ContentslistsavailableatScienceDirect
Journal of Economic Behavior & Organization
jo u r n al ho me p ag e :ww w . e l s e v i e r . c o m / l o c a t e / j e b o
Increasing perseverance in math: Evidence from a field experiment in Norway 夽
Eric Bettinger
a,c, Sten Ludvigsen
b,c, Mari Rege
c,∗, Ingeborg F. Solli
c, David Yeager
c,daStanfordUniversity,UnitedStates
bStenLudvigsen,UniversityofOslo,Norway
cUniversityofStavanger,Norway
dUniversityofTexas-Austin,UnitedStates
a rt i c l e i n f o
Articlehistory:
Received13December2016
Receivedinrevisedform18August2017 Accepted30November2017
Availableonline5December2017
Keywords:
Non-cognitiveskills Realeffortexperiment Growthmindset Perseverance Fieldexperiment Education
a b s t ra c t
Researchbypsychologistsandeconomistsdemonstratesthatmanynon-cognitiveskills aremalleableinbothchildrenandadolescents,butwehavelimitedknowledgeonwhat schoolscandotofostertheseskills.Inafieldexperimentrequiringrealeffort,weinvestigate howschoolscanincreasestudents’perseveranceinmathbyshapingstudents’beliefsin theirabilitiestolearn,aconceptreferredtobypsychologistsas“mindset.”Usingprotocols adaptedfrompsychology,weexperimentallymanipulatestudents’beliefsintheirability tolearn.Threeweeksafterourtreatment,wefindpersistenttreatmenteffectsonstudents’
perseveranceandacademicperformanceinmath.Wheninvestigatingsubsamples,wefind thatstudents,whopriortotheexperimenthadlessofabeliefintheirabilitytolearn, generatethetreatmenteffect.Thefindingssuggestthatalow-costinterventionfocusedon students’mindsetcanimprovestudents’engagementandperformance.
©2017ElsevierB.V.Allrightsreserved.
1. Introduction
Non-cognitiveskills,suchasself-controlandperseverance,predictsuccessineducationandinlabormarkets(Borghans etal.,2008;Heckmanetal.,2006;Robertsetal.,2007).Whileresearchersarestilltryingtounderstandthecausalmechanisms, thereisstrongevidencethatindividualswithhighernon-cognitiveskillsaremorelikelytograduatefromhighschool,have higherratesofcollegeattendanceandcompletion,higherwagesandbetteremployment,andevenbetterhealthoutcomes (Carneiroetal.,2007;Kautzetal.,2014).Moreover,researchbypsychologistsandeconomistsdemonstratesthatmany non-cognitiveskillsaremalleable(Alanetal.,2016;Durlaketal.,2011;Kautzetal.,2014).Still,however,wehavelimited knowledgeonwhatschoolscandotofostertheseskills.
Whilenon-cognitiveskillsmaybeacquiredthroughmotivationorself-regulationprogramsrelyingondirectinstruction andrepetitivepractice,meta-analysesfindthatthisapproachhasmixedresults.Itworksreasonablywellwithyoungchildren,
夽 WearegratefultoElinSvensenintheRogalandCountyschooldistrictwhohasfacilitatedtheRCTandgivenvaluablesuggestionsregardingintervention designandimplementation.WearealsogratefultoLeighLauritzenandNettop-UiSformakingthecomputerprogramforthefieldexperiment,andtothe principal,teachersandstudentsatthehighschoolatwhichthefieldexperimentwasimplemented.Finally,weacknowledgetheRCN(227004and260407) andtheNorwegianMinistryofEducationforfunding.
∗ Correspondingauthorat:UniversityofStavanger,4036Stavanger,Norway.
E-mailaddress:mari.rege@uis.no(M.Rege).
https://doi.org/10.1016/j.jebo.2017.11.032 0167-2681/©2017ElsevierB.V.Allrightsreserved.
butshowsweakornulleffectswithadolescents(Durlaketal.,2011;HeckmanandKautz,2013;Yeageretal.,inpress).An alternativeapproachtodevelopingnon-cognitiveskills,testedhere,focusesonstudents’beliefs.Ratherthanthinkingof perseveranceasafixedability,wethinkofperseveranceasaseriesofrepeatedchoices.Wedefineastudentwithhigh perseveranceassomeonewhoconsistentlychoosestoexerthigheffort–shestaysfocusedonchallengingtasks,works hard,anddoesnotgiveup.Asachoice,perseveranceshouldbesensitivetobeliefsaboutthepayoffofeffort.Therefore, changingbeliefsabouteffort–inawaythatlastsovertime–mightresultinchangesinperseveranceevenafterarelatively minimalbelief-changeintervention.
Anextensiveliteratureinpsychologyandlaboreconomicshasdemonstratedthatstudents’beliefs,suchasself-efficacy, locusofcontrolorgrit,predicttheirperseveranceandchoices(AlanandErtac,2015;Bandura,1977;Cobb-Clark,2015;Cobb- ClarkandSchurer,2013;ColemanandDeLeire,2003;Duckworthetal.,2007;DuckworthandSeligman,2005).Although informative,thisresearchiscorrelational.Ouridentificationstrategyistorandomlyassignstudentstoawell-validated psychologicalinterventionthatisknowntochangebeliefsabouteffort:agrowthmindsetintervention(Dweck2006;Yeager andDweck2012),alsoreferredtoasan“incrementaltheoryofintelligence”intervention(DweckandLeggett,1988).Growth mindsetinterventionsshapestudents’beliefsintheirabilitytolearn,andcauselastingimprovementsinschooloutcomes (Aronsonetal.,2002;Blackwelletal.,2007;Goodetal.,2003;Pauneskuetal.,2015;Yeageretal.,2016).Intheinterventions studentslearnaboutthebrain’spotentialtogrowandchangethroughdedication,hardworkonchallengingtasks,finding therightlearningstrategies,andseekingassistancefromothers.Thegrowthmindsetinterventionisdesignedtocounteract the“fixedmindset,”whichisthebeliefthatintelligenceortalentsarefixedtraits.Bycontrast,studentswitha“growth mindset”believethattheirabilitiescanbedeveloped.
Inafieldexperimentrequiringrealeffort,weinvestigatehowschoolscanincreasestudents’perseveranceinmathby shapingstudents’beliefsintheirabilitiestolearnorintheirpotentialtobenefitfromeffort.Werelyontheweb-based mindsetinterventioninYeageretal.(2016),whichwasbasedonpriorworkbyAronsonetal.(2002)andPauneskuetal.
(2015).WeadaptedtheinterventiontotheNorwegianlanguage,cultureandcontext.Thetreatmentconditionexposes studentstogrowthmindsetthroughonlinereadingandwritingexercises.Theseexercisesfocuson(1)howintellectual abilitiesaremalleableandaccordinglyhowthebraincangrowandchange;(2)howhardworkonchallengingexercises improvestheneuralconnectionsinone’sbrain;(3)howgrowthmindsetaidsincopingwithconfusionanddifficulty;and (4)howgrowthmindsetcanbeusedtostrivetowardpersonalgoals,givingmorepurposetoexerteffortindifficulttasks (Yeageretal.,2014).Thecontrolconditionhasanalogousactivities,whichteachstudentsfactsaboutmemoryandbrain functioning,butdoesnotaddressthemalleabilityofintellectualability.
