(2005) a synthesis of research on language of reading - slavin & cheung

Upload: horian49

Post on 03-Jun-2018

219 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    1/39

    A Synthesis of Research on Language of Reading Instruction for English Language LearnersAuthor(s): Robert E. Slavin and Alan CheungSource: Review of Educational Research, Vol. 75, No. 2 (Summer, 2005), pp. 247-284Published by: American Educational Research AssociationStable URL: http://www.jstor.org/stable/3516050.

    Accessed: 14/02/2014 21:26

    Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at.http://www.jstor.org/page/info/about/policies/terms.jsp

    .JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

    content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

    of scholarship. For more information about JSTOR, please contact [email protected].

    .

    American Educational Research Associationis collaborating with JSTOR to digitize, preserve and extend

    access toReview of Educational Research.

    http://www.jstor.org

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/action/showPublisher?publisherCode=aerahttp://www.jstor.org/stable/3516050?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/3516050?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=aera
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    2/39

    Reviewof EducationalResearchSummer 005, Vol.75,No. 2, pp. 247-284

    A Synthesis of Research on Language of ReadingInstruction for English Language LearnersRobert E. SlavinJohns Hopkins UniversityAlan CheungSuccess for All Foundation

    This article reviews experimental studies comparing bilingual and English-only readingprogramsfor English language learners. The review method isbest-evidence synthesis, which uses a systematic literature search, quantifi-cation of outcomes as effect sizes, and extensive discussion of individualstud-ies that meet inclusion standards. A total of 17 studies met the inclusionstandards. Among 13 studies focusing on elementary reading for Spanish-dominant students, 9favored bilingual approaches on English reading mea-sures, and 4found no differences,for a median effect size of + 0.45. Weightedby sample size, an effect size of +0.33 was computed,which is significantlydifferentfrom zero (p < .05). One of two studies of heritage languages(French and Choctaw) and two secondary studies favored bilingualapproaches. The review concludes that although the numberof high-qualitystudies is small, existing evidence favors bilingual approaches, especiallypaired bilingual strategies thatteach reading in thenative language and Eng-lish at different times each day. However,further research using longitudi-nal, randomizeddesigns is needed to determine how best to ensure readingsuccess for all English language learners.

    KEYWORDS:ilingualeducation,English languagelearners, anguageof instruc-tion,reading,researchreview.The readingeducation of English language learners(ELLs)has become oneof the most importantissues in all of educational policy and practice. As thepace of immigrationto the U.S. andotherdeveloped countrieshas acceleratedin recent decades, increasing numbersof children in U.S. schools come fromhomes in whichEnglishis not theprimary anguage spoken.As of 1999, 14 mil-lion Americans aged 5-24, or 17%of that age group, spoke a language otherthanEnglish at home. This is more than twice the numberof such individualsin 1979, when only 9% of Americans aged 5-24 spoke a language other thanEnglish athome (NCES, 2004). Althoughmanychildren of immigrant amiliessucceed in reading,too many do not. In particular,Latino and Caribbeanchil-dren are disproportionately ikely to performpoorly in readingand in school.As No Child Left Behind and otherfederal and state policies begin to demandsuccess for all subgroups of children, the reading achievement of Englishlanguage learners is taking on even more importance.Thousands of schoolscannot meet their adequate yearly progress goals, for example, unless their247

