Academic Resilience in Challenging Contexts:Evidence From Township and Rural Primary Schools

in South Africa

GABRIELLE WILLSHELEEN HOFMEYR

Stellenbosch Economic Working Papers: WP18/2018

www.ekon.sun.ac.za/wpapers/2018/wp182018

November 2018

KEYWORDS: Student achievement; Exceptional performance; Literacy;South Africa

JEL: J20; J24; J25

ReSEP (Research on Socio-Economic Policy)http://resep.sun.ac.za

DEPARTMENT OF ECONOMICSUNIVERSITY OF STELLENBOSCH

SOUTH AFRICA

A WORKING PAPER OF THE DEPARTMENT OF ECONOMICS AND THEBUREAU FOR ECONOMIC RESEARCH AT THE UNIVERSITY OF STELLENBOSCH

www.ekon.sun.ac.za/wpapers

1

ACADEMICRESILIENCEINCHALLENGINGCONTEXTS:EVIDENCEFROMTOWNSHIPANDRURALPRIMARY

SCHOOLSINSOUTHAFRICA

GabrielleWillsandHeleenHofmeyr1,2

2November2018

Abstract:

Povertyisconsideredariskfactorthatjeopardizeschildren’sacademicperformance.However,evenin high-poverty contexts there are students who manage to achieve consistently good academicresults. This paper uses a resilience framework to identify and describe the characteristics ofstudents fromSouthAfricanruralandtownshipprimaryschoolswhoperformabovedemographicexpectations in reading comprehension.Weuse a rich longitudinal datasetof over 2600Grade6students that contains information on institutional and individual protective factors, includingstudents’ socio-emotional skills (perseverance, aspirations, and attitudes toward school). Thelongitudinaldimensionof thedataprovidesauniqueopportunity tonotonly identify consistentlyhigherachievers,butalsostudentswhoseliteracyskillsimprovedsignificantlymorethantheirpeersduring a school year. After accounting for differences in socio-economic status and other homebackground factors such as English language exposure, we find that resilient students differsignificantly from their lower-achievingpeers along variousdimensions. Students’ socio-emotionalskills emerge as particularly strong correlates of academic resilience. Although individual-levelprotectivefactorsappeartobethestrongestdeterminantsofacademicresilience,classroomfactorssuchtime-on-taskandtheavailabilityoftextsalsoplayanimportantrole.Thesefindingsaddtoourunderstandingoffactorsassociatedwithacademicsuccessinchallengingcontexts.

1GabrielleWills andHeleenHofmeyr are both researchers at ReSEP (Research on Socio-Economic Policy), StellenboschUniversity. Postal address: Department of Economics, Schumann Building, De Beer Street, Stellenbosch University,Stellenbosch7600.Email:heleenh24@gmail.com.

2TheauthorswouldliketothankDavidCarelforhisexcellentmanagementoffieldworkforthisprojectandhisinclusionofquestions on student aspirations and attitudes in data collection instruments.We acknowledge the comments ofServaas van der Berg andNic Spaull on earlier drafts of this paper. Funding from the Economic and Social ResearchCouncil(ESRC)isherebyacknowledged[grantES/N01023X/1].OpinionsexpressedandconclusionsarrivedatarethoseoftheauthorsandarenotnecessarilyattributedtotheESRC.

2

I. IntroductionEducational outcomes and socio-economic status (SES) are strongly linked across countries andeducationsystems(OECD2011b).Despitethisstrongandpersistentassociation,agrowingbodyofliterature documents the existence of students from disadvantaged socio-economic backgroundswhomanage to achieve exceptional academic results (OECD 2011a; Agasisti & Longobardi 2017;Agasisti et al. 2018; Vera et al. 2015; Erberber et al. 2015; Cheung 2016). These students can beclassified as resilient since theymanage to overcome adversity to achieve academic success. Theliteratureonacademicresilienceisprimarilyconcernedwith identifying individualandinstitutionalprotectivefactorsthatunderliethisresilience.Suchstudiesaboundindevelopedcountries(Erberberet al., 2015; OECD, 2011b; Agasisti et al., 2018; Agasisti and Longobardi, 2017), however thequantitative evidence-base for academic resilience in developing countries is considerably smaller(Veraetal.2015).

Our study, located in the South African context, uses a unique dataset to contribute to ourunderstandingofacademicsuccessinchallengingcontexts.4Ourfirstresearchobjectiveistoidentifyacademic resilience in township and rural primary schools in three provinces. We consider howresilientstudentsaredistributedacrossschoolsofdifferingquality,andhowtheyperformrelativeto international samples of students who wrote the same test. Our second research objectiveexploresthewaysinwhichresilientstudentsdiffersystematicallyfromtheirlower-achievingpeers.Ouranalyticalstrategyaimstoidentifythemechanismsthatallowstudentstoobtainunusuallyhighresults relative to their peers even in the absence of crucial inputs such as an affluent socio-economicbackground.Wefindthatoneofthestrongestpredictorsofacademicresilienceissocio-emotional skills. This facet of learning currently receives very little attention in South Africaneducationpolicy.Our findingssuggest that futureresearchstandsmuchtogain from investigatingsocio-emotionalskillsascrucialinputsineducationproduction.

The next section providesmore background to the international literature on academic resilienceandresilienceinthelocalcontext.Section3describesthedataandanalyticalstrategyusedtobothdefine academic resilience in our study and identify its correlates. The results are presented inSection4.The researchandpolicy implicationsof the resultsarediscussed inSection5.Section6concludes.

II. BackgroundPsychologistsfirstusedthetermresiliencetorefertopositiveadaptationinthefaceofadversityinthe 1960’s (Masten, 2012). This termwas subsequently applied in education research to refer tostudents who manage to achieve good academic results despite considerable risks to theireducational success. We situate our analysis in a social ecology view of human development(Bronfenbrenner,1994),wherebyresilienceisthoughttoresultfromthedynamicinteractionofrisksand protective factors that exist at the level of the individual, the family, and the broadercommunity.Thistheoryofresilienceinformstheanalysispresentedinthispaper.

There are two strands of literature that underpin our adoption of this theoretical position. Onestrand can broadly be termed literature on academic resilience and encompasses contributionsmainly from educational psychology. Studies in this strand typically employ qualitative researchmethodologies. The second strandof literature comprises studies that use large-scale assessment

4Thedatacomes fromtheproject “Understanding resilienceandexceptionalism inhigh-functioning townshipand ruralprimaryschoolsinSouthAfrica”,orbrieflyreferredtoasthe“Leadershipforliteracy”project.ThisprojectisfundedbytheEconomicandSocialResearchCouncil[grantES/N01023X/1].

3

data such as the Programme for International Student Assessment (PISA), the Programme toImproveReadingLiteracy(PIRLS)andTrendsinMathematicsandScienceStudy(TIMSS).Thelatterbodyofwork reliesonquantitativeanalysesofacademic resilience.Findings frombothstrandsofthisliteraturesuggestprotectivefactorsattheleveloftheindividual,homeandbroadercommunityare important for fostering resilience, providing empirical support for the social ecology view ofresilience.

Thevalueofself-confidence,aspirations,andattitudestowardsschool

Thetwostrandsofliteratureconvergeonthenotionthatsocio-emotionalskillsareamongthemostimportantprotectivefactorsforfosteringresilience.Anumberofstudiesineducationalpsychologyfindthatstudents’beliefintheirself-efficacyandself-confidencearestrongpredictorsofacademicresilience(Cappella&Weinstein2001;Shumowetal.1999;Gizir&Aydin2009;Borman&Overman2004;Werner 1997). Furthermore, positive attitudes toward school are generally associatedwithacademicresilience(Finn&Rock1997;Moller1995;Das2018;Waxmanetal.1997;Brookoveretal.1978;Wright&Masten2015;Borman&Overman2004).Havinghigheducationalaspirationsisalsoconsideredimportantforresilience(Benard1997;Cappella&Weinstein2001;Gizir&Aydin2009).

Similarly,large-scalequantitativestudiesofresiliencefindthatself-efficacybeliefs,self-confidence,educational aspirations, and attitudes toward school are the strongest and most consistentpredictorsofresilienceamongsocio-economicallydisadvantagedstudents.ThiswasfirstconfirmedintheOECDstudy“AgainsttheOdds:DisadvantagedstudentsWhoSucceedinSchool”(OECD,2011)that used PISA data. The study showed that evenwhen augmenting resiliencemodels to includeindividual-level factors and school-level factors, student self-efficacy remains themost consistentpredictorofresilienceinalmostallcountriesparticipatinginPISA(OECD,2011).Sandoval-Hernandez&Cortes(2012)conductasimilaranalysiswithPIRLS2006dataandfindthatself-confidencewasthestrongest predictor of resilience in reading literacy across the participating countries. Having apositive attitude towards reading and higher educational aspirations also emerged as significantcorrelatesofresilienceinSandoval-Hernandez&Cortes’study.

Using TIMSS 2011 data, Erberber et al. (2015) find that in themajority of participating countries,students’ educational aspirations are the strongest and most consistent predictor of academicresilienceofallprotectivefactorsconsideredinthestudy.Sandoval-Hernandez&Bialowoski(2016)alsouseTIMSS2011dataandlikeErberberetal.,theyfindthateducationalaspirationsarestronglycorrelatedwith resilience in the five countries they consider (Singapore, SouthKorea,HongKong,ChineseTaipei,andJapan).

Ratherthanusingcross-countryanalyses,anumberofstudieshaveusedthePISAandPIRLSdatatoinvestigate predictors of resilience among socio-economically disadvantaged students in countryspecificcontexts(seeforexampleVeraetal.(2015)forananalysisofresilientstudentsinChileandCheung (2016) for resilience inmathematics in China, Singapore, Hong Kong, Taiwan and Korea).Thesestudiesalsopointtotheimportanceofsocio-emotionalskillsforresilience.

ResilienceinSouthAfrica

South Africa presents a particularly interesting context for studying academic resilience becausecomparedtoothercountries, itexhibitsoneofthestrongestcorrelationsbetweenstudents’homebackgroundandacademicperformance(Taylor&Yu,2009).Shepherd(2016)findsthatdifferencesin home background explain themajority of the learning gap (roughly 60%) between poorer andwealthierSouthAfricanstudents.This strongcorrelationhas itsorigins in theapartheideducationsystem,which consisted ofmultiple racially defined departments of education. Apartheid policies

4

deliberately aimed to deliver inferior quality education to black students compared with whitestudents, and low quality persists today across former black schools (Spaull, 2013; Van der Berg,2008).Encouragingly,however,despitehistoricaldisadvantage,povertyandaccesstolowerqualityschools,asmallnumberofstudentsstillbeattheodds.PrePIRLS2011datarevealsthatabout5-6%ofSouthAfricangrade4childrenwritingAfrican language tests reachhighbenchmarks in reading(550 points or above). There are literally one or two of these higher achievers in over half of allclassroomsorschoolstestedinAfricanlanguages(Wills2017).

