Peer effects in the individual and group

Findings indicate the presence of a strong literacy- based peer effect in mathematics, reading, and science even after controlling for individual demographic differences.. For this study

PEER EFFECTS IN THE INDIVIDUAL AND GROUP LITERACY ACHIEVE- MENT OF HIGH-SCHOOL STUDENTS

IN A BI-DIALECTAL CONTEXTS.Joel Warrican, Melissa L Alleyne, Patriann Smith, Jehanzeb

Cheema, and James R King

QUERY SHEETThis page lists questions we have about your paper The numbers displayed

at left can be found in the text of the paper for reference In addition, pleasereview your paper as a whole for correctness

Q1 The reference “Ferreira (2013); Government of the Republic of

Trinidad and Tobago (2010); United Nations Population Division(2013); Kachru (1986); Nero (2006); Pratt-Johnson (2006);Williams & Carter (2005); Burke and Sass (2013); Moll andGreenberg (1990); Street (1995)” is cited in the text but is notlisted in the references list Please either delete the in-text cit-ation or provide full reference details following journal styleQ2 Please provide complete details for Author (2011, 2013, 2015,

2017, 2018)

Q3 The ORCID details of the authors have been validated against

ORCID registry please check the ORCID ID details ofthe authors

Q4 please advise the running head of this article

TABLE OF CONTENTS LISTINGThe table of contents for the journal will list your paper exactly as it

PEER EFFECTS IN THE INDIVIDUAL AND GROUP LITERACYACHIEVEMENT OF HIGH-SCHOOL STUDENTS IN A BI-DIALEC-TAL CONTEXT

S.Joel Warrican, Melissa L Alleyne, Patriann Smith, Jehanzeb Cheema,and James R King

PEER EFFECTS IN THE INDIVIDUAL AND GROUPLITERACY ACHIEVEMENT OF HIGH-SCHOOL STUDENTS

IN A BI-DIALECTAL CONTEXT

S.JOEL WARRICAN Professor and Director of Academic Programming and Delivery Division, Office of the Principal, University of the West Indies, Barbados

MELISSA L ALLEYNE Planning and Institutional Research Officer, University of the West Indies,

Cave Hill Campus, Barbados

Assistant Professor Department of Curriculum and Instruction Language, Diversity and Literacy Studies Texas Tech University, Lubbock, TX, USA

JEHANZEB CHEEMA Term Assistant Professor Department of Information Systems and Operations Management, School of Business, George Mason University, Fairfax, VA, USA

JAMES R KING University of South Florida Childhood Education and Literacy Studies,

Tampa, FL, USA

The theory of peer effect posits that students who find themselves in the pany of high or low performing peers tend to exhibit better academic perform- ance than what is solely attributable to their own individual characteristics.

com-In this study, we investigate peer effects within literacy achievement among Trinidadian and Tobagonian youth, using a nationally representative sam- ple of 15-year-old students Findings indicate the presence of a strong literacy- based peer effect in mathematics, reading, and science even after controlling for individual demographic differences Peer effect alone explained between 30% and 67% of the total variation in literacy Implications are discussed.

Curriculum and Instruction Language, Diversity and Literacy Studies Texas Tech

patriann.smith@ttu.edu

1

ISSN: 0270-2711 print / 1521-0685 online

DOI: 10.1080/02702711.2019.1571545

IntroductionOver the years, people with a vested interest in education havebeen seeking means of optimizing the returns on moniesinvested in education One of the areas that have been investi-gated along these lines is the concept of peer effects (Hoxby,

2000) The theory of peer effects suggests that ‘good’ peershave a positive spillover on others while ‘bad’ peers have thereverse effect In the classroom context, high-achieving peersare expected to influence their fellow students to the pointwhere there are learning gains, while exposure to low-achievingpeers is likely to produce lower performances among students(Gorman, 2018) While there is some ongoing debate on theinfluence of peer effects on academic achievement, there isapparent agreement, regardless of philosophy, that school andclass composition matter It therefore seems worthwhile toinvestigate issues related to school and class composition such

as peer effects In highlighting the importance of peer effectsstudies, Hoxby (2000) states that the expectation is that theyprovide policy makers with information that can assist them inmaking decisions regarding an efficient distribution of students

