowthattheaverage annualtotalearningswillincreaseby$1,097.24about22.8percentofthemean averageannualtotalearningsforeverysemesterthatastudentenrollincommunity collegewhiletheaverageannualtotalearningsarehigherby$20,419.8percent ofthemeanaverageannualtotalearningsforstudentswhocompletedassociatesin artsdegree.Therangeoftheestimatedreturnstothenumberofsemestersenrolled incollegeisquitewidethatIcannotruleoutthezeroeffects.Also,therangeofthe estimatedreturnstocompletinganassociatesinartsdegreehasawidecondence intervalthatincludestheOLSestimatesatleastatthevepercentlevelofsignicance. Althoughthisrangeiswide,Icanruleoutthezeroreturnstocompletinganassociates inartsdegree. 13 Thisprovideasuggestiveevidencethatthecollegestudentswho completedanassociatesinartsdegreehavehigherearningspotential. Theaboveresultspointtothefollowingconclusions.First,theIVestimatesofthe returnstothecommunitycollegeoutcomesareuniformlylargerthantheOLSestimates. Second,becausetheIVestimateofthereturnstothenumberofsemestersenrolled incommunitycollegeisimpreciselyestimated,therangeofthisIVestimateislarge withwhichzeroeffectscannotberuledout.Ontheotherhand,althoughIestimatedan effectinthereturnstocompletinganassociatesinartsdegreearewithinthe95percent condenceinterval,therangeofthisIVestimateremainlargebutIcanruleoutthezero returnstocompletinganassociatesinartsdegree. TheIVestimatesinTable3-13gaveusthepictureofthereturnstocommunity collegeoutcomesforthefullsample.Itispossiblethattheseestimatesmaskdifferences 13 Relativetothemeanaverageannualtotalearnings,the95percentcondence intervalfortheIVestimatesofthereturnstoassociatesinartsdegreecompletionis [1.2%,849%] 136

PAGE 137

forthedifferentdemographicgroups.Forexample,thereturnscouldbebiggerformales thanforfemalesmaybebecausemaleshavestrongerattachmenttothelabormarket. Also,thereturnsforwhitescouldbebiggerthananyoftheracialgroupsbecause theytypicallyhavemoreresourcesandprobablyhaveabettereducationalfoundation tostartwith.Further,studentswhoreached10thgradeontimeearnmorepossibly becausetheyhavebetterskillsandhavehighermotivation.Table3-14presentsthe IVestimatesofthereturnstocommunitycollegeoutcomesbydemographicgroups race,gender,free/reducedlunchstatus,andhavingreached10thgradeontime.The specicationsoftheIVregressionmodelsarethesamewiththemodelsinTable3-12. ColumnpresentstheIVestimatesofthereturnstothenumberofsemestersenrolled incommunitycollegewhilecolumnpresentsestimatesofthereturnstocompleting anassociatesinartsdegree.Iusethesamesetofcontrolvariablesasintheregression specicationsabove.Ialsousetherobuststandarderrorsthatareclusteredbyyearand communitycollegeafliationsofthestudents. IntheIVregressionmodelsincolumn,onlytheestimatesofthereturnstothe numberofsemestersenrolledincollegeforthesamplewhitesandforthesampleof studentswhodidnotparticipateinthefree/reducedlunchprogramarestatistically signicantatthevepercentlevel.Foreverysemesterastudentstayedincommunity college,Indthattheaverageannualtotalearningswillincreaseby$1,769.4inthe sampleofwhitesandby$1,194.4inthesampleofstudentswhodidnotparticipate inthefree/reducedlunchprogram. 14 Theestimatesofthereturnstocompleting anassociatesinartsdegreeincolumnarepositiveandstatisticallysignicant onlyinthesampleofwhites,inthesampleofstudentswhodidnotparticipatein free/reducedlunchprogram,andinthesampleofstudentswhoreached10thgrade ontime.Completingassociatesinartsdegreewillincreasetheaverageannualtotal 14 TheseestimatesarelargerthantheestimatesfoundincolumnofTable3-13. 137

PAGE 138

earningsby$33,347.3about680percentofthemeanaverageannualtotalearnings inthesampleofwhites,by$18,333.0about375percentofthemeanaverageannual totalearningsinthesampleofstudentswhodidnotparticipateinfree/reducedlunch program,andby$17,704.6about368percentofthemeanaverageannualtotal earningsinthesampleofstudentswhoreached10thgradeontime. 15 Theestimates incolumnsandpointtooneconclusion:therearepositivereturnstothenumber ofsemestersenrolledincollegeandtocompletinganassociatesinartsdegreeatleast inthesampleofwhites,studentswhodidnotparticipateinfree/reducedlunchprogram, andthestudentswhoreached10thgradeontime.Althoughtherangeoftheestimates inTable3-14arewide,Icanstillruleoutzeroreturnstocommunitycollegeoutcomesfor thesesubgroups. 3.6FalsicationTests Inthissection,Ishowthatthefractionofthecostofattendingcommunitycollege isalegitimateinstrumentalvariableforthecommunitycollegeoutcomes.Iconducteda numberoffalsicationteststoseeifthefractionofthecollegecostaffectsonlythe communitycollegeoutcomesandnottheotheroutcomesbeyondthecommunitycollege andtoseethatthethefractionofthecollegecosthasnodirecteffectonearnings exceptthroughthecommunitycollegeoutcomes. Intherstfalsicationtest,Ishowthattheinstrumentalvariabledoesnotaffect theoutcomesbeyondcommunitycollege.InthisapplicationIusefourstateuniversity systemSUSoutcomes: 1.SUSonlyattendancethisincludesthehighschoolstudentswhoattendedan SUSaftercompletinghighschoolandhaveneverattendedacommunitycollege; 2.thenumberofsemestersenrolledwhileattendinganSUS; 15 ComparedtotheestimatesincolumninTable3-13,theestimatesarelargerfor whitesbutsmallerforstudentswhoreached10thgradeontime. 138

PAGE 139

3.thebachelor'sdegreecompletioninSUSwhetheradegreeiscompletedwhile attendinganSUS; 4.everattendedanSUSthisincludesthestudentswhoonlyattendedanSUSand thestudentswhotransferredtoanSUSfromacommunitycollege. Ifthefractionofthecollegecostisalegitimateinstrument,thenitshouldonlyaffect thecommunitycollegeoutcomesandnottheseSUSoutcomes.Incarryingoutthe falsicationtests,IrunregressionsoftheSUSoutcomesonthefractionofthecollege costcontrollingforstandardizedmathematicsandreadingtestscores,indicatorsof gender,race,free/reducedlunchstatus,limitedEnglishprociency,specialstudent, reachedgrade10ontime,yeareffects,communitycollegedummies,andinteraction terms.Inalltheregressionspecications,Iusetherobuststandarderrorsthatare clusteredbyyearandcommunitycollegeafliationofindividualsinthesample.The coefcientestimateofthefractionofthecollegecostshouldbestatisticallynotdifferent fromzerotoestablishnocorrelation.Table3-15presentstheestimatesfromtheOLS regressionmodelswithcontrolvariables.Exceptforthemarginallysignicantcoefcient estimateincolumnfortheSUSTerms,allofthecoefcientestimatesarestatistically notdifferentfromzero,with t ratiosbelowtheconventionallevels,implyingnocorrelation betweenthefractionofthecommunitycollegecostandtheSUSoutcomes.Coupled withtherststageresultsinTable3-11thatshowcorrelationsbetweenthefractionof thecollegecostandthecommunitycollegeoutcomes,theresultsfromthesefalsication testsindicatethatthefractionofthecollegecostaffectsonlythecommunitycollege outcomesandnottheSUSoutcomeswhicharebeyondthestudent'scommunity college-goingbehaviorandthedesiretocompleteanassociatesinartsdegree. Thesecondfalsicationtestaimstoshowthatthefractionofthecommunity collegecostdoesnotdirectlyaffectearningsbutaffectsearningsindirectlythroughthe communitycollegeoutcome.Inotherwords,thisfalsicationtestisatestwhetheror notthefractionofthecollegecostcanbeexcludedinthestructuralmodelofearnings. Sincethestudentsfromthelowincomehouseholdswouldbeaffectedmorecompared 139