TherearefiveexistingstudiesonacademicresilienceinSouthAfricawhicharemostlyqualitativeintheirdesign.Dass-Brailsford (2005) studies first-yearuniversity studentswhocome fromtownshipschoolsinKwa-ZuluNatal,andfindsthattheindividualcharacteristicsofbeinggoal-oriented,havinginitiative and motivation, and having an internal locus of control were common in all herparticipants.Participantsalsoidentifiedreceivingemotionalsupportandhavingpositiverolemodelsas important in fostering resilience.Mampane&Bouwer (2006) find that resilientGrade 8 and 9students in township schools demonstrated independence, responsibility, assertiveness, sense ofcontrol, self-efficacy,planningand resourcefulness. In laterwork, theseauthorsposit that schoolsplayadistinctiveanddeterminingroleinfosteringresiliencebycreatingasupportiveteachingandlearningenvironment(Mampane&Bouwer,2011).Importantly,resilientstudentsweremorelikelythantheirnon-resilientpeerstoutilisethissupport.Mampane(2014)studiesresilienceamong291students in township high schools and finds these students attribute their resilience to self-confidence, an internal locus of control, commitment, being achievement-oriented, and havingaccesstosocialsupport.

Whileprovidingvaluable insight intosomeof the individualandcontextualprotective factors thatwereimportantinfosteringresiliencefromstudents’ownperspective,availablestudiesarelimitedin scope and sample size. The present study is the first to consider the correlates of academicresilienceamongalargersampleofstudentsinSouthAfrica.

III. DataandmethodData

ThedatausedwasgatheredforanESRC/DFID-fundedprojectentitled“Understandingresilienceandexceptionalism in high-functioning township and rural primary schools in South Africa”. In 2017,literacy tests scores were gathered from over 2600 Grade 6 students in 60 rural and townshipprimary schools. Although the student sample is not nationally representative, it is instructive forour purposes. The schoolswhere studentswere testedwere purposively selected to ensure theywerepositionedinchallengingcontexts.ExceptforfiveschoolschargingfeesoflessthanZAR2000peryear,theschoolsselectedareno-feeschools5.Theschoolswerealsoselectedtoreflectarangeof variation in schoolquality across townshipand rural areas. Theyarealsowidely geographicallylocatedinthreeprovinces(KwaZulu-Natal,LimpopoandGauteng).

Mostquantitativestudiesonacademicresilienceusetestdata fromonlyonepoint in time,whichmayresult in includingstudents inaresilientsamplewhomayhave justhadunusualsuccessonatest day. A key advantage of our data is its longitudinal dimension, which offers a uniqueopportunity to identify consistently higher achievers. A silent reading comprehension test andvocabularytestwereadministeredatthebeginningofthe2017schoolyearandagaintowardsthe

5No-feeschoolsconstitutethepoorest60%ofschoolsinSouthAfrica,basedonincome,unemploymentandilliteracyofthesurroundingarea.

5

endofthesameschoolyeartoanentireclassofgrade6studentsineachschool.Oftheoriginalpre-testsampleof2656students,2383wrotethepost-test,indicatingalowattritionrateof11%.

Thetwocomprehensiontestsconsistedofreleaseditemsfrompreviousroundsofthegrade4PIRLSassessment.PermissionwasreceivedfromtheIEAfortheiruse.Thereliabilityoftheseassessmentsisreflectedinahighcorrelationbetweenpre-testandpost-testscores6.Theuseofaninternationalassessment allows us to benchmark student proficiencies in our sample against internationalstandards.These testsare inEnglishand in this respectacademic resilience in this study is largelydefined in terms of exceptional performance in English. This may be criticised as a measure ofoverall academic resilience since studentsmayperformbadly in English reading butwell in othersubjects or languages. However, we maintain that English language proficiency is a necessaryconditionforacademicsuccessinSouthAfricanschools.Schoolsareatlibertytochooseoneofthe11officiallanguagesastheirdominantmediumofinstructioninfoundationphasegrades(gradesR-3),butareallrequiredtoteachinEnglishorAfrikaansfromgrade4onwards.IfchildrencannotreadandwriteinEnglishbytheendofgrade3itisverydifficultforthemtoaccessthecurriculum.

Inadditiontothecomprehensiontests, theresearchteamalsoadministeredsurveystoprincipals,teachers,andstudents.Informationfromthesewereusedtoderivetheindividual,homeandschoolvariables that we include in our analysis. The student survey included a grit scale, adapted fromDuckworthandQuinn(2009).Grit,or“perseveranceandpassionforlong-termgoals”isarelativelynewconstruct inpositivepsychology introducedbyDuckworthetal. (2007).Grit isahigher-orderconstruct which consists of two first-order constructs, namely consistency of interest andperseveranceofeffort.Itisdistinguishablefromresilienceinthatgritisdefinedasacharactertrait(Duckworth&Quinn2009),whileresiliencedenotesaprocessofpositiveadaptationinthefaceofadversity.Grit isthereforeoneindividual-levelprotectivefactorwhichmaycontributetoacademicresiliencealongwithotherindividual-,home-,andcommunity-levelprotectivefactors.

Althoughtheliteratureonthelinkagesbetweengritandacademicachievementisstillinitsinfancy,existing studies suggest it may be a strong determinant. In developed countries a positiveassociation is usually identified between grit and academic performance (Duckworth et al., 2007;Duckworth&Quinn, 2009;Duckworth et al., 2011;Mason, 2018; Reraki et al., 2015; Lee& Sohn,2017; Akin et al., 2014;Rojas and Usher, 2012; and O’Neal, 2017, 2018). To our knowledge onlythreestudiesexplorethis topic indevelopingcountries (Datuetal.2016;Orioletal.2017,Mason2018).Mason(2018)administertheGritScaletoasampleoffirst-yearuniversitystudentsinSouthAfricaandfindthatself-reportedgritwaspositivelyassociatedwithacademicperformance.Orioletal (2017) do the same with primary and secondary school students in Peru and find that grit ispositivelyassociatedwithself-efficacy,whichisanimportantdeterminantofacademicachievement.

Afindingemergingfromtwoofthesethreestudies isthatthe‘consistency-of-interest’subscaleofgrit might be less relevant for predicting academic outcomes than ‘perseverance-of-effort’ indevelopingcountrycontexts.7For this reason, inaddition tootherconsiderations,we includeonlytheperseverance-of-effort subscale inour analysis (Datuet al. 2016;Mason2018). The scalewas

6ThePearsoncorrelationstatisticbetweenthepre-testandpost-testwasalmost0.90.7 Datu et al (2016) find that only the perseverance-of-effort subscale of grit is correlated with academic achievement

among high school and university students in the Philippines. InMason’s study the consistency-of-interest subscaleaccountedfor3%ofthevarianceinstudents’academicachievementscores,whiletheperseverance-of-effortsubscaleaccountedfor9%ofthevarianceinacademicachievement.

6

adjustedfortheSouthAfricancontextandtheagegroupofstudentsinthesample.ThequestionscanbefoundintheAppendix.

Method:Definingacademicresilience

Masten and Obradovic (2008) provide a useful typology for different approaches to definingacademic resilience. Initially, they conceptualise resilience as reaching achievement levels abovethoseexpected foraparticularat-riskgroup.This isacommondefinitionalapproachalsousedbyotherquantitativestudiesonacademicresilience(Sandoval-Hernandez&Bialowoski2016;Agasistiet al. 2018; Vera et al. 2015). Applied to our data, this at-risk group are socio-economicallydisadvantaged students from township and rural schools that consistently exceed sampleexpectations in a test administered at the beginning and end of the school year. In a seconddefinition (Masten&Obradovic2008), resilience is conceptualised in termsof recovery tonormalachievementlevelsinthefaceofadversity.Inthisvein,conceptualisingacademicresilienceintermsoflearningimprovementsratherthanjustintermsoflevelsisappropriate.Withtwopointsofdata,wecanidentifystudentswhodisplayunusualimprovementsinliteracyoverthecourseofaschoolyear.

To identify resilient studentwe regress students’ test scores on student SES, as per the followingequation:

!"# = &' + &)*+*"# + ,"#(1)Here!"#representseitherthestandardizedpre-testorpost-testscoresoftheithstudentinschools8.*+*"# isavectorofthestudents’backgroundcharacteristics includingameasureofwealthanditssquare9;thenumberofbooksinthestudent’shome;whetherornottheyspeakEnglishathomeandthe rural or urban status of the school they attend. The residuals of the equation, ,"# , (thedifference between the actual value of Y and the predicted value of Y) can be used to identifystudentswhoconsistentlyperformabovesampleexpectationsafteraccountingforaselectnumberof socio-economicbackground factors.Academically resilient studentsaredefinedas thosewhoseresidualperformance lies twostandarddeviationsabove theirpredictedperformanceatboth testpoints.

In conceptualising a comparison group against which to compare resilient students, it wasappropriatetoconsideratypicallyperforminggroupofstudentswithinthesample.Thereasonforthis is that backgrounddata among average performerswith some level of literacy is likely to bemore reliable, and less likely to be subject to non-response bias, than for a veryweak sample ofstudents with low levels of literacy. In this regard, the comparison group is initially defined asperformingwithinone standarddeviationof the sampleaverageof residualperformance (that is,performanceafteraccountingfortheeffectofSES).Twovariationsondefinition1arealsousedinthe analysis to consider the sensitivity of the results to the exclusion of weaker students and toaddressconcernsthatthechoiceofcut-offpointsforselectingresilientstudentsmaybearbitrary.

Theseconddefinitionofacademicresilienceconsidersperformanceintermsofliteracygainswithinaschoolyear(ratherthanlevels)thatexceedsampleexpectations.Toobtainthisindicatorasimilarequationisusedexceptthatpre-testscoresarecontrolledforinthisvalue-addedmodel:

-./001/0"# = &) + &2-3101/0"# + &4*+*"# + ,"#(2)

8Post-testscoresarestandardizedintermsofthepre-testscoremeanandstandarddeviation.9 The wealth index is derived from questions on asset ownership in the student’s home constructed using principal

componentsanalysis.