in schools and classrooms which will in turn inform the humancapital investments with a view towards macroeconomic growth.Hoxby (2000) cautions, however, about thinking that thetheory of peer effects is simply about academic spillovers frompeers teaching each other This is indeed part of it, but itincludes, for example, the influence of innate ability on peers

to the point where students in a class simply want to operate at

a standard (be it high or low) similar to that of their colleagues.Peer effect is also linked to environmentally determined behav-ior where, for example, the level of discipline of some studentswill have an effect on their peers to the point where it can pro-duce either a disruptive or productive classroom Hoxby alsoposits that peer effect may follow lines of disability, race, genderand family income For example, a disabled child can poten-tially demand more of the teacher’s time in the classroom,which in turn can affect overall learning Additionally, racial orgender tension in a classroom can have an impact on learning,

or parents with greater financial resources can, for example,

purchase learning materials that eventually spread over a room, resulting in greater academic benefits.

class-Recognizing that there has been a plethora of studies inpeer effects in the broad field of education, we decided to studythe phenomenon in a specific education context where there is

an apparent absence of such research For this study, we gate how peer effects impact students’ literacy performance onthe Program for International Student Assessment (PISA) inthe Caribbean setting of Trinidad and Tobago We recognizethat peer effect could present itself in varying forms, but here,

investi-we concentrate on the ultimate effect of literacy performance

in a context where the linguistic diversity is not given recognition

in the collection of PISA data The twin-island of Trinidad andTobago, once a colony of Britain, has as its official languageTrinidad and Tobago’s Standard English (TSE), but also twounofficial varieties that are widely used by most of its citizens:Trinidadian English-lexicon Creole (TEC) and TobagonianEnglish-lexicon Creole (TOB) (Ferreira, 2013; Government ofthe Republic of Trinidad and Tobago, 2010)

Reports indicate that the youth of Trinidad and Tobago(age 10–24) constituted 22% (0.3 million) of the nation’s popu-lation as of 2013 (United Nations Population Division, 2013).According to Craig (2006) and Roberts (2007), the majority ofthese youth predominantly speak TEC or TOB and thereforeuse language daily in ways that differ significantly, both syntac-tically and semantically, from Standard English Regardless,Trinidadian and Tobagonian youth perceive of themselves asStandard English speakers As a result, they tend to identify asfirst language (L1) speakers of [Standard] English, that is, thestandard and accepted form of English in formal settings(Ferreira, 2013) Scholars have observed that the bi-dialectalyouth of Trinidad and Tobago become aware of their categor-ization as non-L1 English speakers only in situations where theyare faced with negative reactions of others to their use ofEnglish, which often happens when they migrate to a new coun-try (Canagarajah, 1999; Kachru, 1986; Nero, 2006) In theabsence of this recognition, the youth remain relativelyunaware that TEC and TOB are considered lower on the creolecontinuum, despite their acknowledgement of the prestige

received by Trinidadian Standard English (Pratt-Johnson,2006) In doing so, the bi-dialectal adolescents of Trinidad andTobago reflect a “rather subtle internalization of [English]

‘monolingualism’” (Williams & Carter, 2005, p 238)

We posit that peer effect brings with it further complicationsbeyond that of or concomitant with a bi-dialectal language situ-ation that is not formally addressed by the PISA test administra-tion We recognize as a limitation that we cannot investigate theeffects of the bi-dialectal factor with peer effects since the stu-dents were not given the option of identifying TEC and TOB astheir home languages Even with this recognition that we cannotaccount for the bi-dialectal factor, we think it important enough

to highlight in this study as it signals the need for inclusion infuture research-related activities such as PISA cycles

Peer Effects in Educational ContextsExamining peer effects in educational contexts poses seriouschallenges to researchers who work within this area of inquiry.Staging experimental studies within this context, especially atthe classroom level, is usually not feasible and is in fact quitescarce Researchers generally use large, extant data sets in posthoc analyses The majority of these studies rely on statisticallygenerated proxy variables, and as Hoxby (2000) points out,researchers have broad empirical obstacles to work through intheir efforts to determine the presence of peer effects Forexample, they have to ensure that what is presented as evidence

of peer effects is not biased by group selection; that is, sayingthe effect is caused by the group when it could instead be a nat-ural characteristic of the group