Inthespringof2016,385Norwegianhighschoolstudentsparticipatedinthefieldexperiment.Duringnormalschool instructionaltime,weintroducedstudentstoawebsite.Eachstudentloggedinindividuallytothewebsiteonpersonal laptops.Oncestudentsloggedin,oursoftwarerandomlyassignedthemtoeitherthemindsetorcontrolconditions.The studentsworkedontheonlinereadingandwritingexercisesduringtwosessionsof45min,twoweeksapart.Inthethird session,wegavestudentsarealefforttask,designedtocaptureafamiliarschoolactivity,inwhichperseveranceiscriticalto learnandsucceed.Specifically,studentsreceivedaseriesof34multiple-choicealgebraquestionssequentially.Thealgebra questionswerechallenging,possiblymakingmanystudentsfrustratedandtemptedtogiveup.Wetoldthestudentsthat wewouldlikethemto“dotheirbest”andthatthey“maylearnsomethingfromworkingonthemathquestions.”Wealso explainedcarefullythatstudents’answerswouldbekeptconfidentialandthattheirperformancewouldnotaffecttheir grade.Asmanystudentsdidnothavetimetofinishallthequestions,or“gaveup”–clickingfastthroughthequestions– welookathowmanycorrectanswersstudentshadonthefirst10,20andall34questions.
Theexperimentalresultsdemonstratethattreatedstudentshavesignificantlymorecorrect answersonthefirst10 questionscomparedtostudentsinthecontrolgroup(19percentofthestandarddeviation),butthereisnosignificant differencebetweentreatedandcontrolwhenlookingatthefirst20orall34questions.Subsampleanalysisdemonstrates thatthetreatmenteffectisentirelydrivenbystudentswhoinitiallyhadafixedmindset.Forstudentwhoenteredthe experimentwithapre-existingfixedmindset,thetreatmenteffectislargeandsignificantforallthreeoutcomemeasures;
treatedstudentsscored35,34and29percentofastandarddeviationhigherthancontrolstudentsonthefirst10,20andall 34questions,respectively.Forstudentswhoalreadyhadagrowthmindsetpre-treatment,therearenosignificanttreatment effects.Ourdescriptivestatisticsdemonstratethatpriortotheexperiment,studentswithlowgradepointaverage(GPA) andstudentsinvocationaltrackshavelessofagrowthmindset.Investigatingthesesubsampleswefindlargeandsignificant treatmenteffects.Amongstudentsinvocationaltracks,thetreatmentincreasedthescoreonall34questionsby25percent ofastandarddeviation.Ourresultsareconsistentwiththehypothesisthatitispossibletoincreasestudents’perseverance inmathbyshapingtheirbeliefsintheirabilitiestolearn,inparticularamongstudentswhoinitiallyhadafixedmindset.
Thispaperrelatestoseveralstrandsofliteratureineconomics.First,asnotedabove,ourworkbuildsontheemerging literatureonnon-cognitiveskillsbyinvestigatingwhetherschoolscanfosterstudents’perseverancebyshapingtheirbeliefs intheirabilitiestolearn.Thisisimportantbecausethestudyoffersanidentificationstrategyfortestingtheeffectofnon- cognitiveskillsonlifeoutcomes.Second,ourworkalsorelatestorecentdevelopmentsinbehavioraleconomicsofeducation (Kochetal.,2015;Lavecchiaetal.,2014)whichattempttounderstandhowlow-costbehavioralorpsychologicalinter- ventionscanhelpstudentsbetterutilizethelearningopportunitiesalreadywithintheeducationalsystem(e.g.Bettinger etal.,2012;Carrelletal.,2016;CastlemanandPage,2015).Wecontributetothisliteraturebyinvestigatinghowabrief, low-costpsychologicalinterventioncanleadstudentstoincreasetheireffortinalearningtaskthreeweekslater.Third,our mindsetexperimentbuildsonotherbehavioraleconomicsexperimentsdesignedtounderstandindividual’smotivationand
performanceinrealeffortchoicesinthelaborfield(AzmatandIriberri,2010;Bradleretal.,2016;Erikssonetal.,2009;Koch etal.,2015;Kvaløyetal.,2015).
Finally,ourexperimentcontributestothepsychologicalliteratureonmindset.Thepsychologicalmechanismsformindset effectsareclearlydefinedandstronglyillustratedinlaboratoryexperiments:growthmindsetinterventionschangestudents’
effortbeliefs,theirattributions,andtheirgoals(Burnetteetal.,2013).YetasarguedbyWilsonandButtrick(2016)andMiller etal.(2017),thebehavioralmechanismsthatexplainhowachangeinbeliefscantranslateintoachangeingradesmonthsor yearslater,arenotfullydocumented.Thismakes“longlag”interventioneffectsseem“magical”(YeagerandWalton,2011).
Weshinesomelightintothisbehavioral“blackbox”,bydemonstratingthatthemindsetinterventionaffectsperseverance inarealeffortlearningtaskwhichtookplacethreeweeksaftertheintervention.Whileourstudyfocusesonmindset,it buildsupontheextantliteraturedemonstratingthatbehavioraltrainingcaninfluenceacademicoutcomes(e.g.Alanetal., 2016;DuckworthandSeligman,2005;Durlaketal.,2011).
2. Background
2.1. Sampleandinstitutionalcontext
TheparticipantsinourexperimentareNorwegianfirstyearhighschoolstudentsinRogalandCounty.WefocusonNorway forthreedistinctreasons.First,todate,theexistingresearchonmindsetfocusesontheUnitedStates.Usingasampleoutside theUnitedStatescanadvanceunderstandingofmindset.ThecapabilityoftheNorwegiangovernmenttotrackstudents throughouttheirlifetime,givesthepossibilitytomeasurethelong-runimpactsofgrowthmindsets.Second,Norway,like manycountries,divideshighschoolstudentsintovocationalandacademictracks.Thistrackingisoftencorrelatedwithprior achievement,andasweshowbelow,thevocationalstudentsinoursamplehavesignificantlylowerincidenceofgrowth mindsets.Third,oursampleisalsodrawnoutofconvenience.MultiplecoauthorsliveintheRogalandregionofNorwayand haveexistingpartnershipsinimplementingeducationalinterventions.Atthetimeoftheresearch,educationalpolicymakers throughoutNorwaywerediscussingtheacademicresearchongrowthmindsetanditsimplicationsforNorway.
InNorway,students starthighschoolaroundage16 aftertenyearsofcompulsoryschoolingsimilarforeverybody (primaryandmiddleschool).1Astudentcanapplytoanyhighschoolinhercounty.Whenapplyingastudenthastodecide whethertoenrollinavocationaltrack,whichleadsdirectlytoemployment,oranacademictrack,whichpreparesstudents toattendcollegeafterhighschoolcompletion.Studentsrankthreedesiredchoices,andacceptanceisbasedonstudents’
GPAfrommiddleschool.Allstudentsareguaranteedacceptanceintoahighschoolinthecounty.
