Gender, Self-concept and Mathematics and Science Performance of South African Grade 9 Students DEBRA LYNNE SHEPHERD Stellenbosch Economic Working Papers: WP11/2017 www.ekon.sun.ac.za/wpapers/2017/wp112017 October 2017 KEYWORDS: gender inequality, STEM, stereotype, performance, self-concept, development JEL: C21, I21, I24, J16 ReSEP (Research on Socio-Economic Policy) http://resep.sun.ac.za DEPARTMENT OF ECONOMICS UNIVERSITY OF STELLENBOSCH SOUTH AFRICA A WORKING PAPER OF THE DEPARTMENT OF ECONOMICS AND THE BUREAU FOR ECONOMIC RESEARCH AT THE UNIVERSITY OF STELLENBOSCH www.ekon.sun.ac.za/wpapers
Transcript
Gender, Self-concept and Mathematics and SciencePerformance of South African Grade 9 Students
Despite improvementsover thepastdecade, SouthAfricanwomencontinue tobeunderrepresentedintertiarystudiesandprofessionalcareersinthefieldsofscience, technology, engineering and math. This has implications not only foreconomic development and growth, but also for social inequality as womencontinuetohaveloweraccesstohigherpayingemploymentopportunities.UsingdatafromtheTrendsinMathematicsandScienceStudyof2011,thispaperfindsthatwhilstgrade9girlsinthepoorest80%ofSouthAfricanschoolsexperiencenodifferenceindomainspecificperformance,self-conceptandmotivation,girlsin thewealthiest subset of schools are found to significantly underperform inbothsubjects,aswellaspossess lowerself-conceptandmotivation,andhigheranxiety.Teachergenderandeducationareshowntocorrelatewiththeseresults;specifically, female teacherswithmath backgrounds negatively influence girls’performancesinwealthyschools.Thisisarguedtobeinkeepingwithstereotypethreat theory (Steele, 2003) whereby women that are highly identified withmath are subject to greater anxiety and concern over their performance. Therelativedifferenceintheperformanceofgirlstaughtbyafemaleversusamaleteacher compared to the performance of boys is smaller when exposed toteacherswitheducationtraining,suggestingthatclassroommethodologyplaysarole in theattainmentof girls. Student fixedeffectsestimation reveals that theteachercharacteristicsmentionedaboveplayimportantrolesformoderatingtherelationshipbetweenstudentself-conceptandperformanceofboysandgirls.
1I gratefully acknowledge funding from the Growth and Economic Opportunities for Women(GrOW) initiative, a multi-funder partnership between the UK’s Department for InternationalDevelopment,theHewlettFoundationandtheInternationalDevelopmentResearchCentre.Ialso2Lecturer, Department of Economics, StellenboschUniversity,Private Bag X1, 7602, Matieland, South Africa. [email protected]
2
1. IntroductionMany high-paying, high-skilled occupations in developed countries are in thefields of science, technology, engineering and math (STEM), with the SouthAfricanlabourmarketbeingnoexception.This isnotsurprisingasscienceandtechnology have been the engines of profound social change and economicgrowth(AtkinsonandMayo,2010).Consequently,theycanplayasignificantrolein the economic empowerment of women. However, women continue to beunderrepresented inSTEMfields.Genderequity inSTEMeducationand jobs isneeded not only because the education of women in STEM fields would giveaccess to a larger number of highly-skilled (and productive) labour, therebymeetingthescientificandtechnologicalneedsofaneconomy,butalsobecausesocietycanbenefitfromtheinsightsandknowledgethatwomenpossess/bringtothemannerinwhichscienceandtechnologyareused.
At the level of tertiary education in South Africa, less than 30% ofstudents enrolled for undergraduate degrees in the fields of engineering andcomputer science in 2012 were women, whilst about half of scienceundergraduate enrolments were women; this is compared to 65% of allundergraduatestudentsbeingwomen.WhilsttherehasbeensomeimprovementintherepresentationofwomeninSTEMfields–forexample,theproportionoffemaleenrolmentsforengineeringandcomputerscienceundergraduatedegreesincreasedfrom16.4%to24.8%between2007and2012,agrowthrateof50%–women continue to be underrepresented, particularly atmasters and doctorallevels.Approximately35%ofallprofessionaloccupationsovertheperiod2000to2014havebeeninthefieldsofscienceandtechnology,withwomenholdingapproximately45%ofprofessionalscienceandtechnologyjobs(StatisticsSALF32000 – 2007; Statistics SA QLFS42008-2014). Removing nursing occupationsfromthenumerator,approximately25%to30%ofscienceandtechnology(non-nursing)jobsareheldbywomen,indicatingthatapproximately35%to40%ofprofessionalSTEMwomenareinnursing.ThisgendergapinSTEMoccupationscan contribute significantly to the overall gender wage gap as well as socialinequality(Barres,2006).
Oakes(1990)pointstowardsthreefactorsthatarecriticaltoattainmentinSTEM fields:opportunities to learnscienceandmathematics;achievement inthesesubjects;andthedecisiontopursuethem(choice).Womencanlosegroundonall threeof these factorsduring theireducation,suchthat theSTEMgendergapapparentinuniversityandthelabourmarketcanbeginatthestagesofbasiceducation,evenwhengirlsareperformingatthesamelevelorbetterthanboysinscienceandmathematics.Overall,genderdifferencesineithermathematicsorscience achievement tend not to appear until high school levels (c.f. Lindberg,
3LabourForceSurvey4QuarterlyLabourForceSurvey
3
Hyde, Petersen& Linn, 2010), although recent research has shown the gap toemergeasearlyasthegrades1and2(Levineetal,2005;Rathbunetal,2004).Educationalfoundations,whichincludesocialcomparisonsofperformancewithpeers, can have important implications for teacher expectations, studentmotivationandfutureacademicachievement(Farkas,2003).
ResearchongenderdifferencesinbrainstructureandinherentbiologicalabilityhasbeeninconclusiveinexplainingdifferencesingenderparticipationinSTEMfields(Ceci,Williams&Barnett,2009).Otherresearchhaspointedtotheexistenceofnegativestereotypeswithinsocietythataffectthemathandscienceperformance of female students through undermining their confidence (Hill,Corbett&Rose,2010;SteeleandAronson1995).Whilstchildrenmayviewboysandgirlsasequallyableinmathematics,theyviewadultmenasbeingbetteratmaths than adult women (Steele, 2003); similar tendencies to stereotype thedomainofmathematicsasamaleonehavebeenshowntoexistamongstparents(c.f.Frome&Eccles,1998)andteachers(c.f.Helwig,Anderson&Tindal,2001).
Itisalsoimportanttotakecognisanceofthefactthatgenderinequalitiesdonotoperatewithinavacuum,but interactwith, forexample, raceandclass.Whilst previous studies have shown that gender gaps in achievement interactwith race and SES (e.g. Jencks & Phillips, 1998; Lee & Burkam, 2002), limitedresearch(e.g.Fin& Ishak,2012;Goni&Bello,2016;Vyas&Choudhary,2016)has examined how gender differences in self-concept andmotivation differ bysocioeconomic status; and this despite the increasing cultural diversity of theschool-aged population. It is crucial that we understand how students ofdifferent genders and socio-economic backgrounds construct their academicidentities, particularly in the context of gender and cultural norms, aswell associalization experiences. Although numerous studies have investigated theeffectofvariousconventionalschoolinginputsoneducationaloutcomesinSouthAfrica, peer effects and (to a greater degree) school composition and socialcomparisonhavebeeninvestigatedtoalimitedextent.
Thisstudyaimstoanalysegenderdifferencesinself-concept,motivationand subsequent performance in mathematics and science amongst grade 9students in South Africa. To this end, the Trends in Mathematics and Sciencestudies (TIMSS) of 2011 is used to construct measures of self-concept andmotivation that are then regressed onto TIMSS mathematics and scienceperformanceusingclassroomandstudentfixedeffectsestimation.Fixedeffectsregression models are estimated separately for the wealthiest and poorersubsetsoftheSouthAfricanschoolingsysteminordertoidentifywhetherornotcompetencebeliefsandmotivation/attitudesplaydifferentrolesindeterminingperformance depending on the social context of the school environment, andwhether a gender gap in STEM achievement exists after controlling for thesefactors.Finally, theanalysis investigatestheroleof teachergenderandteacher
Thepaperproceedswithareviewofthetheoryandexistingliteratureongender,self-belief(efficacy)andmotivationinsection2,followedbyadiscussionofsocialcomparisontheoryandstereotypethreatinsection3.Sections4and5describe the data and the constructs to be used in the analysis and themethodological approach, respectively. Section 6 provides a discussion of theempiricalresults,andsection7concludes.
Self-beliefs such asself-conceptandself-efficacy have received prominence intheoriesaboutthemotivationalsourcesofindividualdifferencesinperformance,for example self-regulation theory (Carver & Schreier, 1981),social learning(cognitive)theory(Bandura,1986)andself-determinationtheory(Deci&Ryan,1985). The effect of self-concept on student achievement has received mostattention(Valentine,DuBois&Cooper,2004;Marsh&Craven,2006).Althoughknowntobedistinctconstructs,self-conceptandself-efficacyarerelatedbothintheir conception and in their effects on student achievement. Pivotal toBandura’s (1986) social cognitive theory is the concept of self-efficacy, whichrefersto“thebeliefsinone’scapabilitiestoorganizeandexecutethecoursesofactiontoproducegivenattainments”(Bandura,1997:3).Centraltothistheoryisthe reciprocity between self-efficacy beliefs, classroom structures and socialinteractions with peers, and the mediating effects that these can have on astudent’s performance and motivation. For example, 20 years of self-efficacyresearchhasshownthatpeoplewithaugmentedself-efficacybeliefstendtohavehigheracademicachievementthanthosewithlowerself-efficacybeliefs(Jinks&Morgan,1999;Pajares&Schunk,2001;Usher&Pajares,2008),aswellasbetterengagement(Schunk&Mullen,2012).