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    3/39

    Slavin & CheungEnglish languagelearnersaredoing well in reading.Moreimportant,Americansociety cannot achieve equal opportunityfor all if its schools do not succeedwith the children of immigrants.Sixty-fivepercentof non-English-speakingmmigrantsnthe United Statesareof Hispanicorigin(NCES,2004), and this is also one of the fastestgrowingof allgroups.HispanicshaverecentlysurpassedAfricanAmericansas thelargestminor-ity groupin the United States. Hispanicstudentsas a whole, includingEnglish-proficientchildren n the secondgenerationandbeyond,scoresignificantly owerin reading than other students. On the National Assessment of EducationalProgress NAEP;Grigg,Daane,Jin,& Campbell,2003), whichexcludes childrenwith the lowest levels of English proficiencyfrom testing, only 44% of Latinofourthgradersscored at or above the basic evel, in comparisonwith 75%ofAnglo students.Only 15%of Latinofourthgraders coredat proficient rbetter,ascomparedwith 41% of Anglos.Furthermore,1% of studentswho speak Span-ish at home fail to complete high school, as comparedwith 10%of studentswhospeak only English(NCES,2004).There is considerable controversy, among policymakers, researchers,andeducators,about how best to ensure the reading success of English languagelearners.Many aspectsof instructionareimportantn thereadingsuccess of Eng-lish language learners,but one question has dominatedall others:What is theappropriaterole of the native language in the instructionof English languagelearners?In the 1970s and 1980s, policies andpracticefavoredbilingualeduca-tion, in which children were taught partiallyorentirelyin theirnative language,andthentransitionedat some pointduringthe elementarygradesto English-onlyinstruction.Such programsare still widespread,but fromthe 1990s to the pres-ent, thepoliticaltide hasturnedagainstbilingualeducation;California,Arizona,Massachusetts,and other stateshave enactedpolicies to greatlycurtailbilingualeducation.Recent federalpolicies restrict the amountof time thatchildrencanbe taught n theirnativelanguage. Among researchers, hedebatebetweenadvo-cates of bilingualandEnglish-onlyreadinginstructionhas been fierce, andide-ology hasoften trumpedevidence on both sides of the debate(Hakuta,Butler,&Witt, 2000).Thisarticlereviewsresearchon thelanguageof reading nstruction orEnglishlanguage earners n anattempt o applyconsistent,well-justifiedstandards f evi-dence to drawconclusionsaboutthe role of nativelanguage n reading nstructionfor these children.The review appliesa techniquecalled best-evidencesynthe-sis (Slavin, 1986), which attempts o use consistent,clear standards o identifyunbiased,meaningfulinformation rom experimentalstudies andthen discusseseach qualifying study, computing effect sizes but also describingthe context,design, and findings of each study. Best-evidence synthesis closely resemblesmeta-analysis,but it requiresmoreextensivedescriptionof key studies.Detailsofthis procedurearedescribedbelow. The purposeof this review is to examinetheevidence on languageof instruction n reading programsfor English languagelearners o discoverhow muchscientificbasisthere is forcompetingclaimsaboutthe effects of bilingualas opposed to English-only programs.Such a review isneededto informpractitioners,policymakers,and researchersaboutthe currentstate of theevidenceon thistopicandaboutgapsin theknowledgebasein need offurther cientificinvestigation.248

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    4/39

    Language of InstructionFor many years, researchers,educators,and policymakershave debatedthequestionof theappropriateanguageof reading nstructionor childrenwho speak

    languagesotherthanEnglish. Proponentsof bilingual nstructionarguethatwhilechildrenare earning o speakEnglish,theyshouldbe taught o read n theirnativelanguagefirst,to avoid thefailureexperiencethat s likely if childrenareasked tolearn bothoralEnglishandEnglishreadingat the same time. Programsbased onthisphilosophytransition hildren o English-onlyinstructionwhen theirEnglishis sufficient o ensuresuccess,usually n thirdorfourthgrade.Alternatively,manybilingualprograms eachyoungchildren o read bothin theirnativelanguageandin English,at different imesof thedayoron alternating ays.There s a greatdealof evidence thatchildren'sreadingproficiency n theirnativelanguage s a strongpredictorof their ultimateEnglish readingperformance(Garcia,2000; Lee &Schallert,1997; Reese, Garier, Gallimore,& Goldenberg,2000) and thatbilin-gualism itself does not interfere with performance n either language (Yeung,Marsh,& Suliman,2000). Bilingualadvocatesalso arguethatwithout nativelan-guageinstruction,English anguage earnersare ikelyto lose theirnative anguageproficiency,or fail to learnto read n theirnativelanguage, osing skills thatare ofeconomic and social value in the worldtoday. Opponentsof bilingualeducation,on the otherhand,arguethat nativelanguage nstructionnterfereswithor delaysEnglishlanguagedevelopment,andrelegateschildrenwho receive such instruc-tion to a second-class,separate tatuswithintheschooland,ultimately,within soci-ety. They reason that more time on English readingshould translate nto morelearning see Rossell & Baker,1996).Reviews of the educational utcomesof native anguagenstruction avereachedsharplyconflictingconclusions. In a meta-analysis,Willig (1985) concludedthatbilingual education was more effective than English-only instruction.Wong-FillmoreandValadez(1986) cameto the same conclusion.However,a reviewbyRossell and Baker (1996) claimed that most methodologically adequatestudiesfound bilingualeducationto be no more effective thanEnglish-only programs.Greene 1997)re-analyzedhestudiescitedby RossellandBakerandreportedhatmanyof the studies heycited ackedcontrolgroups,mischaracterizedhetreatments,or hadotherseriousmethodologicallaws.Amongthe studies hatmetanacceptablestandardof methodologicaladequacy, ncludingall of the studiesusing randomassignmentto conditions,Greene found that the evidence favoredprograms hatmadesignificantuse of native anguage nstruction.AugustandHakuta 1997)con-cludedthatalthoughresearchgenerally avoredbilingualapproaches,henatureofthemethodsused andthepopulationso whichtheywereappliedwere more mpor-tant han helanguageof instruction erse. Quantitativeesearch n the outcomesofbilingualeducationhasdiminished n recentyears,butpolicy andpracticeare stillbeinginfluencedby conflicting nterpretationsf researchon thistopic.The follow-ingsectionssystematicallyxamine his evidencetoattempto discoverwhatwe canlearn romresearch o guidepoliciesin thiscontroversial rena.