Norwegianhighschooltypicallylastthreetofouryearsdependingonthespecificprogram.Only70percentofstudents completehighschoolwithinfiveyears.Forvocationaltrackstudents,completionratesareparticularlylow–only55percent completewithinfiveyears.WhileNorwegianpolicymakershaveaimedseveralreformsatimprovinghighschoolcompletion rates,therehavebeennosystematiceffortstoalterstudents’beliefsintheirabilitiestolearnasameansofimproving educationaloutcomes.
2.2. Conceptualframework
Wedefineastudentwithhighperseveranceassomeonewhoconsistentlyexertshigheffort–shestaysfocusedon task,workshardonchallengingyetpotentiallyrewardingtasks,anddoesnotgiveup.Belowwepresentastylizedmodel illustratinghowshapingstudents’beliefsintheirabilitiestolearncanaffecttheirperseverance.
Consideralearningsituation.Astudentchooseshowmuchefforttoexert.Bystayingfocusedontask,workinghardand notgivingup,thestudentcanbenefitfromlearning.However,theremaybeanopportunitycostofexertinghigheffort–for example,thestudentcannotcheckherphone,daydream,ortalktoaclassmate.
LetpB(e) representthestudent’sexpectedbenefitoflearningwhenexertingeffort,e,whereB>0,B<0andpisthe student’sbeliefinherabilitiestolearn,i.e.howefforttranslateintobenefitoflearning.LetC(e)representtheopportunity costofexertingeffort,whereC>0andC>0.Autilitymaximizingstudentchooseseffortlevele*suchthatC(e∗)=p·B(e∗).
Fromthisfirstordercondition,itfollowsthat dedp∗ >0andd2e∗
d2p <0.Thus,wehavethefollowingtwoconjectures:
Conjecture1. Itispossibletoincreaseastudent’seffortbyincreasingherbeliefinherabilitytolearn.
Conjecture2. Theeffort-effectofincreasingthestudent’sbeliefinherabilitiestolearnishigher,thelowerthestudent’s initialbelief.
2.3. Academicmindsetinterventions
Inpsychology,astudent’sbeliefinherabilitiestolearnisreferredtoasheracademicmindset(Dweck,2006).Students witha“fixedmindset”believetheirintelligenceortalentsarefixedtraits.Studiesusingsurveymeasuresofmindsetsand
1About9percentofstudentsinoursamplewereolderthan16,meaningthattheyeitherdelayedkindergartenentry,experiencedgraderepetitionat somepointintheiracademiccareer,or“tooktimeoff”aftercompulsoryschooling.
Fig.1. ContentofComputerProgram.
experimentalmanipulationsofmindsetsinlaboratoryandschoolsettingssuggestthatafixedmindsetshapesstudents’
academicachievementsinmanyways(Dweck2006;YeagerandDweck2012;YeagerandWalton,2011).First,students withafixedmindsetavoidacademicchallenges.Theywanteasierproblemsthatwillmakethemlookandfeelsmart(Mueller andDweck1998;Yeageretal.,2016).Second,afixedmindsetleadstounproductivebeliefsaboutefforts.Forexample,a studentwithafixedmindsetmightsay“IfIhavetotryhardatmath,I’mnotsmartatmath”(Blackwelletal.,2007).Last, fixed-mindsetstudentsarelessresilient.Instead,theyhidesetbacksanddeficiencies,notwantingpeopletoseethemas havinglowability.Theyfailtoaskforhelpandsometimesevenlieaboutlowscores(MuellerandDweck,1998).
Bycontrast,students with“growthmindsets”believethat intelligencecangrowand improveinresponsetoeffort, goodstrategies,andhelpfromothers.Fromthisperspective,anacademicchallengeisnotathreattoone’sability;itisan opportunityforlearningandimprovement.Inagrowthmindset,effortisagoodthing:astudentmightsay“tryingharder makesyousmarter”(Blackwelletal.,2007).Inthefaceofadifficultproblem,agrowthmindsetstudentismoreresilient, seeksappropriatehelp,orswitchesstrategies.Thestudentdoesnothideconfusion.Comparedtoafixedmindset,holdinga growthmindsetpredictsmorelearning,betterlearningstrategies,andhighergradesovertime,providedthattheschoolwork ischallenging(Blackwelletal.,2007;YeagerandDweck2012).
Thereissubstantialevidencesuggestingthatparentsandteacherssocializechildren’smindsetsthrougheverydaycom- munication(KaminsandDweck1999;MuellerandDweck1998;Rattanetal.,2015).Subtleverbalfeedbackfromadults canputchildreninafixedmindsetandundermineinternalmotivation.Thiscanhappenevenfromvaluedcaregiverstrying toencouragechildren.Forexample,aclassicpaperbyMuellerandDweck(1998)showedthatpraisingyoungadolescents fortheirintelligence–tellingthemtheywere“smart”whentheydidwell−createdafixedmindsetandunderminedtheir resilienceinthefaceoflaterstruggle.Incontrast,praisingstudents’“processes”(effortsorstrategies)putchildrenina growthmindsetandfosteredresilience.
Recentlyseveralstudieshavedemonstratedthatprecisetheory-basedinterventionscancommunicateagrowthmindset toyouthsandproducelastingimprovementsinstudents’grades(Aronsonetal.,2002;Blackwelletal.,2007;Goodetal., 2003;Pauneskuetal.,2015;Yeageretal.,2016).Theseinterventionsappealtoneuroscienceandevidenceonthemalleability ofthebrain.Tocommunicatethemalleabilityofintelligence,theseinterventionsusephysicalexerciseasametaphorfor growthmindset.Theinterventionsteachthestudentstothinkoftheirbrainsasmuscles,whichgetstrongerasoneexercises them.Theinterventiondepictsnewneuronalconnectionsgrowingasstudentscompletechallengingmathproblems.
3. Experimentaldesign 3.1. Interventionandmeasures
Wedevelopacomputerprogramwiththreeonlinesessions,eachlastingabout45min.Webasecontentandvisuallayout inSessions1and2ontheinterventioninYeageretal.(2016)(alsoseePauneskuetal.,2015).However,bythemeansofa professionaltranslatorandinterviewswithseveralfocusgroupsofNorwegianhighschoolstudents,wecarefullyadapted thematerialtotheNorwegianlanguage,cultureandcontext.Session3consistsofarealefforttaskinwhichthestudents havetosolveaseriesofalgebraquestions.Fig.1illustratesthecontentofthethreesessions.
InSession1studentsfirstanswersurveyquestionsdesignedtomeasurestudents’mindsetsatbaseline.2Inparticular, weaskhowmuch,onascalefrom1to6,thestudentdisagrees(1)oragrees(6)withthefollowingstatements(assigned variablenameinparenthesis):
2 Below,weonlylistsurveyquestionsusedforthispaper.Thestudentsreceivedothersurveyitems,designedtoanswerdifferentresearchquestions.
Allsurveyquestionswereidenticalfortreatedandcontrol.
Fig.2.ScreenShotfromComputerProgram.