Althoughlinkedtojudgmentsofone’scapabilitiesinagivendomain,theconstructofself-conceptrepresentsaquitedifferentviewofselftoself-efficacy.Whereas self-efficacy beliefs revolve around questions of “can” (confidence),self-conceptbeliefs reflectquestionsof “being” (self-worth) (Pajares&Schunk,2001). Prior to the 1980s, over 90 percent of studies linking self-concept toachievement reportedmoderate toweak correlations (Byrne, 1984).However,this was because global rather than domain-specific measures of self-conceptwere being compared to achievement. When domain-specific (for example,academic)self-concept is comparedwithachievement in thesamedomain, therelationshipisevidencedtobepositiveandsignificant(Marsh,1993).Academicself-conceptcanfurthermorebedefinedas“anevaluativeself-perceptionthatis
5
formed through experience with, and interpretation of one’s schoolenvironment”(Guay,etal.,2004:53).5
Gender differences of mathematics self-concept have generally beenconsistent with traditional gender role expectations and stereotypes, withresearchshowinghighermathematicsself-conceptscoresformales(Ecclesetal,1993; Marsh & Yeung, 1997) and higher mathematics self-efficacy scores formales (Pajares & Miller, 1994; OECD, 2004). As early as school-entry age,children have been shown to make distinct judgments about their abilities indifferent domains (Eccles, Wigfield et al, 1993). These competency beliefsdecline over the course of schooling, although the rate of change differs bygender(Jacobsetal,2002)andethnicity(Okeke,2009).Higherself-perceptionofmathematics ability in boys has been found to be relatively independent ofperformancehistory(e.g.,Frome&Eccles,1998),achievementlevel,andability(Holling&Preckel,2005).2.2 Achievementmotivation
Self-DeterminationTheory (SDT;Deci andRyan, 1985) laid downmuch of thefoundationforcontemporaryresearch.Basedonthedifferentgoalsthatgiveriseto action, SDT distinguishes between different types of motivation, the mostbasic distinction being between intrinsic (doing something because it isinherently interesting or enjoyable) and extrinsic (doing something because itleads to an independent outcome, such as good performance). A natural(intrinsic)motivationaltendencyiscriticaltocognitiveandsocialdevelopment;however, intrinsic motivation does not only exist within individuals, but canexist in relation between individuals or in the activity itself (Skinner, 1953).Because most activities people do are not intrinsically motivated, such ascompletingnon-intrinsicallyinterestingtasksatschool,intrinsicmotivationdoesappeartobecomeweakerasindividualsmoveupgradelevels.
SDTsuggests thatextrinsicmotivationcanvaryaccording to its relativeautonomy.Forexample,achilddoingtheirmathhomeworkonlyoutof fearofthe parental sanctions that would result from not doing it is very differentlyextrinsicallymotivated to a child that believes that completing the samemathhomeworkisvaluableforgettingintouniversity.DeciandRyan(1985)describewithinSDTaprocessofpromotingtheinternalisationandintegrationofvaluesand behavioural regulations that motivate students value and self-regulateeducational activities.Whilst internalisation is the process of taking in a value(e.g. mathematics is necessary for getting into the university of your choice),integrationistheprocessbywhichindividualstransformthisvalue/regulation5The term ‘academic self-concept’ can be characterised by two elements consistent with theShavelson model: first, academic self-concept reflects descriptive (e.g., I like math) as well asevaluative(e.g.Iamgoodatmath)aspectsofself-belief;andsecond,self-beliefstendtofocusonachievementinschoolratherthanattitudes(Reyes,1984).
6
into their own (identification and integrated regulation). Increasedinternalisation of values and commitment to these values leads to greaterpersistenceandbetterengagement(Deci&Ryan,2000).Thiscanbecontrastedtoamovitation(lackofanyintentiontoactresultingfromnotvaluinganactivityand not feeling competent to do an activity), external regulation (activitiesperformed to satisfy an external demand) and introjected regulation (activitesperformedtoavoidanxietyortoattainego-enhancements).Studiesconcerningthe different types of extrinsicmotivation have shown thatmore autonomousextrinsic motivation is related to greater engagement, better performance(Miserandino, 1996), lower drop-out (Vallerand & Bissonnette, 1992), higherquality learning (Grolnick & Ryan, 1987) and greater psychologicalwell-being(Sheldon&Kasser,1995).
Studiesconductedsincetheearly1990shaverevealednoclearpatternofgender differences in students' achievement goal orientations andmotivation.Wheredifferenceshavebeenfound,theseareusuallymoderatedbyability,race,andclassroomcontext.Severalstudiesineconomicshaveexaminedtheeffectofhaving a female teacher on academicperformance, especially inmath, and thechoice of a math and science major of female students in high school (c.f.Ehrenberg,Goldhaber,andBrewer,1995;Dee,2005,2007;Wintersetal.,2013,Antecoletal,2014).Thesestudieshaveoftenfoundthathavingafemaleteacherhaspositiveornoeffectsonthetestscoresofgirls.Antecoletal(2014),ontheother hand, find that having a female teacher lowers themath scores of girlsattendingprimaryschools indisadvantagedneighbourhoods; this iseliminatedwithastrongermathbackgroundofthefemaleteacher.ThefindingsofAntecolet al (2014) are in agreement with research in the educational psychologyliterature that finds that primary school classroomswithmore anxious femalemath teachersaswell asagreaterendorsement for stereotypes that “boysaregoodsatmath”isrelatedtolowermathachievementsofgirlsrelativetoboys.Itis therefore important to understand the role that teacher gender plays as amechanism linking lower self-concept amongst girls to lower achievement inSTEMsubjects.
3. SocialcomparisonandstereotypethreatSocialisation and achievement experiences play a pivotal role in the
development of gender differences in motivation. Social comparison isconcernedwith theprocesses involved in comparingourselveswith others, orthinkingabouttheself inrelationtoothers.Thiscanincludecomparisonswithstereotypes and hypothetical characters (Festinger, 1954;Wood, 1996;Webb-Williams, 2006). Festinger (1954) defined social comparison as a deliberateprocessofselectingsocialinformationtoevaluateone'sopinionsandabilities,aswell as to reduce uncertainty with regard to beliefs of self-worth. Socialcomparisonswithpeerscanbemarkedlyincreasedbyteachers,parentsandthe
7
studentsthemselves,especiallyinrelationtothedomainsofschoolachievement.Asmentionedabove,socialcognitivetheoryhypothesizesthatthecognitionsofindividuals regarding performance are influenced by social-contextual factorsinter alia messages from teachers about task difficulty, perceived abilities ofclassroom peers, and information about the importance of learning (Bandura,1986;Dweck&Leggett,1988).Motivationandself-efficacyand-conceptbeliefs,therefore,emergefromtheinteractionbetweenagentswithinthesocialcontextsofthehome,classroomandschool(Bandura,1986;Pajares,1996;Schunk,1984;Schunk&Miller,2002).
Social interactions in schools can takemany forms,having complexandvaryingeffectsonstudents’academicandself-efficacybelief.Untilrecently, theimpact of social comparisononperformancehadnotbeen thoroughly studied,whichissurprisinggiventhatstudentstendtocomparetheirownperformancetothatoftheirpeersratherthantotheirpastperformances.Socialcomparisonsare critical to the development of self-concept and self-efficacy beliefs; forexample, the Big-Fish-Little-Pond-Effect describes how students form self-concept beliefs in part by comparing their own academic ability with theperceivedabilitiesofother students in theirpeergroup.Both self-conceptandself-efficacyresearchagrees thatsocial-comparativeschoolpracticescanassistindestroyingtheself-beliefsofweakerperformer,whilstsocialcomparisonscanbeminimizedthroughindividualizedclassroomstructures(Marsh,1993).
Becausegenderdifferencesarefoundsoearlyindevelopment,thehomeenvironmentplaysan important role in the shapingof competencybeliefsandinterests. At school, childrenwill then have an opportunity to validate, refine,and enact their already learned gender beliefs and behaviour. According tothesocialcognitivemodelofEccles,Adler,Futterman,Goff,Kaczalaetal(1983),culture, parents and teachers contribute to shaping gender differences incompetency, value beliefs and in several ways, including modelling sex-typedbehaviour, communicating different expectations and goals for boys and girlsand encouraging different activities and skills. The development of maleadvantaged inmathematics has also been shown to varywith socio-economicstatus (Levine et al, 2005), suggesting that gender segregation in STEMoccupation may be related to gender dynamics in middle and upper classfamilies.
Researchhasshownthatculturalstereotypes(e.g.boysarebetterinmathandscience)influenceparents’perceptionsoftheirchild’sabilities,whichinturnaffects a child’s perception of their own abilities (Parsons, Adler, et al, 1982;Jacobs&Eccles,1992).Parentalinfluencehasalsobeenshowntohaveenduringinfluences on achievement and attitudes through its impact a child’s careerinterestsandchoices(Bleeker&Jacobs,2004;Jacobs,Chhin&Bleeker,).Atthelevelof theschool, socialisation in the formofgendernormsandrolesmaybeeven more rigid and polarised than what is found in broader society; for
Curriculumandteachingmaterialscanalsoreinforcegenderrolelessons.Classroom interactions that follow gendered differentiated patterns (e.g. boysreceivingmorepraiseorbeingcalledontoanswermoredifficultquestions)alsoservetocommunicatedifferentlearningexpectationsforboysandgirls(Brophy& Good, 1974). These gender-differentiated interaction patterns appear to bemorepronounced in stereotypicallymale sex-typed subjects such asmath andscience(Jones&Dindia,2004;Kahle&Meece,1994),butcanbemoderatedbyclassroom structures and environments (e.g. degree of competitiveness andteachercontrol).Ashighschoolteacherstendtousewhole-classinstructionanddiscussion, and boys are more likely to take an active role in these kinds ofclassroom settings, the emergence of gender differences in performance infavourofboys(orthenarrowingofgenderdifferencesinperformancefavouringgirls)mayberelatedtoashiftinteachingenvironments.
Value beliefs (the activities that students value) are also evidenced tofollow gender and stereotype norms, and decline in relation to achievementvalues of time. Socialization processes that lead to internalization andacceptanceofstereotypesresponsible forgroupdifferences incompetencyandvalue beliefs can therefore have significant effects for engagement andperformance(Ecclesetal,1983).However,itneedstoberecognisedthatgenderdifferences can vary across ethnic and socioeconomic groups; along withgendered expectations, black students must also contend with stereotypes oftheir intellectual inferiority and discrimination (Spencer et al, 1991; Steele,1992). Black boys, in particular, experience more academic and disciplineproblemsthantheir femalecounterparts; thiscan leadtogenderdifferences invalueandself-efficacybeliefs.
Finally, identity processes play a central role in the development ofmotivation.As shown in the seminal work of Steele, Spencer and Aronson,membersof negatively stereotyped groups (for example, girls inmathematics)mayworry that their weak academic achievement could confirm the negativestereotypeabout their group; the resultant stress leads toworseperformance.Even if the individual does not believe the stereotype, the threat of beingevaluated in terms of the stereotype could be felt nonetheless. If thepsychologicalenvironmentisabletorenderthestereotype/sirrelevant,thenitstandstoreasonthatperformanceamongstthesegroupswouldimprove.Steele(1997)proposedthatindividualswhoaremostaffectedbystereotypethreatcanbe categorised as being ‘highly identified’ with the academic domain; forexample, a woman who identifies as being good at mathematics. Therefore,stereotype threat is “feltmost by peoplewho care orwho are invested in thedomainwherethestereotypethreatapplies”(Crocker,Major&Steele,1998).
The data usedwere taken from the Trends inMathematics and Science Study(TIMSS)for2011.ThefocusofTIMSSis toassessthemathematicsandscienceknowledge of students in the fourth and eighth grade, respectively (TIMSSInternational Report, 2011). For South Africa, the TIMSS 2011 survey wascarriedoutin285schools,among11969grade9students(HSRC,2011).TIMSSwas specifically chosen for the analysis of this paper, as the data is rich incontextual information regarding student attitudes towards school, theirteachersandlearning.