    EnglishImmersionandBilingual ProgramsWhen a child enters kindergartenor first grade with limited proficiency inEnglish,the school faces a seriousdilemma.How can thechildbe expectedto learnthe skills andcontent aught n theearlygradeswhile he or she is learningEnglish?

    249

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    5/39

    Slavin&CheungTheremay be manysolutions,but two fundamental ategoriesof solutionshavepredominated:Englishimmersionandbilingualeducation.EnglishImmersionInimmersion trategies,English anguage earnersareexpectedto learn n Eng-lishfromthebeginning,andtheirnative anguageplayslittleor no roleindailyread-ing lessons. Formalor informalsupport s likely to be given to ELLsto helpthemcope in anall-Englishclassroom.Such supportmay includehelp froma bilingualaide who providesoccasionaltranslation rexplanation,a separate lass inEnglishas a SecondLanguage o helpbuild oralEnglishskills,or use of a carefulprogres-sion fromsimplifiedEnglishto full Englishas children'sskills grow. TeachersofEnglishlanguage earnersmayuse languagedevelopmentstrategiessuch as totalphysical response (acting out words) and realia (concrete objects to representwords)to help students nternalizenew vocabulary.Teachersmay simplify theirlanguageandteachspecific vocabulary hat s likely to be unfamiliaro ELLs(seeCalder6n,2001; Carlo et al., 2004). Immersionmay involve placing Englishlanguage earnersmmediatelynclassescontainingEnglishmonolingual hildren,or it mayinvolveteachingELLsin a separate lass for sometimeuntilthechildrenarereadyto be mainstreamed.These variationsmay well have importancen theoutcomesof immersionstrategies,but theirkey commonfeature s the exclusiveuse of Englishtexts,withinstruction verwhelminglyorentirely n English.Manyauthorshave madedistinctionsamongvariousformsof immersion.Onetermoften encountered s submersion, primarilyused pejorativelyto refertosink or swim strategies n which no special provisionis made for the needs ofEnglish anguage earners.Thisis contrastedwith structured nglish mmersion,which refersto a well-planned,gradualphase-inof Englishinstruction elyingini-tiallyon simplificationandvocabulary-buildingtrategies.Inpractice, mmersionstrategiesarerarely pure types; andin studiesof bilingualeducation, mmersionstrategies are rarely described beyond their designation as the English-onlycontrolgroup.BilingualEducationBilingual educationdiffers fundamentally rom English immersionin that itgives English language learnerssignificant amounts of instructionin readingand/orother subjects in their native language. In the United States, the over-whelmingmajorityof bilingualprograms nvolve Spanishbecause of the greaterlikelihood of acriticalmass of studentswho areSpanish-dominantnd thegreateravailabilityof Spanishmaterials.Therearebilingualprogramsn Portuguese,Chi-nese, andotherlanguages,but these are rare.In transitionalbilingual programs,childrenaretaughtto readentirelyin theirnative languagethroughthe primarygrades; heytransition o Englishreading nstruction omewherebetween the sec-ondandfourthgrades.Englishorallanguageproficiency s taught rom thebegin-ning, andsubjectsother thanreadingmay be taught n English.But the hallmarkof transitionalbilingualeducation s theteachingof reading n the nativelanguagefor a periodof time. Suchprograms an be early-exit models,with transition oEnglishcompletedin secondor thirdgrade,or late-exit models, in whichchil-drenmay remainthroughout lementaryschool in native-languagenstruction oensuretheirmasteryof readingand content beforetransition see Ramirez,Pasta,250

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    6/39

    Research on Languageof ReadingInstructionforELLsYuen, Billings, & Ramey, 1991). Alternatively, pairedbilingual models teachchildren o read n bothEnglishand theirnativelanguageatdifferent imeperiodseach day or on alternatingdays. Withina few years,the nativelanguage readinginstructionmaybe discontinued,as childrendevelopthe skills to succeedin Eng-lish.Willig (1985) calledthis model alternativemmersion, because childrenarealternativelymmersed n nativelanguageandEnglishinstruction.Two-waybilingualprograms, lso called dual anguage or dual mmersion,providereading nstruction n thenative anguage(usuallySpanish)and nEnglishbothto ELLsandto Englishspeakers Calderon& Minaya-Rowe,2003;Howard,Sugarman,& Christian,2003). For the ELLs,a two-way program s like a pairedbilingual model, in that they learn to read both in English and in their nativelanguageatdifferent imes eachday.A special case in bilingualeducation is the kind of program hatis designedmore to preserveor show respectfor a given languagethan to help childrenwhoare genuinely strugglingwith English. For example, Morgan (1971) studied aprogram n Louisiana or childrenwhose parentsoften spokeFrenchathomebutgenerallyspokeEnglishas well. Such heritage anguage programsare includedin this review if the outcomevariable n the studyis anEnglishreadingmeasure.However,these programs hould be thoughtof as addressinga differentproblemfromthataddressedbybilingualorimmersionreading nstructionorchildrenwhoarelimited in Englishproficiency.