PAGE 140

tothestudentsfromthehighincomehouseholdsfacingthesamehighfractionof thecollegecost,thissuggeststhatafractionofthecollegecostmusthavelarger effectsonthecollegeeducationoutcomesofthestudentscomingfromthelowincome households.Themodelfortheschoolingchoicebecomes: S = X 1 + 1 Fraction + 2 Ratio P + u where X includesallcontrolvariablesotherthan Fraction P denotesanindicatorfor lowfamilyincome, 1 1 ,and 2 arepositivecoefcients.If Fraction isincludedinthe earningsequation, E = 1 X + 1 Fraction + 1 S + v theinteractionterm Fraction P inEquation3canbeusedasaninstrumental variableforcommunitycollegeeducation S .Theidenticationconditionisthatthedirect earningseffectassociatedwiththe Fraction variabledoesnotvarywiththeunobserved factors. 16 Sincethereisnodirectinformationorindicatorforlowfamilyincomeinthe data,Iusedseveralobservedfactorsthataredirectlyrelatedtobeingalowincome household: 1.ifthestudent'sraceisblack;and 2.ifthestudentparticipatedinthefree/reducedlunchprogram. Table3-16showsthecoefcientestimatesfromtheinstrumentalvariablesestimation basedontheinteractiontermofthefractionofthecollegecostandanindicatoroflow familyincomeastheinstrument.Thefractionofthecollegecostisincludedbothinthe schoolingequationrststageandintheearningsequationndstage.Thecoefcient estimatesofthefractionofthecollegecostvariableshouldappearstatisticallynot 16 Forexample, Fraction shouldnotinanywayvarywithunobservedabilityorthe unobservedgeographicwagedifferentials 140

PAGE 141

differentfromzerotoclaimthatthefractionofthecollegecostcanbeexcludedinthe earningsequation.Ifthisisthecasethenwecansupporttheclaimthatthefractionof thecollegecostdoesnotdirectlyaffecttheearningsequation.Also,theIVestimate basedontheinteractiontermshouldbesmallerthantheIVestimatebasedonthe fractionofthecollegecost.Thisisconsistentwiththeideathatahigherfractionof thecollegecosthasagreaternegativeimpactonthecommunitycollegeoutcomes ofstudentsfromthelowincomefamilyandthus,willyieldlowerearningsestimates. TheresultsinTable3-16conrmthis.Allofthecoefcientestimatesofthe Fraction ofCollegeCost incolumnarestatisticallynotdifferentfromzerointhestructural modelofearningsequationwith t ratiosbelowtheconventionallevels.Moreover, theIVestimatesofthereturnstothenumberofsemestersenrolledincollegeandto completinganassociatesinartsdegreebasedontheinteractiontermofthefraction ofthecollegecost,black,andthelunchstatusarealsosmallerthantheIVestimates basedonthefractionofcollegecostinTable3-13implyingthatstudentsfromlow incomefamiliesareaffectedmorebyahigherfractionofthecommunitycollegecost. 3.7Discussion Inthispaper,Iconsiderusingthefractionofthecostofattendingcommunitycollege asthesourceofexogenousvariationinthecommunitycollegeoutcomestoestimate thereturnstothenumberofsemestersenrolledincommunitycollegeandthereturns tocompletinganassociatesinartsdegree.Theanalysissuggeststhatthefractionof thecollegecosthasanegativeimpactbothonthenumberofsemestersenrolledin collegeandontheassociatesinartsdegreecompletion.Theseareassociatedwith positivereturnsbothtothenumberofsemestersstudentswereenrolledincommunity collegei.e.,theschoolingeffectandtocompletinganassociatesinartsdegreei.e., thesheepskineffect.InthissectionIdiscusswhatmightexplaintheseresults. Thendingsinthispapersuggestthatcommunitycollegesgeneratespositive returnstothecommunitycollegeeducation.Theresultsabovealsosuggestthatthat 141

PAGE 142

theseestimatesarequitelargerelativetothesizeoftheestimatesfromtheOLS regression,withtheIVestimatesofthesheepskineffecti.e.,AAdegreecompletion ofabout19timeslargerandtheschoolingeffecti.e.,thenumberofsemesters enrolledincommunitycollegeofabout3timeslargerthantheOLSestimates.To puttheseestimatesintoperspective,weneedtoknowatleasttwothingsaboutthese IVestimates.First,wecannotdirectlycomparetheIVandOLSestimatessincethey measuredifferentthings.Ingeneral,theIVestimatesapplyonlytothegroupofstudents inthesamplewhosecollege-goingdecisionsareshiftedbythefractionofthecollege costImbensandAngrist,1994.Forexample,byassumingthatthestudentswho arepresentedwithalowfractionofthecollegecostcompletedAAdegreesandthe studentswhoarepresentedwithahighfractionofcollegecostdidnotcompletetheAA degree,thenthesheepskineffectappliesonlytothestudentswhocompletedtheir AAdegreesbecausetheyarepresentedwithalowfractionofthecollegecost.Thus, theeffectappliesonlytosomegroupsinthesamplewhoareshiftedbytheinstrument. Inthesample,however,thisgroupisverysmall.Recallthattheinstrumenthasa negativeeffectontheAAdegreecompletionrateofonlytwopercentagepoints.Inentire sample,theAAcompletionrateisonlyaboutfourpercentorabout23,457students. AdecreaseinthefractionofthecollegecostfromonetozerowillonlypushtheAA degreecompletionratebyatmost50percentorabout11,700students,whichisabig effectrelativetothefourpercentaverageAAdegreecompletionrateinthesample. Second,weneedtounderstandwhytheseestimatesareparticularlybigforthis smallgroupofstudentsinthesample.BecausetheIVestimatescanbeinterpretedas theweightedaverageofthecausaleffectoftheAAdegreecompletion,theremightbe groupsinthesamplewhosereturnstocompletinganAAdegreeareuniquelylarge. Forinstance,theresultsaboveconrmedthattheIVestimatesofthesheepskinand theschoolingeffectsareparticularlylargeinthesampleofwhites,studentswhodidnot 142

PAGE 143

participateinthefree/reducedlunchprogram,andstudentswhoreachedgradetenon time. Inthesample,thestudentswhocompletedanAAdegreearemostlywhites percent,studentswhodidnotparticipateinthefree/reducedlunchprogram percent,andthestudentswhoreached10thgradeontimepercent.Mostofthe variationsintheeffectoftheAAdegreecompletionandthenumberofsemesters enrolledincollegeareduetotheheterogeneouseffectsassociatedwiththesegroups. InthereducedformmodeloftheAAdegreecompletionandsemestersstudents enrolledincollege,theestimatedeffectsoftheinstrumentforthesegroupsarealso smallerrelativetotheestimatedeffectsintheversionthatusesthefullsample.There isatleastonereasonwhythesegroupofstudentschoosetocompleteanAAdegree regardlessofthehighcostofattendingacommunitycollege.Conditionalonability, studentsthatbelongtothesegroupsmighthavemoreaccesstocreditorhavemore familyresourcesavailabletothemtodefraythehighcostofattendingcommunity college. Theestimatesfromthereducedformmodelofearningsindicatethattheinstrument hasthelargesteffectontheearningsamongthesegroupsofstudents.Thereare reasonswhythesestudentswhowerepresentedalowfractionofcollegecostsearned morelaterinlife.Itispossiblethatthesestudentsdifferfromtheircounterpartsthatthey havemorefamilyresources,strongerattachmenttothelabormarket,higherlevelsof motivation,andhavemoresocialcapital.Thesefactorsprovidethemtheedgeinthe labormarket.Itispossiblethattheearningsopportunitiesofthesestudentsarebetter evenwithoutobtainingcollegeeducation.BecausewecaninterprettheIVestimates astheratiooftheestimatesfromthereducedformmodelsofearningsandAAdegree completionorthenumberofsemestersstudentswereenrolledincollege,wecan seethatthelargeIVestimateofthereturnstocompletinganAAdegreeandalsothe numberofsemestersenrolledincollegecanbeattributedtothesesmallfractionof 143