•“Youhaveacertainamountofintelligence,andyoureallycan’tdomuchtochangeit”(FixedMindset1);
•“Yourintelligenceissomethingaboutyouthatyoucan’tchangeverymuch”(FixedMindset2);
•“Beinga‘mathperson’ornotissomethingthatyoureallycan’tchange.Somepeoplearegoodatmathandotherpeople aren’t”(FixedMindsetMath);and
•“Whenyouhavetotryreallyhardinasubjectinschool,itmeansyoucan’tbegoodatthatsubject”(FixedMindsetEffort).
Thesemindsetmeasureshavebeenusedandvalidatedinnumerousstudies,demonstratingthattheystronglypredict gradesandperformanceonbehavioraltasks(seee.g.Burnetteetal.,2013;Yeageretal.,2016)
Afterrespondingtothesurveyquestions,thestudentsreceivetheintervention.Thecomputerprogramrandomlyallocates studentstoeitherthetreatmentorcontrolconditions.Thetreatedstudentshavetodothreecognitiveexercises.First, studentshavetoreadanarticleaboutresearchinneurosciencethatdemonstratesthebrain’spotentialtogrowandchange, originallywrittenfortheexperimentinBlackwelletal.(2007)andsubstantiallyrevisedinYeageretal.(2016).Thearticle presentationrunsoverseveralscreens(oneofwhichappearsina singlescreenshotinFig.2)andhasastylizedvisual layoutwithillustrations.Italsousesthemetaphorthatthebrainislikeamusclethatgrowsinresponsetochallenging learningexperiences.Second,studentsareaskedtosummarizethearticleandexplainhowitsmessagerelatestotheirown lives.Linkinginformationtotheselfinthiswaymakesitmoreself-relevantandeasiertorecall(BowerandGilligan,1979;
HullemanandHarackiewicz,2009).Third,studentsareaskedwhatgrowthmindsetadvicetheymightgivetoafriendwho wasstrugglinginschool.Providingadvicetoothersisa“saying-is-believing”tactictoencouragestudentstointernalizethe ideasbyendorsingthemtosomeoneelse(seee.g.Aronsonetal.,2002).
Studentsinthecontrolcondition,likethoseinthetreatmentcondition,readabriefarticleaboutthebrainandanswer reflectivequestions.However,theydonotlearnaboutthebrain’smalleability.Instead,theylearnaboutbasicbrainfunctions andtheirlocalization,forexample,thekeyfunctionsassociatedwitheachcorticallobe.Theexperimentalconditionsare designedtolookverysimilartodiscouragestudentsfromcomparingtheirmaterials.Itinvolvesthesametypeofgraphic art–e.g.imagesofthebrain,animations−aswellascompellingstories.
InSession2,theinterventionusesinsightsfromthesocialpsychologyofattitudechangeandpersuasion,whatwecall
“supportivepsychologies”,toallowabrief,onlineinterventiontoachievelastingeffects.Thepurposeofthe“supportive psychologies”istocreateagestaltimpressionthatthegrowthmindsetmessageismemorable,credible,normal,andimpor- tant.Specifically,Session2repeatsthemetaphorfromSession1,thatthebrainislikeamusclethatgrowsinresponse tochallengingtasks(memorable);Itincludesquotesfromscientistsandcelebritieswhoendorsedthenotionthatthebrain developswhenitlearnsandexplainedhowastrongerbraincouldleadtoimprovementsinpersonalandsocialwelfare(cred- ible);Itleveragessocialnormsbyincludingquotationsfrompastparticipantsendorsingtheprogrammessages(Cialdiniand Goldstein,2004)(normal);Finally,Session2emphasizesapurposeforlearning(Yeageretal.,2014)byincludingprosocial
beyond-the-selfmotivesforadoptingandusingagrowthmindset(Grant2013;Yeageretal.,2016)(important).Forexample, onescreenintheinterventionreadsasfollows:“Peopletellusthattheyareexcitedtolearnaboutagrowthmindsetbecause ithelpsthemachievethegoalsthatmattertothemandtopeopletheycareabout.Theyusethemindsettolearninschool sotheycangivebacktothecommunityandmakeadifferenceintheworldlater.”
Inadditiontoreinforcingthebriefmindsetintervention,the“supportivepsychologies”areimportantforthefollowing reason:Thisstudyisoneofthefirstpapersineconomicsfocusingonmindset.Oneofourgoalsistoreplicatemindsetas truetotheoriginalpsychologyliteratureaspossible.Somepriorworkbyeconomists(e.g.Dee2014)hasfailedtoreplicate psychologicalinterventionspossiblyinpartbecauseofsubstantialdeviationsfromthepsychologicalintervention.Weare relyingontheexactscript(adaptedtoNorwegianlanguageandcontext)ofthelargestandmosteffectivemindsetstudyto date.
The“supportivepsychologies”dopresentsomeissuesintheinterpretationoftheresults.Inparticular,ourintervention inSession2couldaffectacademicperformancethroughsocialpressure:Theuseofcrediblerolemodelsandsocialnorms maycreatesocialpressureforstudentstoadoptgrowthmindsets.Weareunabletodisentanglewhethertheimpactsderive primarilyfromeducationaboutthemind(Session1)orfromthesocialpressureexertedinSession2.Bothsessionspoint studentstowardimprovinggrowthmindset,butwedonotidentifytherelativestrengthoftheeffectfromthetwosessions.
SimilartoSession1,thecontrolstudents’activityinSession2isdesignedtobeparalleltothetreatmentactivity.Students learnmoreaboutthebrain,butnotaboutitsmalleabilityandgrowthmindset.Ingeneral,wetookeveryprecautiontomake surethattherewasminimaltonocontaminationacrosstreatmentcategoriesduringimplementation.Ifsomecontamination occurredafterimplementation(forexampleifstudentstalkedtoeachotherabouttreatmentmaterial),thisislikelytobias ourestimatedimpactsdownward.
Aftertheinterventionmaterial,Session2providesallthestudentswiththesameseriesofsurveyquestions,measuring students’mindsets,asatthestartofSession1.Thereafterweincludeameasureofchallengeseekingwhichpriorresearch hasassociatedwithgrowthmindsets(Blackwelletal.,2007;MuellerandDweck1998).SimilartoYeageretal.(2016)we letstudentscreatetheirownmathworksheetwhichtheywillhavetoworkoninSession3.Studentscanpickfromeasy questionsfromwhichtheylikelywillnotlearnnewskills,orhardquestions,whichmayrequiremoreeffortbutprovide morelearningopportunities.Asmeasuresofstudents’challengeseeking,weusenumberofveryhardquestionsselected andnumberofveryhardorsomewhathardquestionsselected.ThequestionswereprovidedbyTheNorwegianDirectorate forEducationandTraining,andwecategorizedthemintoeasy,hardandveryhardbasedonpreviousstudents’scoreson eachquestion.