7

The advantage of this approach is that one can factor out historical accumulation of exposure tohome background, teacher and school quality factors to determine which contemporaneousindividualandinstitutionalfactorsareassociatedwithhigherthanexpectedlearning.Academicallyresilientstudents,whendefined intermsoftestscoregains,arethosewhoseresidualgainsfall inthe90thpercentileofthedistributionofgainscores.Theselearnersarecomparedtotheirpeerswhohadimprovementsintheircomprehensionscoresthatlaybetweenthe25thandthe75thpercentileofthesampledistribution,aftercontrollingforSES.

A limitation of both approaches to defining resilience is that they rely on relative performancemeasures. InSection IVweconsiderhowtheperformanceof these resilientstudentscompares inabsolutetermstointernationalsamplesofstudents.

Analyticalstrategy

Alogisticregressionisusedtodeterminewhichindividualandinstitutionalfactorsarelinkedtotheprobabilityofbeingacademically resilientasopposed to typicallyperformingorunderperforming.Wemodeltheprobabilityofbeingresilientasafunctionofindividual,householdandschoolfactorsthataretheoreticallylinkedtoacademicachievement.TablesA1andA2intheAppendixprovideadescription of all the variables included in the estimation and the questions informing theconstructionof indices.Weprovidedescriptions of the explanatory variables included in the logitmodelinthenextsection.

IV. ResultsIdentifyingresilientstudents

Theresultsoftheestimatingequations1and2areshowninTable1.Inthemodelsusingtestscorelevels rather than a value-added approach, the wealth index is shown to be positively andsignificantlyrelatedtoEnglish literacyscores. Inaddition,students inruralsettingsscore loweronthetestthantheirurbancounterparts.MorefrequentuseofEnglishathomeisalsopositivelyandsignificantly associated with reading test scores, but having more books at home provides littleadditional explanatory power to the models. In the value-added model, with estimation resultsshown in the last column of Table 1, the pre-test score,which absorbswealth effects on Englishreadingperformance,explainsmostoftheperformancevariation(80%)inpost-testscores.

Theresidualsfromestimationsusingpre-testandthenpost-testscorelevelsareplottedinFigure1(page 9). Using our first approach to defining resilience, we identify 87 resilient students in oursampleof2383.Graphicallythese87studentsarerepresentedbythesolidgreenscattereddotsinFigure 1. Initially the group to which they are compared, the 1 563 students performing onestandarddeviationabout theaverage,arerepresentedbytheorangescatteredpoints inFigure1.Variations on definition 1 are used in the analysis to consider the sensitivity of the results to theexclusionofweakerstudentsandtoaddressconcernsthatthechoiceofcut-offpointsforselectingresilientstudentsmaybearbitrary.Definition1aexpandsthecomparisongrouptoincludeallweaklyperforming students (i.e. even thoseperformingone standarddeviationbelow theaverage in thepre-test and post-test). Definition 1b maintains the original comparator group but expands thegroup of resilient students from 87 to 158 by including those with residual performance levelsbetween 1.5 and two standard deviations above average. These students are represented by thegreen outlined scatter points in Figure 1. Residuals from estimations of equation 2 are used toidentifystudentswithliteracygainsinthe90thpercentile.Thesestudentsareconsideredresilientintermsofdefinition2.ThesecondpanelofFigure1overlayspointsinblue,whichrepresentstudentsin the samplewith residual gains in the 90th percentile. Academically resilient learners defined in

8

these terms have varying levels of initial pre-test scores,with a strong representation of learnerswithlowinitialpre-testscores.

Table1Estimationoftotalmarks(expressedinz-scores)forEnglishcomprehensiontest

Pre-test Post-test Value-

addedWealthindex(assetownership) 0.295*** 0.320*** -0.0213

(4.64)

(4.26) (-1.45)

Wealthindex(assetownership)squared 0.0971** 0.0744 -0.0381**

(2.31) (1.53) (-3.04)

Booksathome: Afewbooks(10) -0.00456 -0.0812 -0.0759**

(-0.10)

(-1.35) (-2.16)

Enoughtofilloneshelf(20) -0.0431 -0.113 -0.0632

(-0.43) (-0.86) (-1.15)

Enoughtofillonebookcase(50) -0.0514 -0.113 -0.0530

(-0.42)

(-0.58) (-0.63)

Enoughtofilltwoormorebookcases(100) -0.0409 -0.370 -0.323**

(-0.20) (-1.44) (-2.21)

Location: Rural(1=Rural,0=Urban) -0.241** -0.309** -0.0299

(-2.11) (-2.06) (-0.63)

Englishspokenathome: Sometimes 0.307*** 0.483*** 0.127***

(5.70)

(5.83) (3.64)

Always-almostalways 0.346** 0.492** 0.0914

(2.55) (2.57) (1.46)

Pre-testscore: 1.158*** (50.25)

Constant -0.191** 0.291** 0.512***(-2.07) (2.23) (9.10)

Observations 2379 2379 2379

AdjustedR-squared 0.156 0.130 0.798

Source: Leadership for Literacy dataset. Notes: Reference categories include: no books at home, never speaks English at home, rural

location. Standarderrorsare inparenthesesand clusteredat the school level. Statistically significantat the following levels: *p<0.01,

**p<0.05,***p<0.001.

Table2(page9)describesaveragepre-testscores,post-testscoresandgains(expressedinz-scores)aswell as sample sizes for each group using the different definitions of resilience. The table alsoindicates the number of schools in which they are located. Considering how these academicallyresilient studentsaredistributedacrossschools reveals important findings.The87students in theresilientgroup(definition1)arescatteredovermorethanhalfoftheschoolsample-36ofthe60schools. When defined in terms of exceptional gains at or above the 90th percentile, resilientstudentsareevenmorespread,identifiedin51ofthe60schools.However,theseschoolsmayvarynotablyintermsofquality.Figure2ranksthegrade6classesacrossthe60schoolsinourstudyinorderofthetestscoreofthemiddlelearnerineachclass(i.e.medianclassperformance).Thefigurealsoshowsthehighestandlowestmarkineachclass.Onthesecondreversedaxis,theyellowbarsindicatethepercentageofresilientlearnersidentifiedineachschoolusingdefinition1.Itisevidentthatresilientlearnersaremorelikelytobeidentifiedinschoolsthatarebetterperformingoverall.For example, in the school classwith the highestmedianmark, 30% of learners are identified as

9

resilient.Butthereareclearexceptionstothistrend.Academicallyresilientlearnersarealsolocatedinschoolclasseswithverylowmedianlevelsofperformance.

Figure1residualperformanceinthepre-testandpost-test

Table2Academicresilienceindicators:Samplesizesandaverageperformance

Residualz-scoreAcademicallyresilient Comparisongroup Totalstudents

(%ofavailablesample)Mean Standard

DeviationNumberofstudentsa Mean Standard

DeviationNumberofstudentsa

Definition1(levels)

Pre-test 2.78 0.7487(36)

-0.16 0.421563(60)

1650(0.69)

Post-test 3.40 0.77 -0.22 0.57Gains 1.16 0.63 0.44 0.46

Definition1a(levels)

Pre-test 2.78 0.7487(36)

-0.33 0.531941(60)

2028(0.85)

Post-test 3.40 0.77 -0.44 0.71Gains 1.16 0.63 0.39 0.49

Definition1b(levels)

Pre-test 2.34 0.75158(42)

-0.16 0.421563(60)

1721(0.72)

Post-test 2.96 0.87 -0.22 0.57Gains 1.17 0.65 0.44 0.46

Definition2(gains)

Pre-test 0.31 1.02238(51)

-0.17 0.821191(60)

1429(0.60)

Post-test 1.48 1.23 -0.22 0.98

Gains 1.71 0.39 0.45 0.26

Source:Residualsareestimatedfromeitherequation(1)or(2).aValuesinbracketsshowthenumberofschoolsinwhichtheresilientstudentsarelocated.

10

Figure2Thepresenceofresilientlearnersingrade6schoolclassroomswithdifferentlevelsofmedianliteracytestscores

Source:Leadershipforliteracydataset.Note:Marksareshownastotalscores,notnormalisedvalues.Academicallyresilientlearnersaredefinedusingdefinition1.

11

Descriptiveevidenceofcompetence:Internationalcomparison

Our resilient studentshavebeendefined in termsof exceeding sampleperformanceexpectationsgiven their level of socio-economic status. But are these levels of performance educationallymeaningful in any way? From a human capital perspective, how does the performance of thesestudents compare to the competencies of students in other countries? Given the internationalnatureofthePIRLStests,wecanbenchmarkoursample’sperformanceagainstothercountries.Wecomparetheperformanceofouracademicallyresilientandtypicallyperformingstudents(identifiedusing definition 1) to studentswhowrote the same released PIRLS comprehension tests in othercountries in PIRLS 2011. Figure 3 shows the percentage of items on a PIRLS passage correctlyansweredbystudentsatthe10th,50thand90thpercentileinlowertomiddleincomecountriesandthreewealthiercountries.Mostof thePIRLScountriesadministeredthetestsat thegrade4 level,thus our grade 6 resilient students’ performance is compared with the performance of grade 4studentsinothercountries.

The median performance of our 87 resilient students is comparable to performance at the 90thpercentileforother lower-to-uppermiddle-incomecountries ingrade4.Furthermore,ourresilientstudents’performanceatthe10thpercentileiscomparabletoAustralia’smedianperformance,andcomparableatallpercentilestoFinland’sperformance.EventhoughAustraliaandFinland’sstudentsarealmosttwogradesaheadofourresilientstudents,thisresultisencouragingnonetheless,giventhehighlychallengingandunder-resourcedenvironmentsfromwhichourresilientstudentscome.

Figure3Internationalcomparisonofreadingperformance

Source:LeadershipforliteracydatasetandPIRLS2011releaseditems,owncalculations.Resultsaresortedbythe50thpercentile.

Descriptivedifferencesacrossresilientandtypically-performingpeers

The explanatory variables used in the logit estimations for the set of resilient and typically-performingstudentsareshowninTable3(usingtheresilienceindicatorbasedondefinition1).Theresults in the table indicate that resilient students exhibit very different individual and homebackground characteristics compared to their typically-performing peers. They are wealthier,younger,more likely tobegirls, andmore likely tohaveattendedgradeR.Thegendergap in theprobabilityofbeingresilientinfavourofgirlsisnotsurprisingconsideringfindingselsewhereinthe

12

SouthAfricaneducationliteraturethatgirlsgenerallyoutperformboysbysignificantmarginsalonganumberofdimensions(suchasdropoutandrepetitionrates,matricresultsandsuccessatuniversity(VanBroekhuizen&Spaull,2017)).