PAGE 144

studentswhohavebetterearningsprospectsandwhosedecisionstocompleteanAA degreeortoattendmoresemestersincollegearelesssensitivetothecollegecost. ItisalsointerestingtoknowwhethertheseIVestimatesofthereturnstocommunity collegeeducationareconsistentwiththeestimatesfoundintheliterature.Thereare afewresearchstudiesthatinvestigatetheeffectsofcommunitycollegeeducation, sub-baccalaureate,ortwo-yeardegreeontheeconomicoutcomesofthestudentssuch asearnings.Estimatesofthereturnstoeducationfromtheseresearchstudiesare oftengeneratedfromsurveydatasetsthatarettedinOLSregressionmodels,with testscoresandfamilybackgroundvariablesincludedtocontrolfortheabilitybias.Also, theuseofthelogearningsinthesestudiesmakestheestimatesfromthesestudies conditionalonthelaborforceparticipation.ThestudyofKaneandRouse1995isthe mostinuentialone.Conditionalontestscoresandfamilybackgroundcharacteristics, theyestimatedtheannualearningsdifferentialassociatedwiththecreditscompleted andtheearningsdifferentialassociatedwiththedegreesreceivedintwo-yearandfour yearcolleges.Theirestimatessuggestthatthereturnstothecreditscompletedranged fromfourtotenpercentoftheannualearningswhilethereturnstoassociatesinarts degreerangedfrom24to31percentoftheannualearnings.KaneandRouse1995 attemptedtousethedistanceofstudent'shighschoolfromthenearesttwo-yearand four-yearinstitutionaswellastuitionlevelsinthestateasinstrumentsforthecollege attendance.Marcotteetal.2005estimatedthemodelsthatyieldreturnsassociated withthenumberofyearsoffulltimeequivalentFTEpost-secondaryeducationamong studentswhodidnotreceivecredentialsandthereturnstocompletinganassociatesin artsdegreeamongstudentswhocompletedpost-secondaryeducation.Theestimated returnstoattendingatwo-yearcollegewithoutearningadegreeoracredentialisabout sixtoelevenpercentoftheannualearnings.Moreover,theirestimatesindicatethat thereturnstocompletinganassociatesinartsdegreeisabout14to39percentof theannualearnings.TherecentstudybyJepsen,TroskeandCoomes2009used 144

PAGE 145

theadministrativedatafromKentuckyCommunityandTechnicalCollegeSystem KCTCStoestimatethechangeintheearningsassociatedwithattendanceandthe degreesreceivedfromtheKCTCS.Byincludingthepersonxedeffectstocontrolfor thetimeinvariantmeasuresofinnateability,theirestimatesindicatethatthereturnsto associatesinartsdegreeisabout20to39percentofthequarterlyearnings. Thendingsinthesestudiesareconsistentwiththeresultsinthispaper.Tomake theresultsofthispapercomparablewiththeresultsofthestudiesmentionedabove, Iestimatedthereturnsassociatedwiththenumberofsemestersenrolledincollege andthereturnsassociatedwithcompletingassociatesinartsdegreesusingtheOLS regressionmodelsofthelogaverageannualtotalearningsthatincludetestscores, demographiccharacteristics,cohort,andindicatorsofnearestcommunitycollegeasthe controlvariables.Theestimatesindicatethatthereturnsassociatedwiththenumberof semestersenrolledisabout6.3percentoftheaverageannualtotalearningsandabout 22percentforcompletinganassociatesinartsdegree.Apparently,theseestimates arecomparableinmagnitudewiththeestimatesinthestudiesmentionedabove.In theIVregressionversionofthelogearnings,theestimatessuggestthatreturnstothe semestersenrolledisabout26percentwhilethereturnstoassociatesinartsdegree completionisabout400percent. 17 3.8Conclusion InthispaperIexploretheuseofthefractionofthecostofattendingcommunity collegeastheexogenoussourceofvariationoftwocommunitycollegeoutcomesthe numberofsemestersenrolledincollegeandtheassociatesinartsdegreecompletion. UsingrichadministrativedatainFloridaandthecommunitycollegecostdatafromthe IntegratedPostsecondaryEducationDataSystemIPEDS,theanalysisofaverage annualtotalearningsandthecommunitycollegeeducationoutcomessuggeststhata higherfractionofthecollegecostwilldecreaseboththenumberofsemestersenrolled andtheassociatesinartsdegreecompletionrates,andtheaverageannualtotal 145

PAGE 146

earningsarehigheramongstudentswhoenrolledmoresemestersincollegeand amongstudentswhocompletedanassociatesinartsdegree.Becauselivingexpenses accountforalargeportionofthecostofattendance,theresultsisconsistentwith studentswhodonotliveathome.Theevidencesupportsthattherearesomefractionof studentswhodonotliveathomewhileattendingcommunitycollege. WhiletheIVestimatesshowsuggestiveevidenceoflargepositivereturns,these estimatesarelocaltothecommunitycollegestudentswhosedecisionstoacquire communitycollegeeducationareaffectedbythefractionofthecollegecost.The estimatedreturnstocommunitycollegeoutcomesareconcentratedamongwhite students,studentswhodidnotparticipateinthefree/reducedlunchprogram,andthe studentswhoreached10thgradeontime.Criticsmightarguethatthesegroupof studentsmaynotbeonthemarginofdroppingoutofcollege.Thesecommunitycollege students,however,arestillmoresensitivetothecostofcollegeeducationcompared tothestudentsinthefouryearcolleges.Becausecommunitycollegestudentsare sensitivetothechangingcostofthecollegeeducation,ahighercostofcollegeeducation wouldlikelyhindertheirdesirestoattendandnishacollegedegree.WhileIhavenot comparedstudentsfromthecommunitycollegesandthefouryearcollegesinterms oftheirsensitivitytothecostofhighereducation,theevidenceinthispapersuggests thattherearesomegroupsofstudentswhoareaffectedbythehighcostofcommunity collegeeducation. Caveatsshouldbeappliedtothesendings.First,becausetheIVestimatesare localtothecommunitycollegestudentswhosecollegeeducationdecisionsareshifted bythefractionofthecollegecost,Icannotassessthereturnstocollegeeducationfor therestofthestudentsinthesample.Assuch,theanalysisandtheinterpretationof thendingsholdforstudentsinthesamplewhochoosetoattendcommunitycollege whenthefractionofthecollegecostislowandchoosenottoattendcollegewhen thefractionofthecollegecostishigh.Second,Icannotaddresstheadvantagesand 146

PAGE 147

disadvantagesofattendingcommunitycollegerelativetothetraditionalfour-year colleges.Theinstrumentusedinthispaperissetupwiththecommunitycollege studentsinmind.Theinstrumentisexpectedtoaffectonlythecommunitycollege-going decisionsandnottheuniversityattendancedecisions.Third,sinceexpensesonroom andboardaccountforalargeportionofthecostofattendance,theestimatesare consistentwithstudentswhodonotliveathome.Theevidenceshowsthatthere arestudentswhodonotliveathomewhileattendingcommunitycollege. 18 Despite thesecaveats,theseresultssuggestthatcommunitycollegecostsaffectcollege-going decisions,degreecompletion,andearnings. Promotingabetteraccesstocommunitycollegeeducationisonewaytonarrow theexistinggapinincomeinequality.Asoundpolicytoincreasecollegeattendanceand degreecompletionisneededtohelpstudentsespeciallythosefromthelowincomeand disadvantagedfamilieswhoarelesslikelytoattainabetterstandardofliving.Since gettingacollegeeducationisincreasinglybecomingaprerequisitetojoinatleastthe ranksofthemiddleclass,accesstocollegeeducationmustbegivenagreatdealof attentionandmustbeinthelistofthepolicymakers'agenda. 18 Theestimatedeffectsarereducedby50percentwhentheroomandboard expensesarenotincludedaspartofthecostofattendance. 147