    Problemsof Researchon Language of InstructionResearchon theachievementeffects of teaching n the child's nativelanguage,in comparisonwith teachingin English,suffersfrom a numberof inherentprob-lems beyondthose typicalof other researchon educationalprograms.First,thereareproblemsconcerning heages of thechildren nvolved,the lengthof timetheyhave been taughtin their firstlanguage,and the length of time they have beentaught in English. For example, imagine that a transitionalbilingual programteachesSpanish-dominanttudentsprimarilyn Spanish n GradesK-2 and thengradually ransitions hem to English by fourthgrade.If thisprogram s comparedwith anEnglishimmersionprogram,at whatgradelevel is it legitimateto assessthechildren nEnglish?Clearly,a test in secondgrade s meaningless,as the bilin-

    gualchildrenhave notyetbeentaught o read n English.At the end of thirdgrade,the bilingualstudentshave partially ransitioned,buthave they hadenoughtimeto become fully proficient?Some would arguethat even the end of fourthgradewouldbe too soon to assess the children airlyin sucha comparison,as the bilin-gualchildrenneedareasonable imeperiod n which to transfer heirSpanishread-ing skills to English(see, forexample,Hakuta,Butler,& Witt,2000).A relatedproblemhas to do withpretesting. maginethata studyof a K-4 tran-sitional Spanishbilingualprogrambegan in thirdgrade.Whatpretestwould bemeaningful?AnEnglishpretestwouldunderstatehe skillsof thetransitional ilin-gual students,while a Spanish test would understatethe skills of the Englishimmersionstudents.Forexample,Valladolid(1991) comparedgainsfromGrades3 to 5 for childrenwho had been in eitherbilingualor immersionprograms incekindergarten.Thesechildren's pretest cores arein factposttestsof verydiffer-ent treatments.Yet studiescomparingtransitionalbilingualand immersionpro-gramsaretypicallytoobrief to havegiventhe students n thetransitional ilingual251

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    7/39

    Slavin& Cheungprogramsenoughtime to have completedthe transition o English. In addition,many studies begin after students have alreadybeen in bilingual or immersiontreatments or severalyears.The studiesthatdo look at4- or5-yearparticipationsnbilingualorimmersionprogramsare usually retrospective(i.e., researcherssearch recordsfor childrenwho havealreadybeenthrough heprogram).Retrospective tudiesalsohavechar-acteristicbiases,in thattheybeginwiththechildrenwho endedupin one programoranother.Forexample,childrenwho areremoved rom agiventreatmentorsys-tematicreasons,such as Spanish-dominanttudentsremoved romEnglish mmer-sionbecause of their ow performancehere,cangreatlybias aretrospective tudy,makingthe immersionprogramook moreeffective thanit was in reality.Manyinherentproblemsrelateto selectionbias.ChildrenendupintransitionalbilingualeducationorEnglishimmersionby many processesthat couldbe highlyconsequential or the outcomes. Forexample,Spanish-dominanttudentsmay beassignedto Spanishor Englishinstructionon the basis of parentpreferences.Yetparentswho would select Englishprogramsare surelydifferentfrom those whowouldselect Spanish n waysthatmatter or outcomes.A parentwho selectsEng-lishmaybe moreor less committed o education,maybe less likelyto be planningto return o a Spanish-speaking ountry,ormayfeel morepositiveaboutassimila-tion. Thomas and Collier (2002) reportedextremely low scores for Houstonstudentswhose parentsrefusedto have theirchildrenplacedin eitherbilingualorEnglishas a SecondLanguageprograms.Are those scoresdue to relativelyposi-tive effects of bilingual and ESL programs,or are there systematicdifferencesbetween childrenwhose parentsrefused bilingual or ESL programsand otherchildren?It is impossibleto say, as no pretestscoreswerereported.Bilingualprogramsaremorelikely to exist in schools with very high propor-tions of Englishlanguage earners,andthis is anotherpotentialsourceof bias.Forexample, Ramirez et al. (1991) found that schools using late-exitbilingualpro-gramshadmuchhigherproportions f ELLsthandidearly-exitbilingualschools,andEnglishimmersionschools had the smallestproportionof ELLs.This meansthatwhatever he languageof instruction,children n schools withveryhigh pro-portionsof ELLsareconversing ess with nativeEnglish speakersbothin and outof school thanmightbe the case in anintegrated choolandneighborhoodhatusesEnglishfor all studentsbecause its proportionof ELLs is low. Most problemati-cally, individualchildrenmaybe assignedto nativelanguageorEnglish programsbecauseof theirperceivedor assessedcompetence.Nativelanguage nstructionsoften seen as an easier,moreappropriate lacementfor ELLswho arestrugglingto read n theirfirst anguage,whereasstudentswho areverysuccessfulreaders ntheirfirstlanguageor are felt to have greaterpotentialareplacedin English-onlyclasses. Thisselectionproblem s mostvexingatthepointof transition, sthemostsuccessfulstudents n bilingualprogramsgo through he transition arlier hantheleast successful children. A studycomparingbilingualand immersionprogramsinvolvingthirdorfourthgradersmaybe seriouslybiasedby thefact thatthehigh-est-achievingbilingualstudentsmayhavealreadybeentransitioned,withthe resultthattheremainingstudentsarethe lowest achievers.A source of bias not uniqueto studies of bilingualeducationbut very impor-tant n this literature s the filedrawer problem: he fact that studiesshowingnodifferences areless likely to be publishedor to otherwisecome to light.This is a252