Table 3 also shows that at home, resilient students are more likely to be exposed to protectivefactors that support literacydevelopment: Theyaremore likely tohaveemployedparents and tohave their own story books to read. These students are also more likely to ‘always’ or ‘almostalways’speakEnglishathomeandtoattendaschool inanurbanratherthanruralarea.Theyareless likelytoreportgettinghelpwiththeirhomeworkfromsomeoneathome.Theirscoresonthegrit index arenotablyhigher, theyhavemore aspirations for academic success, and report betterattitudestowardsschoolandtheirteacher.

Theschoolsattendedbyresilientstudentsinthissamplearealsosystematicallydifferentfromthoseattended by the comparison student group. The schools they attend have relatively wealthierstudentsandaremore likely touseEnglishas themediumof instruction in the foundationphase.There is evidence of more time-on-task in the classrooms of resilient students, as seen by thedifferences in student-reported teacher absence between resilient and typically-performingstudents.Theclassroomsof resilient students contain significantlymore texts thatare relevant toteaching English and their schools aremore likely to spend at least 33%of state allocated schoolbudgetsonbooks.Theirlanguageteachersarealsomorelikelytohavelanguagespecialisations.

Table 3 further shows that these observed differences between resilient and average-performingstudentsdefinedaccordingtotestscorelevelsalsoemergewhenusingaresilienceindicatorderivedfromtestscoregains.That is, thesedifferencesemergeevenwhenwecontrol forpre-testscores,which may be a proxy for an accumulated history of inputs underpinning students’ academicpreparedness. Importantly, this suggests that many of the individual, home and particularlyclassroom protective factors that are associated with better overall academic preparedness alsofunctionduringtheschoolyeartopromoteliteracydevelopment.

13

Table3Descriptivestatistics:Resilientstudentsvs.comparisongroupusingdefinitions1and2

Definition1(Levels) Definition2(Gains) Typical Resilient Typical ResilientIndividualfactors:

StudentSESindex(z-score) -0.26*** 0.55 -0.11*** 0.25

Age(inyears) 12.1*** 11.8 12.1*** 11.8

Girl(%) 51.4** 62.1 48.8*** 58.4

AttendedGradeR(%) 89.1*** 98.9 89.8 91.6

Gritindex(z-score) -0.08*** 0.72 -0.1*** 0.4

Gritindeximputed(%) 2.7 1.1 4.1 3.4

Aspirationsindex(z-score) -0.04*** 0.38 -0.1*** 0.2

Aspirationsindeximputed(%) 7.5 4.6 8.4 8.0

Attitudetoschoolindex(z-score) -0.05*** 0.25 -0.07*** 0.30

Attitudetoschoolimputed 8.6* 4.6 9.4 8.4

Homefactors:

Getshelpathomewithhomework(%) 93.3*** 81.6 93.4 90.3

Livewithmother(Referencecategory:“No”)(%) 77.0 81.6 79.8 79.4

Livewithfather(Referencecategory:“No”)(%) 45.1 43.7 46.7 45.0

Parentemployment:Neitherparenthasajob(%) 16.4*** 6.9 16.7*** 10.5

Parentemployment:Oneparenthasajob(%) 47.7** 36.8 46.5 42.9

Parentemployment:Bothparentshaveajob(%) 30.6*** 55.2 31.9*** 42.4

Parentemployment:Notknown(missing)(%) 5.4*** 1.1 4.87 4.20

Studenthasownstorybooks(%) 42.4* 51.7 43.1 47.9

Englishspokenathome:Never(%) 22.6*** 6.9 20.7*** 10.5

Englishspokenathome:Sometimes(%) 70.6 74.7 71.8*** 80.3

Englishspokenathome:Alwaysoralmostalways(%) 6.8*** 18.4 7.6 9.2

Rural(Reference:Urban)(%) 62.6*** 42.5 57.3*** 37.8

Schoolfactors:

SchoolSESindex-averageSESofclass(z-score) -0.18*** 0.43 -0.11*** 0.17

Classsize 45.9 47.5 46.3 47.2

School'slanguageofinstructionGr1-3:English(%) 13.6*** 44.8 14.6** 21.8

Timeontask:

Teachermissingfromclass:Never(%) 23.7* 33.3 25.7 28.6

Teachermissingfromclass:Sometimes(%) 50.2* 59.8 48.1 53.4

Teachermissingfromclass:Often(%) 18.9*** 2.3 17.7*** 10.9

Teachermissingfromclass:Missing(%) 7.17 4.6 8.5 7.1

Classeswithpresentteacher(%) 83*** 88 83.1*** 86.3

Disruptionsindex 15.1 15.3 15.3** 15.9Availabilityanduseoftext:

%ofteachersindicatingpupilsuselibraryweekly/daily 27.3* 33.7 27.2*** 33.9

%ofschoolsallocatingsufficientbudgettobooks 15.1*** 32.2 16.1*** 25.2IndexofbookpresenceinGr.6classroom(z-score) -0.11*** 0.67 -0.1*** 0.1Teacherquality:

Languageteacherswithlanguagespecialisations(%) 47.9*** 52.6 48.0*** 52.0

Gr.6teachervocabularytestscore(z-score) -0.08*** 0.27 -0.02 -0.05

N(students) 1563 87 1191 238

Source:Leadershipforliteracy.Notes:Significantusing*90%confidenceinterval,**95%confidenceinterval,***99%confidenceinterval.

14

Estimatingresiliencedefinedusingtestscorelevels

Havingobservedsystematicdifferencesinthecharacteristicsofresilientstudentsandtheirtypically-performing peers, the logistic regression results in Table 4 explore which of the factors remainsignificant in a multivariate context when resilience is operationalised according to definition 1.Coefficientsarepresentedasoddsratioswhichreflecttheoddsofbeingresilientforeachone-unitincrease in the predictor variable. For example, the coefficient of 1.64 on being a girl in the firstmodel (Table4) indicatesthatgirlsare1.64timesas likelyasboys (that is,64%more likely) toberesilient.Similarly,thecoefficientof0.25ongettinghelpwithhomeworkinthesecondmodel(Table4)indicatesthatstudentswhogethelpwithhomeworkare75%lesslikelytoberesilientthanthosewhodonotgethomeworkhelp.

Individualandhome-levelprotectivefactors

Attheindividuallevel,genderremainsasignificantpredictorofresilienceevenaftercontrollingforother relevant factors. The influence of having attended grade R plays out very strongly,with anoddsratioindicatingstudentsare7.5to112timesaslikelytobeclassifiedasacademicallyresilientiftheyhadattendedgradeR.AlthoughattendingGradeRisincludedasanindividual-levelprotectivefactor, this is alsoan indicatororhomebackground; reflecting thedecisionsofparents toexposetheirchildrentoearlyschooling.

The multivariate regression results are suggestive of the importance of socio-emotional skills inpromoting academic resilience. The odds of being classified as academically resilient are about 3timeshigherwhenastudentscoresonestandarddeviationabovethemeanonthegrit index.Theinfluence of aspirations is not as strong, but significant,with a studentwho scored one standarddeviationabovethemeanontheaspirations indexbeing1.6timesmore likelytoberesilientthantherestofthesample.Althoughstudentwealthinitiallyappearstobeaprotectivefactorpromotingresilienceinmodels1to 3, this association is no longer statistically significant when accounting for school-levelcharacteristics. Most of the home background factors are insignificant in the multivariateestimations.While this finding contrasts against international studies in which home backgroundfactorsarestrongdeterminantsofresilience(Cheung2016),itisconsistentwiththeexistingschooleffectiveness studies in South Africa (Van der Berg & Louw 2006), which suggest that SES worksthroughtheleveloftheschoolratherthanattheindividuallevel.Thatis,SESimpactsonschoolingoutcomesprimarilythroughdeterminingwhichschoolsstudentsattend(Taylor&Yu2009).FrequencyofEnglishuseathomeremainsassociatedwithacademic resilience in thismultivariatecontext.ComparedtostudentswhoneverspeakEnglishathome,studentswhooftenspeakEnglishathomeareabout4to6timesmorelikelytobeidentifiedasresilient.Thefactthatthelanguageofinstruction of the school does not consistently predict resilience suggests the large coefficient onfrequencyofEnglishuseathomedoesnotsimplyreflectEnglishlanguageexposure.Itmaycaptureotherunobservedhomebackgroundfactors.ConsistentwiththedescriptivedifferencesinTable4,studentswhogethelpwiththeirhomeworkfromanyoneathomearelesslikelytobeidentifiedasresilient. The negative association is puzzling, and could suggest students who struggle aremorelikely to get homework help, while high-ability students may be more able to complete theirhomeworkunassisted.

15

Schoolfactors

Only some of the differences in school-level factors emerge as important for resilience in themultivariateestimation.SchoolSESremainssignificant,withaonestandarddeviationincreaseintheschoolSESindexassociatedwithbeing2to4timesmorelikelytoberesilient.Thisisconsistentwithfindings in the international literature. For example, Agasisti et al. (2018) find a consistentrelationship between school SES and resilience across 50 countries participating in PISA. Teacherabsenteeismasreportedbythestudentalsoremainssignificantinthemultivariateestimations,withresilient students’ teachers less likely to be ‘often’ absent from the classroom. Being in a schoolenvironmentwithmoreliteracymaterialsalsoappearstopromoteresilience,asdoesmanagementprioritisationofbudgetstosupportatext-richenvironment.Resilientlearnersarealsomorelikelytobelocatedinschoolswithhigherpercentagesoflanguageteacherswithlanguagespecialisations.

Estimatingresilienceusingtestscoregains

The first set of estimation results (Table 4) consider resilience from the perspective of achievinghigher-than-expectedtestscorelevels.Resultsarenowshownforestimatesofresiliencedefinedintermsofachievinghigher-than-expectedtestscoregains(i.e.literacyimprovementsduringaschoolyear). Table 5 shows that it is largely individual factors that explain higher-than-expected gains.Classroomandschoolcharacteristicsdonotemergeasparticularly important intheseestimations,although the percentage of teachers in the school with a language specialisation is positive andsignificant. Itmust be kept inmind that 51 out of the 60 schools in the sample had at least oneresilientstudentwhenresiliencewasdefinedintermsoftestscoregains(Table5).Thenineschoolsthat did not have any resilient students according to this definition likely constitute a sample toosmallforthelogitregressiontopickupsignificanteffectsforschool-levelvariables.

Attheindividuallevel,gritemergesagainaspositivelyassociatedwithresilience:theoddsofbeingresilientareone-and-a-halftimeshigherifastudent’sgritindexisonestandarddeviationabovethemean. Attitudes toward school rather than aspirations are now significant, and getting helpwithhomework remains significantly negatively associated with resilience. Student wealth is slightlypositively associated with the probability of resilience. Age now emerges as significant andnegativelyassociatedwithresilience,suggestingthatrepeatingstudentsarelesslikelytoexperienceunusualimprovementsinliteracy.