ThesamplingandassessmentdesignsforTIMSSposesomecomplexitiesforanalysis.Specifically, theassessmentdesign isabalanced incompleteblockdesign that is used to increase content-area coverage without simultaneouslyincreasing the assessment time demanded of students. As a result, studentachievementinmathematicsandscienceisrepresentedbyfiveplausiblevaluesforeachstudent.Forpurposesofthisstudy,onlythefirstplausiblevalueisused,whichisRasch-scaledtoaninternationalmeanof500andstandarddeviationof100. Furthermore, sampling weighting is applied in all model estimation andclustering at classroom level is taken into account for the computation ofstandarderrors. As interest is specifically in determining the relationship betweenperformanceandself-conceptandmotivationwhilstcontrollingfortheculturalandgenderdiversityoftheSouthAfricanschoolingsystemandclassrooms,table1 summarises descriptive statistics for the sample by school socioeconomic(SES)quintile, the genderof the student, gender compositionof the classroomand the gender of the teacher. In linewith existing research, large disparitiesbetweentheaverageperformancesofstudentsinthepoorest60to80percentofschools(Q1to4)andthewealthiest20percent(Q5)areevident;studentsinthetop quintile of school SES attained average scores close to two internationalstandarddeviationshigherthanstudentsinthe1stquintile. Thedistributionofgirlsacrossschoolsandclassroomsarefairlyuniformacrossallquintiles,exceptin the case of single-sex schools where representation is largely concentratedwithinthetopquintileofschools.Incontrast,approximatelyathirdofallgrade9mathematics teachers in the top school SES quintile are male, whereasapproximately60percentofteachers intheremainingschoolSESquintilesaremale.ThegenderdistributionamongstscienceteachersisfairlyuniformacrossschoolSESquintiles.
10
4.2 Measuresofacademicself-conceptandmotivation
Thispaperreliesontheuseoftwokeyconstructs,namelyacademicself-conceptand motivation. These constructs are generated from student responses to anumber of survey questions as they appear in the TIMSS 2011 studentquestionnaire.StudentsrespondedonaLikertscale from1(“Agreea lot”) to4(“Disagreealot”).Inordertogeneratethesemeasures,polychoricfactoranalysiswas used, which serves as an alternative approach to the analysis of discretedata when interest is in the computation of correlations between ordinalvariables,suchasinthecaseofLikertscaletyperesponses.Thefactorweightsassigned to each item and their distribution across the three measures areshownintable2.Ascanbeseen,similarfactorweightsareassignedtothesameitemacrossthedomainsofmathematicsandscience.
Table 1: descriptive statistics (mean) for performance and gender distribution ofstudentsandteachers,byschoolSESquintile
Note:mathematics and science scores aremeasured by the first plausible value (scaled to aninternationalM=500andSD=100).SchoolSESquintileiscalculatedusingtheaverageSESofstudentsineachschool,whereSESiscomputedusingprincipalcomponentanalysisof11homepossessions/assets.
The sample (Cronbach’s alpha) reliabilities of the self-concept andmotivation composites for each subject are shown in table 3. There is somevariation in the internal consistency of particularly the self-concept andamotivation constructs across school SES. Closer inspection of item-test anditem-restcorrelationswithinschoolSESquintilesrevealsthatitems8,9and15in table 2 do not, as expected, correlate negatively with the remaining self-concept items in quintile 1 schools. This suggests that students in the poorestschools are possibly receiving poor signals/ feedback regarding theirperformance such that they can effectively evaluate their own performance
11
against thatof theirpeers,or conversely,performance isgenerallyvery low inthis context such that assessment of own performance in one subject againstanother and in comparisonwith peers is impractical. A correlation analysis ofthethreeconstructs(seebottompaneloftable3)indicates,asexpected,positivecorrelationsbetweenself-conceptandmotivationandnegativecorrelationswithamotivation.
5. Methodological approach: classroom and within-student fixed effectsestimation
Inordertomakeinferencesofcausality,self-conceptresearchshouldincludeasufficientlylargeanddiversesample.Furthermore,domain-specificself-conceptandachievementshouldbemeasuredat least twiceso thatareciprocaleffects(non-recursive)model that allows for feedback betweenmultiple endogenousvariablescanbeestimated(Pajares&Schunk,2001).WhilsttheTIMSSdatadoessatisfytheconditionofalargeanddiversesample,thecross-sectionalnatureofthe data prevents the estimation of a non-recursivemodel.However, evidencedoes suggest that self-conceptmayplay a stronger causal role at higher-gradelevels(Skaalvik&Hagtvet,1990).Therefore,thispapermodelstherelationshipbetweenself-concept,motivationandacademicperformanceasarecursiveone,withtheself-conceptandmotivationconstructsassumedtobeexogenous.
where𝑦𝑦!"# is the test score for student i, in subject t and in classroomc.F is adummyvariablerepresentingwhetherthestudentisagirl,andSC,MandAMarethe self-concept, motivation and amotivation constructs, respectively.𝑿𝑿!"# is avectorofstudenti'sspecificcharacteristics,includinglanguagespokenathome,exposure to English and home socio-economic status.𝜗𝜗!" represent classroomfixedeffects.Giventhepotentialforerrorcorrelationacrossstudentsinthesameclass,allstandarderrorsarecorrectedtoreflectclassroomclustering.
13
There are two important identification issues that may cause bias inconventional ordinary least squares estimation. First, unobserved teacher andstudent traits thatarecorrelatedwithstudent test scoresmay furtherbias theconventionalOLSestimates.Theclassroom fixedeffectsare includedsuch thatunobserveddifferences(commontreatments)acrossclassroomsandschoolsaredealtwith.6Furthermore,controllingforclassroomspecificfixedeffectsexploitsfor the natural variation in student self-concept and motivation and cohortcompositions within classrooms in order to identify the relationship betweenthesefactorsandperformanceinmathandscience.Second,studentsmaynotbeassigned randomly across and within schools (c.f. Clotfelter, Ladd & Vikdor,2006;Kane,Taylor,Tyler&Wooten,2011). Ifbetter-performingstudentswithhigher self-concept andmotivation select into, for example, quality schools orclassrooms taughtbybetter-educated/higher-quality teachers, the coefficientsofinterestwillbeoverstated.Commonpracticefordealingwiththisissueistocontrolforstudents’priorachievement(c.f.Hanushek&Rivkin,2010).
Awithin-student,between-subjectestimationprocedurewouldbeabletodeal with non-random sorting and subject-invariant student and teacher/classroom unobservables. However, any student and classroom characteristicsthat are invariant across subjects, such as student gender and the gendercompositionoftheclass,cannotbeincludedinthemodel.Separationofthefullgrade 9 sample by school SES and a suitable set of student and home levelcharacteristicsascontrols in themodelmaycorrect fornon-randomsortingofstudents across schools. Restricting the model further to single-class schoolseliminateswithinschoolsorting.7
As there are two observations for each student (math and science),within-studentvariationinperformanceandself-conceptandmotivationcanbeexploited to identify δ, γ and ρ after accounting for subject-invariantunobservable student and home background traits. Following Clotfelter et al(2007),thepreferredmodelnowtakesthefollowingform:(2) 𝑦𝑦!",! − 𝑦𝑦!",! = 𝛼𝛼! − 𝛼𝛼! + 𝛿𝛿(𝑆𝑆𝑆𝑆!",! − 𝑆𝑆𝑆𝑆!!,!)+ 𝛾𝛾(𝑀𝑀!",! −𝑀𝑀!",!)+𝜌𝜌 (𝐴𝐴𝐴𝐴!",! − 𝐴𝐴𝐴𝐴!",!)+ 𝜀𝜀!"(𝑠𝑠)− 𝜀𝜀!"(𝑚𝑚)
where𝜑𝜑! refers to a set of student specific fixed effects. As student priorachievements are not available, the assumption ismade that the student fixedeffect captures any overall ability or achievement level, and furthermore thatability is independent of subject. Finally, the effects of student gender, self-
680 percent of the schools surveyed only had one grade 9 classroom, with the remaining 20percentofschoolshadpredominantly2classrooms.Therefore,theclassroomfixedeffectisalsoaschooleffect.7However,single-classschoolsarenotrandomlydistributedacrosstheschoolsystem.
14
concept, motivation and amotivation are assumed to be subject-invariant, forexample,𝛽𝛽𝛽𝛽!",! = 𝛽𝛽𝛽𝛽!",!.Itcanbeshownthatequation(2)isequivalentto:(3) 𝑦𝑦!",! − 𝑦𝑦! ∗ = 𝛿𝛿 𝑆𝑆𝑆𝑆!",! − 𝑆𝑆𝑆𝑆! ∗ + 𝛾𝛾(𝑀𝑀!",! −𝑀𝑀!
∗)+ 𝜌𝜌 (𝐴𝐴𝐴𝐴!",! − 𝐴𝐴𝐴𝐴! ∗)
+ (𝜀𝜀!,! − 𝜀𝜀! ∗)+ (𝜀𝜀!",! − 𝜀𝜀!,!)
where thevariableswithasterisksare student-specificmeans,(𝜀𝜀!,! − 𝜀𝜀! ∗)referstoastudent-specificerrortermthatvariesacrosssubjectand(𝜀𝜀!",! − 𝜀𝜀!,!)refersto a subject-specific error term that varies with unobservable student (andclassroom)characteristics.Astudent’sachievementinsubjectt(inclassroomc)isthereforemeasuredrelativetotheaverageoftheirachievementbasedonbothtests.Similarly,teacherandclassroomcharacteristicswillbemeasuredrelativetotheaverageclassroomandteachercharacteristicsofthatstudent.
Model (3) will provide unbiased estimates of δ, γ and ρ if neither(𝜀𝜀!,! − 𝜀𝜀! ∗)nor(𝜀𝜀!",! − 𝜀𝜀!,!) are correlated with the demeaned variables. Taking𝜀𝜀!,!torepresentthestudent’sabilityinsubjectt,theterm(𝜀𝜀!,! − 𝜀𝜀! ∗)wouldequalzeroifstudentabilitydoesnotdifferacrossmathandscience,andthereforenostatistical issueswould arise. The second error term,(𝜀𝜀!",! − 𝜀𝜀!,!), accounts forthe effects on student achievement of unobservable student and classroomcharacteristics,suchasglobalself-esteemandlearningstrategies.Thistermwillnot bias the coefficients of interest if these characteristics are randomlydistributed among students and classrooms. It cannot be proven conclusivelythattheabove-mentionedconditionshold,andthereforetheanalysisconductedinthispapercannotbeinterpretedascompletelyfreefrombias.
One consequence of fixed effects estimation is that any student andclassroomcharacteristicsthatareinvariantacrosssubjectscannotbeincludedinthe model. As student gender is fixed within a student, student fixed effectsestimation prevents estimation of the relationship between gender andperformance;however, it ispossible toestimateandcompare the results fromseparatemodelsformaleandfemalestudents.