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    8/39

    Researchon Languageof ReadingInstructionor ELLsparticularproblem n studieswith small samplesizes, which areveryunlikelytobepublished f theyshowno differences.Thebestantidote o the file drawerprob-lem is to search for dissertationsand technicalreports,which aremorelikely topresenttheirdataregardlessof the findings(see Cooper,1998).Finally,studiesof bilingualeducationoftensaytoo little about hebilingualandimmersionprograms hemselves or thedegreeorqualityof implementation f theprograms.Yet bilingualmodelscanvary substantiallyn quality,amountof expo-sureto Englishin andout of school, teachers'languagefacility, time duringtheschoolday,instructional trategiesunrelatedo languageof instruction,andso on.Becauseof these inherentmethodologicalproblems,anadequate tudycompar-ingbilingualand mmersionapproacheswould(a)randomlyassigna largenumberof children o be taught n Englishor theirnative anguage; b)pretest hem n theirnativelanguagewhen they begin to be taughtdifferentially,eitherin theirnativelanguageorin English(typically nkindergarten);c) follow themlongenoughforthe latest-transitioninghildren n the bilingualconditionto have completedtheirtransition o Englishandhave been taught ong enoughin Englishto makea faircomparison; nd(d) collectdatathroughoutheexperiment o document he treat-mentsreceived in all conditions.Unfortunately, nly a few, very small studies ofthiskindhave everbeencarriedout. As a result,the studiesthatcomparebilingualandEnglish-onlyapproachesmust be interpretedwithgreatcaution.

    Review MethodsThis section focuses on researchcomparing mmersionandbilingualreadingprogramsappliedwithEnglishlanguage earners,withmeasuresof Englishread-ing as the outcomes. The review method,best-evidencesynthesis(Slavin, 1986),uses thesystematic nclusioncriteriaandeffect size computations ypicalof meta-analyses(see Cooper,1998;Cooper& Hedges, 1994)butdiscusses thefindingsofcritical studies in a form more typical of narrativereviews. This strategy isparticularlywell suitedto the literature n readingprogramsorEnglish languagelearners,becausethe studiesarefew in numberandaresubstantivelyand method-ologically diverse.Insuch a literature,t is particularlymportanto learnas muchas possible fromeach study,notjust to averagequantitativeoutcomes and studycharacteristics.

    LiteratureSearchStrategyThe literaturesearchbenefited from the assistanceof the federallycommis-sioned NationalLiteracyPanelon theDevelopmentof LiteracyAmong LanguageMinorityChildrenandYouth,chairedby Diane AugustandTimothyShanahan.The firstauthorwas initiallya memberof the panelbutresignedin June2002 toavoid a 2-yeardelay in publicationof the presentarticle. The article, however, isindependentof thepanel'sreportanduses differentreview methodsandselectioncriteria. ResearchassistantssearchedERIC,Psychological Abstracts,andotherdatabases for all studies with the following descriptors: language minoritystudents,Englishlanguagelearners,bilingualeducation,bilingualstudents,bilin-gualism, English as a second language,English immersion,dual language, andtwo-way bilingualeducation.Citationsfrom other reviews and articleswere alsoobtained.Inparticular, very effort was made to find all studies cited in previousreviews. Fromthis set, we selected studies that metthecriteriadescribedbelow.253