Finally,inSession3studentsfirsthavetosolvetworandomlydrawnquestionsfromtheworksheettheycreatedin Session2.3Aftertheworksheetquestions,thestudentshavetoparticipateinarealefforttask,consistingof34multiple choicealgebraquestions,givensequentially.4Thealgebraquestionswerechallenging,andseveralstudentsdidnothave timetoworkonallthequestions.Onaverage,thestudentsansweredcorrectly45,41,and37percentofthefirst10,20and 34questions,respectively.
Theon-screenintroductiontothealgebraquestionstellsthestudentsthattheywillbegivenaseriesofalgebraquestions, andthattheyshouldtrytodotheirbesttofindthecorrectanswer.Moreover,itexplainsthatstudentsmightlearnsomething fromworkingonthemathquestions,butthattheirperformancewillnotaffecttheirgrade.Asmanystudentsgaveuporran outoftimeonthelastquestions,weusenumberofcorrectquestionsonthefirst10,20,andall34questionsasmeasuresof students’effortonthealgebraquestions.
Notably,thestudentsdidnotknowthattheywouldreceivealgebraquestionsinSession3,sotherewasnowaytoprepare.
Moreover,algebrawasnotonthecurriculuminschoolbetweenSession1andSession3.Thus,ifwefindatreatmenteffect onthesemeasures,itisreasonabletointerpretitasaneffectofeffort(studentsaremorefocusedontask,workharderand donotgiveup)andnotthattreatedstudentshaveactuallybecomebetterinalgebra(althoughwecannotruleoutthatsome studentsmayhavebeensufficientlymotivatedtopursueoutsidelearningexperience).
Apossibleconcernwithourrealefforttaskisdemandinducedeffects,thatstudentsintheexperimentalcondition increaseefforttocomplywiththeexperimenters’wishes,andnotbecausetheyhavemoreofagrowthmindset.Tolimit thispossibility,weprovidenopromptsbeforepresentingthetestquestionsinSession3,andtheformatofthesessionhas minimalremindersofthepriorsessions:Experimenterswerenotpresentintheclassroom,andthegraphicsonthetest questionsdidnotreiteratetheexperiment.Furthermore,effectscausedbyachangeinmindsetareexpectedtodifferacross subsamples(seeHypothesis2below),whereastherearenoclearreasontoexpectdifferentialdemandinducedeffects.
Asasecondarymatter,ourdataalsoallowustoinvestigatetimespentoneachalgebraquestion.Ontheonehand,we couldimaginethatstudentswithmoreperseverancemanagetostaymorefocusedandworkharderandhenceareableto movefasterthroughtheproblems.Ontheotherhand,giventhedifficultyofthequestions,studentswithmoreperseverance mayhavespentmoretimetryingtosolveaquestionbeforegivingup.Theymayhaveelectedtotrydifferentapproaches
3 Wedidnotusethisdata,asselectioninSession2(treatmentaffectschallengeseeking)mayaffectperformanceonthesequestions.However,giving thestudentstimeinSession3,toworkonsomequestionsfromtheworksheet,isstillanimportantpartofthedesigntoavoiddeceptioninSession2.
4 Thefirst13questionswerethesameforallstudents.Thereafter,thecomputerprogramrandomlyassignedthestudentstooneofthreegroups,and eachgroupreceivedtheremainingalgebraquestionsindifferentorder.ThequestionswereselectedfromthepubliclyreleasedNAEPtestquestionsand theCaliforniaStandardsTestsquestions.
insteadofmakingarandomguessandmovingontothenextquestion.Assuch,wedonothaveaclearhypothesis,asto howourinterventionaffectedtimeuse.Wheninvestigatingtimespentonthefirst10,20or32min(resultsavailablefrom authorsonrequest),wefindnosignificanttreatmenteffectsneitheronthefullsamplenoronrelevantsubsamples.
3.2. Hypotheses
RecallfromthestylizedmodelinSection2.2,ifthetreatmentincreasesastudent’sbeliefinherabilitiestolearn,this increaseshermarginalbenefitofeffort,andleadstoanincreaseintheoptimaleffortlevelinalearningsituation(Conjecture 1).Assuch,wehypothesize:
Hypothesis1. Thetreatmenthasapositiveeffectonastudent’seffortintherealefforttaskinSession3.
Thestylizedmodelalsodemonstratedthatanincreaseinastudent’sbeliefinherabilitiestolearn,increasesoptimal effortlevelatadiminishingrate(Conjecture2).Assuch,wehypothesize:
Hypothesis2. Treatmenteffectsoneffortintherealefforttaskarelargerforstudentswhoinitiallyhavelowbeliefsin theirabilitiestolearn.
Hypothesis2isalsoconsistentwithseveralstudiesdemonstratingthatinitiallylowperformingstudentsbenefitmore frommindsetinterventions,andthatafixedmindsetismorecommonamongthesestudentspriortotheintervention (Pauneskuetal.,2015;Yeageretal.,2016).
4. Sampleandprocedures
Inthespringof2016,allfirstyearstudentsatapublichighschoolinruralNorwayparticipatedinthefieldexperiment.
Astheschoolservesalargeregion,itislargeandoffersbothavocationalandacademictrack.Participationwasmandatory aspartoftheschoolinstruction,butstudents hadtoconsenttoparticipateintheresearchproject.Whenthestudent loggedontothefirstsession,theyreceivedinformationabouttheresearchprojectandhadtomaketheirconsentdecision.
Wehad458studentsparticipateinthefirstsession,amongwhom385studentsconsentedtoparticipateintheresearch project.
Afterastudenthadmadetheconsentdecision,werandomlyassignedthestudenttoeitherthemindset(treatment)or controlcondition.Amongthestudentswhoconsented,22studentshadmissingregistrydataonmiddleschoolgrades.
Another9 studentswere olderthan20 years old.5 Droppingthesestudents fromoursample resultedin aSession 1 sampleof354students.AbsenceisamajorconcerninNorwegianhighschools,andweexperiencedsomeattritionin Sessions2and3.FromourSession1sample,289and254studentsparticipated inSessions2and3,respectively.Our balancetest(seeTable1)demonstratesthatattritionwasnotsignificantlycorrelatedwithtreatmentstatus.Notably,we collecteddataforstudentsinSession3eveniftheyhadnotparticipatedinSession2,i.e.receivedthetreatmentreinforce- ment.
Weimplementedallsessionsinthestudents’classroomduringschoolhours.Thestudentsusedtheirownlaptopcom- putersandheadset.6Atthebeginningofeachsession,wereadabriefscripttoallthestudents.Wetoldthestudentsthat theywereabouttologontoacomputerprogramdesignedtolearnaboutthebrainandreflectonlearning.Weaskedthe studentstoworkindependentlyandnottalktootherstudents.Wealsoemphasizedthatstudentsshoulddotheirbestand thattheiranswerswouldbekeptanonymousandnotaffecttheirgrades.Weassuredthemthattheirteacherorschool wouldneverseetheirindividualanswers.Finally,wetoldthestudentsthatthesessionwouldlastfor45minandprovided themwithlogoninformation.InSession1andSession2,membersofourresearchteamadministeredtheprotocolwiththe teacherpresentintheclassroom.InSession3,theteacherswereresponsibleforimplementation.Weprovidedascriptto them.Membersofourresearchteamwerestillpresentattheschoolincasetheteachershadanyquestionsortechnical challenges.Inallthreesessions,ifstudentsfinishedpriortothe45min,theywereaskedtoworkonotherschoolwork.There arefouropen-endedquestionsintreatedSession1.Thefrequencyofstudentsputtingeffortintotheopen-endedquestions suggestthatmoststudentspaidattentiontothetreatmentmaterial.About85percentansweredsubstantively,whichwas definedasanyattemptofasincereanswer.