Meansandclusterrobuststandarderrorsforgirls'andboys'achievement(testscores), self-concept,motivation and amotivation/ anxiety inmathematics andscience across school SES quintile are indicated in table 4 below. Whilst nosignificant gender difference in mathematics and science performance isobserved for the full sampleof schools, a significantdifference inmathematicsscoresof33points(approximately0.4quintile5standarddeviations)infavourofboysisobservedforQ5schools,andsimilarlyasignificant(atthe10percentlevel) difference of 33 points for science; no significant gender difference isevidencedfortheremainingQ1to4schools.
15
With regards to self-concept, motivation and amotivation, againsignificant gender differences in favour of boys are found for Q5 schools.Therefore,thefindingsfromtheinternationalliteratureonlyappeartoholdforthewealthiest subset of the SouthAfrican schooling system. In addition to thegenderdifferencesacrossschoolSESquintile, it is interestingtopointoutthat,amongst all school quintile and gender groupings, girls attending Q5 schoolshavethelowestexpectedmathematicsandscienceself-conceptwhilsttheirmalecounterparts have the highest. Girls attending Q5 schools are also the leastmotivatedandmostamotivatedofthestudentgroups.
SchoolSES Quintile1-3 Quintile4 Quintile5 All girls boys girls boys girls boys girls boys
MathematicsTestscore 327.6
(3.10)325.3(2.88)
368.7(5.53)
364.7(5.48)
477.8(6.93)
509.4**(10.23)
354.0(4.03)
352.1(4.60)
Self-concept 0.09(0.03)
0.12(0.03)
-0.20(0.05)
0.03**(0.05)
-0.39(0.07)
0.09***(0.07)
-0.03(0.03)
0.10**(0.02)
Motivation 0.07(0.03)
-0.04(0.04)
0.06(0.04)
0.20*(0.04)
-0.45(0.06)
-0.18***(0.05)
-0.01(0.03)
-0.01(0.03)
Amotivation -0.05(0.03)
-0.01(0.02)
-0.03(0.05)
-0.10(0.05)
0.27(0.05)
0.04**(0.05)
0.00(0.03)
-0.02(0.03)
ScienceTestscore 297.0
(4.80)291.9(4.27)
363.1(7.73)
353.5(6.95)
495.3(7.56)
528.2*(10.18)
333.8(5.73)
328.0(6.04)
Self-concept 0.04(0.03)
0.06(0.03)
-0.06(0.08)
0.09(0.05)
-0.50(0.09)
-0.08**(0.07)
-0.05(0.03)
0.05(0.03)
Motivation 0.08(0.03)
0.10(0.03)
-0.06(0.08)
0.08(0.05)
-0.62(0.09)
-0.31***(0.07)
-0.04(0.04)
0.05(0.02)
Amotivation 0.03(0.04)
0.10(0.03)
-0.11(0.06)
-0.08(0.05)
0.10(0.07)
-0.01(0.05)
0.01(0.03)
0.06(0.03)
Note:mathematics and science scores aremeasured by the first plausible value (scaled to aninternationalM=500andSD=100).SchoolSESquintileiscalculatedusingtheaverageSESofstudentsineachschool,whereSESiscomputedusingprincipalcomponentanalysisof11homepossessions/assets.Mathematics and science self-concept is computedusingpolychoric factoranalysis on 7 items, motivation is computed using polychoric factor analysis on 7 items, andamotivation/ anxiety is computed using polychoric factor analysis on 3 items. Cluster(classroom) robust standard errors are shown in parentheses. Sampleweighting is taken intoaccount.***p=0.01,**p=0.05,*,p=0.10.
16
6.2 Classroomfixedeffectsresults
Tables5and6summarisethemodelresultsformathandscienceperformanceof the full sample of students, respectively. Self-concept, motivation andamotivation have the expected (and statistically significant) relationshipswithperformance. However, controlling for all, domain-specific motivation isnegatively (and significantly) related to science performance. Furthermore, asignificantgenderdifferenceinfavourofboysofapproximately3to4pointsisfound for both subjects, although gender is not a significantmoderator of therelationshipbetweenself-conceptandperformance.Controlling further for thegendercompositionoftheclassroom,thecoefficientonstudentgenderbecomeslarger (although turns insignificant for science). This indicates that the gendereffect in favour of boys is related to gender compositions in the classroom;specifically,girlstaughtinclassroomsthatareeithermaleorfemaledominatedperform significantly lower than their male counterparts, controlling for self-conceptandmotivation.Wheregendercompositionsaremoreequal,thereisnosignificant gender difference in performance when comparing students withsimilarlevelsofself-conceptandmotivation.8However,itshouldbepointedoutthat larger gender performance gaps in gender-skewed classrooms areaccompaniedbyhigheraverageperformanceforbothgenders.
Tables7to11reportresultsforthesamemodelspecificationestimatedforthesamplesofQ1to4studentsandQ5studentsseparately. It isnoteworthythat motivation and amotivation appear to have different relationships withperformance (controlling for self-concept) in Q5 andQ1to4 schools.Whilst allthree of the self-belief and attitude measures appear to be correlated in Q5schools(that is, theeffectofself-conceptcapturestheeffectofmotivation),theroleofself-conceptandmotivationonperformanceinthepoorerschoolsappeartobedistinct;a studentwhohasa loworaverageself-concept inmathematicscan still experience a positive performance effect through high motivationand/orinterestinthesubject.Inscience,aswiththefullsample,therelationshipis negative. Closer analysis of the components of the motivation constructindicates that external regulation, which may induce feelings of anxiety, isnegatively related to performance, whilst integrated regulation is positivelyrelatedtoperformance.
Oneresultthat iscommontobothschoolsamplesisastrongernegativerelationship between performance and amotivation/ anxiety for science thanwhat isestimated formathematics.Conversely,asignificantgendergap isonlyestimated forQ5 schools,which is only eliminated inmathematics classroomswithmoreequaldistributionsofboysandgirls.Thecoefficientonself-conceptinmathematics in thesampleofQ5schools isestimated tobesignificantly largerthanthecoefficientonscienceself-conceptinthesamesample,aswellasinboth8Thisresultremainsevenwhenremovingsingle-sexclassrooms.
Note:mathscoreismeasuredbythefirstplausiblevalue(scaledtoaninternationalM=500andSD=100).Self-concept,motivationandamotivation/anxietyarecomputedusingpolychoric factor analysis and z-scored. Cluster (classroom) robust standard errors are shown inparentheses. Sampleweighting is taken intoaccount.Studentcontrolsincludewhetherthestudenthastheirownbooksathome,howoftenthestudentspeaksthelanguageofthetest,socio-economicstatusofthehousehold,mother’seducationandfather’seducation.***p=0.01,**p=0.05,*,p=0.10.
6.3 Teacher gender and training as potential mechanisms for the testperformancesoffemalestudentsrelativetomalestudents
One environmental factor that might influence the relationship between theconstructs of self-concept andmotivation and performance, as well as genderand performance, is the gender of the teacher. Asmentioned in the literaturereview,classroominteractions,whichmayormaynotdifferbyteachergender,can allow for gender-differentiated learning expectations and experiences. Forexample, a teacher’s expectations based on stereotypical gender beliefs canbecome self-fulfilling prophecies (e.g. Cushman, 2010). The teaching styles ofmaleandfemaleteacher’scanalsoconformtonormativegenderroles,withmaleteacherspreferring compliant studentsanda teacher-centred instruction style,whilstfemaleteachersmayrelyonmorecollaborativeclassroomenvironments(Brophy,1985).
In order to formally determine the effect of female teachers on studentachievement,thefollowingregressionisestimatedformathematics:9(4) 𝑦𝑦!" = 𝛼𝛼 + 𝜋𝜋!Girl!" + 𝜋𝜋!FemaleTeacher!"+ 𝜋𝜋!Girl ∗ FemaleTeacher!" + 𝜃𝜃𝑿𝑿!" + 𝜗𝜗𝑻𝑻𝑻𝑻!" + 𝜀𝜀!"
where the interaction allows for teacher gender to differentially affect femaleand male students, the vector of student characteristics,𝑿𝑿, now includes theconstructs for self-concept, motivation and amotivation, and𝜀𝜀!",!is defined aspreviously. Note that this model does not include classroom fixed effects, butotherteacher/classroomcontrols,suchasteacherage,teachingexperienceandjob satisfaction,10 are included in the vector𝑻𝑻𝑻𝑻 . Single-sex classrooms areexcludedandstandarderrorsareclusteredattheclassroomlevel.
TheresultsfortheQ5andQ1toQ4schoolsamplesarepresentedintable12. Focusing first on the results for Q5, relative to male students taught byfemale teachers, female students score worse on the mathematics test,approximately9%of a sample standarddeviation.Thisdifference is, however,insignificant at conventional levels. Similarly, girls taughtbymale teachers arenot estimated to perform significantly better or worse than boys taught by a9Thispartoftheanalysisislimitedtomathematicsgiventhatthenegativecoefficientongirlwaslargelyconfinedtothisdomain.10Teacher age is a categorical variable that takes on 6 values: younger than 25 years; 25-29years; 30-39 years; 40-49 years; 50-59 years; and 60 years or older. Teaching experience is acategoricalvariablethattakeson4values:0-5years;6-10years;11-15years;and16yearsormore.Jobsentimentisacontinuousvariable(z-scored)thatisgeneratedusingpolychoricfactoranalysisof6surveyitemsrelatingtoteachersentimentsabouttheirprofessione.g.“Ihadmoreenthusiasmwhen I began teaching than I have now.” A higher value of this variable implies amorepositivesentiment.
24
maleteacherinQ5schools.Thedifference-in-differencescoefficient(𝜋𝜋!),whichgivestherelativedifferenceinmathematicstestscoresbetweenfemalestudentswith female versusmale teachers andmale studentswith female versusmaleteachers, is large (-11.1, which is approximately 15% of a sample standarddeviation) and statistically significant at the 10% level. These results are verysimilartothoseofAntecoletal(2014),andtakentogethersuggestthat femalegrade9mathematicsteachersrelativetomaleteachersinQ5schoolsadverselyinfluence themathematicsoutcomesof female studentsbutnotmale students.WithregardstoQ1toQ4schools,therelativedifferenceinthemathperformanceof female and male students taught by a female teacher is very large andsignificant(-38points,or59%ofastandarddeviation).Therelativedifferenceinscores of female andmale students taught by amale teacher is similarly largeand significant (-30 points), but, as indicated by the difference-in-differencescoefficient, is significantly smaller than the relative difference under a femaleteacher.Therefore,inQ1toQ4schools,bothmaleandfemaleteachersadverselyinfluencethemathematicsoutcomesofgirls,butmoresointhecaseofthelatter.
Studentcontrols Yes Yes Yes YesTeachercontrols Yes Yes No NoClassroomfixedeffects No No Yes YesR-squared 0.239 0.472 0.131 0.335Observations 6861 1422 7492 1571
Note:mathscoreismeasuredbythefirstplausiblevalue(scaledtoaninternationalM=500andSD = 100). Cluster (classroom) robust standard errors are shown in parentheses. Sampleweightingistakenintoaccount.***p=0.01,**p=0.05,*,p=0.10.