16

Table4Logisticregressionsofacademicresilience(Levels)

Logisticregression Model1 Model2 Model3 Model4 Model5 Model6 Model7Individualfactors

StudentSESindex 3.20*** 2.69*** 2.15*** 1.17 1.18 1.17 1.14(0.74) (0.60) (0.49) (0.26) (0.25) (0.25) (0.23)

Age 0.82 0.81 0.85 0.78 0.77 0.77 0.74(0.12) (0.13) (0.13) (0.13) (0.14) (0.14) (0.15)

Girl 1.64** 1.54* 1.38 1.61** 1.68** 1.72** 1.70**(0.36) (0.36) (0.31) (0.34) (0.35) (0.37) (0.39)

AttendedGradeR 7.98** 7.51* 7.69** 9.27** 9.89** 10.40** 11.72**(8.25) (7.75) (7.75) (10.13) (10.86) (11.67) (13.45)

Homebackgroundfactors

Helpwithhomework 0.249*** 0.214*** 0.196*** 0.198*** 0.199*** 0.204***

(0.08) (0.07) (0.06) (0.07) (0.07) (0.08)

Livewithmother 1.29 1.27 1.25 1.33 1.30 1.26

(0.37) (0.39) (0.40) (0.45) (0.45) (0.45)

Livewithfather 0.72 0.76 0.77 0.76 0.76 0.74

(0.21) (0.23) (0.23) (0.23) (0.23) (0.23)

Eithermotherorfatherhaveajob 1.55 1.53 1.45 1.31 1.14 1.20

(0.73) (0.74) (0.70) (0.64) (0.56) (0.61)

Bothmother&fatherhaveajob 2.312* 2.260* 2.193* 2.00 1.74 1.78

(1.05) (1.04) (0.99) (0.95) (0.83) (0.87)

Jobnotknown

0.36 0.30 0.17 0.15 0.119* 0.124*

(0.40) (0.32) (0.20) (0.18) (0.14) (0.14)

Learnerhasownstorybook 0.82 0.76 0.72 0.76 0.77 0.79

(0.19) (0.21) (0.20) (0.23) (0.24) (0.25)

Englishspokenathome:sometimes 2.385** 1.884* 1.57 1.64 1.59 1.56

(0.92) (0.70) (0.63) (0.66) (0.64) (0.62)

Englishspokenathome:alwaysoralmostalways

6.368*** 5.070** 4.291** 4.667** 4.616** 4.627**

(3.30) (2.71) (2.29) (2.57) (2.56) (2.49)

Perseverance,aspirations,attitudes

Gritindex(perseverance) 3.262*** 3.283*** 3.185*** 3.350*** 3.492***

(0.95) (0.96) (0.93) (0.95) (1.01)

Aspirationsindex 1.525** 1.676** 1.656** 1.674** 1.670**

(0.24) (0.30) (0.30) (0.32) (0.32)

Attitudetoschoolindex 1.11 1.12 1.14 1.19 1.19

(0.26) (0.25) (0.26) (0.26) (0.26)

Schoolfactors

Rural(Urbanref) 1.19 1.02 0.71 1.02

(0.64) (0.58) (0.42) (0.62)

SchoolSESIndex

3.928** 3.562** 2.388* 1.98

(1.88) (1.81) (1.26) (0.97)

Classsize

1.00 0.99 0.99 1.01

(0.01) (0.02) (0.02) (0.02)

Languageofinstruction-English 1.14 1.07 0.99 0.73

(0.41) (0.39) (0.32) (0.23)

Teachermissingfromclass:sometimes 0.92 0.95 0.94

(0.29) (0.30) (0.28)

Teachermissingfromclass:often 0.133** 0.120** 0.118**

(0.11) (0.09) (0.09)

Teachermissingfromclass:non-response

0.99 0.98 1.05

(0.64) (0.64) (0.68)

%ofclasseswithpresentteacher 0.84 1.56 0.65

(0.90) (1.47) (0.68)

Disruptionsindex 0.96 0.97 1.00

(0.03) (0.03) (0.04)

%educatorswithlanguagespecialisations

1.024** 1.030**

(0.01) (0.01)

StandardizedvaluesofGr.6teachertestscore

1.242* 0.99

(0.16) (0.14)

%ofteachersindicatingpupilsuselibrary

1.01

(0.00)

Strategicallocationofbudgettobooks 2.322**

(0.88)

Indexofbookpresenceingrade6classroom

1.465** (0.27)

Observations 1650 1650 1650 1650 1650 1650 1650Wald!"(p-value) 37(0.000) 93(0.000) 206(0.000) 382(0.000) 500(0.000) 663(0.000) 928(0.000)Source:Leadershipforliteracy.Notes:Coefficientsexpressedinoddsratios.Standarderrorsareinparenthesesandclusteredattheschoollevel.Estimationscontrolforprovincialindicatorsaswellasthreedummyvariablestoindicateifperseverance,attitudinalandaspirationindicesaremissingforthestudentbutthesearenotshown.Significantat*p<0.1;**p<0.05;***p<0.001.

17

Table5Logisticregressionsofacademicresilience(Gains)

Model1 Model2 Model3 Model4 Model5 Model6 Model7Individualfactors

StudentSESindex1.342** 1.230* 1.17 1.17 1.179* 1.17 1.17(0.17) (0.15) (0.14) (0.12) (0.12) (0.12) (0.12)

Age 0.67*** 0.67*** 0.72** 0.71*** 0.71*** 0.72*** 0.72***(0.07) (0.07) (0.07) (0.07) (0.07) (0.07) (0.07)

Girl 1.30 1.25 1.10 1.10 1.10 1.12 1.13(0.27) (0.26) (0.23) (0.23) (0.24) (0.24) (0.23)

AttendedGradeR 1.00 0.96 0.91 0.92 0.94 0.92 0.92(0.34) (0.33) (0.31) (0.31) (0.32) (0.31) (0.31)

Homebackgroundfactors

Helpwithhomework 0.566** 0.589** 0.589** 0.578** 0.582** 0.591**

(0.14) (0.15) (0.15) (0.15) (0.15) (0.15)

Livewithmother 0.86 0.87 0.86 0.86 0.85 0.86

(0.16) (0.16) (0.16) (0.16) (0.15) (0.15)

Livewithfather 0.94 0.97 0.95 0.94 0.96 0.95

(0.13) (0.13) (0.12) (0.13) (0.12) (0.12)

Eithermotherorfatherhaveajob 1.22 1.19 1.18 1.17 1.19 1.21

(0.27) (0.27) (0.27) (0.26) (0.27) (0.26)

Bothmother&fatherhavejob 1.570* 1.483* 1.477* 1.465* 1.47 1.46

(0.37) (0.35) (0.34) (0.34) (0.35) (0.35)

Jobnotknown 1.12 1.24 1.22 1.19 1.24 1.25

(0.45) (0.54) (0.52) (0.51) (0.54) (0.54)

Learnerhasownstorybook 0.95 0.90 0.92 0.93 0.90 0.90

(0.14) (0.13) (0.13) (0.13) (0.13) (0.13)

Englishspokenathome:sometimes 1.711** 1.508* 1.480* 1.42 1.41 1.43

(0.41) (0.36) (0.34) (0.32) (0.33) (0.33)

Englishspokenathome:alwaysoralmostalways

2.378** 2.092** 2.061** 1.944** 1.894** 1.953** (0.77) (0.66) (0.66) (0.62) (0.62) (0.64)

Perseverance,aspirations,andattitudes

Gritindex(perseverance) 1.545*** 1.548*** 1.533*** 1.550*** 1.538***

(0.16) (0.17) (0.17) (0.18) (0.18)

Aspirationsindex

1.16 1.15 1.14 1.16 1.16

(0.13) (0.12) (0.12) (0.12) (0.12)

Attitudetoschoolindex 1.215* 1.211* 1.232** 1.233** 1.236**

(0.13) (0.13) (0.13) (0.13) (0.12)

Schoolfactors

Rural(Urbanref) 0.75 0.80 0.66 0.72

(0.32) (0.38) (0.31) (0.32)

SchoolSESIndex

0.88 0.83 0.78 0.75

(0.25) (0.23) (0.22) (0.22)

Classsize

1.00 1.00 1.00 1.00

(0.01) (0.01) (0.01) (0.01)

Languageofinstruction-English 1.08 1.08 0.91 0.87

(0.39) (0.37) (0.27) (0.25)

Teachermissingfromclass:sometimes 1.09 1.14 1.11

(0.22) (0.21) (0.22)

Teachermissingfromclass:often 0.71 0.72 0.70

(0.21) (0.21) (0.22)

Teachermissingfromclass:non-response 0.85 0.87 0.90

(0.39) (0.41) (0.43)

%ofclasseswithpresentteacher 2.49 2.89 1.95

(2.38) (2.54) (1.78)

Disruptionsindex 1.02 1.02 1.02

(0.05) (0.05) (0.05)

%educatorswithlanguagespecialisations 1.027** 1.027**

(0.01) (0.01)

StandardizedvaluesofGr.6teachertestscore 0.90 0.84

(0.11) (0.10)

%ofteachersindicatingpupilsuselibrary 1.00

(0.00)

Strategicallocationofbudgettobooks 1.46

(0.63)

Indexofbookpresenceingrade6classroom 1.08

(0.13)Observations 1429 1429 1429 1429 1429 1429 1429Wald!"(p-value)

62(0.000)

93(0.000)

232(0.000)

273(0.000)

290(0.000)

375(0.000)

415(0.000)

Source:Leadershipforliteracy.Notes:Coefficientsexpressedinoddsratios.Standarderrorsareinparenthesesandclusteredattheschoollevel.Estimations control for provincial indicators aswell as threedummyvariables to indicate if perseverance, attitudinal and aspiration indices aremissingforthestudentbutthesearenotshown.Significantat*p<0.1;**p<0.05;***p<0.001.

18

Sourcesofconfoundedness:Unobservedinnateability

Onemajor limitationof theanalysispresentedhere is thatwedonot control for innateability. Itmaybethathigh-abilitystudentsalsohavewell-developedsocio-emotionalskillswhereatwo-waycausal relationship exists between the two. For example, if persevering in academic pursuits is amoreproductiveactivityforhigh-abilitystudentsthanforlow-abilitystudents,perseveringinschoolwouldbemorerewardingforhigh-abilitystudents,causingthemtobemoreperseverant.Ifthereissuchacorrelationbetweenthesetwotraits,ourmodelsmaysimplybepickinguptheeffectofhighabilityandnotsocio-emotionalskills.Teachersandparentsmayalsoreinforcehigherattitudesandaspirations for those students who display high-ability resulting in a two-way casual relationshipbetweenacademicachievementandtheseindividualtraits.A related concern is that self-reported questions require not only socio-emotional skills to reflectandanswerthequestionsbutalsoliteracyskills.AsDuckworth&Yeager(2015:240)argue,“Beyondvocabulary, it cannot be assumed that students always understand the pragmatic meaning—theintended idea—of questionnaire items”. In this respect disentangling socio-emotional skills fromacademic ability can be difficult. The positive relationship between grit and test scores may besimplybedetectinghigherliteracyorability.