PAGE 148

Table3-1.EstimatesoftheCostofAttendingCommunityCollegesinFlorida TuitionandFeesBooksHousingandTransportationOtherPersonalEffects CommunityCollege2003200420052006200320042005200620032004200520062003200420052006 Brevard1,3561,4281,5421,6267008008008006,0008,0008,0008,0002,0242,4242,4242,424 Broward1,5361,7551,5131,6759901,2968641,0008,7659,9289,5699,1474,6905,0084,5323,995 CentralFlorida1,3811,4501,5691,6319849841,2301,2305,1125,1125,5624,8963,2123,2123,3865,012 Chipola1,7551,8702,0202,1378008008008004,9744,9744,9744,9742,8962,8962,8963,642 DaytonaState1,7391,8252,0032,1034638008008007,67613,40013,40013,4002,2262,6242,6242,624 EdisonState1,7471,8331,9832,0916736966966963,6453,6453,6454,6803,0292,4872,4872,701 FloridaKeys1,9001,9802,0792,2502,1052,3002,3922,48813,59114,15714,72315,3129449831,9031,980 FloridaSC1,5101,6151,6381,7148008008008007,9507,9507,9508,2603,1103,1003,1003,590 GulfCoast1,6721,7501,9162,0069009009001,0356,1006,1507,2007,8502,3002,3502,4002,450 Hillsborough1,7471,8331,9832,3801,9502,0481,9501,9505,9646,2625,9645,9644,8605,1034,8605,475 IndianRiverSC1,6501,7401,6621,7569009001,1001,0005,4005,4005,7005,4008508501,0201,020 LakeCity1,7061,7901,8872,9798608778779005,5105,5185,5765,799445445450465 Lake-Sumter1,7311,7851,9322,0088008008508506,1926,1926,1926,1922,9752,9752,9952,995 MiamiDade1,5831,8752,0221,6551,3681,5001,5001,2006,9928,1008,10014,5104,4102,9702,9706,610 NorthFlorida1,6711,7401,8601,8607007008008005,4005,4005,4005,4002,5562,3002,3002,300 NorthwestFloridaSC1,5501,6291,6641,7631,4001,4751,5001,6509,0009,0009,12610,4264,5004,5002,6703,294 PalmBeachSC1,9502,0882,1002,3007508109001,0005,6506,2757,5008,0001,7502,0002,2502,500 Pasco-Hernando1,5611,6441,7221,9671,2001,2001,2001,2004,9504,9504,9504,9502,9712,9712,9712,971 PensacolaJC1,3371,4041,4961,5817757757751,1005,0855,0855,0855,0852,0862,0862,3752,375 PolkSC1,6931,7551,8411,8901,3491,4841,7101,7006,8007,6006,8006,4003,0003,0003,0003,000 St.JohnsRiver1,5851,7321,7321,9741,0361,0361,0361,0626,5776,5776,5776,7412,6752,6752,6752,745 SantaFe1,3381,4041,5211,6047007007007005,3845,3925,5606,0803,1442,9283,0482,880 SeminoleSC1,7471,9312,0812,1807751,2001,2001,2005,4006,0306,0306,0322,6102,6102,6102,610 SouthFlorida1,6951,8201,8922,0488128248408753,6293,6843,7573,7571,5231,5461,5771,577 St.Petersburg1,7471,8331,9832,0918001,0001,0001,1005,2605,2605,2605,4602,7802,7802,7803,080 Manatee1,7411,8331,9832,0911,0991,1271,1641,20210,08110,08110,41410,7583,6713,7633,8874,015 Tallahassee1,3201,4241,4761,6458008008008005,0005,0006,0006,0001,8001,8001,8001,800 Valencia1,7291,8001,9832,1001,0001,0001,0001,0008,8008,8008,8008,8002,3002,8002,8002,800 Note:DataistakenfromtheIntegratedPost-secondaryEducationSurvey. 148

PAGE 149

Table3-2.EstimatedTotalCost,AverageFinancialAid,andtheFractionofCost TotalCostofAttendanceTotalFinancialAidFractionofCost CommunityCollege200320042005200620032004200520062003200420052006 Brevard10,08012,65212,76612,8505,2094,0395,4294,7150.480.680.570.63 Broward15,98117,98716,47815,8174,9815,1484,9955,7050.690.710.700.64 CentralFlorida10,68910,75811,74712,7692,1244,0785,2802,8980.800.620.550.77 Chipola10,42510,54010,69011,5536,7687,1347,5647,9070.350.320.290.32 DaytonaState12,10418,64918,82718,9275,2145,7625,7186,0910.570.690.700.68 EdisonState9,0948,6618,81110,1685,8195,7765,7916,7100.360.330.340.34 FloridaKeys18,54019,42021,09722,0303,9212,8504,6634,6960.790.850.780.79 FloridaSC13,37013,46513,48814,3644,5204,8524,7685,3920.660.640.650.62 GulfCoast10,97211,15012,41613,3415,6096,0275,8266,3160.490.460.530.53 Hillsborough14,52115,24614,75715,7695,1345,0425,2415,3980.650.670.640.66 IndianRiverSC8,8008,8909,4829,1766,1628,3385,3162,6970.300.060.440.71 LakeCity8,5218,6308,79010,1435,8765,9235,5816,8050.310.310.370.33 Lake-Sumter11,69811,75211,96912,0455,4015,7554,8774,6400.540.510.590.61 MiamiDade14,35314,44514,59223,9755,6545,7315,7455,3110.610.600.610.78 NorthFlorida10,32710,14010,36010,3606,2004,7874,7682,4290.400.530.540.77 NorthwestFloridaSC16,45016,60414,96017,1331,4151,4611,6291,9470.910.910.890.89 PalmBeachSC10,10011,17312,75013,8002,8852,5003,6515,0860.710.780.710.63 Pasco-Hernando10,68210,76510,84311,0884,1884,3654,8656,6510.610.590.550.40 PensacolaJC9,2839,3509,73110,1415,5385,8566,1946,6390.400.370.360.35 PolkSC12,84213,83913,35112,9905,0555,4865,9786,3360.610.600.550.51 St.JohnsRiver11,87312,02012,02012,5226,3086,3766,9665,8440.470.470.420.53 SantaFe10,56610,42410,82911,2645,9215,8765,9726,2760.440.440.450.44 SeminoleSC10,53211,77111,92112,0225,3112,2865,0035,5270.500.810.580.54 SouthFlorida7,6597,8748,0668,2575,2095,5435,7846,2640.320.300.280.24 St.Petersburg10,58710,87311,02311,7312,9062,9763,0873,3460.730.730.720.71 Manatee16,59216,80417,44818,0665,5296,1146,4195,9410.670.640.630.67 Tallahassee8,9209,02410,07610,2452,0945,1655,1655,5880.770.430.490.45 Valencia13,82914,40014,58314,7005,5195,7145,8655,8940.600.600.600.60 Note:DataistakenfromtheIntegratedPost-secondaryEducationSurvey. 149

PAGE 150

Table3-3.SummaryStatistics CohortYear FullSample2000200120022003 PercentofTotalObservations21.8723.9626.5227.65 Gender% Male48.5047.7048.3048.8048.80 Female51.5052.3051.7051.2051.20 Race% White54.1056.6054.9053.1052.60 Black20.7020.3020.4021.2020.10 Hispanics20.2018.5019.8020.7021.20 Others5.004.604.905.006.10 Free/ReducedLunchStatus27.7026.8024.9027.4031.20 LimitedEnglishProcient18.6017.6018.5018.9018.90 Special8.707.608.308.909.50 Grade10onTime76.7077.5076.7075.6077.10 Age16.0416.0216.0416.0516.04 TotalCostofAttendance12,985.9012,214.8012,936.8012,984.0013,639.70 NetCostofAttendance7,680.407,118.807,707.407,595.308,182.50 FractionofCostNet/TotalCost0.560.560.560.550.57 Observations584,300127,745139,996154,980161,579 150

PAGE 151

Table3-4.NumberofSemestersEnrolledinCommunityCollege CohortYear FullSample2000200120022003 Mean1.752.572.141.570.94 Gender Male1.502.211.851.340.80 Female1.982.902.411.791.07 Race White1.902.702.281.721.06 Black1.372.101.711.210.71 Hispanics1.642.582.081.460.81 Free/ReducedLunchStatus Yes1.412.231.721.250.77 No1.882.702.281.691.02 LimitedEnglishProcient Yes1.562.451.971.370.76 No1.792.602.181.620.98 Special Yes1.111.751.411.010.59 No1.82.642.211.620.98 Grade10onTime Yes1.872.722.291.701.01 No1.332.061.631.150.70 Observations584,300127,745139,996154,980161,579 151

PAGE 152

Table3-5.AssociatesinArtsDegreeCompletionRates% CohortYear FullSample2000200120022003 Mean4.018.755.932.040.49 Gender Male3.107.124.641.440.39 Female4.8410.247.142.620.61 Race White5.0410.417.332.720.67 Black1.653.822.530.700.18 Hispanics3.358.235.101.490.31 Free/ReducedLunchStatus Yes2.456.233.611.020.29 No4.619.686.702.430.60 LimitedEnglishProcient Yes2.917.074.311.240.24 No4.269.116.302.230.56 Special Yes1.313.641.840.580.09 No4.279.176.302.190.54 Grade10onTime Yes4.529.696.662.370.57 No2.355.503.521.020.25 Observations584,300127,745139,996154,980161,579 152

PAGE 153

Table3-6.AverageAnnualTotalEarningsinYear2000$ CohortYear FullSample2000200120022003 Mean4,7958,1205,8023,8352,213 Gender Male4,8737,9315,6963,7622,195 Female4,7218,3265,9173,9122,233 Race White5,0588,4916,0463,9602,306 Black4,2767,1835,2153,5081,994 Hispanics4,8708,3676,0044,0342,319 Free/ReducedLunchStatus Yes4,6397,9125,7303,9152,267 No4,8558,1965,8263,8052,189 LimitedEnglishProcient Yes4,5327,6935,5543,7122,128 No4,8558,2115,8593,8642,233 Special Yes4,4887,7085,7733,7732,132 No4,8248,1535,8053,8412,222 Grade10onTime Yes4,8928,2835,9113,8972,250 No4,4767,5565,4453,6432,092 Observations584,300127,745139,996154,980161,579 Note:Theaverageannualtotalearningsiscalculatedasthemeanof the2ndto6thyearearningsafterhighschoolgraduation. 153