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    9/39

    Criteria or InclusionThe best-evidencesynthesisfocused on studiesthatmet minimalstandards fmethodologicaladequacyandrelevanceto thepurposesof the review.Thesewereas follows:

    Criterion1.Thestudiescompared hildrenwhoweretaughtreading n bilin-gualclasses withthosetaught n Englishimmersionclasses, as definedearlier.Criterion2. Eitherrandomassignment o conditionswas used,orpretestingor othermatchingcriteriaestablished the degreeof comparabilityof bilingualandimmersiongroupsbefore the treatmentsbegan.If thesematchingvariableswere not identicalat pretest,the analysesmadeadjustments or pretestdiffer-ences, or else datapermittingsuch adjustmentswere presented.Studies with-out controlgroups,such as pretest-posttestcomparisonsor comparisonswithexpected scores or gains, were excluded. Studies with pretestdifferencesexceedingone standard eviationwere excluded.Thosewithpretestdifferencesless thanone standard eviationwereincluded f theresearchers ad carriedoutappropriatetatisticaladjustments.A specialcategoryof studieswasrejectedon thebasisof therequirement fpretestmeasurementbefore treatmentsbegan.Those werestudiesin which thebilingualand mmersionprogramswerealreadyunderwaybeforepretestingormatching.Forexample,Danoff, Coles, McLaughlin,andReynolds(1978), in awidely cited study,compared1-yearreadinggains in many schools by usingbilingualor immersionmethods. The treatmentsbeganin kindergarten r firstgrade,but the pretests(and, later,posttests)were administered o childreninGrades2-6. Because thebilingualchildrenweretaughtprimarilyn theirnativelanguage in K-1 and the immersion children were taught in English, theirpretests n secondgradewouldsurelyhavebeen affectedbytheir reatment on-dition. Meyer andFeinberg(1992, p. 24) notedthe same problemwith refer-ence to theGrades 1-3 componentof theRamirezet al. (1991) study: It s likewatchinga baseballgamebeginningin the fifthinning:If you arenot told thescore from the previousinnings,nothingyou see can tell you who is winningthegame. Similarly,several studiestestedchildren nupperelementaryor sec-ondarygradeswhohadexperiencedbilingualorimmersionprogramsn earlieryears.These studies were includedif premeasureswere availablefrom beforethe programsbegan,but in most cases suchpremeasureswere notreported, othere was no way to know if the groups were equivalentbeforehand(e.g.,Thomas& Collier,2002; Curiel,Stenning,& Cooper-Stenning,1980).Criterion3. The subjectswere English languagelearners n elementaryorsecondary schools in English-speakingcountries. Studies that mixed ELLsand English monolingual students in a way that did not allow for separateanalyses were excluded (e.g., Skoczylas, 1972). Studiesof childrenlearninga foreign languagewere not included. However, Canadianstudies of Frenchimmersionhave been widely discussed and arethereforediscussed in a sepa-rate section.Criterion4. Thedependentvariables ncludedquantitativemeasuresof Eng-lish reading performance,such as standardized ests and informal readinginventories.If treatment-specificmeasureswereused, theywere includedonlyif therewas evidence thatall groupsfocused equally on the same outcomes.

    254

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    10/39

    Research on Languageof ReadingInstructionor ELLsMeasures of outcomesrelated to reading,such as languagearts,writing,andspelling,were not included.Criterion5. The treatmentdurationwas at least 1 school year.For the rea-sons discussed earlier,even 1-yearstudies of transitionalbilingualeducationareinsufficient,because students aught n their nativelanguageareunlikelytohave transitioned oEnglish.Studiesshorterhan1yeardo notaddress heques-tion in a meaningfulway.

    Studies thatpassedaninitialscreeningforgermaneness o the topic, includingallstudies cited by Rossell and Baker(1996) or by Willig (1985), are listed in theAppendix,whichindicates whethereach studywas includedand,if not, the mainreasons for exclusion.Limitations

    Itis importanto notethat hereview methodsapplied nthis best-evidencesyn-thesis have some importantimitations.First,in requiringmeasurableoutcomesand control groups, the synthesis excludes case studies and qualitativestudies.Many such descriptivestudiesexist and are valuable in suggestingprogramsorpracticesthatmightbe effective. Descriptionalone, however, does not indicatehow much children earned n a given programor whatthey wouldhave learnedhadthey not experiencedthatprogram.Second, it is possible that a program hathas no effect on readingachievementmeasuresmightnevertheless ncrease chil-dren's interest n readingor readingbehaviors outside school. However, studiesrarelymeasure such outcomesin any systematicor comparativeway, so we canonly speculateaboutthem.Finally,it is importanto notethatmanyof the studiesreviewedtookplacemany yearsago. Socialandpoliticalcontexts,aswell as bilin-gualand mmersionprograms,havechanged,so it cannotbe takenforgranted hatoutcomesdescribedherewouldapplyto outcomesof bilingualand mmersionpro-grams today.