Although we cannot test for ability with the available data, the results from our second set ofregressions, that is, themodelswhereacademicresilience isdefined in termsofgains rather thanlevels,suggestwearenotsimplypickingupinnateability.Theseresults(Table5)showthatevenfora sample characterised by very low pre-test scores, socio-emotional skills matter for learningimprovements over time. In other words, the results suggest it is possible for a lower-abilitystudent’s learningoutcomesto improvebetweenthepre-andpost-test,andthatsocio-emotionalskillsexplainalargepartofthisimprovementovertime.Studentswithlowscoresarealsolesslikelyto receive affirmation for their performance than higher performing students ruling out that areversecausalrelationshipentirelydrivestheresultsweidentify.

Onewaytotestwhetherhighergritscoresaresimplyreflectiveofhigherliteracyistocomparethedistribution of negatively-worded grit scores for the resilient students and rest of the sample,respectively.Thesenegatively-wordedquestions(largelyreflectingtheconsistency-of-interestscale)aremorechallengingtoanswer.Wearguethatstudentswithmoreliteracyskillswillbemorepronetothebiasesassociatedwithself-reportedquestionnaires,sincetheyunderstandthecontentofthequestionsbetterthantheirlessliteratecounterparts.Thus,wemightexpectmorebiasamongmoreliterate students. In this sense, being more prone to acquiescence (being agreeable to thestatementsregardlessoftheircontent(Diers1964))andsocialdesirabilitybiasmaypotentiallyproxyforhigherliteracyinthesample.

We can use this to test for the possibility that higher grit merely reflects higher ability. If moreliterate students aremore prone to cognitive biases, then the distribution of grit scores (derivedfrom negatively worded questions) for more literate students should lie to the left of typically-performingstudents’distribution,since lowerscoresareattachedtotherightsideofthescalefornegativelywordedquestions.Figure3belowshowsthereisnosystematicdistributionaldifference,suggesting higher scores on the grit scale among resilient students do not simply reflect betterliteracyamongthesestudents.

While we cannot rule out that innate abilitymight be driving some of the effects we see, thesefindings provides some evidence that innate ability is not all we are picking up, and that socio-emotionalskillsplayadistinctiveroleinexplainingacademicresilience.

19

Figure3:Distributionofnegativelywordeditemsingritindex

V. DiscussionApplying a resilience framework to exceptional academic performance among students in SouthAfrican township and rural schools adds to our understanding of the factors that likely protectstudentsfromthemultitudeofriskstotheiracademicsuccess.However,animportantlimitationofthis research is that the relationships we find between the probability of resilience and theprotective factors considered are correlational rather than causal. We cannot infer casualrelationships, or the direction of causality, between a student’s probability of being academicallyresilient and associated protective factors. In this sense, the results presented here should beconsideredexploratoryinnature,withacalltofutureresearchtofocusongaininganunderstandingof the aetiology of resilience. Qualitative research is likely to aid in developing such anunderstanding(Wright&Masten,2015).However,wemakeacasethattheassociationsobservedarenotmerelydrivenbydifferencesinstudents’unobservedability.

Despite these limitations, our results constitute an important first step in understanding thecorrelatesofacademicresilienceinchallengingschoolcontextsinSouthAfrica.Weshowthatsocio-emotional skills such as perseverance and aspirations are strong predictors of being academicallyresilient. These results are supported by an established literature on the importance of “non-cognitive” skills in determining academic success (see Kautz et al. (2014) for a comprehensivesummary of this literature).Moreover, our results suggest socio-emotional skills are important inchallengingschoolcontexts,wherestudentsfaceamultitudeofriskstotheiracademicsuccess.Thisnotionissupportedbyotherstudiesinthebroaderresilienceliterature,wheresocio-emotionalskillsare found to be particularly important for enabling children to successfully navigate challengingcircumstances(Varelaetal.2013).Fromaresearchperspective,ourresultspointtotheimportanceof includingmeasures of socio-emotional skills in future analyses of academic performance, sincetheseskillsmayexplainpartoftheperformancedifferentialsbetweenstudents.

It is important to stress thatour resultsdonot imply resilience is locatedwithin the individual.Alargebodyofresilienceliteratureconvergesontheviewthatresilienceisnotatrait;rather,itresultsfrom “ongoing interactions among complex systems within the person and between person andenvironment” (Masten, 2012: 211). Our results can be explainedwithin this framework: The two“individual-level”variablesthathavethestrongestassociationswiththeprobabilityofbeingresilient

20

(perseverance and aspirations) are shaped by the social contexts (including families, schools, andbroadercommunities)withinwhichchildrenareraised.

Policy implications emerge from these results, for as Kautz et al. (2014: 2) explain, “Skills are nottraitsset instoneatbirthandsolelydeterminedbygenes.Theycanbefostered.”Thisconclusion,basedondecadesofresearchonsocio-emotionalskills, inconjunctionwiththeresultswepresenthere,mayholdpromiseforeducationpolicynot just inSouthAfrica,butforalleducationsystemswhere students face challenging and under-resourced environments. Our results suggest thatpoliciesaimedatfosteringsocio-emotionalskillsamongstudentsinsuchcontextsmayhavealotofleverageinchangingthelearningtrajectoriesofthesestudents.Ourresultsalsoimplythatschoolsandteachershavearoletoplayinpromotingexceptionalachievementthroughmanagingtime-on-task,allocatingschoolbudgetstobooksandhiringsubjectspecialists.

Afurtherimportantresultemergesfromtheanalysispresentedhere,namelytheexistenceofverydifferent levelsof learningwithinthesameclassroom.12Suchdifferential learninglevelswithinthesame classroom have been the subject of growing research interest in recent years, especially indeveloping country contexts, including sub-Saharan Africa (see for example Duflo et al. (2011)).Policyinterventionsmusttakeaccountofthisvariationinpreparedness,andlooktostudiesoftheeffectivenessofprogrammesaimedspecificallyatdealingwiththisheterogeneity.

VI. ConclusionInthispaper,we identifiedresilientstudents inasampleof townshipandruralprimaryschools inSouthAfrica.Threekey findingsemerged.First, althoughacademically resilient studentsaremorelikelytobeidentifiedinbetterperformingschools,theyarealsopresentinverylow-qualityschools.Second, academically resilient studentsdiffer systematically from their non-resilientpeers alonganumber of dimensions. Third, socio-emotional skills such as perseverance and aspirationsmay beparticularlyimportantforfosteringacademicresilienceintheSouthAfricancontext.Understandinghow some studentsmanage to overcome the risks to their academic success that result not onlyfromtheir socio-economicallydisadvantagedhomebackgrounds,butalso thepoorqualityschoolsthey attend,may aid in developing knowledgeof how to supportmore students to do the same.

12 Our results echo those of Van der Berg and Louw (2006), which suggest a similar situation exists in South African

classrooms.

21

VII. ReferencesAgasisti,T.etal.,2018.AcademicResilience:Whatschoolsandcountriesdotohelpdisadvantaged

studentssucceedinPISA.OECDEducationWorkingPapers.

Agasisti, T. & Longobardi, S., 2017. Equality of Educational Opportunities, Schools’ Characteristicsand Resilient Students: An Empirical Study of EU-15 Countries Using OECD-PISA 2009 Data.SocialIndicatorsResearch,134(3),pp.917–953.

Akin,A.,Arslan&Serhat,2014.TheRelationshipsbetweenAchievementGoalOrientationsandGrit.EducationandScience,39(175),pp.267–274.

Benard,B.,1997.FosteringResiliencyinKids.EducationalLeadership.

Borman,G.D.&Overman,L.T.,2004.AcademicResilienceinMathematicsamongPoorandMinorityStudents.TheElementarySchoolJournal,104(3),pp.177–195.

Bronfenbrenner,U.,1994.Ecologicalmodelsofhumandevelopment. In InternationalEncyclopediaofEducation.Oxford:Elsevier,pp.37–43.

Brookover,W.B.etal.,1978.ElementarySchoolSocialClimateandSchoolAchievement.AmericanEducationalResearchJournal,15(2),pp.301–318.

Cappella, E. & Weinstein, R.S., 2001. Turning Around Reading Achievement: Predictors of HighSchoolStudents’AcademicResilience.JournalofEducationalPsychology,93(4),pp.758–771.

Cheung, K., 2016. The effects of resilience in learning variables on mathematical literacyperformance:astudyoflearningcharacteristicsoftheacademicresilientandadvantagedlowachieversinShanghai,Singapore,HongKong,TaiwanandKorea.EducationalPsychology,37(8),pp.965–982.

Colosi,R.,2000.NegativelyWordedQuestionsCauseRespondentConfusion.ASASectiononSurveyResearchMethods,20746,pp.2896–2903.

Das, D., 2018. Academic Resilience Among Children from Disadvantaged Social Groups in India.SocialIndicatorsResearch.

Dass-Brailsford, P., 2005. Exploring resiliency: Academic achievement among disadvantaged blackyouthinSouthAfrica.SouthAfricanJournalofPsychology,35(3),pp.574–591.

Datu, J.A.D., Valdez, J.P.M. & King, R.B., 2016. Perseverance Counts but Consistency Does Not!ValidatingtheShortGritScaleinaCollectivistSetting.CurrentPsychology,35(1),pp.121–130.

Diers, C.J., 1964. Social desirability and aquiescence in response to personality items. Journal ofConsultingPsychology,28(1),pp.71–77.

Duckworth, A.L. et al., 2007. Grit: Perseverance and Passion for Long-Term Goals. Journal ofPersonalityandSocialPsychology,92(6),pp.1087–1101.

Duckworth, A.L.&Quinn, P.D., 2009. Development and validation of the short Grit Scale (Grit-S).JournalofPersonalityAssessment,91(2),pp.166–174.

Duckworth, A.L. & Yeager, D.S., 2015. Measurement Matters: Assessing Personal Qualities OtherThan Cognitive Ability for Educational Purposes. Educational Researcher, 44(4), pp.237–251.Availableat:http://edr.sagepub.com/cgi/doi/10.3102/0013189X15584327.