Perhaps one of the most cited studies in area of peereffects is Hoxby (2000) In studying the phenomenon of peereffects in Texas, Hoxby sought to eliminate the empiricalobstacles that she cites as plaguing studies of this nature Toaddress this issue, the researcher implemented two empiricalstrategies that she reported as being reasonably free of selectionbias based on the idea that, although parents may deliberatelychoose a school for their children based on certain perceivedcharacteristics, and schools may assign students based on ability,

there are still idiosyncratic variations within grade cohorts thatare beyond the control of the parents and schools Throughthese empirical strategies, Hoxby (2000) sought to: (a) identifyidiosyncratic variation by comparing adjacent cohorts along thelines of gender and race and (b) identify the idiosyncratic com-ponent of each group’s achievement and determine whetherthe components are correlated (p 36).

Hoxby’s results revealed the existence of significant peereffects For example, she found that having a group where themajority of the students are females results in both males andfemales reflecting higher scores in reading and mathematics Interms of race, she found that in Texas, Black and Hispanic stu-dents tend to enter school with low initial achievement As ageneral finding she discovered that when students are exposed

to unusually low performing peers their performance is affectednegatively This she found to be particularly evident amongBlack students when they are placed in groups of weak peers intheir own racial group On the other hand, she found thatHispanic students fare better when placed with a majorityHispanic cohort, even if many in the group are weak Herexplanation as it relates to the Hispanic phenomenon is that itcould be that for the students learning English they are morelikely to get support from their peers Although she found littleevidence of a general asymmetry such as low achievers havinggreater gains by being grouped with high achievers and thathigh achievers lose when grouped with low achievers, Hoxbynoted that a student’s test score is estimated to rise by0.10–0.55 points when he or she is grouped with peers whoscore one point higher Hoxby’s (2000) finding about Hispanicpeers in classrooms, with her explanation regarding the peereffects, also raises questions more broadly relating to peereffects in language learning classrooms

There have been some studies that sought to establish peerinfluence on, for example, English language learners (ELL).For example, Cho (2012) reports on a study that utilized anationally representative data set of students in early elemen-tary grades to study the effects of ELL students on their non-ELL peers in the U.S She found that the presence of an ELLclassmate during kindergarten and first grade increased the

likelihood of lower test scores in reading, but not necessarily inmath for non-ELL students whose L1 was English These nega-tive effects for reading test score gains were largest for low-income and for female students Similar findings were observed

by Lavy, Passerman, and Schlosser (2008), who conducted astudy in Israel to examine the effects of low-ability (enrolled atleast 1 year behind their age) students on their classmates Thestudy found that low-ability students affected the academic per-formance of regular students negatively and that the negativeimpact was more concentrated among students of low socioeco-nomic status Lavy et al (2008) highlighted that disruption ofclasses was one of the effects of the presence of low-ability stu-dents in their study They also noted that a major drawback oflow-ability students in the study is that they diverted teacherattention from regular students to struggling ones

Another study that can have implications for peer effects

in a possible language learning context was conducted byEntorf and Lauk (2008) In their study, the researchers specific-ally investigated the impact of social integration of migrants onpeer effects for both natives and migrants using data from 11countries that participated in the 2000 OECD PISA test Theyexamined the impact on peer effects within and betweengroups of migrants and found the presence of peer effects.However, they found peer effects to be greater among native-to-native and migrant-to-migrant groups when they are in systemswhere schools are comprehensive (mixed-ability groups) versusthose that are non-comprehensive (tracked groups) In systemsthat were predominantly non-comprehensive, the researchersalso found enrollment of students of low socioeconomic stand-ing to be higher in the lower tracks than in those that were pre-dominantly comprehensive They also found that in the non-comprehensive system, it was more likely than in the compre-hensive system that the students’ home language would differfrom that of the school, and by extension, on formal testingsuch as that of PISA From their findings, Entorf and Lauk(2008) surmised that non-comprehensive school systems appear

to magnify existing educational inequalities between studentswith lower socioeconomic status and those from more privi-leged backgrounds