Studentsloggedonwithauniquestudentnumberandpasswordassignedtoeachstudentbytheschooldistrictadminis- tration.Teacherswereunawareofstudents’treatmentstatus.Forthestudentswhoconsented,theschooldistrictprovidedus withregistrydatautilizingthesameuniquestudentnumber.Theschooldistrictde-identifiedthedatabeforetheyprovided thedatatoourteam.Fromtheregistrydata,weemploythefollowingvariables:GPAandmathgrade7frommiddleschool,
5Thisimplieslaggingbehindregularschoolprogressionbyatleastfouryears.
6Alaptopismandatoryforschoolwork.Allthestudentshaveschooldistrictsubsidizedlaptops.Wehadsomeextraheadsetstolendtostudentswho didnothaveheadset.
7Sixstudentswereregisteredwithoutamathgrade.Thesemissingobservationswerereplacedbypredictedvalues(predictionbasedonbaselinemindset measures,GPA,genderandvocationaltrack).Thecorrelationbetweenpredictedandobservedvaluesare0.72.Resultsarerobustwithandwithoutthese students.
Table1
DescriptiveStatisticsandBalanceTest.
Session1 Session2 Session3
Control Treatment Difference Control Treatment Difference Control Treatment Difference
(1) (2) (3) (4) (5) (6) (7) (8) (9)
GPA 0.083 −0.081 −0.165 0.165 0.092 −0.072 0.224 0.141 −0.083
(0.916) (1.072) (0.106) (0.809) (0.899) (0.101) (0.892) (1.029) (0.121)
Mathgrade 0.066 −0.064 −0.130 0.130 0.062 −0.067 0.172 0.141 −0.031
(0.970) (1.026) (0.106) (0.940) (1.005) (0.115) (0.956) (1.029) (0.125)
Vocationaltrack 0.583 0.664 0.082 0.566 0.645 0.080 0.489 0.561 0.073
(0.494) (0.473) (0.051) (0.497) (0.479) (0.057) (0.501) (0.498) (0.063)
Female 0.440 0.525 0.085 0.441 0.548 0.107+ 0.511 0.595 0.084
(0.497) (0.500) (0.053) (0.498) (0.499) (0.059) (0.511) (0.492) (0.062)
Olderthan16 0.091 0.078 −0.013 0.055 0.027 −0.027 0.075 0.057 −0.017
(0.289) (0.269) (0.030) (0.304) (0.164) (−0.023) (0.265) (0.234) (0.031)
FixedMindset1 0.049 −0.047 −0.096 0.008 −0.062 −0.071 0.036 −0.108 −0.144
(0.996) (1.004) (0.106) (0.977) .(983) (0.115) (0.988) (0.993) (0.124)
FixedMindset2 0.041 −0.040 −0.081 0.046 −0.050 −0.097 0.027 −0.047 −0.074
(0.993) (1.008) (0.106) (0.988) (0.988) (0.116) (0.957) (0.979) (0.121)
FixedMindsetMath 0.047 −0.045 −0.092 0.040 −0.109 −0.150 0.029 −0.034 −0.063
(1.013) (0.987) (0.106) (0.974) (0.968) (0.114) (0.968) (0.951) (0.120)
FixedMindsetEffort −0.035 0.034 0.070 −0.111 0.000 0.112 −0.090 0.044 0.135
(0.965) (1.034) (0.106) (0.885) (0.987) (0.110) (0.906) (1.004) (0.121)
BaselineGrowthMindset −0.038 0.036 0.075 0.001 0.079 0.078 −0.005 0.054 0.059
(.1.020) (0.982) (0.106) (0.968) (0.936) (0.112) (0.984) (0.957) (0.122)
N 175 179 354 145 144 289 133 121 254
Notes:Foreachsessionsample,columnsprovidethemean(andstandarddeviation)forthecontrolgroupandthetreatedgroup,andtheestimatedcoefficient (robuststandarderror)fromregressingeachcovariateagainsttreatmentstatus.+p<0.10,*p<0.05,**p<0.01.
Table2
CorrelationbetweenPre-TreatmentMindsetMeasures.
FixedMindset1 FixedMindset2 FixedMathMindset FixedEffortMindset
FixedMindset2 0.690**
FixedMathMindset 0.281** 0.425**
FixedEffortMindset 0.212** 0.322** 0.299**
BaselineGrowthMindset −0.765** −0.845** −0.696** −0.600**
Notes:+p<0.10,*p<0.05,**p<0.01.Session1sample(n=354).
highschooltrack(indicatorforvocational),gender(indicatorforfemale)andage(indicatorforlaggingbehindregularschool progression,i.e.beingolderthan16years).
5. Results
5.1. Balancetestanddescriptivestatistics
Table1presentsourdescriptivestatisticsandbalancetests.ThefourFixedMindsetmeasurescorrespondstothefour surveyquestionsasreportedinSection3.1.Thesevariablesarestandardizedwithmeanzeroandstandarddeviationone usingtheSession1sample,andapositivescoreindicatesafixedmindset.TheBaselineGrowthMindsetmeasureisthe meanofthefourfixedmindsetmetrics,wherethescaleisreversedandstandardized.Hence,apositivescoreindicatesa growthmindset.Fortheeaseofinterpretation,wehavealsostandardizedGPAandmathgradefrommiddleschoolusingthe Session1sample.Olderthan16isanindicatorforlaggingbehindregularschoolprogression.ThecolumnslabeledControl andTreatmentprovidesummarystatisticsforthecontrolandtreatmentgroupineachsession.IntheSession1control group,wecanseethatabout44percentofthestudentsarefemale;58percentareinthevocationaltrack;and9.1percent ofstudentsareoneortwoyearsolderthantheon-trackageof16.Allothercovariatesarestandardized.Whencomparing acrosssessions,thenumbersindicatethattheprobabilityofcompletingallsessionsaresomewhathigherforthosewith ahighGPA,beingonacademictrack,females,andhavingapre-interventiongrowthmindset.Weregresseachcovariate againsttreatmentstatus,andpresenttheresultingcoefficientandrobuststandarderrorincolumnslabeledDifference.We findthattherearesignificantlymorefemalesinthetreatedgroupinSession2(p<0.1),butallothercharacteristicsarewell balancedacrosstreatmentstatus.WeconcludethatrandomizationwassuccessfulandthatattritioninSessions2and3did notleadtosignificantdifferencesintreatmentstatus.
Table3
PredictorsofBaselineGrowthMindset.