The model results in columns (3) and (4) control for classroom fixedeffects in order to ensure that the results are not driven by unobservable
25
differencesbetweenmaleandfemaleteachers.11TherelativedifferencebetweenfemaleandmalestudentstaughtbyafemaleteacherinQ5schoolsisnowlargernegative (-10.95 points, or 14.4% of a standard deviation) and statisticallysignificant. Girls taught by male teachers are also estimated to perform 8.79points (approximately 12% of a standard deviation) worse relative to boystaught by male teachers, implying that, after controlling for unobservableteacher characteristics, the mathematics performance gap in favour of boysundera female teacherdoesnotappear tobe relatively larger than that foundunder a male teacher in wealthier school settings. In poorer school settings,accounting for teacher unobservables dramatically reduces the male-femaleperformance gap under a female teacher to a coefficient that is now also 10.6points(16%ofastandarddeviation),asinthecaseofQ5schools.Girlstaughtby male teachers are estimated to perform 4.93 points worse, although thiscoefficient is not statistically significant. Therefore, the mathematicsperformance gap in favour of boys under a female teacher is relatively largerthanthatfoundunderamaleteacherinpoorerschoolsettings.
Itisconceivablethattheteachingstylesofteachersmaydifferdependingonthestrengthoftheirbackgroundsinthesubjectofteaching,astoocouldtheteaching styles between male and female teachers, and this may affect theacademic achievements of male and female students differently. Antecol et al(2014)posit threeways inwhichaweaker test score for girls in, for example,mathematicscouldarise.First,teacherswithoutstrongsubjectbackgroundsmayadoptamoremechanicalteachingstylethatisfineformalestudentsbutnotforfemalestudents.Alternatively,femaleteachersmayteachSTEMsubjectsjustaswell as male teachers, but female students may respond more positively(negatively)toamale(female)teacheratcertainages.Bothofthesereasonsarejudged as unlikely by Antecol et al (2014), and rather they find suggestiveevidence in support of math anxiety, which may be reduced the stronger themathbackgroundofthefemaleteacherand/orthelessheldstereotypicalbeliefsare in the classroom. Recent evidence from the educational psychologyliterature, inparticular thatofBeilock,Gunderson,RamirezandLevine(2010),findsthathighermathanxietyinfemaleprimaryschoolteachershurtsthemathperformancesof femalestudentsbutnotthatofmalestudents.Specifically, themoreanxiousfemaleteachersareinmathematicsclasses,themorelikelyfemalestudents are to endorse the stereotype thatboysarebetter atmath thangirls.Thisseemstosuggestthatstereotypethreatdoesnotonlyaffectthegirlstudentsin a classroom, but female teachers as well. Interestingly, the genderachievement gap is closed when students are taught by female teachers inclassrooms with more equal distributions of student gender. This could be
suggestive of an environment in which previously held stereotypes can bechallenged.
The main area of study of the teacher is observed from the TIMSScontextualsurveys,withsomeinterestingdifferencesinthedistributionofpost-secondary study-area across subjects and teacher gender emerging. Table 13indicates the distribution of teacher study area by school SES quintile, genderandsubject.ItisevidentthatthemajorityofmaleandfemaleteachersinQ1toQ4schools tend to be specialists in their subjects, with math teachers showingslightlymoregeneralisationwithtraininginbothmathematicsandsciences.Thismay correlate todifferences in thequalityof teachingacrossmathematics andscience classrooms. Overall, there do not appear to be any important genderdifferences in the distribution of focal areas across grade 9 science andmathteachers inQ1toQ4 schools. Interestingdifferences, however, are found forQ5schools.Malemathematicsandscience teachers inQ5schools tend tobemoregeneralist than their female counterparts; whilst just more than half of malemathematics teachersandclose to40percentofmalescience teachersstudiedbothmathematics and science, just less than a quarter of femalemathematicsand science teachers have similar focal areas. A fifth of female mathematicsteachers inQ5 schools donot denote their post-secondary focal area as eithermathematicsorscience,orevengeneraleducation.
Following Antecol et al (2014), a teacher’s training background iscontrolledforusingthefollowingregressionspecificationformathematics:12
where all variables are defined as previously, single-sex classrooms are againexcluded,andclassroomfixedeffectsareadded.13Thevariable‘training’iscodedto take three values: diploma or degree inmathematics; diploma or degree inmath and science; and a diploma or degree with neither a mathematics norscience focus. The distribution of these post-secondary qualifications acrossschoolSES,teachergenderandsubjectareindicatedinthesecondpaneloftable13. Asnotall teachersstudiedadegreeordiplomawitha focusoneducation,equation(5)isestimatedseparatelyfortwogroupsofteachers,thosewithpost-secondarystudiesthatincludedafocusoneducationandthosethatdidnot.The12ThesamemodelisestimatedforscienceandshownintableA3oftheappendix.However,theresults are not as easily interpretable, especially for the Q5 sample, given the substantialdifferences in the distributions ofmale and female teachers across tertiary training categories(seetable13).13Thecoefficientsubscriptshavebeenintentionallychosentocomparewithequation(4).
results arepresented in table14.Three coefficients of interest are included inthe table. (𝜋𝜋! + 𝜋𝜋!) indicates the relative difference in the scores of girls andboys taught by female teachers that studied a diploma or diploma without amathfocus;(𝜋𝜋! + 𝜋𝜋!,!)indicatestherelativedifferenceinthemathscoresofgirlstaughtby female versusmale teachers andboys taughtby female versusmaleteachers with mathematics as a main area of post-secondary study; similarly,(𝜋𝜋! + 𝜋𝜋!,!) and𝜋𝜋!indicate the relative performances of girls and boys whentaughtbya femaleversusmale teacherwhosepost-secondarystudies includedmathematicsandscienceandnomathematics,respectively.
Withinpoorer(Q1toQ4)schoolcontexts,girlsrelativetoboystaughtbyafemaleteacherwithateachingqualificationwithnomathfocusperformworse(-19.18or29.4%ofasamplestandarddeviation),andthiscoefficientissignificantatthe5%level.Thegenderperformancegapforstudentstaughtbyasimilarlyqualifiedmale teacher is largenegative (-16.81or25.8%ofa samplestandarddeviation)andstatisticallysignificant.Thecoefficients(𝜋𝜋! + 𝜋𝜋!,!)and(𝜋𝜋! + 𝜋𝜋!,!)are not estimated to be significantly different from zero, indicating that, inQ1toQ4 schools, girls taught by female teachers with education qualificationswithout a focus on math do not have significantly lower math achievementoutcomes than girls taught by similarly educated female teachers with mathbackgrounds.However, it canbenoted that thecoefficient𝜋𝜋!,!is largepositive(11.29or17.3%ofasamplestandarddeviation).Eventhoughtheresult isnotsignificant at conventional levels, girls see an improvement in their relativeperformancewhen taughtbya female teacherwithamathematicsbackgroundcomparedtobeingtaughtbyafemaleteacherwithnomathematicsbackground.
Whenattentionisturnedtoteacherswithouteducationqualifications,therelativedifferenceinscoresbetweengirlswithfemaleversusmaleteachersandboys with female versus male teachers is large and negative when teachersreporthavingstudiedbothmathandscienceattertiarylevel.Whentaughtbyateacher reportingnooronlya focusonmathematics, the relativeperformancedifference is not significantly different; the test score of girls remains atapproximately 20-24% of a sample standard deviation lower than boys whentaught by a female teacher.However,when taught by teacherswith a broaderstudy focus (mathand science), the relativegender gap isdoubled. It isworthnoting that the coefficientπ!,!is large and positive, indicating that there is abenefitforgirlsbeingtaughtbyfemaleteacherswithtertiaryqualificationswithafocusonmathematics.Thesmallsamplesizeofthegroupoffemaleandmaleteachers in Q1toQ4 schools that studied a diploma or degree that had nomathematics focus may be generating this result, especially if there aresignificantdifferencesinthesubject/sthatformedthefocusofthequalificationacrossgender.
29
In the caseofwealthier school contexts, adissimilar result is found.Nostatistically significant difference in the relative performances of female andmale students in Q5 schools is estimated when the gender and relativemathematics background of the teachers are changed. It is interesting to note,however, thattherelativeperformancegapbetweengirlsandboystaughtbyafemale versus a male teacher reported to have majored in both math andeducation is 18.3%of a standard deviation larger (not statistically significant)thantherelativeperformancegapwhencomparingtheperformanceofgirlsandboys taught by female versus male teachers with education training but nomathematics focus. Finally, when comparing Q5 teachers that did not reporteducationasamajorareaofstudy, therelativeperformancegapbetweengirlsandboyswhentaughtbyafemaleteacherversusamaleteacherincreaseswiththemathbackgroundof the teacher.Aswith theQ1toQ4 sample, these resultsmay be driven by differences in the distribution of male and female teachersacrossthedifferencedomainsofpost-secondarystudy.
An alternative interpretation for the result that female teachers versusmaleteachersadverselyinfluencethemathematicsperformanceofgirlsbutnotthatofboysisthatmaleteachersrelativetofemaleteacherspositivelyinfluencethe performance of girls. Re-estimations ofmodel (5) for separate samples ofmale and female teachers are shown in tablesA1andA2of the appendix.Theadverse effect of female teachers on the mathematics performances of girlspersists evenwhenmale teachers are excluded from the analysis. No positiveinfluenceofmaleteachersongirlsissuggestedfromtheestimates,exceptinthecaseofmaleteachersinQ5schoolsthatstudiedaneducationdegreewheregirlsareestimatedtoperformsignificantlybetterthanboys.Incombinationwiththefindings of table 14, this suggests a higher relative performance of girls in Q5schools taught by teachers with university qualifications in education, andproposesthatdifferencesinthesocialisationofgirlsandboysmayinteractwiththe styles and/or modes of teaching that may be utilised by teachers withdifferent training rather thanmathematicsbackgrounds. In the caseofQ1toQ4schools, the relative performance of girls in comparison to boys is improvedwhentaughtbya femaleteacherwithqualifications ineducation,althoughthispositiveinfluenceiscancelledoutwhentheteacherhasauniversitydegree.
ThesefindingsaredifferenttothoseofAntecoletal(2014) inthattheydonotprovide, at least direct, evidence in support of the math anxiety hypothesis.However, where suggestive evidence may exist is in the comparison of therelativeperformanceofgirlsassignedto femaleversusmaleteachersandboysassignedtofemaleversusmaleteacherswhenteachersdonotreporthavinganeducationqualification.Whenassignedtoafemaleteacherwithhigherexposureto mathematics at post-secondary levels, the math performances of girls inrelation to that of boys worsen when compared with assignment to maleteacherswith(assumedly)similarbackgroundsinmathematics.
30
Table 14: Classroom fixed effects model of math achievement controlling for teachereducation,byschoolSES (1) (2) (3) (4) Educationincludedin
Studentcontrols Yes Yes Yes YesTeachercontrols No No No NoClassroomfixedeffects Yes Yes Yes YesR-squared 0.133 0.325 0.143 0.378Observations 4310 963 3182 608Note:mathscoreismeasuredbythefirstplausiblevalue(scaledtoaninternationalM=500andSD = 100). Cluster (classroom) robust standard errors are shown in parentheses. Sampleweightingistakenintoaccount.***p=0.01,**p=0.05,*,p=0.10.