    Computation f EffectSizesIf possible, effect sizes were computedfor each study.These were computedas the experimentalmean minus the control mean, with the result divided by apooled standarddeviation. Wheninformationwas lacking,however,effect sizeswere estimated using exact t or p values or other well-established estimationmethods(see Cooper, 1998; Cooper& Hedges, 1994;Lipsey & Wilson, 2001).For studies that lacked means and standarddeviations and reportedno signifi-cant difference betweenthe experimentaland controlgroupsand didnotindicatethe direction of the effect (e.g., Cohen, 1975), an estimatedeffect size of zerowas used. Only Englishreadingmeasureswere used in determiningeffect sizes,even if other measureswerementioned n the text. No studywas excludedsolelyon the groundsthatit did not providesufficient information or computationofan effect size.

    Data AnalysisAll datawereentered nto the beta versionof theComprehensiveMeta-AnalysisProgramBorenstein, 005)to estimate he effect sizes of eachstudy, ocalculate heoverallmeanweightedeffectsizes,andtotest whether hemeanweightedeffect size

    255

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    11/39

    Slavin & Cheungwasderived roma homogeneous et (Q statistic).Theweighting actorwas samplesize, so thateffect sizes fromlargersamplescontributedmoreto themeanthandidthose fromsmaller amples.Eachstudycontributed singleeffect size to the overallmeanweightedeffect size. Forstudies hathadmorethanone independent rouporone independent utcomemeasure,effect sizes werecalculated eparatelyor eachgroupandmeasure.These effect sizes werethenaveraged o createone effect sizeforthe study.Forlongitudinal tudies,the lasttime-pointwas usedto estimatetheoveralleffect of thestudy.Forexample, f a study olloweda groupof children romGrade1to Grade5, theoutcomemeasuresor fifthgraderswere usedtogenerate heeffect sizes.

    Previous Quantitative ReviewsThe debate aboutempiricalresearch on language of instructionfor Englishlanguage learners has largely pitted two researchers, Christine Rossell andKeith Baker, against several other reviewers. Rossell and Baker have carriedouta series of reviews andcritiques arguingthatresearchdoes not supportbilin-gualeducation(see Baker & de Kanter,1981, 1983;Baker, 1987;Rossell, 1990;Rossell & Baker, 1996; Rossell & Ross, 1986). The most comprehensive andrecent version of their review was publishedin 1996. Incontrast,Willig (1985)carried out a meta-analysisandconcluded that researchfavoredbilingual edu-cation, after controls were introduced for various study characteristics.Otherreviewersusingnarrativemethodshave agreedwithWillig (e.g., Wong-Fillmore& Valadez, 1986). Baker (1987) and Rossell and Baker (1996) criticized theWillig (1985) review in detail,andWillig (1987) respondedto the Baker(1987)criticisms.In areviewcommissionedby theTomasRiveraCenter,JayGreene 1997) care-fullyreexamined heRossell and Baker(1996) review.WhereasRossellandBakeruseda vote-counting method n whichthey simplycounted henumbersof stud-ies that avoredbilingual, mmersion,orotherstrategies,Greene 1997) carriedouta meta-analysis in which each study produced one or more effect sizes, theproportion f a standard eviationseparating ilingualandEnglish-onlyprograms.Greenecategorizedonly 11of the 72 studiescitedby RossellandBakeras method-ologically adequate,andamongthose he calculatedaneffect size of +0.21 favor-ing bilingualoverEnglish-onlyapproacheson Englishreadingmeasures.Amongfive studiesusing randomassignment,Greenecalculatedan effect size of +0.41on Englishreadingmeasures.As partof thisreview,we attempted o obtainevery studyreviewedby RossellandBaker(1996) andby Willig (1985), as well as additionalstudies, and inde-pendently reviewed each one against the set of standardsoutlined previously.Consistent with Greene, we found that the Rossell and Baker (1996) reviewacceptedmanystudiesthat lackedadequatemethodology.The Appendixlists allof the reading studies cited by Rossell and Baker according to categories ofmethodologicaladequacyoutlinedin this article,which closely follow Greene'scategorization.As is apparent rom the Appendix,only a few of the studiesmetthe most minimal of methodologicalstandards,and most violated the inclusioncriteriaestablishedby Rossell and Baker(1996) themselves.We found,however,thatmost of the 16 studies cited by Willig also did not meet these minimal stan-dards.These are also notedin the Appendix.In itself, this does not meanthat the