In another study conducted over a 6-year period, Burkeand Sass (2013) measured peer effects on classroom achieve-ment using a comprehensive data set comprising data from theFlorida Comprehensive Assessment Test (FCAT), a testdesigned to ensure that students in elementary through highschool are learning the state’s standards After controlling forthe fixed inputs of student ability and teacher effectivenessacross school level (elementary school, middle school and highschool) as they measured for peer effects on academic achieve-ment at classroom level in the areas of mathematics and read-ing, Burke and Sass found a small but significant effect in thelinear model However, they also report that this seemingly sig-nificant effect runs the risk of obscuring subtleties in the inter-actions For example, through the analyses from a non-linearmodel, the researchers found that the students who wereranked as low ability seemed to benefit less from their high-abil-ity colleagues but more from their average ability peers In add-ition, the students at the middle of the distribution (averageability) had greater academic spillovers from being in groupswith their high-ability peers Interestingly, Burke and Sass(2013) observed that high achievers preferred to be in a groupwith similar academic performers to themselves, but given thechoice, they preferred low-ability peers to middle ones.

Unlike most studies on the benefits of peer effects, Duflo,Dupas, and Kremer (2011) were able to study the impact of peereffects on achievement in a quasi-experimental setting Theresearchers compared students’ abilities in a tracked and mixedelementary Kenyan classroom They found that students in track-ing schools, regardless of distribution by ability, did significantlybetter than those in non-tracking ones, a result that they foundstill present 1 year later when they revisited the schools Duflo

et al (2011) also found that students in non-tracking schoolsscored better when, through random assignment, they wereplaced with peers with higher initial scores Additionally, whilepeer effects appeared very strong for the high performing stu-dents they seemed to be nonexistent for the middle studentsand positive but not as strong for low performing students

In light of the backdrop of the international literature sented here, we thought it necessary to conduct a study on peer

effects in the Caribbean setting of Trinidad and Tobago Wenoted that no research of this nature is recorded for theCaribbean region In conceptualizing the study, we recognizedthe importance of having a reliable data source such as thatprovided through the OECD PISA database We noted thatTrinidad and Tobago is the only Caribbean country to date thathas participated in the PISA offering, and as such, that influ-enced our decision to select it for this study Specifically, wehypothesized that peer effects, that is, the impact of peers oneach other’s performance, may cause a shift in results for stu-dents of Trinidad and Tobago As was mentioned before, most

of the youth of Trinidad and Tobago are bi-dialectal, but withthe limitation mentioned earlier of this data not being capturedthrough the PISA exercise, we believe that their true perform-ance on this literacy measure may indeed be unknown We arealso concerned that we cannot measure the contribution of thebi-dialectal factor on peer effects in this study, but despite this

we thought it necessary to highlight it so as to draw attention to

it for future considerations

The Program for International Student AssessmentPISA is an international literacy measure engineered by theOrganization for Economic Cooperation and Development(OECD) to assess the performance of 15-year old youth in theareas of mathematics, science, and reading literacy across mem-ber and nonmember OECD nations (OECD, 2014) The examaims to determine ‘how well students can apply the knowledgeand skills they have learned at school to real-life challenges’(OECD, 2014, para 3) In relation to PISA, the OECD definesliteracy as a ‘student’s ability to apply knowledge and skills inkey subject areas and to analyze, reason and communicateeffectively as they examine, interpret and solve problems’(OECD, 2014, para 1) This definition aligns, to some degree,with descriptions of literacy as socially and culturally situated(Moll & Greenberg, 1990; Street, 1995)

Successful performance on PISA requires that adolescentyouth are highly functional in foundational literacy skills, butthat they also possess the prerequisite skills critical for responding

to discipline-specific questions (Faggella-Luby, Graner, Deshler,

& Drew, 2012; Lee & Spratley, 2010) Discipline-specific contentliteracy is more challenging for students’ comprehension thangeneral literacy (Fang & Schleppegrell, 2010) and evidenceshows that such requirements can place adolescents at a disadvan-tage because of the density of the information involved (Jetton &Shanahan, 2012) In general, OECD countries tend to performbetter than non-OECD countries in the literacy areas assessed,with a number of explanations proffered for this disparity inperformance