(1) (2) (3) (4) (5) (6)
GPA 0.250** 0.139+
(0.052) (0.077)
Mathgrade 0.260** 0.179*
(0.051) (0.075)
Vocationaltrack −0.282* 0.053
(0.109) (0.129)
Female 0.151 0.082
(0.106) (0.105)
Olderthan16 −0.157 0.107
(0.191) (0.193)
R-squared 0.060 0.065 0.016 0.003 0.003 0.066
Notes:+p<0.10,*p<0.05,**p<0.01.Dependentvariable:BaselineGrowthMindset.Eachcolumnpresentsaseparateregressionandreportstheestimated coefficient(robuststandarderror)forallincludedcovariates.Session1sample(n=354).
Table4
TreatmentEffectonPost-TreatmentMindsetandChallengeSeekinginSession2.
Post-TreatmentGrowthMindset Choosing“VeryHard”
ChallengeQuestions
Choosing“Hard”or“Very Hard”ChallengeQuestions
(1) (2) (3) (4) (5) (6)
Treatment 0.557** 0.545** 0.234* 0.285** 0.240** 0.300**
(0.113) (0.090) (0.117) (0.110) (0.117) (0.111)
GPA 0.055 0.040 0.119
(0.085) (0.105) (0.105)
Mathgrade 0.228** 0.378** 0.407**
(0.068) (0.084) (0.084)
Vocationaltrack 0.006 0.159 −0.006
(0.114) (0.140) (0.140)
Female 0.002 −0.468** −0.417**
(0.092) (0.113) (0.114)
Olderthan16 0.323 −0.204 −0.094
(0.225) (0.277) (0.277)
BaselineGrowth 0.522** 0.045 0.020
Mindset (0.048) (0.059) (0.059)
R-squared 0.075 0.435 0.010 0.144 0.011 0.141
Notes:+p<0.10,*p<0.05,**p<0.01.Firstrowliststhedependentvariable.Eachcolumnpresentsaseparateregressionandreportstheestimatedcoefficient (robuststandarderror)forallincludedcovariates.Session2sample(n=289).Forcolumns(1)and(2):n=288.
InTable2wepresentacorrelationmatrixofourmindsetmeasurespresentedinSection3.1.FixedMindset18andFixed Mindset2representtwodifferentwordingsofthesamequestion,and itshouldnotbesurprisingthattheyarehighly correlated.Wecanalsoseethatthesedirectmeasuresofafixedmindsetarestronglycorrelatedwithourmeasuresof havingafixedmindsetwhenitcomestoMath(FixedMindsetMath)andEffort(FixedMindsetEffort).Thesearemoreindirect measureswhichmeasuretheconsequencesofhavinglessofagrowthmindset.ThroughoutwewilluseBaselineGrowthas ourpreferredmindsetmeasure.
InTable3weinvestigatehowourpre-treatmentcovariatespredictsagrowthmindset.WecanseeinColumns1–3 thatthepresenceofgrowthmindsetseemstobesignificantlymorelikelyforstudentswithhighGPA/mathgradeandfor studentswhodonotattendvocationaltracks.VocationaltracksinNorwaygenerallyincludestudentswithloweracademic credentials.InColumn5weaddallpredictorstothesamemodelandwefindthatthereisonlyasignificantrelationship betweenGPA/mathgradeandgrowthmindset.
5.2. Treatmenteffects
InTable4weinvestigatetreatmenteffectsonoutcomemeasuresgatheredattheendofSession2.First,weseeifthe treatmentaffectedthemeasureofgrowthmindset.Wegatheredthesamemeasuresofmindsetpost-treatment aswe didatbaseline.Post-TreatmentGrowthMindsetisthenconstructedidenticallytoourBaselineGrowthMindsetvariable,and standardizedontheSession2sample.WecanseefromColumn1thatthereisalargeandsignificanteffectoftreatmenton
8Onestudentdidnotrespondtothefirstfixedmindsetquestion.Thismissingobservationwasreplacedbythepredictedvalue(predictionbasedon baselinemindsetmeasures,GPA,genderandvocationaltrack).Thecorrelationbetweenpredictedandobservedvaluesare0.70.Resultsarerobustwith andwithoutthisstudent.
Table5
TreatmentEffectonEffortinSession3.
ScoreonFirst10Questions ScoreonFirst20Questions ScoreonAll34Questions
(1) (2) (3) (4) (5) (6)
Treatment 0.121 0.193* 0.057 0.134 0.012 0.089
(0.126) (0.089) (0.126) (0.085) (0.126) (0.086)
GPA −0.036 0.004 −0.028
(0.072) (0.068) (0.069)
Mathgrade 0.542** 0.582** 0.551**
(0.067) (0.063) (0.064)
Vocationaltrack −0.643** −0.542** −0.602**
(0.110) (0.103) (0.106)
Female −0.081 −0.134 −0.176*
(0.092) (0.086) (0.088)
Olderthan16 0.492** 0.464** 0.304+
(0.184) (0.172) (0.176)
BaselineGrowth 0.021 0.068 0.102*
Mindset (0.048) (0.045) (0.046)
R-squared 0.000 0.506 0.003 0.567 0.004 0.544
Notes:+p<0.10,*p<0.05,**p<0.01.Firstrowliststhedependentvariable.Eachcolumnpresentsaseparateregressionandreportstheestimatedcoefficient (robuststandarderror)forincludedcovariates.Allincludedcovariatesarereported,exceptfromincolumns2,4and6whereweadditionallycontrolfor questionassignmentorder(twoindicators).Session3sample(n=254).
growthmindset.Indeed,treatmentincreasesthescoreby56percentofastandarddeviation.InColumn2wecanseethat thisfindingisrobusttocontrollingforourpre-interventionvariables,includingbaselinegrowthmindset.
Incolumns3–6,weinvestigatehowtreatmentaffectedchallengeseekingwhenstudentshadtocreatetheirownmath worksheetinSession2.AsexplainedinSection3.1,challengeseekingismeasuredasthenumberof“veryhard”or“somewhat orveryhard”questionschosen.Thechallengeseekingmeasuresarestandardizedwithmeanzeroandstandarddeviation oneusingtheSession2-sample.InColumn3,weseethattreatedstudentschosesignificantlymore“veryhard”questions comparedtostudentsinthecontrolgroup.Wecanseethattheestimateisrobusttocontrollingforourpre-intervention variablesinColumn4.Itsuggeststhattreatmentincreaseschallengeseekingby29percentofastandarddeviation.In Columns5and6,weusenumberof“somewhathard”or“veryhard”questionsastheoutcomemeasure,andwegetasimilar treatmenteffectonthismeasure.
TheeffectsizesinTable4areslightlylargerthansimilareffectsizesobtainedinevaluationsofthesamematerials conductedintheUnitedStates.Arecentevaluationwith76highschoolsandover16,000studentsfoundaneffectonself- reportedmindsetequalto34percentofastandarddeviationandaneffectonchallenge-seekingof23percentofastandard deviation(Yeageretal.,2016).Anevaluationin11highschools with3276studentsshowedaneffectonself-reported mindsetof30percentofastandarddeviation(Pauneskuetal.,2015).Thepresenteffectsseemtoexceedtheeffectsizesin theUnitedStates,perhapsbecausestudentspaidgreaterattentiontothematerialsintheNorwegiancontext.