One explanation for this might be found in stereotype threat theory(Steele,1997),wherebywomenthatidentifystronglywithaparticulardomain,in this casemathematics,may experiencemore anxiety or concernwhen theyhavethepotentialtoconfirmanegativestereotypeabouttheirsocialgroup,suchas “womenareworseatmath thanmen”.This implies that the ‘threat’maybestronger for highly identified women. The threat associated with domainidentificationmayoperatedifferentlyacrossthedifferentschoolSEScontexts.Inrelativelypoorer school contextswhereperformance isdramaticallyweaker, a
31
woman who has obtained a university degree (academically identified) mayexperiencegreaterthreatastheexpectationsofperformance,andhenceanxiety,aremuchgreater.
Thelevelof(stereotype)threatcould,amongstotherthings,beinfluencedbytherelativedistributionsofboysandgirlsinclassrooms,aswellasthegenderdistribution of other mathematics and science teachers in their school. Fromtable 14 we see that girls perform relatively better in gender equalenvironments, when the teacher has an education qualification. Therefore,teaching styles and classroom interactions that generate more equity in theperformances of boys and girls are likely to arise from particular educationtraining, as well as classrooms in which representation is more equal. TIMSSunfortunatelydoesnotprovide informationon thegenderdistributionofmathandscienceteachersineachschool,makingitdifficulttoassumethatthesampleofteachers(ortheteacherselectedfromeachschool)isnecessarilyrandom.Forexample,inacountrysuchasSouthAfricawheretheschool-exit(matriculation)examdeterminesastudent’saccesstoandplacementinuniversity,someschoolsmaydecidetoallocatehigherqualitymathteachers,whichtheperson/smakingthehiring-decisionmightassumetobemen,atgrades10andabove.
6.4 Studentfixedeffectmodelresults
In order to determinewhether or not the gender andqualification of teachersplays a role on performance through the self-concept of a student, a within-studentbetween-subjectmodelisestimated.Thisallowsforbiasdrivenbynon-randomsortingofstudentsintoandwithinschoolstobeeliminated.Astherearetwoobservationsforeachstudent(mathandscience),within-studentvariationinperformanceandself-conceptandmotivationcanbeexploitedtoidentifyδ,γand ρ after accounting for subject-invariant unobservable student and homebackground traits. Following Clotfelter and Ladd (2007), themodel now takesthefollowingform:(6) 𝑦𝑦!",! − 𝑦𝑦!",! = 𝛼𝛼! − 𝛼𝛼! + 𝛿𝛿(𝑆𝑆𝑆𝑆!",! − 𝑆𝑆𝑆𝑆!",!)+ 𝛾𝛾(𝑀𝑀!",! −𝑀𝑀!",!)+𝜌𝜌 (𝐴𝐴𝐴𝐴!",! − 𝐴𝐴𝐴𝐴!",!)+ 𝜀𝜀!"(𝑠𝑠)− 𝜀𝜀!"(𝑚𝑚)
where𝜑𝜑! refers to a set of student specific fixed effects. As student priorachievements are not available, the assumption ismade that the student fixedeffect captures any overall ability or achievement level, and furthermore thatability is independent of subject. Finally, the effects of self-concept,motivationand amotivation are assumed to be subject-invariant, for example,𝛽𝛽𝛽𝛽!",! =𝛽𝛽𝛽𝛽!",!.
where thevariableswithasterisksare student-specificmeans,(𝜀𝜀!,! − 𝜀𝜀! ∗)referstoastudent-specificerrortermthatvariesacrosssubjectand(𝜀𝜀!",! − 𝜀𝜀!,!)refersto a subject-specific error term that varies with unobservable student (andclassroom)characteristics.Astudent’sachievementinsubjectt(inclassroomc)isthereforemeasuredrelativetotheaverageoftheirachievementbasedonbothtests.Similarly,teacherandclassroomcharacteristicswillbemeasuredrelativetotheaverageclassroomandteachercharacteristicsofthatstudent.
Model (7) will provide unbiased estimates of δ, γ and ρ if neither(𝜀𝜀!,! − 𝜀𝜀! ∗)nor(𝜀𝜀!",! − 𝜀𝜀!,!) are correlated with the demeaned variables. Taking𝜀𝜀!,!torepresentthestudent’sabilityinsubjectt,theterm(𝜀𝜀!,! − 𝜀𝜀! ∗)wouldequalzeroifstudentabilitydoesnotdifferacrossmathandscience,andthereforenostatistical issueswould arise. The second error term,(𝜀𝜀!",! − 𝜀𝜀!,!), accounts forthe effects on student achievement of unobservable student and classroomcharacteristics,suchasglobalself-esteemandlearningstrategies.Thistermwillnot bias the coefficients of interest if these characteristics are randomlydistributed among students and classrooms. It cannot be proven conclusivelythattheabove-mentionedconditionshold,andthereforetheanalysiscannotbeinterpretedascompletelyfreefrombias. As gender is fixedwithin a student and across subjects, equation (7) isestimatedseparatelyforboysandgirls.Furthermore,interactionsbetweenself-concept and the gender and training of the teacher, specifically whether theteacherstudiedaneducationqualification,areincludedasfollows:
where SC, M and AM are as previously defined, and TC, a vector of teachercharacteristics, includesteacherage, teachingexperience, jobsentimentandanindicatorofwhethertheystudiedatuniversity. Theestimationresultsofmodel(8)areshownintable15.Focusingfirston the results for theQ5sample, thecoefficientson thesubjectof focus in thepost-secondarystudiesofteachersagreewiththosefoundintable14andtableA2 of the appendix; controlling for learner subject-invariant characteristics,subject focus is not found to have a significant effect on the performance ofstudents,althoughamorenarrowfocusonthesubjectofteachingasopposedto
33
abroadSTEMfocusispositivelyrelatedtoperformance.Thesignificantlyhigherperformance(atthe10%level)ofboystaughtbyfemaleteacherssuggeststhatthe negative influence of female teachers on girlsmay be driven by a positiveinfluenceof female teachershaveonboys’performance. It ispossible that thiscoefficient arises from the fact that significantly fewermalemath and scienceteachers inquintile5 schoolshavepost-secondary education that is not in thesubject of teaching (with no indication of the relative quality of thesequalifications across male and female teachers). As these teachers are notspecialised in thesubject that isbeing taught, the teachingstyleadoptedby, inthis case, predominantlywomen teachers,may bemoremechanical in nature,whichmightbelessharmfultoboysthangirls.
Apositivelarge(approximately12-17%ofasamplestandarddeviation)and significant relationship is estimated between student performance andteacher training thatdidnot includeeducation.This isaccompaniedbya largepositive and statistically significant influence of university teacher training forgirls (13.96 points, or 17% of a sample standard deviation). This result is animportant one, as it provides additional evidence that the gender gap inperformance (or relativeperformanceofgirls) is smaller (higher)when taughtbyteacherswithuniversitytraining.
When the above is combinedwith education training as amoderatorofself-concept,theresultisevenmorepowerful.Thecoefficientsatthebottomoftable15,namely𝛿𝛿!,𝛿𝛿! + 𝛿𝛿!, 𝛿𝛿! + 𝛿𝛿!,!and𝛿𝛿! + 𝛿𝛿! + 𝛿𝛿!,!represent theestimatedrelationship between self-concept and test scores for students taught bymaleteachers trained in (at a minimum) education, males teachers not trained ineducation, female teachers trained in (at a minimum) education and femaleteachersnottrainedineducation,respectively.Theeffectofself-conceptontestscoresispositiveandstatisticallysignificant(atthe5%and10%level)forgirlstaught by teachers whose training included a focus on education; for each 1standarddeviationincreaseinself-concept,testscoresofgirlsimproveby6.5to8.5percentofasamplestandarddeviation.Theself-concepteffectforgirlswhentaught by teachers without education training is not statistically significantlydifferent from zero. For boys, the only similar positive self-concept effect isestimatedwhen taught by female teacherswith education training (8.8% of asample standarddeviation).A largepositive, but statistically insignificant, self-concept effect is also estimated when boys are taught by male teachers nottrainedineducation.Itisrelevanttopointoutthattheeffectsofself-conceptaresubstantiallysmallerthanthoseestimatedunderaclassroomfixedeffectsmodel,suggesting a positive relationship between student and home backgroundunobservablesandself-concept.
Note: test scores are measured by the first plausible value (scaled to an international M = 500 and SD = 100). Single-sex classrooms are excluded. Cluster(classroom) robust standard errors are shown in parentheses. Sample weighting is taken into account. *** p = 0.01, ** p = 0.05, *, p = 0.10.
35
Turning the focus tostudents inQ1toQ4schools,bothboysandgirlsdosignificantlybetter(19%to23%ofasamplestandarddeviation)whentaughtbya teacher that ismorebroadly trained inbothmathandscienceasopposed toneitherormathalone.Nosignificanteffectofuniversityandeducationtrainingis estimated. The coefficient on self-concept for girls and boys is very largepositive(13-15pointsor16-19%ofastandarddeviation)andhighlysignificant;thiscoefficientisnotalteredmuchwheninteractedwiththetrainingandgenderoftheteacher.Itisclearthatreportedself-conceptofstudentsinQ1toQ4schoolsis negatively correlated with unobserved student and home backgroundcharacteristics. This might suggest that a relatively small proportion of (mostlikely) higher ability students in poorer school settings are able to correctlygauge their academic efficacy against that of their peers. The feedbackmechanisms,suchas frequentandreliableassessments,maybeofsuchapoorquality that students are unable to properly rate their abilities within theclassroom.Therefore, the largesignificantcoefficientonself-conceptestimatedintable15indicatesthat,accountingforstudentability,astrongself-conceptisrelatedtoaugmentedtestperformance,albeititfromalowbase.
7. ConcludingremarksandpolicyrecommendationsDespite recent improvements, the under-representation of women in tertiaryenrolments in STEM fields of study, and subsequently professional STEMcareers, continues to be an issue faced by the South African economy. To theknowledgeoftheauthor,thisisthefirstpaperthathasattemptedtounderstandthe roleof self-conceptonobservedachievementpatterns inmathematics andscience in South Africa. The analysis of this paper and interpretation of theresults borrows considerably from the social psychology and educationalpsychologyliteratures,specificallytheroleofstereotypesinformulatingtheself-efficacy and value beliefs of children. The TIMSS 2011 dataset was used, as amultitude of questions related to student attitudes towards mathematics andscience were integrated into the contextual surveys, allowing for a number ofconstructstobedevelopedthroughfactoranalysis. Insummary,thispaperfindssignificantandlargegapsindomainspecificself-concept,motivationandanxietyforgirlsandboys,aswellasinmathematicsandscienceperformance,forthewealthiestsubsetofschools(Q5);nosignificantdifferences are found for the poorer schoolwealth (SES) quintiles. Controllingfor student observable characteristics and the domain specific constructs inclassroom fixed effects models of mathematics and science achievement, aperformance gap of 13-15 points (14-19% of a sample standard deviation) infavourofboysisestimatedforthesampleofQ5schools,andonlyeliminatedinclassroomswhere thedistributionofboysandgirls ismoreequal.The findingthat, conditional on the same level self-concept and motivation, a substantialgender performance gap remains is concerning. The relationship between the
36
achievementoutcomesofgirlsandboysandteachergenderisexaminedthroughthe inclusion of interactions between student and teacher gender in theclassroom fixed effects model. Controlling for teacher and classroomunobservables,asignificantgendergapinmathtestscoresofapproximately11pointsisestimatedforstudentstaughtbyfemaleteachers.Asizeablegapisalsoestimated for male teachers, indicating that teacher gender alone does notexplaintheperformancegap.