A key argument arising recently that affects students’ tioning in non-OECD countries relates to the language ofassessment used on PISA Findings from recent analyses(Author,2017,2018) indicate that students who take the test in

func-a lfunc-angufunc-age thfunc-at is not their home lfunc-angufunc-age perform cantly lower than their counterparts This issue of assessment isfurther complicated in countries that are bi-dialectal where theofficial dialect, though not foreign, is not native to the students.For example, in the English-based Caribbean, Standard English

signifi-is the official language (and the language of high standing) butfor the most part an English vernacular is generally the homelanguage and by extension the language of wide use In thisCaribbean context, citizens self-identify as English speakers,and on tests such as PISA that are administered in the officiallanguage, students identify their home language as StandardEnglish, which can then potentially conflate the results of theirassessments with the results of students who are actuallyStandard English L1 speakers On such tests, there are no for-mal checks to identify which students are indeed StandardEnglish or vernacular L1 speakers

In a previous article (Author, 2018), we presented thecase that tests such as PISA do not take into account the non-standardardized English use of students in the Caribbean.Using Trinidad and Tobago’s PISA 2009 assessment data as ourreference point, we argue that such a practice may inadvert-ently reinforce assumptions that privilege Standard English as

a language of assessment and thus devalue other WorldEnglishes, such as the non-standardized varieties in theCaribbean In this article, we use the same PISA 2009 data

representing Trinidadian and Tobagonian students to presentanother variable, peer effects, that when left unattended canpotentially influence students’ literacy performance.

As with our previous article (Author,2018) we continue tostress the importance of the study of the PISA data as it relates

to Trinidad and Tobago since it allows us to gain some insightsinto a context where students are being assessed usingStandard English, which, though having related features, is sig-nificantly different from the students’ home dialect In thiscase, this study provides us with the opportunity to examine theimpact of peer effects on science, math and reading literacy forlearners in this bi-dialectal context On embarking on thisstudy, we recognized that we would not be able to definitivelymeasure the effects of the bi-dialectal variable as PISA doesnot formally differentiate between the related dialects in thiscontext Nevertheless, reporting the findings in this languagecontext will not only bring awareness to how the non-demo-graphic factor of peer effects can impact and confound PISAassessment results, but also that such results obtained from abi-dialectal cohort present their own complexities that should

be given some consideration The findings of this study shouldtherefore provide the OECD with information on other factorsthat are not usually considered in their analyses, in this case,peer effects on the literacy performance of adolescent bi-dialectal learners

Method

As previously mentioned, this study utilizes the extant PISAdatabase From this database, we were able to run analyses tohelp us to draw conclusions relating to peer effects amongstudents in Trinidad and Tobago We wish to stress that thisstudy is correlative in nature and, as such, no causality can

be assumed

Sample and Participants

We used a nationally representative sample of 15-year-old dents from Trinidad and Tobago who participated in PISA

2009 For that year, PISA was administered by respectivenational governments in more than 60 countries around theworld Of these, approximately half the countries were OECDmember states In the PISA 2009 survey, 475,460 students from17,145 schools participated PISA is administered every 3 years

in the three areas of mathematics, reading, and science literacy,but with a special focus on one of the areas in each cycle In

2009, the area of focus was reading literacy This year was thefirst time that Trinidad and Tobago participated in PISA.PISA samples are based on a two-stage stratified, randomselection scheme where schools are randomly selected in thefirst stage, and then students are randomly selected fromwithin each selected school in the second stage When appro-priate sampling weights are used these samples become repre-sentative of their target populations In the PISA 2009administration, 4778 students and 158 schools participatedfrom Trinidad and Tobago (OECD, 2014) These studentsrepresent a population of 14,938 15-year-old students in thecountry The data set was complete for all factors except onevariable of interest, socioeconomic status, used in empiricalanalyses presented later in this study Since the percentage ofmissing data was only 2.8%, we decided to impute missing val-ues For this purpose, we used the single random imputationmethod (Allison, 2001) and imputed missing data using aniterative Markov Chain Monte Carlo (MCMC) procedure(IBM Corp, 2013) that predicted values of socioeconomic sta-tus based on all other variables included in the imputationmodel These predicted values were then used to impute miss-ing values Thus, after imputation we had 4,778 cases withcomplete information

Individual Student Literacy

Student performance in PISA literacy assessments of matics, reading, and science is measured on the basis of studentresponses to assessment items that are delivered in a variety offormats, such as multiple-choice items and constructedresponse questions PISA reports scores for each student as aset of five plausible values for each area that represent random

draws of scores from the distribution of all possible scoresattributable to that student (Mislevy, 1991; Mislevy, Beaton,Kaplan, & Sheehan, 1992) Following Brown and Micklewright(2004), we used one randomly chosen plausible value for allempirical analyses performed in this study For a technical dis-cussion, including psychometric properties of plausible values,and for examples of assessment items we refer the interestedreader to OECD (2012).