PAGE 154

Table3-7.BalancingTest FractionofCollegeCostCoefcients CovariateNoControlsControls Male[mean=0.48]-0.0170.000 std.error0.0070.000 t-stat-2.264-0.055 White[mean=0.54]-0.4370.000 std.error0.1460.000 t-stat-2.977-1.469 Black[mean=0.21]0.2450.000 std.error0.0640.000 t-stat3.826-0.942 Hispanic[mean=0.20]0.1780.000 std.error0.1210.000 t-stat1.470-1.311 LunchStatus[mean=0.28]0.045-0.001 std.error0.0600.002 t-stat0.757-0.434 LEP[mean=0.19]0.1760.003 std.error0.1040.002 t-stat1.6931.628 Special[mean=0.09]-0.0070.000 std.error0.0130.001 t-stat-0.504-0.366 Grade10onTime[mean=0.77]-0.017-0.001 std.error0.0190.001 t-stat-0.913-1.591 Age[mean=16.04]-0.0050.000 std.error0.0210.000 t-stat-0.253-1.016 AgeSquared[mean=257.90]-0.1370.002 std.error0.6780.002 t-stat-0.2031.014 Observations584,249584,249 Note:Thecontrolvariablesincludetestscores,indicatorsof gender,race,lunchstatus,limitedEnglishprociency,special student,reachedgrade10ontime,yeareffects,nearest communitycollege,andinteractionterms.Standarderrors arerobustandclusteredbyyearandnearestcommunity college. 154

PAGE 155

Table3-8.EstimatesoftheRegressionofFCATScoresonFractionofCost DependentVariable MathematicsReading 12 FractionofCollegeCost0.2070.188 std.error0.1340.150 t-stat1.5501.250 Mean14.86013.710 Observations584,249584,249 Note:Thecontrolvariablesincludetestscores,indicatorsof gender,race,lunchstatus,limitedEnglishprociency,special student,reachedgrade10ontime,yeareffects,nearest communitycollege,andinteractionterms.Standarderrors arerobustandclusteredbyyearandnearestcommunity college. 155

PAGE 156

Table3-9.EstimatesoftheRegressionofCurrentYearFractionofCost onPriorYearCountyandCommunityCollegeCharacteristics IndependentVariableEstimate Enrollmentheadcount0.034 std.error0.046 t-stat0.728 CCAnnualExpenditures0.034 std.error0.046 t-stat0.728 CCAnnualInstructionalCosts0.034 std.error0.046 t-stat0.728 CountyMeanAnnualEarnings5.180 std.error6.265 t-stat0.827 CountyMeanAnnualUnemploymentRate6.626 std.error46.201 t-stat0.143 Observations268 Note:Thecontrolvariablesincludetestscores,indicatorsof gender,race,lunchstatus,limitedEnglishprociency,special student,reachedgrade10ontime,yeareffects,nearest communitycollege,andinteractionterms.Standarderrors arerobustandclusteredbyyearandnearestcommunity college. 156

PAGE 157

Table3-10.RegressionEstimatesofCollegeOutcomesonTestScores EffectofMathonEffectofReadingon CCTermsAADegreeCCTermsAADegree 1234 Sub-group BelowMedianFractionofCost0.0320.0010.0320.001 std.error0.0110.0000.0080.000 t-stat3.0002.6403.9701.600 AboveMedianFractionofCost0.0230.0010.0210.000 std.error0.0080.0000.0060.000 t-stat2.9102.2503.7901.230 InteractionTerm FractionofCost Mathematics-0.020-0.001 std.error0.0380.002 t-stat-0.540-0.680 FractionofCost Reading-0.028-0.001 std.error0.0310.002 t-stat-0.910-0.800 Note:Mathandreadingscoresarestandardizedwithmean100andstandard deviation15.Thecontrolvariablesincludetestscores,indicatorsofgender, race,lunchstatus,limitedEnglishprociency,specialstudent,reachedgrade 10ontime,yeareffects,nearestcommunitycollege,andinteractionterms. Standarderrorsarerobustandclusteredbyyearandnearestcommunity college. 157

PAGE 158

Table3-11.FirstStageRegressionModelsReducedFormEstimates ReducedFormModelOLSEstimate EffectofFractionofCostonSemestersEnrolled-0.37 std.error0.07 t-stat-4.86 R-Square0.15 Mean1.75 StandardDeviation2.49 EffectofFractionofCostonAADegreeCompletion-0.02 std.Error0.00 t-stat-3.21 R-square0.10 Mean0.04 StandardDeviation0.20 EffectofFractionofCostonAverageAnnualTotalEarnings$-405.08 std.Error165.95 t-stat-2.44 R-square0.13 Mean4,794.80 StandardDeviation6,249.40 Observations584,249 Note:Thecontrolvariablesincludetestscores,indicatorsofgender,race, lunchstatus,limitedEnglishprociency,specialstudent,reachedgrade10 ontime,yeareffects,nearestcommunitycollege,andinteractionterms. Standarderrorsarerobustandclusteredbyyearandnearestcommunity college. 158

PAGE 159

Table3-12.FirstStageRegressionModelswithHeterogeneousEffects SubGroupCCTermsAADegreeObservations 123 White-0.335-0.018316,418 std.error0.0980.008 Black-0.253-0.009121,117 std.error0.0920.004 Hispanic-0.530-0.028117,977 std.error0.1420.010 Male-0.271-0.015283,010 std.error0.0850.006 Female-0.463-0.025301,290 std.error0.0780.007 Lunch-0.439-0.015162,220 std.error0.1410.006 NoLunch-0.352-0.023422,080 std.error0.0790.007 Grade10onTime-0.437-0.024448,166 std.error0.0730.007 Grade10notontime-0.127-0.006136,083 std.error0.1370.005 Note:Thecontrolvariablesincludetestscores,indicatorsofgender,race, lunchstatus,limitedEnglishprociency,specialstudent,reachedgrade10on time,yeareffects,nearestcommunitycollege,andinteractionterms.Standard errorsarerobustandclusteredbyyearandnearestcommunitycollege. 159

PAGE 160

Table3-13.InstrumentalVariablesEstimatesoftheReturnstoCommunityCollege EducationbasedontheFractionofCostofAttendingaNearbyCommunity College CommunityCollegeOutcomeOLSStructuralEstimatesIV 12 NumberofSemestersEnrolledinCollege323.621097.24 std.error8.37573.00 t-stat38.651.91 AssociatesinArtsDegree1,070.8020,419.80 std.error75.9810,183.50 t-stat14.092.01 Observations584,249584,249 Note:Thecontrolvariablesincludetestscores,indicatorsofgender,race,lunchstatus, limitedEnglishprociency,specialstudent,reachedgrade10ontime,yeareffects, nearestcommunitycollege,andinteractionterms.Standarderrorsarerobustand clusteredbyyearandnearestcommunitycollege. 160

PAGE 161

Table3-14.InstrumentalVariablesEstimateswithHeterogeneousEffects SubGroupCCTermsAADegreeObservations 123 White1,769.433,347.3316,418 std.error839.516,694.7 Black155.14,254.9121,117 std.error717.119,756.1 Hispanic148.62,853.0117,977 std.error447.18,411.0 Male1,843.534,248.2283,010 std.error1,186.619,907.9 Female675.612,661.0301,290 std.error341.46,593.5 Lunch642.618,223.7162,220 std.error601.416,608.7 NoLunch1,194.418,333.0422,080 std.error581.18,571.8 Grade10onTime975.117,704.6448,166 std.error494.68,910.1 Grade10notontime2,047.243,541.7136,083 std.error2,494.449,060.4 Note:EstimatesinYear2000$.Excludingthedemographicgroups usedintheestimationsample,thecontrolvariablesincludetest scores,indicatorsofgender,race,lunchstatus,limitedEnglish prociency,specialstudent,reachedgrade10ontime,yeareffects, nearestcommunitycollege,andinteractionterms.Standarderrorsare robustandclusteredbyyearandnearestcommunitycollege. 161