Following Antecol et al (2014), teacher training and education incombinationwithteachergenderisexplored.Itisconceivablethattheteachingstyles, and hence effectiveness, of teachers may contribute to differentialperformances amongst boys and girls, particularly in the manner in whichclassroom interactions and engagement interplay with different socialisationexperiences of girls and boys, as well as the stereotype beliefs of both thestudentsandtheteacher.Adistinctionismadebetweenthesubjectbackgroundoftheteacherandthetypeoftertiaryqualification,asSouthAfricanhigh-schoolteachers, particularly in the STEM fields, may have completed their educationtrainingeither in combinationwithorafter theirmathematics and/or sciencetraining, or may have transitioned into teaching after more specialist tertiaryeducation.14
Theresultsshowthat therelativeperformancegapbetweenfemaleandmale students is reduced when taught by teachers whose tertiary trainingincluded a focus on education. This is suggestive of differences in teachingmethods thatmaybe lessharmful to girls adoptedby teacherswitheducationbackgrounds; specialist, non-educationalist training may lend itself to, forexample,moremechanicalstylesofteachingthatboysfindlessharmfulormorebeneficial. In the case of poorer schools, some evidence of the math-anxietyhypothesispositedbyAntecoletal(2014)issuggestedbytheresults;thatis,thenegativeeffectofhavingafemaleteacheronthemathscoresofgirlsisreducedfor students taught by female teachers whose training included a focus onmathematics.15
In Q5 schools, however, quite the opposite result is found; the relativedifference in mathematics test scores between girls with female versus maleteachersandboyswithfemaleversusmaleteachersislargerthemoreexposuretheteacherhashadtomathematics.Whilstthisseemstobeincontradictiontothe math-anxiety hypothesis, this paper argues that this result relates well tostereotypethreattheory,wherebystrongdomainidentificationofanindividual14Thissectionoftheanalysisfocusedonlyonmathematicsgivendissimilardistributionsinthetraining of male and female science teachers in Q5 schools, which complicates the ease ofinterpretabilityofthemodelresults.15Theresultsare lessclearwithrespect toQ1toQ4teacherswithoutaneducationbackground,and are possibly related to systematic differences in the quality of training across male andfemaleteachers.
As a final exploration of the mechanisms for female student math andscience test scores, a student fixed effectsmodels that exploitswithin studentvariationinperformance,self-conceptandmotivationacrossthetwosubjectsisestimated.Therelationshipbetweenself-conceptandperformanceisallowedtovarywithteachergenderandeducationqualification.Inpoorerschoolcontexts,the relationship between self-concept and performance does not appear to beinfluencedbytheseteacherobservables.However,giventhatthecoefficientonself-concept is dramatically increased once controlling for studentunobservables, this suggests that the process of social comparison on self-efficacy, from which self-concept derives, may be hampered by the largerdysfunctional nature of schoolingwithin the poorest part of the SouthAfricaneducationsystem. If feedbackonperformance is invalidorunreliable,studentswill struggle to accurately evaluate their performance and define their self-efficacy. It therefore comes as no surprise that, on average, better performingstudentswithhigherabilityorbetterequippedtoevaluatetheirself-efficacy.
Ofmore interest and relevance to theprimaryquestionathandare theresultsestimatedforthesampleofQ5schools.Aswouldbeexpected,andunlikewhatwasfoundforQ1toQ4schools,studentunobservablecharacteristics(oratleast, subject invariant ones) are positively correlated with self-concept andmotivation. Furthermore, controlling for these student unobservables andteacher observable characteristics, girls’ test scores are positively significantlyinfluenced by teacher education, in terms of both the level and type. What isclearly suggested by the results of table 15 is that teacher characteristicsinfluence the relationship between student self-concept and performance.Specifically,girlsexperienceastrongerself-concepttoperformancerelationshipwhen taught by a teacher with a focus on education training than not.Additionally, the self-concept effect is larger when girls are taught by maleteacherswitheducationbackgrounds,whichmight suggest thatgirls,believingin the stereotype that men are better than women at maths and science, arelikely to undervalue their own efficacy when taught by a female teacher. Analternative argument might be that male teachers are able to instil morediscipline in their classrooms, especially at an age where posturing andcompetition amongst boys could occur, which can assist in creating anenvironment that isbeneficial togirls’ learningexperiences. Whilst thismightwork to the favour of girls, it appears to hurt boys; similarly, boys taught bywomen with education training experience a significant self-concept toperformancerelationship.
This latter finding (although based on a relatively small, albeit betterfunctioning, sample of South African high schools), in combination with thefindings from the classroom fixed effect models, provides suggestive evidence
38
that student endorsement of gender stereotypes as well as stereotype threatexperienced by female teachers may be contributing to poorer performance,particularly inmath, among female students.This finding is inagreementwiththe recent findings of Antecol et al (2014) in the economics literature, andstressestheimportanceoffutureworkinthisareaaswellasadvocatesfortheuseofpath-analysismodellingthatcanrevealnotonlythemoderatingrole,butalso themediating role of teacher characteristics on girls’ attainment in STEMsubjects.
However, it needs to be pointed out that, given the differences in thegenderandtraining-backgrounddistributionofteachersacrosssubjectdomain,itisdifficulttoarguethattheestimatedcoefficientsarefreefrombiasdrivenbyteacherunobservables,bothsubject-variantandsubject-invariant.Theteacher-gender dynamics observed here would need to be tested against other data,preferably the TIMSS grade 4 and grade 8 2015 datasets. Modelling similarrelationships in the grade 4 data would serve to understand whether or notindications of differential self-concept linked to performance across boys andgirls emerge at younger ages. This could explain at which stage of the schoolexperience the academic course of students, particularly of girls, is shaped.PoolingtheTIMSS2011and2015datasetswouldhelptodeterminewhetherornottheresultspresentedherearesampledriven.Afurthermediatingfactorthatwas not directly explored by this paper is the SES of the home and parentalinvolvement. The stereotype threat observed within the wealthier part of theschool system, and hence amongst middle class families, may be due todifferencesinthemannerinwhichmathrelated“culturalresources”aresharedamongstboysandgirls(c.f.Muller,1998).
The findings of this paper and the proposed future work present anumberofpolicy implications.Weneedtobetterunderstandatwhichstageofschooling gender differences in performance emerge; knowing that there is agenderdifferenceintheperformanceandself-conceptsofgrade9boysandgirlsinmathandsciencemightbemetwithadesiretobettertargetandinformthesubject choices of girls in high school, but this alone is not likely to solve theproblem.Considerationneedstobegiventothesocialfactorsthatcontributetogenderdifferencesduringearlierstagesofschooling,aswellasperpetuateandextend them as students move through the education system, need to bedeveloped.AsarguedbyPennerandParet(2008:251),“strategiesforreducinggaps… need to be [placed]… within larger structures of inequality... it isimportanttofocusonthewaysinwhichprocessesofacademicachievementareembedded in and shaped by the hierarchical structures of difference andinequality”.
Classroom practices and methodologies that reflect and reinforcestructuresofgenderinequalityneedtobeidentified,andreplacedwithonesthatenhancegenderparity.Forexample,whilstthelearningstylesoffemalestudents
39
tend to emphasise mastery (pursuing work in the hope of understanding thematerial) over performance (focus on one’s marks) in task completion, thereversehasbeenshownformalestudents(Kenney-Bensonetal,2006).Whichofthese are emphasised or encouraged in a classroom setting (or by the schoolsystem in general) may play an important role in the relative performance ofgirls andboys. Similarly, genderdifferences in socialisation that contribute todifferences in activity level and temperament could affect teachers’ subjectiveperceptionsofstudents,andthereforetheirperformances(Bennetetal,1993).Finally, complimentary to the above mentioned could be policies targeted atcreatingandraisingthegenderawarenessofteachers’andparents’ inordertocounteractstereotypesaboutfemalemathinferioritycouldalsobeemphasised.
Studentcontrols Yes Yes Yes YesTeachercontrols Yes No Yes NoClassroomfixedeffects No Yes No YesR-squared 0.273 0.120 0.243 0.143Observations 2713 2713 4148 4204
Note:mathscoreismeasuredbythefirstplausiblevalue(scaledtoaninternationalM=500andSD=100).Studentcontrolsincludewhetherthestudenthastheirownbooksathome,howoftenthe student speaks the language of the test and the socio-economic status of the household.Cluster(classroom)robuststandarderrorsareshowninparentheses.Sampleweightingistakenintoaccount.***p=0.01,**p=0.05,*,p=0.10.
Studentcontrols Yes Yes Yes YesTeachercontrols Yes No Yes NoClassroomfixedeffects No Yes No YesR-squared 0.493 0.360 0.653 0.230Observations 953 953 469 496
Note:mathscoreismeasuredbythefirstplausiblevalue(scaledtoaninternationalM=500andSD=100).Studentcontrolsincludewhetherthestudenthastheirownbooksathome,howoftenthe student speaks the language of the test and the socio-economic status of the household.Cluster(classroom)robuststandarderrorsareshowninparentheses.Sampleweightingistakenintoaccount.***p=0.01,**p=0.05,*,p=0.10.
Studentcontrols Yes Yes Yes YesTeachercontrols No No No NoClassroomfixedeffects Yes Yes Yes YesR-squared 0.145 0.110 0.178 0.142Observations 3574 946 3428 700
Note:sciencescoreismeasuredbythefirstplausiblevalue(scaledtoaninternationalM=500and SD = 100). Cluster (classroom) robust standard errors are shown in parentheses. Sampleweightingistakenintoaccount.***p=0.01,**p=0.05,*,p=0.10.
43
ReferencesAntecol, H., Eren, O. and Ozbeklik, S., 2014. The effect of teacher gender on
Aronson, J. and Steele, C.M. 2005. Stereotypes and the Fragility of AcademicCompetence,MotivationandSelf-Concept.InA.J.Eliot&C.S.Dweck(Eds.),Handbook of Competence and Motivation (pp. 436 – 456). New York:GuilfordPress.
Atkinson, R., and Mayo, M. 2010. Refueling the US Innovation Economy: FreshApproaches to Science, Technology, Engineering and Mathematics (STEM)Education. Executive Summary. Information Technology and InnovationFoundation.
Bandura, A. 1977. Self-efficacy: Toward a Unifying Theory of BehaviouralChange.PsychologicalReview,84(2):191.