Duflo, B.E., Dupas, P. & Kremer, M., 2011. Peer Effects, Teacher Incentives, and the Impact ofTracking: Evidence from a Randomized Evaluation in Kenya. American Economic Review,

22

101(August),pp.1739–1774.

Erberber, E. et al., 2015. Socioeconomically Disadvantaged Students Who Are AcademicallySuccessful: Examining Academic Resilience Cross-Nationally. International Association for theEvaluationofEducationalAchievement,(March),pp.1–12.

Finn,J.D.&Rock,D.A.,1997.AcademicSuccessAmongStudentsatRiskforSchoolFailure.JournalofAppliedPsychology,82(2),pp.221–234.

Gizir,C.A.&Aydin,G.,2009.ProtectiveFactorsContributingtotheAcademicResilienceofStudentsLivinginPovertyinTurkey.ProfessionalSchoolCounselling,13(1),pp.38–49.

Kautz,T.etal.,2014.FosteringandMeasuringSkills:ImprovingCognitiveandNon-CognitiveSkillstoPromoteLifetimeSuccess.IZADiscussionPapersNo8696.

Lee,S.&Sohn,Y.W.,2017.Career-RelatedAttitudesofCollegeStudentsinKorea.SocialBehaviourandPersonality,45(10),pp.1629–1642.

Mampane, M.R., 2014. Factors contributing to the resilience of middle-adolescents in a SouthAfricantownship:Insightsfromaresiliencequestionnaire.SouthAfricanJournalofEducation,34(4),pp.1–11.

Mampane, R. & Bouwer, C., 2006. Identifying resilient and non-resilient middle-adolescents in aformerlyblack-onlyurbanschool.SouthAfricanJournalofEducation,26(3),pp.443–456.

Mampane, R. & Bouwer, C., 2011. The influence of township schools on the resilience of theirlearners.SouthAfricanJournalofEducation,31:pp.114-126.

Mason, H.D., 2018. Grit and academic performance among first-year university students: A briefreport.JournalofPsychologyinAfrica,28(1),pp.66–68.

Masten,A.S.,2012.ResilienceinChildren:VintageRutterandBeyond.InA.M.Slater&P.C.Quinn,eds.DevelopmentalPsychology:Revisiting theClassicStudies. London:SAGEPublicationsLtd,pp.204–221.

Masten, A.S. &Obradovic, J., 2008. Disaster preparation and recovery: Lessons from research onresilienceinhumandevelopment.EcologyandSociety,13(1).

Moller,V.,1995.Homeenvironmentandeducationalachievementamonghighschoolpupilslivinginthree-generationurbanblackhouseholds.SouthAfricanJournalofSociology,26(3).

OECD,2011a.Against theOdds:DisadvantagedStudentsWhoSucceed in School,OECDEducationWorkingPapers.

OECD,2011b.PISAasa Studyof StudentResilience. InAgainst theOdds:DisadvantagedStudentsWhoSucceedinSchool.Paris:OECDPublishing,pp.13–21.

Oriol, X. et al., 2017. The role of self-control and grit in domains of school success in students ofprimaryandsecondaryschool.FrontiersinPsychology,8(OCT),pp.1–9.

Reraki,M.,Celik, I.&Saricam,H.,2015.Gritasamediatoroftherelationshipbetweenmotivationandacademicachievement.OzeanJournalofSocialScience,8(1),pp.19–32.

Rojas, J.P.&Usher,E.L.,2012.ExploringtheCorrelationsAmongCreativity,Grit,andMathematicsAchievement inSocioeconomicallyDiverseSchools.PosterpresentedattheGraduateStudentConferenceforChildrenatRiskinLexington,KY.

Salazar,M.S., 2015. The dilemmaof combining positive and negative items in scales.Psicothema,27(2),pp.192–199.

23

Sandoval-Hernandez, A. & Bialowoski, P., 2016. Factors and conditions promoting academicresilience: a TIMSS-based analysis of five Asian education systems. Asia Pacific EducationReview,17,pp.511–520.

Sandoval-Hernandez,A.&Cortes,D.,2012.Factorsandconditionsthatpromoteacademicresilience:Across-countryperspective,Malmo.

Shepherd,D.,2016.Theopendooroflearning-accessrestricted:SchooleffectivenessandefficiencyacrosstheSouthAfricaneducationsystem.StellenboschEconomicsWorkingPapers.

Shumow, L., Vandell, D.L. & Posner, J., 1999. Risk and Resilience in the Urban Neighborhood:PredictorsofAcademicPerformanceAmong Low-IncomeElementary SchoolChildren.Merill-PalmerQuarterly,45(2),pp.309–331.

Spaull,N.,2013.Poverty&Privilege:PrimarySchoolInequalityinSouthAfrica.InternationalJournalofEducationalDevelopment,33,pp.436–447.

Taylor, S. & Yu, D., 2009. The importance of socio-economic status in determining educationalachievementinSouthAfrica.StellenboschEconomicsWorkingPapers.

VanBroekhuizen,H.&Spaull,N.,2017.The‘MarthaEffect’:ThecompoundingfemaleadvantageinSouthAfricanhighereducation.StellenboschEconomicsWorkingPapers.

VanDerBerg,S.,2008.Howeffectivearepoorschools?PovertyandeducationaloutcomesinSouthAfrica.StudiesinEducationalEvaluation,34(3),pp.145–154.

Van der Berg, S. & Louw, M., 2006. Lessons learnt from SACMEQII: South African studentperformanceinregionalcontext.InJohannesburg:InvestmentChoicesforEducationinAfrica.StellenboschEconomicsWorkingPapers.

Varela,A.D.etal.,2013.Learningandresilience:Thecrucialroleofsocialandemotionalwell-beingincontextsofadversity,Washington,D.C.

Vera, G.G., Valenzuela, J.P. & Sotomayor, C., 2015. Against All Odds: Outstanding ReadingPerformance among Chilean Youth in Vulnerable Conditions.Comparative Education Review,59(4),pp.693–716.

Waxman,H.C.,Huang,S.L.&Wang,M.C.,1997.Investigatingtheclassroomlearningenvironmentofresilientandnon-resilientstudentsfrominner-cityelementaryschools.InternationalJournalofEducationalResearch,27(4),pp.343–353.

Werner, E.E., 1997. Vulnerable but invincible: high-risk children from birth to adulthood. ActaPaediatr,422,pp.103–105.

Wills,G.,2017.Whatdoyoumeanby‘good’?Thesearchforexceptionalprimaryschools inSouthAfrica’sno-feeschoolsystem.StellenboschEconomicsWorkingPapers.

Wright,M.O.&Masten,A.S.,2015.PathwaystoResilienceinContext.InL.C.Theron,L.Liebenberg,&M. Ungar, eds. Youth Resilience and Culture: Commonalities and Complexities. New York:Springer,pp.3–22.

24

Appendix

TableA1:Descriptionofvariablesandtheirformation:Individualfactors

Individualfactors:

StudentSESindex(z-score)Indexof13home-ownershipassetscombinedusingprincipalcomponentsanalysis.AveragetakenacrossFebruaryandOctober.Home-ownershipquestionswerepictorialinnature,reducingtheneedforstrongliteracyskillstoanswerthesequestions.

Age AgeinyearsGirl Indicatorvariableforwhetherstudentisfemale.AttendedGradeR IndicatorvariableforwhetherchildattendedgradeR.

Studenthasownstorybooks Indicatorvariableforwhetherthechildidentifieshavinganyofhis/herownstorybooksathome.

Gritindex(z-score)

Principalcomponentsindexderivedfromtheperseverancesubscaleofthegritindex.Studentswereaskedtorespondtothefollowingstatementsusingaresponsescale:[That'snotatalllikeme","That'snotreallylikeme","That'ssometimeslikeme","That'salotlikeme"]1)"Problemsandchallengesdon’tdiscourageme.WhenImakeamistakeIgetbackupandtryagain."2)"Iworkhardtodothingswell."3)"IfinishwhateverIstart."4)"Icansitstillforlongerthanotherchildrenintheclass."5)Idomyschoolworkcarefully."Cronbach'salphavalueislowat0.51,withanaverageinter-itemcovarianceof0.18.

GritindeximputedIndicatorforwhetherthegritindexwasimputedforthestudent.Usingdefinition1ofresilience,86(5%)non-resilientstudents'scoresimputed,and98(2%)resilientstudents’scoreswereimputed.

Aspirationsindex(z-score)

Principalcomponentsindexderivedfrom3questionsaskedofstudents:i)Howimportantisittogetgoodmarksinschool?[Notimportant,alittleimportant,veryimportant.]Ii)Doyouthinkyou'llpassmatric?[No,Maybe,Yes]iii)Doyouthinkyou'llgotouniversityafterschool?[No,Maybe,Yes]Cronbach'salphavalueis0.59,withanaverageinter-itemcovarianceof0.27

Aspirationsindeximputed(%)Indicatorforwhethertheaspirationsindexwasimputedforthestudent.Usingdefinition1,83(5%)non-resilientstudents'scoresimputed,and4(3%)resilientstudents’scoreswereimputed.

Attitudetoschoolindex(z-score)

Aprincipalcomponentsindexderivedfrom4questionsaskedofstudents:i)Howmuchdoyoulearninschooleveryday?[Notverymuch,Alittle,Alot!]ii)Howexcitedareyoutogotoschooleachday?[Notexcitedatall,Alittleexcited,Veryexcited!]iii)Howmuchdoyouthinkyourteachercaresaboutyou?[Notatall,Alittle,Alot!]iv)Doyoufeellikeyourteacherbelievesinyou?[No,Yes,Idon'tknow].Cronbach'salphavalueis0.46,withanaverageinter-itemcovarianceof0.22

AttitudetoschoolimputedIndicatorforwhethertheattitudesindexwasimputedforthestudent.Usingdefinition1,83(5%)non-resilientstudents'scoresimputed,and4(3%)resilientstudentscoreswereimputed.Cronbach'salphavalueis0.59,withanaverageinteritemcovarianceof0.27

25

TableA2Descriptionofvariablesandtheirformation(Homeandschoolfactors)

Homefactors: Getshelpathomewithhomework

Indicatorvariableforwhetheranyoneinthechild'shouseholdhelpsthemwiththeirhomework.

Liveswithmother Indicatorvariableforwhetherthechildliveswiththeirmother.Liveswithfather Indicatorvariableforwhetherthechildliveswiththeirfather.Parentemployment:

Indicatorvariablesfortheemploymentstatusofthestudent'sparents.