PAGE 162

Table3-15.RegressionEstimatesofSUSOutcomesontheFractionofCost DependentVariable AttendedSUSOnlySUSTermsBADegreeEverAttendedSUS 1234 FractionofCollegeCost0.0240.2350.0060.023 std.error0.0140.1130.0040.013 t-stat1.7602.0801.4901.750 Mean0.0911.0270.0140.196 Observations584,249584,249584,249584,249 Note:Thecontrolvariablesincludetestscores,indicatorsofgender,race, lunchstatus,limitedEnglishprociency,specialstudent,reachedgrade10on time,yeareffects,nearestcommunitycollege,andinteractionterms.Standard errorsarerobustandclusteredbyyearandnearestcommunitycollege. 162

PAGE 163

Table3-16.InstrumentalVariablesEstimatesoftheReturnstoCommunityCollege EducationbasedontheInteractionoftheFractionofCost,LunchStatus,and Black Model FirstStageStructuralIV 12 FractionofCollegeCost Black LunchStatus0.15 std.error0.07 t-stat2.14 FractionofCollegeCost-0.38-270.43 std.error0.03802.51 t-stat-13.59-0.34 NumberofSemestersEnrolledinCommunityCollege364.73 std.error2,117.60 t-stat0.17 ControlVariablesYesYes Observations584,249584,249 FractionofCollegeCost Black LunchStatus0.02 std.error0.00 t-stat3.61 FractionofCollegeCost-0.02-352.28 std.error0.00332.10 t-stat-9.48-1.06 AssociatesinArtsDegreeCompletion2,661.70 std.error14,534.20 t-stat0.18 ControlVariablesYesYes Observations584,249584,249 Note:TheIVEstimatesareinYear2000$.Thecontrolvariablesincludetest scores,indicatorsofgender,race,lunchstatus,limitedEnglishprociency, specialstudent,reachedgrade10ontime,yeareffects,nearestcommunity college,andinteractionterms.Standarderrorsarerobustandclusteredbyyear andnearestcommunitycollege. 163

PAGE 164

APPENDIXA THEROLEPLAYEDBYTHENUMBEROFPOSSIBLETESTATTEMPTS Considerastudentwhohasnarrowlyfailedthersttest.Thestudenthasto decidewhethertotakethetestagain.Supposethemaximumnumberofattemptsis2. Further,supposethereissomebenet B thesignallingvalueofhavinganassociates inartsdegreetopassingthetestandsomecost C oftakingitopportunitycostoftime spentinremediationclasses,etc..Then,theexpectedutilityoftakingthesecondtime is: E [ U take 2 j fail 1, s 1 =0 ] = P pass 2 j fail 1, s 1 =0 B )]TJ/F39 11.9552 Tf 11.955 0 Td [(C Hencewilltakeasecondtimeif: P pass 2 j fail 1, s 1 =0 > C B where P pass 2 j fail 1, s 1 =0 istheprobabilityofpassingthesecondtestconditional onjustfailingthersttest.Thisconditionalprobabilitywilldependonhowthetest scoresevolveovertimee.g.whetherscoresimproveasstudentsspendmoretimein remediationclasses,howreliableisthetest,etc.. Nowconsiderthesamechoicewhenthemaximumnumberoftestsis3.The expectedutilitybecomes: E [ U take 2 j fail 1, s 1 =0 ] = P pass 2 j fail 1, s 1 =0 B + [ 1 )]TJ/F39 11.9552 Tf 11.955 0 Td [(P pass 2 j fail 1, s 1 =0 ] V )]TJ/F39 11.9552 Tf 11.955 0 Td [(C where: V = E [ max f U take 3 j fail 1, fail 2, s 1 =0, s 2 U nottake 3 j fail 1, fail 2, s 1 =0, s 2 g ] istheoptionvalueoftakingthetestthesecondtime.Thisassumesthatifyoudonot takethetesta2 nd timeyoucannottakeitagain.Assumethat U nottake 3 j fail 1, fail 2, s 1 =0, s 2 =0 164

PAGE 165

andassumethat U take 3 j fail 1, fail 2, s 1 =0, s 2 = P take 3 j fail 1, fail 2, s 1 =0, s 2 B )]TJ/F39 11.9552 Tf 11.955 0 Td [(C Then, V = E [ max f E [ U take 3 j fail 1, fail 2, s 1 =0, s 2 ] ,0 g ] Theexpressionin fg isafunctionof s 2 ,whichwillbeknownatthepointatwhichthe studentdecideswhethertotakethetestforthe3 3 rd time.Whenthestudentisdeciding whethertotakethetestforthe2 nd time,however, s 2 isnotknown.Thestudentdoes knowhoweverthathewilltakethetestfora3 rd timeif E [ U take 3 j fail 1, fail 2, s 1 =0, s 2 ] > 0, whichwillbethecaseprovided s 2 > s 2 ,where s 2 isthevalueof s 2 suchthat E [ U take 3 j fail 1, fail 2, s 1 =0, s 2 ] =0 = P pass 3 j fail 1, fail 2, s 1 =0, s 2 = C B Soafterthestudenttakesthetestthersttime,hecangureoutthat: V = P s 2 > s 2 E s 2 > s 2 [ P pass 3 j fail 1, fail 2, s 1 =0, s 2 B )]TJ/F39 11.9552 Tf 11.955 0 Td [(C ] Thisexpression V determinesthedifferencebetween E [ U take 2 j fail 1, s 1 =0 ] when themaximumnumberoftestattemptsis2andwhenthemaximumnumberoftest attemptsis3.Intuitively,since P pass 2 j fail 1, s 1 =0 < 0.5 unlessthereissome learningand/orimprovementbetweentests, V shouldbeveryimportant.Wecan easilygeneralizethemodeltoallowlargenumberoftestattemptsandobtainthesame intuition. 165

PAGE 166

APPENDIXB PROPERTIESOFTHEGRAPHSINFIGURE2-1 Wemakeaseriesofargumentstoestablishthepropertiesofthegraphspresented inFigure2-1. BottomPanel FromEquation2-2andEquation2-3,effortischosentomaximizethevalueof stayinginschoolwhichwedenote V e t 1 : V e t 1 = RF [ e )]TJ/F22 11.9552 Tf 11.955 0 Td [( L )]TJ/F39 11.9552 Tf 11.955 0 Td [(t 1 ] )]TJ/F39 11.9552 Tf 13.151 8.088 Td [(ce 2 2 therst-orderconditionthattheoptimaleffortchoicesmustsatisyis: R c f [ e )]TJ/F22 11.9552 Tf 11.956 0 Td [( L )]TJ/F39 11.9552 Tf 11.956 0 Td [(t 1 ]= e Forany t 1 ,theleft-hand-sideofthisequationisapositivefunctionthattakesamaximum at e = L )]TJ/F39 11.9552 Tf 12.464 0 Td [(t 1 andissymmetricaroundthis.Thisfollowsfromthepropertiesof f .The right-handsideofthisequationisafunction g e forwhich g =0, g 0 > 0 and g 00 =0 Wenowestablishseveralfactsabout e t 1 : 1.Forsomevalueof t 1 whichwecall t M 1 ,wecanshowthat e M = L )]TJ/F39 11.9552 Tf 12.223 0 Td [(t M 1 isaunique solutiontothisequation.Fortheexistenceofsuchasolution,weonlyneedto showthatthereisa t M 1 suchthat f = c R L )]TJ/F39 11.9552 Tf 10.206 0 Td [(t M 1 .Suchavalueis t M 1 = L )]TJ/F40 7.9701 Tf 11.401 4.707 Td [(R c f Foruniqueness,weneedtoshowthatfor t 1 = t M 1 ,therecanbenoother e such that f e t M 1 = g e .Therecanbenoother e > e M ,since f e t M 1 < g e forall e > e M .Thisfollowsbecause f 0 < 0 and g 0 > 0 for e > e M .Therecanbenoother e < e M ,since f e a M > g e forall e > e M .Thisfollowsbecause f a M > 0 and f 00 < 0 and g =0 and g 00 =0 and f and g arecontinuous. 2.For t M 1 < t 1 L ,wecanshowthatthereisauniquevalueof e thatsolvesthis rst-ordercondition,where L )]TJ/F39 11.9552 Tf 12.174 0 Td [(t 1 < e t 1 < e M .Wemusthave e < e M because f e M t 1 < f e M t M 1 = f = g e M .Hence f e M t 1 < g e M andweknow that f 0 < 0 and g 0 > 0 for e > e M sothatno e > e M cansolvethisequation. Wemusthave e > L )]TJ/F39 11.9552 Tf 12.532 0 Td [(t 1 ,since f L )]TJ/F39 11.9552 Tf 12.532 0 Td [(t 1 t 1 = f = g e M .Since t 1 > t M 1 g L )]TJ/F39 11.9552 Tf 12.329 0 Td [(t 1 < g e M ,hence f L )]TJ/F39 11.9552 Tf 12.329 0 Td [(t 1 t 1 > g L )]TJ/F39 11.9552 Tf 12.329 0 Td [(t 1 andweknowthat f 00 < 0 and g =0 and g 00 =0 and f and g arecontinuoussonovalueof e lowerthan L )]TJ/F39 11.9552 Tf 12.072 0 Td [(t 1 cansolvethisequation.Bythecontinuityof f and g ,itfollowsthataunique e solvesthisequation,where L )]TJ/F39 11.9552 Tf 11.955 0 Td [(t 1 < e t 1 < e M 166