Literacy scores in PISA data files are scaled to have a mean

of 500 and a standard deviation of 100 for OECD member tries Although this scheme makes it useful to compare a coun-try’s performance against OECD countries as a group, it makescomparisons within a country relatively difficult to interpret Inorder to avoid this issue, we rescaled the scores for this non-OECD country to have a mean of 500 and a standard deviation

coun-of 100 in our sample coun-of 4778 students Such rescaling represents

a linear transformation of original scores and thus does notaffect unit-independent statistics such as correlations, R2, orCohen's d, and results of tests of hypotheses In our sample, therescaled scores ranged between 162.69 and 811.67 for mathemat-ics, between 166.98 and 825.92 for reading, and between 151.69and 794.88 for science Each of the three literacy scores wereapproximately normally distributed We provide the histogram

of individual mathematics scores as illustration in Figure 1

In addition to the assessment component, the PISA survey alsocollects data on student beliefs, perceptions, and attitudes, aswell as demographic information such as gender, grade, andsocioeconomic status

Group Literacy PerformanceThis variable is a measure of peer performance in literacy andwas obtained for a school as the average literacy of all studentssampled from that school Thus, for a school j that has m stu-dents, group literacy X as a function of student literacy Y isgiven by the expression in (1):

A positive association between individual student literacyand group literacy performance would thus be indicative of theexistence of a literacy-related peer effect For examples of simi-lar applications we refer the interested reader to Burke andSass (2008), Hoover-Dempsey, Bassler, and Brissie (1987), andPerry and McConney (2013) In our sample, group literacy per-formance ranged between 316.76 and 675.97 for mathematics(M¼ 500, SD ¼ 81.83), between 252.55 and 666.72 for reading(M¼ 500, SD ¼ 77.55), and between 284.97 and 653.64 forscience (M¼ 500, SD ¼ 78.43).

Control Variables

In order to control for student demographic differences, weused gender, student grade level at school, and student socioe-conomic status as covariates in our models of individual andgroup literacy Gender is a dichotomous nominal variable thattakes a value of 0 for boys and a value of 1 for girls Of the

4778 students in our sample 2,366 were boys (49.5%) and 2,412(50.5%) were girls Grade is an ordinal variable with valuesranging between 7 and 11 In our sample 100 (2.1%) students

were in Grade 7; 420 (8.8%) were in Grade 8; 1210 (25.3%)were in Grade 9; 2679 (56.1%) were in Grade 10; and 369(7.7%) were in Grade 11 Socioeconomic status is based on theindex of economic, cultural, and social status that is included

in PISA student data files (OECD, 2012) This index has threeunderlying subscales that include parental education, parentaloccupation, and home possessions These factors were com-bined using factor analysis The reliability of this index in oursample was 63 with factor loadings being 74 for parental edu-cation, 79 for parental occupation, and 74 for home posses-sions The values of this variable ranged from a minimum

of3.63 to a maximum of 3.37 (M ¼ 0, SD ¼ 1) in our sample

Analytical Method

We computed descriptive statistics and correlations for ual and group literacy performance in mathematics, reading,and science, and control variables In order to estimate the pro-portion of variation in individual student literacy scores thatcould be explained by group literacy performance, we com-puted three separate ordinary least squares regression modelsfor each of the three literacy areas The first of these modelswas a simple regression model that predicted individual literacyfrom group literacy This model allowed us to calculate theupper limit on the effect of group literacy on individual studentscores The second model was a multiple regression model thatpredicted individual literacy from the three control variables.This model allowed us to estimate the proportion of variation

individ-in individ-individual literacy scores that could be explaindivid-ined by graphic differences among students The third model was also

demo-a multiple regression model thdemo-at combined demo-all predictors fromthe first two models in order to predict individual literacy per-formance The difference in R2 between this model and thesecond model allowed us to calculate the lower limit on theproportion of variation uniquely attributable to group literacyperformance The three regression models are given by expres-sions in (2)–(4)