Bandura, A. 1986. Social Foundations ofThought andAction:A Social CognitiveTheory.EnglewoodCliffs,NJ:PrenticeHall.
MathAnxietyAffectsGirls'MathAchievement, Proceedings of theNationalAcademyofSciences,USA,107(5):1060-1063.
Bennett, R.E., Gottesman, R.L., Rock, D.A. and Cerullo, F. 1993. Influence ofbehaviour perceptions and gender on teachers’ judgements of students’academicskill.JournalofEducationalPsychology,85:347-356.
Bleeker, M.M. and Jacobs, J.E. 2004. Achievement in Math and Science: DoMothers' Beliefs Matter 12 Years Later? Journal of EducationalPsychology,96(1):97.
Crocker, J.,Major,B. andSteele,C.M.1998.Social stigma. InD.T.Gilbert, S.T.Fiske, and G. Lindzey (Eds.), Handbook of Social Psychology (4th ed.,pp.504-553).Boston:McGraw-Hill.
Eccles, J.S.,Midgely,C.andAdler,T.1984.Grade-RelatedChanges intheSchoolEnvironment: Effects on Achievement Motivation. In J. Nicholls (Ed),Advances inMotivationandAchievement:TheDevelopmentofAchievementMotivation(Vol.3,pp.283–331).Greenwich,CT:JAIPress.
Eccles, J., Adler, T.F., Futterman, R., Goff, S.B., Kaczala, C.M., Meece, J.L. andMidgley,C.1985.Self-perceptions,taskperceptions,socializinginfluences,and the decision to enroll in mathematics.Women and mathematics:Balancingtheequation:95-121.
Farkas, G. 2003. Racial disparities and discrimination in education. TeachersCollegeRecord,105:1119-1146.
Festinger, L. 1954. A theory of social comparison processes.Humanrelations,7(2):117-140.
Fin,L.S.andIshak,Z.2012.APrioriModelofStudents’AcademicAchievement:The Effect of Gender as Moderator.Procedia-Social and BehavioralSciences,65:1092-1100.
Fiske, S.T. 1992. Thinking is for Doing: Portraits of Social Cognition fromDaguerreo-TypetoLaserphoto.JournalofPersonalityandSocialPsychology,63:877–889.
Frome,P.M.andEccles,J.S.1998.Parents'InfluenceonChildren'sAchievement-Related Perceptions.Journal of personality and social psychology,74(2):435.
45
Goni, U. and Bello, S. 2016. Parental socio-economic status, self-concept andgender differences on students' academic performance in Borno Statecollegesofeducation:Implicationsforcounselling.JournalofEducationandPractice,7(14):21-27.
Grolnick, W.S. and Ryan, R.M.,1987. Autonomy in children's learning: Anexperimentalandindividualdifferenceinvestigation.Journalofpersonalityandsocialpsychology,52(5):890.
Hanushek, E.A. andRivkin, S.G. 2010. Generalizations about using value-addedmeasures of teacher quality.TheAmericanEconomicReview,100(2): 267-271.
Helwig, R., Anderson, L. and Tindal, G. 2001. Influence of elementary studentgender on teachers’ perceptions of mathematics achievement. Journal ofEducationalResearch,95:93-102.
Hill, C., Corbett, C., and St Rose, A. 2010. Why so few? Women in Science,Technology, Engineering, and Mathematics. American Association ofUniversityWomen,Washington,DC.
Holling,H. andPreckel, F. 2005. Self-estimates of intelligence––methodologicalapproaches and gender differences.Personality and IndividualDifferences,38(3):503-517.
Jacobs,J.E.,Lanza,S.,Osgood,D.W.,Eccles,J.S.andWigfield,A.2002.Changesinchildren’s self-competence and values: Gender and domain differencesacrossgradesonethroughtwelve.Childdevelopment,73(2):509-527.
Kahle, J.B.,Meece, J. and Scantlebury, K. 2000. UrbanAfrican-Americanmiddleschool science students: Does standards-based teaching make adifference?.JournalofResearchinScienceTeaching,37(9):1019-1041.
Kane, T.J., Taylor, E.S., Tyler, J.H. and Wooten, A.L. 2011. Identifying effectiveclassroom practices using student achievement data.Journal of humanResources,46(3):587-613.
46
Kenney-Benson, G.A., Pomerantz, E.M., Ryan, A.M. and Patrick, H. 2006. Sexdifferences in math performance: The role of children’s approach toschoolwork.DevelopmentalPsychology,42:11-26.
Levine, S.C., Vasilyeva, M., Lourenco, S.F., Newcombe, N.S. and Huttenlocher, J.2005. Socioeconomic status modifies the sex difference in spatial skill.PsychologicalScience,16:841-845.
Marsh, H.W. 1993. The multidimensional structure of academic self-concept:Invariance over gender and age.American Educational ResearchJournal,30(4):841-860.
Marsh,H.W.,Hau, K. T., Artelt, C., Baumert, J., and Peschar, J. L. 2006.OECD'sBriefSelf-reportMeasureofEducationalPsychology'sMostUsefulAffectiveConstructs:Cross-cultural,PsychometricComparisonsAcross25Countries.InternationalJournalofTesting,6(4):311-360.
Marsh, H.W. and Craven, R.G. 2006. Reciprocal Effects of Self-Concept andPerformance From a Multidimensional Perspective: Beyond SeductivePleasure and Unidimensional Perspectives. Perspective Psychology, 1(2):133-163.
Marsh, H. W., and O'Mara, A. 2008. Reciprocal Effects Between Academic Self-concept,Self-esteem,Achievement,andAttainmentOverSevenAdolescentYears:UnidimensionalandMultidimensionalPerspectivesofSelf-concept.PersonalityandSocialPsychologyBulletin,34(4):542-552.
Marsh, H.W. and Yeung, A.S. 1997. Academic Self-Concept andAchievement.AmericanEducationalResearchJournal,34(4):691-720.
Marsh, H.W. 1996. Positive and Negative Global Self-Esteem: A SubstantivelyMeaningful Distinction or Artifactors? Journal of Personality and SocialPsychology,70:810–819.
Marsh, H.W. and Craven, R.G. 2006. Reciprocal effects of self-concept andperformance from a multidimensional perspective: Beyond seductivepleasure and unidimensional perspectives.Perspectives on psychologicalscience,1(2):133-163.
Meece, J.L. and Kurtz-Costes, B.E. 2001. Introduction: The Schooling of EthnicMinorityChildrenandYouth.EducationalPsychologist,36(1):1-7.
Muller, C. 1998. Gender differences in parental involvement and adolescents’mathematicsachievement.SociologyofEducation,71:336-356.
Oakes, J. 1990.Chapter3:Opportunities, achievement, and choice:womenandminority students in science and mathematics.Review of Research inEducation,16(1):153-222.
47
Okeke,N.A. 2009.Race stereotypes, academic self-concept and racial centralityin African American adolescents(Doctoral dissertation, The University ofNorthCarolinaatChapelHill).
Pajares,F.,andMiller,M.D.1994.RoleofSelf-efficacyandSelf-conceptBeliefsinMathematical Problem Solving: A Path Analysis. Journal of EducationalPsychology,86(2):193.
Pajares,F.andSchunk,D.H.2001.Self-BeliefsandSchoolSuccess:Self-Efficacy,Self-Concept, and School Achievement. Chapter in R. Riding & S. Rayner(Eds),Perception(pp.239–266).London:AblexPublishing.
Pajares, F. and Urdan, T. 2006. Adolescence and Education: Vol. 5. Self-EfficacyBeliefsofAdolescents.Greenwich,CT:InformationAge.
Parsons, J.E., Adler, T.F. and Kaczala, C.M. 1982. Socialization of achievementattitudesandbeliefs:Parentalinfluences.Childdevelopment:310-321.
Penner,A.M.&Paret,M.2008.Genderdifferencesinmathematicsachievement:Exploring the early grades and the extremes. SocialScienceResearch, 37:239–253,
Schunk, D.H. and Mullen, C.A. 2012. Self-efficacy as an engaged learner.InHandbookofresearchonstudentengagement(pp.219-235).SpringerUS.
Sheldon, K.M. and Kasser, T. 1995. Coherence and congruence: two aspects ofpersonality integration.Journalofpersonalityandsocialpsychology,68(3):531.
Skaalvik, E.M. andHagtvet,K.A. 1990.Academic achievement and self-concept:An analysis of causal predominance in a developmentalperspective.JournalofPersonalityandSocialPsychology,58(2):292.
Skinner,B.F.1953.Scienceandhumanbehavior.SimonandSchuster.Slee, R., Weiner, G. and Tomlinson, S. 1998. School Effectiveness for Whom?
Challenges to the School Effectiveness and the School ImprovementMovements.Londer,FalmerPress.
Spencer, S.J., Fein, S., Wolfe, C.T., Fong, C. and Duinn, M.A. 1998. AutomaticActivation of Stereotypes: The Role of Self-Image Threat.Personality andSocialPsychologyBulletin,24(11):1139-1152.
Steele, C.M. 1992. Race and the Schooling of Black Americans.The AtlanticMonthly,269(4):68-78.
Steele, C.M. 1999. Thin Ice: “Stereotype Threat” and Black College Students.AtlanticMonthly,284(2):44–54.
Steele, C.M., andAronson, J. 1995. StereotypeThreat and the Intellectual TestPerformance of African Americans. Journal of Personality and SocialPsychology,69(5):797.
Thrupp, M. and Lupton, R. 2006. Taking School Contexts More Seriously: TheSocialJusticeChallenge.BritishJournalofEducationalStudies,54(3):308-328.
Thrupp, M., Lauder, H and Robinson, T. 2002. School Composition and PeerEffects.InternationalJournalofEducationalResearch,37(5):483-504.
Usher,E.L.andPajares,F.2008.SourcesofSelf-EfficacyinSchool:CriticalReviewof the Literature and Future Directions. Review of Educational Research,78(4):751-796.
Valentine,J.C.,DuBois,D.L.andCooper,H.2004.Therelationbetweenself-beliefsand academic achievement: A meta-analytic review. Educationalpsychologist,39(2):111-133.
Vallerand, R.J. and Blssonnette, R. 1992. Intrinsic, extrinsic, and amotivationalstyles as predictors of behavior: A prospective study.Journal ofpersonality,60(3):599-620.
Vyas, S. and Choudhary, G. 2016. Study habits of sr. sec. school adolescentstudentsinrelationtotheirsocio-economicstatus.IJAR,2(6):134-139.
Webb-Williams, J. 2014. Gender differences in school children's self-efficacybeliefs: Students' and teachers' perspectives.Educational Research andReviews,9(3):75.
Winters,M.A.,Haight,R.C.,Swaim,T.T.andPickering,K.2013.Theeffectofsame-genderteacherassignmentonstudentachievementintheelementaryandsecondary grades: Evidence from panel data. Economics of EducationReview,34:69–75.
Wood, J.V. 1996. What is social comparison and how should we studyit?.PersonalityandSocialPsychologyBulletin,22(5):520-537.