PAGE 167

3.For t M 1 < t 1 L ,wecanshowthattheuniquevalueof e thatsolvesthisrst-order conditionisdecreasingin t 1 .Thisfollowsfromdifferentiationoftherst-order condition.Thisyields e t 1 = f 0 c R )]TJ/F40 7.9701 Tf 6.587 0 Td [(f 0 < 0 since f 0 < 0 for t M 1 < t 1 L and L )]TJ/F39 11.9552 Tf 11.955 0 Td [(t 1 < e t 1 < e M 4.For t < t 1 < t M 1 ,wecanshowthatthereisauniquevalueof e thatsolvesthis rst-ordercondition,where 0 < e a < e M .Wemusthave e < e M because f e M t 1 < f e M t M 1 = f = g e M asina.Wemusthave 0 < e t 1 because f > g =0 .Again,bycontinuityandthepropertiesof f and g ,itfollowsthat 0 < e t 1 < e M 5.For t < t 1 < t M 1 ,wecanshowthattheuniquevalueof e thatsolvesthisrst-order conditionisincreasingin t 1 .Thisfollowsfromdifferentiationoftherst-order condition.Thisyields e t 1 = f 0 c R )]TJ/F40 7.9701 Tf 6.587 0 Td [(f 0 > 0 since f 0 > 0 and c R )]TJ/F39 11.9552 Tf 11.986 0 Td [(f 0 > 0 for t < t 1 < t M 1 and 0 < e t 1 < e M 6.For t 1 = L ,denoteoptimaleffort e L .For t 1 = t ,denoteoptimaleffort e 0 .Itfollows that e 0 < e L provided f e L L )]TJ/F39 11.9552 Tf 11.956 0 Td [(t 1 < f e L ,0 .Weassumethisistrue. 7.From-andthecontinuityof f and g ,theoptimaleffortfunctioncanbe depictedasinthebottompanelofFigure2-1. MiddlePanel FromEquation2-2,theequilibriumpassprobabilityi.e.,thepassprobabilitygiven t 1 andoptimaleffortis: P t 1 = F [ e t 1 )]TJ/F22 11.9552 Tf 11.955 0 Td [( L )]TJ/F39 11.9552 Tf 11.955 0 Td [(t 1 ] where e t 1 isdenedinthebottompanelofFigure2-1.Notethat: 1.Itissimpletoshowthat P t M 1 =0.5 2.Since e t 1 > 0 for t < t 1 < t M 1 then P t 1 > 0 for t < t 1 < t M 1 3.For t M 1 < t 1 L f 0 < 0 hence j e a j = j f 0 c R )]TJ/F40 7.9701 Tf 6.586 0 Td [(f 0 j < 1 .Itfollowsthat P t 1 > 0 4.Itmustbethecasethat P L < 1 since F e L < 1 TopPanel Themaximizedvalueofstayinginschoolcanbewritten: V t 1 = RP t 1 )]TJ/F39 11.9552 Tf 13.151 8.088 Td [(ce t 1 2 2 167

PAGE 168

Fromtheenvelopetheorem,thismaximizedvalueisincreasingin t 1 ,.decreasingin L increasingin R anddecreasingin c .Thatfollowsfromthedifferentiationof V t 1 with respecttothoseparameterstreatingeffortasxed.Intuitively,thevalueofstayingin schoolincreaseswithabilityrelativetothestandardandwiththereturntograduation relativetothecostofgraduation. 168

PAGE 169

REFERENCES Betts,JulianR. 1998.TheImpactofEducationalStandardsontheLeveland DistributionofEarnings. AmericanEconomicReview ,88:266. Cameron,StephenV.,andJamesJ.Heckman. 1998.LifeCycleSchoolingand DynamicSelectionBias:ModelsandEvidenceforFiveCohortsofAmericanMales. JournalofPoliticalEconomy ,106:262. Cameron,StephenV.,andJamesJ.Heckman. 1999.CanTuitionPolicyCombat RisingWageInequality?In FinancingCollegeTuition .,ed.InMarvinKosters, 76.Washington,DC:AEIPress. Card,David. 1995.UsingGeographicVariationinCollegeProximitytoEstimatethe ReturntoSchooling.In AspectsofLaborMarketBehaviour:EssaysinHonourof JohnVanderkamp .,ed.E.K.GrantInL.N.ChristodesandR.Swidinsky.Toronto, Canada:UniversityofTorontoPress. Card,David. 1999.TheCausalEffectofEducationonEarnings.In HandbookofLabor Economics .Vol.3of HandbookofLaborEconomics ,,ed.O.AshenfelterandD.Card, Chapter30,1801.Elsevier. Card,David. 2001.EstimatingtheReturntoSchooling:ProgressonSomePersistent EconometricProblems. Econometrica ,69:1127. Carneiro,Pedro,andJamesJ.Heckman. 2002.TheEvidenceonCreditConstraints inPost-secondarySchooling. EconomicJournal ,112:705. Cohen,ArthurM.,andFlorenceB.Brawer. 2003. TheAmericanCommunityCollege. .4thed.,SanFrancisco,CA:Jossey-Bass. Costrell,RobertM. 1994.ASimpleModelofEducationalStandards. American EconomicReview ,84:956. Davenport,ErnestC.,MarkDavison,NohoonKwak,MargaretIrish,andChiKeungChan. 2002.MinnesotaHighStakesHighSchoolGraduationTestand CompletionStatusfortheClassof2000.OfceofEducationalAccountability,College ofEducationandHumanDevelopment,UniversityofMinnesota,Minneapolis,MN. Dee,ThomasS.,andBrianA.Jacob. 2006.DoHighSchoolExitExamsInuence EducationalAttainmentorLaborMarketPerformance?NationalBureauofEconomic ResearchWorkingPaper12199,Cambridge,MA. Dee,ThomasS.,andBrianA.Jacob. 2007.DoHighSchoolExitExamsInuence EducationalAttainmentorLaborMarketPerformance?In Standards-BasedReform andChildreninPoverty:EvidencefromNoChildLeftBehind .,ed.A.Gamoran. Washington,D.C.:BrookingsInstitution. 169

PAGE 170

DesJardins,StephenL.,DennisA.Ahlburg,andBrianP.McCall. 2002.Simulating theLongitudinalEffectsofChangesinFinancialAidonStudentDeparturefrom College. JournalofHumanResources ,37:653. Dougherty,Kevin. 1987.TheEffectsofCommunityColleges:AidorHindranceto SocioeconomicAttainment? SociologyofEducation ,60:86. Dynarski,SusanM. 2003.DoesAidMatter?MeasuringtheEffectofStudentAidon CollegeAttendanceandCompletion. AmericanEconomicReview ,93:279. Dynarski,SusanM. 2008.BuildingtheStockofCollege-EducatedLabor. Journalof HumanResources ,43:576. Ellwood,DavidT.,andThomasJ.Kane. 2000.WhoisGettingaCollegeEducation? FamilyBackgroundandtheGrowingGapsinEnrollment.In SecuringtheFuture .,ed. InS.DanzigerandJ.Waldfogel.NewYork:RussellSageFoundation. Fischhoff,Baruch,AndrewM.Parker,WandiBruinedeBruin,JulieDowns,Claire Palmgren,RobynDawes,andCharlesF.Manski. 2000.TeenExpectationsfor SignicantLifeEvents. PublicOpinionQuarterly ,64. FloridaDepartmentofEducation. 2005 a .Assistanceto11thand12thGrade StudentsthatHaveNotEarnedPassingScoresonGrade10FCATfor2004-2005. Availableonline. FloridaDepartmentofEducation. 2005 b .FCATHandbook-AResourcefor Educators.Availableonline. Greene,JayP.,andMarcusA.Winters. 2009.TheeffectsofexemptionstoFlorida's test-basedpromotionpolicy:Whoisretained?:Whobenetsacademically? EconomicsofEducationReview ,28:135. Grubb,W.Norton. 1993.TheLong-RunEffectsofProprietarySchoolsonWagesand Earnings:ImplicationsforFederalPolicy. EducationalEvaluationandPolicyAnalysis 15:17. Grubb,W.Norton. 1997.Thereturnstoeducationinthesub-baccalaureatelabor market,1984-1990. EconomicsofEducationReview ,16:231. Grubb,W.Norton. 2002.Learningandearninginthemiddle,partI:nationalstudiesof pre-baccalaureateeducation. EconomicsofEducationReview ,21:299. Imbens,Guido,andJoshuaAngrist. 1994.IdenticationandEstimationofLocal AverageTreatmentEffects. Econometrica ,62:467. Imbens,GuidoW.,andThomasLemieux. 2008.Regressiondiscontinuitydesigns:A guidetopractice. JournalofEconometrics ,142:615. 170