InTable5,weinvestigatehowtreatmentaffectedeffortinthealgebraquestionsinSession3.Sincesomestudentsdid nothavetimetofinishallthequestions,or“gaveup”−clickingfastthroughthequestions,9welookathowmanycorrect answersstudentshadonthefirst10,20andall34questions.Themeasuresarestandardizedwithmeanzeroandstandard deviationoneusingtheSession3sample.InColumn1,weseenosignificantdifferenceineffortonthefirst10questions.
However,whencontrollingforbaselinevariablesinColumn2,wecanseethattreatedstudentshavesignificantlymore correctanswersonthefirst10questionscomparedtostudentsinthecontrolgroup.ConsistentwithHypothesis1,the estimatesuggeststhattreatedstudentsscored19percentofastandarddeviationhigherthancontrolstudents.Therewere nosignificantdifferencesbetweentreatedandcontrolwheninvestigatingtreatmenteffectsonthefirst20orall34questions inColumns3–6,suggestingnostrongtreatmenteffectontheoverallsample.
InTable6weinvestigatetreatmenteffectsfordifferentsubsamples.First,wecharacterizethestudentstohaveeithera fixedoragrowthpre-interventionmindsetbysplittingthesampleatthemeanofourmindsetmeasure.Consistentwith Hypotheses2,PanelsAandBdemonstratethatthetreatmenteffectdetectedinColumn2ofTable5isentirelydrivenby studentswhoinitiallyhadafixedmindset.Forthesestudents,thetreatmenteffectislargeandsignificantforallthree outcomemeasures.Theestimatessuggestthattreatedstudentsscored35,34and29percentofastandarddeviationhigher thancontrolstudentsonthefirst10,20andall34questions,respectively.Forstudentswhoinitiallyhadagrowthmindset, thereisnosignificanttreatmenteffect;theestimatedcoefficientisevennegativeinallbutonecolumn.
IntheremainingpanelsofTable6weinvestigateifreadilyobservablevariablesfromregistrydatacanhelpusidentify studentsparticularlyresponsivetotreatment.RecallthatTable3demonstratedthat,priortotreatment,afixedmindsetis
9 Eighteenstudents(7percent)didnotfinishallquestions.Amongthe236studentswhofinishedallquestions,theaveragetimeperquestiondeclined.
Thefirst13questionstookjustunder50sperquestion;questions14–20tookonaverageabout40sperquestion;questions21–27tookabout30sper question,andstudentsspentjustover20sperquestionontheremainingquestions.
Table6
TreatmentEffectonEffortinSession3.Subsampleanalyses.
ScoreonFirst10Questions ScoreonFirst20Questions ScoreonAll34Questions
(1) (2) (3) (4) (5) (6)
PanelA:Pre-TreatmentFixedMindset(n=130)
Treatment 0.297+ 0.348* 0.299+ 0.335** 0.245 0.285*
(0.178) (0.135) (0.170) (0.117) (0.170) (0.116)
AdjR-squared 0.014 0.434 0.016 0.538 0.008 0.540
PanelB:Pre-TreatmentGrowthMindset(n=124)
Treatment −0.094 0.022 −0.227 −0.059 −0.266 −0.129
(0.165) (0.114) (0.172) (0.119) (0.169) (0.121)
AdjR-squared 0.005 0.553 0.006 0.563 0.012 0.530
PanelC:VocationalTrack(n=133)
Treatment 0.234+ 0.272* 0.217+ 0.253* 0.222+ 0.251*
(0.133) (0.128) (0.117) (0.109) (0.119) (0.114)
AdjR-squared 0.016 0.131 0.018 0.189 0.018 0.143
PanelD:AcademicTrack(n=121)
Treatment 0.172 0.088 0.056 −0.029 −0.043 −0.110
(0.162) (0.122) (0.178) (0.116) (0.173) (0.121)
AdjR-squared 0.001 0.449 0.008 0.584 0.008 0.524
PanelE:LowGPA(n=129)
Treatment 0.180 0.286* 0.156 .226* 0.121 0.191+
(0.135) (0.128) (0.115) (0.108) (0.114) (0.106)
AdjR-squared 0.006 0.146 0.007 0.165 0.001 0.171
PanelF:HighGPA(n=125)
Treatment 0.148 0.074 0.053 −0.030 −0.006 −0.078
(0.165) (0.125) (0.169) (0.123) (0.174) (0.131)
AdjR-squared 0.002 0.458 0.007 0.496 0.008 0.459
Controlvariablesincluded No Yes No Yes No Yes
Notes:+p<0.10,*p<0.05,**p<0.01.Firstrowliststhedependentvariable.Eachpanelrepresentsadifferentsample.Forgivensample,eachrowpresentsa separateregressionandreportstheestimatedtreatmentcoefficient(robuststandarderror)andadjustedR-square.Lastrowsindicatewhethertheregression includesthecontrolvariables.ThecontrolvariablesareGPA,mathgrade,vocationaltrack(indicator),female(indicator),olderthan16(indicator),and questionassignmentorder(twoindicators).
particularlyprevalentamongstudentswithlowGPAandstudentsinthevocationaltrack.Thisisalsoclearlycommunicated inthetabulationsinTablesA1andA2intheAppendixA.Thus,weinvestigatetreatmenteffectssplittingthesamplebased ontrackandGPA(atthemedian).10
InPanelC,wecanseethatamongstudentsinthevocationaltrackthereisalargeandsignificanttreatmenteffect.Looking atthefirst10,20andall34questions,thetreatmenteffectis27,25and25percentofastandarddeviation,respectively.
Studentsintheacademictrackscoredhigherthanotherstudents,andasweshowinPanelD,thereisnosignificanttreatment effectforthem.
InPanelEweinvestigatetreatmenteffectsforstudentswithalowpre-treatmentGPA.Weseethatthereisasignificant treatmenteffectonallthreeoutcomemeasures.PanelFdemonstratesthatthereisnosignificanttreatmenteffectonthe studentswithahighpre-treatmentGPA.
SinceourmostinterestingresultsareinthevocationaltrackandlowGPAsamples,AppendixATablesA3–A5providethe balancetests,inadditiontoeffectestimatesonSession2outcomes,forthesesubsamples.FromthebalancetestsinTables A3andA4wefindthatbothsubsamplesarewellbalancedacrosstreatmentstatusonallcharacteristics,withtheexception ofsignificantdifferencesinFemalesandOlderthan16inthevocationaltracksample,andinFemaleandVocationaltrackin thelowGPAsample.Hence,allourreportedestimatesarecontrolledforallobservables.TableA5demonstratesthatalsoin thesesubsamplestherearelargeandsignificanteffectsonthemeasuresofgrowthmindsetandchallengeseekinginSession 3,andthemagnitudesaresimilartothemagnitudesestimatedforthefullsample.
10ThecorrelationbetweenhavingahighGPAandchoosingtheacademictrackinhighschoolis0.56.Evenifthiscorrelationishigh,theoverlapisbyno meansperfect,andconsequentlyinformativetosplitthesamplebothbasedontrackandGPA.