Neitherparenthasajob(Referencecategory)OneparenthasajobBothparentshaveajobNotknown(missing)Englishspokenathome:

IndicatorsforhowoftenthechildreportsspeakingEnglishathome.Never(Ref.category)SometimesAlways/almostalwaysRural(Referencecategory:Urban)

Indicatorvariableforruralvs.urbanstatusoftheschool'slocationobtainedfromEMIS/Censusdata.

Schoolfactors: SchoolSESindex(z-score) ClassaverageoftheStudentSESindexClasssize Numberofstudentswhowrotethepre-testineachclass.School'slanguageofinstructionGr1-3:English(%)

Theschools'languageofinstructioninGrades1-3beforeacompulsoryswitchtoEnglishinGrade4.

Teachermissingfromclass:Never(Ref.category)

Theseindicatorvariablesvaryatthestudentlevelandwerederivedfromaquestionnaireaskingeachstudenttoidentify"Howoftenisyourteachingmissingfromclass?".

SometimesOftenMissing

Proportionofutilisedclassroomsintheschoolwithapresentteacher(teachingorclassengaged).

Fieldworkerswalkedaroundeachschoolidentifyingthenumberofclassroomswithi)apresentteacherwhowasteaching,ii)apresentteacherwhowasnotteachingbutstudentsinhis/herclasswereengagedinalearningactivity,iii)apresentteacherbutstudentswerenotengagedinanylearningactivity,iv)absentteachersbutstudentsintheclassandv)classroomsthatwerenotutilisedatall.

Disruptionsindex

Thedeputyprincipal,principalandgrade6teacherwereaskedtoidentifyonascaleof1-5(“Notatall”=1to“Ahugeamount”==5),howmuchtheschoolhasbeenaffectedinthepastyearbyanyofthefollowing:a)Damagestoschoolinfrastructure;b)Negativechangestoschoolmanagement;c)Watersupplydisruptions;d)Communityprotests;e)Strikeactionbyteachers;f)Disruptionstoworkduetorules/go-slow/downtoolsbyteachers;g)Conflictamongstaff;h)Violenceamongststudents;i)Break-ins,robberiesorothercriminalactivity.Theaverageresponseacrossthethreeteacherstoeachofthesub-questionsissummedtoobtainthedisruptionsindex.

Proportionoflanguageteacherswithlanguagespecialisations

Proportionofteachersinaschoolidentifyingthattheyhave"completedanACEorshortcourseinteachinglanguageorreading"orthat"EnglishorAfricanlanguagewasoneofmysubjectmajors"

Gr.6teachervocabularytestscore(z-score)

Standardizedscoresofteacher'sresponsestoavocabularytest.Thetestconsistedof72questionsrangingfrommosttolessfrequentlyusedwordsinEnglish.Thiswasthesamevocabularytestprovidedtostudentsalongwiththereadingcomprehensiontest.Thelowestscorewas16.5andhighestscore66.5(maximumpossiblescore72).

Proportionofteachersindicatingpupilsuselibraryweekly/daily

Proportionofteachersinaschoolresponding"weekly"or"daily"tothequestion"Howoftendostudentsinyourclassvisittheschoollibrary?"

Percentageofschoolsallocatingsufficientbudgettobooks(%)

Anindicatorforwhether33%ormoreof2016or2017financialyearprovincialallocationtotheschoolwasspentonbooks.

Indexofbookpresenceingrade6classroom(z-score)

Aprincipalcomponentsindexderivedfrom6questionsaskedofthegrade6Englishlanguageteacheronaccesstobooks/resourcesaswellasanobservationalquestiononthenumberofdifferenttypesofbooksobservedinthegrade6classroombythefieldwork[i)Englishtextbooks;ii)DBEworkbooks;iii)Gradedreaderforgrade6;iv)Dictionaries;v)Novels;vi)Booksofshortstories;vii)Poetrybooks;viii)Dramas/plays;ix)Newspapersormagazines].

26

TableA3:Sensitivityanalysis–excludingEnglishLOLTsamplesandusingdifferentcut-offpointsindefiningresilience.Logisticregressionofacademicresilience(OddsRatios)

Definition1

ExcludeEnglishschools

Definition1a

Definition1b

Model6 Model7 Model6 Model7 Model6 Model7 Model6 Model7Individualfactors:

StudentSESindex 1.166 1.142 1.016 0.978 1.041 1.024 0.837 0.822(0.73) (0.66) (0.06) (0.09) (0.19) (0.12) (1.21) (1.39)

Age 0.769 0.737 0.676** 0.662** 0.705* 0.682* 0.713** 0.700**(1.43) (1.47) (2.05) (2.07) (1.93) (1.84) (2.52) (2.66)

Girl1.715** 1.702** 2.389** 2.365** 1.707** 1.693** 1.802** 1.812**(2.48) (2.31) (2.25) (2.17) (2.47) (2.25) (3.05) (3.03)

AttendedGradeR10.40** 11.72** 3.162 3.355 10.75** 11.83** 4.116** 4.370**(2.09) (2.14) (1.07) (1.11) (2.13) (2.17) (2.26) (2.32)

Perseverance,aspirations,andattitudes:

Gritindex(perseverance) 3.350*** 3.492*** 4.956*** 5.165*** 3.774*** 3.821*** 3.270*** 3.361***(4.26) (4.33) (3.78) (3.73) (4.61) (4.55) (5.80) (5.98)

Aspirationsindex1.674** 1.670** 2.004** 1.954** 1.702** 1.679** 1.598*** 1.605***(2.74) (2.67) (2.42) (2.35) (2.77) (2.56) (3.70) (3.71)

Attitudetoschool&teacherindex1.185 1.189 1.800 1.798 1.226 1.231 1.400* 1.410*(0.78) (0.80) (1.34) (1.33) (0.94) (0.99) (1.85) (1.95)

Homefactors:

Helpwithhomework0.199*** 0.204*** 0.221** 0.203** 0.206*** 0.217*** 0.261*** 0.263***(4.47) (4.28) (2.88) (2.94) (4.78) (4.47) (3.97) (3.87)

Livewithmother1.296 1.257 1.284 1.182 1.177 1.149 0.903 0.866(0.75) (0.64) (0.62) (0.42) (0.50) (0.40) (0.42) (0.58)

Livewithfather0.760 0.743 0.471 0.486 0.842 0.790 0.817 0.811(0.89) (0.95) (1.58) (1.50) (0.53) (0.73) (0.87) (0.89)

Eithermotherorfatherhaveajob1.137 1.201 0.835 0.844 1.139 1.223 0.965 0.961(0.26) (0.36) (0.35) (0.32) (0.27) (0.42) (0.09) (0.10)

Bothmother&fatherhavejob1.743 1.775 1.842 1.759 1.745 1.776 1.717 1.678(1.17) (1.17) (1.11) (1.00) (1.19) (1.20) (1.48) (1.41)

Jobnotknown 0.119* 0.124* 0.371 0.317 0.114* 0.119* 0.136** 0.134**(1.88) (1.85) (1.02) (1.19) (1.95) (1.90) (2.60) (2.57)

Learnerhasownstorybook 0.770 0.785 0.578 0.634 0.772 0.780 0.930 0.958(0.86) (0.77) (1.45) (1.19) (0.84) (0.79) (0.35) (0.21)

Englishspokenathome:Sometimes 1.585 1.558 1.427 1.333 1.364 1.337 1.266 1.220(1.14) (1.11) (0.79) (0.63) (0.75) (0.71) (0.73) (0.62)

Englishspokenathome:Always/almostalways4.616** 4.627** 2.621* 2.877* 4.215** 4.356** 3.035** 2.983**(2.76) (2.85) (1.74) (1.90) (2.54) (2.70) (2.58) (2.62)

Schoolfactors:

Rural(Urbanref)0.713 1.020 1.969 2.768 0.843 1.092 0.486 0.597(0.58) (0.03) (0.86) (1.22) (0.28) (0.15) (1.44) (1.07)

SchoolSESIndex2.388* 1.983 2.793 2.977** 2.854* 2.152 2.825** 2.649**(1.65) (1.40) (1.57) (2.20) (1.91) (1.55) (2.49) (2.74)

Classsize0.993 1.006 1.005 1.014 0.992 1.008 0.993 0.999(0.42) (0.29) (0.27) (0.67) (0.48) (0.39) (0.56) (0.07)

Languageofinstruction-English0.992 0.729 N.A. N.A. 0.946 0.728 0.767 0.623(0.03) (0.99) (0.16) (1.05) (0.84) (1.49)

Teachermissingfromclass:Sometimes 0.948 0.944 0.802 0.785 0.969 0.968 0.761 0.750(0.17) (0.19) (0.55) (0.60) (0.10) (0.11) (1.23) (1.37)

Teachermissingfromclass:Often0.120** 0.118** 0.0587** 0.0611** 0.107** 0.106** 0.0802*** 0.0797***(2.72) (2.81) (2.62) (2.55) (2.78) (2.85) (3.87) (3.93)

Teachermissingfromclass:Non-response0.980 1.053 0.383 0.396 0.725 0.790 0.528 0.553(0.03) (0.08) (1.13) (1.15) (0.54) (0.41) (0.82) (0.78)

%ofclasseswithpresentteacher1.559 0.648 0.384 0.203 1.802 0.617 1.733 0.820(0.47) (0.41) (0.76) (1.19) (0.61) (0.45) (0.67) (0.22)

Disruptionsindex0.965 1.002 1.034 1.080 0.967 0.997 0.990 1.017(1.00) (0.05) (0.75) (1.39) (0.91) (0.09) (0.27) (0.45)

%educatorswithlanguagespecialisations1.024** 1.030** 1.034** 1.034** 1.025** 1.031** 1.023** 1.026**(2.06) (2.42) (2.66) (2.53) (2.19) (2.57) (2.72) (2.89)

StandardizedvaluesofGr.6teachertestscore

1.242* 0.993 0.670* 0.652** 1.205 0.958 1.155 0.993(1.67) (0.05) (1.95) (2.09) (1.41) (0.33) (1.16) (0.05)

%ofteachersindicatingpupilsuselibrary 1.007

1.012*

1.009*

1.008*

(1.55)

(1.89)

(1.79)

(1.82)

Strategicallocationofbudgettobooks 2.322**

1.314

2.605**

1.630

(2.23)

(0.58)

(2.50)

(1.43)

Indexofbookpresenceingrade6classroom 1.465**

1.395

1.495**

1.329**

(2.10)

(1.53)

(2.26)

(1.97)

Observations 1650 1650 1398 1398 2028 2028 1721 1721Wald!"(p-value)

663(0.000)

928(0.000)

694(0.000)

932(0.000)

698(0.000)

981(0.000)

998(0.000)

1007(0.000)

27