PAGE 171

Jacobson,Louis,RobertLaLonde,andDanielG.Sullivan. 2005.Estimatingthe returnstocommunitycollegeschoolingfordisplacedworkers. JournalofEconometrics ,125-2:271. Jepsen,Christopher,KennethTroske,andPaulCoomes. 2009.TheLabor-Market ReturnstoCommunityCollegeDegrees,Diplomas,andCerticates.University ofKentuckyCenterforPovertyResearchDiscussionPaperSeries,Universityof Kentucky. Kane,ThomasJ. 1994.CollegeEntrybyBlackssince1970:TheRoleofCollege Costs,FamilyBackground,andtheReturnstoEducation. JournalofPoliticalEconomy ,102:878. Kane,ThomasJ. 1995.RisingPublicCollegeTuitionandCollegeEntry:HowWell DoPublicSubsidiesPromoteAccesstoCollege?NationalBureauofEconomic ResearchWorkingPaper5164. Kane,ThomasJ. 1999. ThePriceofAdmission:RethinkingHowAmericansPayfor College. Washington,DC:BrookingsInstitution. Kane,ThomasJ.,andCeciliaElenaRouse. 1995.Labor-MarketReturnstoTwo-and Four-YearCollege. AmericanEconomicReview ,85:600. Kane,ThomasJ.,andCeciliaElenaRouse. 1999.TheCommunityCollege: EducatingStudentsattheMarginbetweenCollegeandWork. JournalofEconomic Perspectives ,13:63. Lee,DavidS.,andThomasLemieux. 2009.RegressionDiscontinuityDesigns inEconomics.NationalBureauofEconomicResearchWorkingPaper14723, Cambridge,MA. Leigh,DuaneE.,andAndrewM.Gill. 1997.LaborMarketReturnstoCommunity Colleges:EvidenceforReturningAdults. JournalofHumanResources 32:334. Leppel,Karen. 2002.SimilaritiesandDifferencesinCollegePersistenceofMenand Women. TheReviewofHigherEducation ,25:433. Leslie,LarryL.,andPaulT.Brinkman. 1988. TheEconomicValueofHigherEducation. .1sted.,NewYork:CollierMacmillanPublishers. Leuven,Edwin,HesselOosterbeek,andBasvanderKlaauw. 2010.TheEffect ofFinancialRewardsonStudents'Achievement:EvidencefromaRandomized Experiment. JournaloftheEuropeanEconomicAssociation ,8:1243. Long,BridgetTerry. 2004.DoestheFormatofaFinancialAidProgramMatter?The EffectofStateIn-KindTuitionSubsidies. TheReviewofEconomicsandStatistics 86:767. 171

PAGE 172

Manski,CharlesF. 1989.Schoolingasexperimentation:areappraisalof thepostsecondarydropoutphenomenon. EconomicsofEducationReview 8:305. Marcotte,DaveE.,ThomasBailey,CareyBorkoski,andGregS.Kienzl. 2005.The ReturnsofaCommunityCollegeEducation:EvidencefromtheNationalEducation LongitudinalSurvey. EducationalEvaluationandPolicyAnalysis ,27:157. Martorell,Francisco. 2005.DoHighSchoolGraduationExamsMatter?ARegression DiscontinuityApproach.PhDdiss.UniversityofCalifornia,Berkeley,CA,Unpublished Dissertation. Martorell,Francisco. 2010.TheImpactofHighSchoolExitExamsonHighSchool GraduationandPersistence. mimeo ,RAND. Martorell,Francisco,andDamonClark. 2010.TheSignallingValueofHighSchool Diploma.PrincetonUniversity,IndustrialRelationsSectionWorkingPaper557. McCrary,Justin. 2008.ManipulationoftheRunningVariableinRegression DiscontinuityDesign:ADensityTest. JournalofEconometrics ,142:698. NationalCenterforEducationStatistics. 2010.DigestofEducationStatisticsAnnual Report.Washington,DC. NationalCenterforEducationStatistics. 2011.TheConditionofEducationAnnual Report.Washington,DC. OfceoftheProgramPolicyAnalysisandGovernmentAccountability. 2003. CollegeAttendanceCostsVaryandResultFromHigherTuition,Room,andBoard. OPPAGAInformationBriefReportNo.03-33. OfceoftheProgramPolicyAnalysisandGovernmentAccountability. 2007. HighSchoolCompletionRateIncreasedSlightlyAfterFCATGraduationPolicy Implemented.OPPAGAInformationBriefReportNo.07-27. Ou,Dongshu. 2010.ToLeaveorNottoLeave?ARegressionDiscontinuityAnalysis oftheImpactofFailingtheHighSchoolExitExam. EconomicsofEducationReview 29:171. Papay,JohnP.,RichardJ.Murnane,andJohnB.Willett. 2008.Theconsequences ofHighSchoolExitExaminationsforStrugglingLow-IncomeUrbanStudents: EvidencefromMassachusetts.NationalBureauofEconomicResearchWorking Paper14186,Cambridge,MA. Perna,LauraW.,andChunyanLi. 2006.CollegeAffordability:ImplicationsforCollege Opportunity. JournalofStudentFinancialAid ,36. 172

PAGE 173

Reardon,SeanF.,AllisonAtteberry,NicoleArshan,andMichalKurlaender. 2009.EffectsoftheCaliforniaHighSchoolExitExamonStudentPersistence, Achievement,andGraduation.InstituteforResearchonEducationPolicyand PracticeWorkingPaper2009-12,StanfordUniversity,CA. Rouse,CeciliaElena. 1994.WhattodoafterHighSchool?TheTwo-Yearversus Four-YearCollegeEnrollmentDecision.In ChoicesandConsequences:ContemporaryPolicyIssuesinEducation .,ed.RonaldInEhrenberg,59.CornellUniversity Press. Rouse,CeciliaElena. 1995.DemocratizationorDiversion?TheEffectofCommunity CollegesonEducationalAttainment. JournalofBusinessEconomicStatistics 13:217. St.John,EdwardP. 1990.Priceresponseinenrollmentdecisions:Ananalysisof theHighSchoolandBeyondSophomorecohort. ResearchinHigherEducation 31:161. St.John,EdwardP.,andJohnnyB.Starkey. 1995.AnAlternativetoNetPrice: AssessingtheInuenceofPricesandSubsidiesonWithin-YearPersistence. The JournalofHigherEducation ,66:156. St.John,EdwardP.,andSandraC.Andrieu. 1995.TheInuenceofPriceSubsidies onWithin-YearPersistencebyGraduateStudents. HigherEducation ,29:143. Tyler,JohnH. 2004.BasicSkillsandtheEarningsofDropouts. EconomicsofEducationReview ,23:221. U.S.DepartmentofEducation. 2006. CommissiononFutureHigherEducationReport. Warren,JohnR.,KristaN.Jenkins,andRachaelB.Kulick. 2006.HighSchoolExit ExaminationsandState-LevelCompletionandGEDRates,1975Through2002. EducationalEvaluationandPolicyAnalysis ,28:131. 173

PAGE 174

BIOGRAPHICALSKETCH EdwardC.SeewasborninManila,Philippines.Hisparents,DoryandLuisSee, spentmostoftheircareersinbusiness.EdwardgraduatedwithhonorsfromDeLa SalleUniversityManilain1998withamajorinappliedeconomics.In2004,heearned aMasterofScienceinEconomicsdegreefromDeLaSalleUniversityManila.He continuedhisgraduateworkatMississippiStateUniversitywhereheearnedaMaster ofArtsinEconomicsdegreein2007.HereceivedhisPhDinEconomicsfromthe UniversityofFloridainthesummerof2012.Edward'sresearchinterestsincludepublic economicswithemphasisontheeconomicsofeducation,appliedeconometrics,labor economics,appliedmicroeconomics,andtheeconomicsofconsumptionandsaving. 174