    256

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    12/39

    Research nLanguagef ReadingnstructionorELLsoverall conclusions of eitherreview areincorrect,but it does meanthat the ques-tionof effects of languageof instructiononreadingachievementmust be exploredwith a different set of studies than the ones synthesized by either Rossell andBaker or Willig. The Rossell andBakerandWillig studies canbe categorizedasfollows (following Greene,1997):

    1. Methodologicallyadequatestudies of elementaryreading.These studiescomparedEnglish anguage earnerswhoweretaught o read hroughbilin-gualstrategieswiththose whoweretaught hroughEnglish-onlystrategies,with randomassignmentorwell-documentedmatchingonpretestsor otherimportantvariables. All of these studies focused on Spanish-dominantstudents.2. Methodologicallyadequatestudiesof heritagelanguage programs.Twostudies,one involvingChoctaw nMississippiandone involvingFrench nLouisiana,evaluatedbilingualprogramswith childrenwhogenerallyspokeEnglish but were expected to benefit from introductionof their culturallanguage.3. Methodologicallyadequatestudies of secondary programs. We put twosecondaryschool studies (Covey, 1973; Kaufman,1968) in this separatecategory.4. Canadian studiesof French immersion.Severalstudies (e.g., Lambert&Tucker, 1972; Genesee & Lambert,1983) evaluatedFrenchimmersionprograms n Canada.However, because they comparedimmersionwithmonolingual English instruction or with brief French-as-a-second-languageclasses,these studieswerenot evaluationsof bilingualeducation.5. Studies nwhich thetarget anguagewasnot the societal language.In addi-tion to Canadianstudies of French immersionin non-Francophone reas(e.g., Day & Shapson, 1988), there was a studyby Ramos,Aguilar,andSibayan 1967) thatexaminedvariousstrategies orteachingEnglish nthePhilippines.6. Studiesof outcomes other than reading.A few studies (e.g., Lum, 1971;Legarreta,1979;Pena-Hughes& Solis, 1980) assessedonly orallanguageproficiency,notreading.7. Studies in which bilingual treatments nvolvedlittle use of native lan-guage reading instruction. A few studies (e.g., EducationalOperationsConcepts,199la, 1991b)evaluatedprograms hatmaybe calledbilingualbut that n factmadeonly incidentaluse of the nativelanguageanddid notuse nativelanguagereading exts.8. Studies n whichpretestingtookplace aftertreatmentswere underway.Asnotedearlier,manystudies(e.g., Danoff et al., 1978;Rosier&Holm, 1980;Rossell, 1990;Thomas& Collier,2002;Valladolid,1991)comparedgainsmade in bilingualandimmersionprogramsafterthe programswere wellunder way. Both the Willig review and the Rossell and Baker reviewincluded such studies, and Greene (1997) accepted some of them asmethodologicallyadequate ;but we would arguethatthey addlittle tounderstandingheeffects of bilingualeducation.9. Redundantstudies. Rossell and Baker included many studies that wereredundantwithotherstudies n theirreview. Forexample,one longitudinal257

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    13/39

    Slavin&Cheungstudy(El Paso, 1987, 1990, 1992) issued threereportson the sameexper-iment,but it was countedas threeseparate tudies.Curiel's 1979 disserta-tionwas published n 1980, and both versionswere counted.10. Studies with no evidenceof initial equality.Several studieseither lackeddataon initialachievement,beforetreatments egan,orpresenteddata ndi-cating pretestdifferences in excess of one standarddeviation.11. Studies with no appropriate comparison group. Many of the studiesincluded by Rossell and Baker had no control group. For example,Burkheimer,Conger,Dunteman,Elliott,andMowbray 1989) andGersten(1985) used statisticalmethods to estimate where children should havebeenperformingand thencompared hatestimatewiththeiractualperfor-mance.Rossell andBakerset standards equiring acomparisongroupofLEPstudentsof the sameethnicityand similar anguagebackground, etthey included many studies that did not have such comparisongroups.Furthermore,manystudies ncludedby Rosselland Baker ackedanyinfor-mation aboutthe initialcomparabilityof childrenwho experiencedbilin-gual or English-only instruction(e.g., Matthews, 1979). These includestudies thatretroactivelycomparedsecondarystudentswho had partici-pated in bilingual or English-only programsin elementaryschools butfailed to obtain measures of early academic ability or performance(e.g., Powers, 1978;Curielet al., 1980).Otherstudiescomparedobviouslynoncomparablegroups. For example, Rossell (1990) compared 1-yeargains of English languagelearners n Berkeley, California,who were inSpanishbilingualor English immersionprograms;yet 48% of the ELLs,all in theEnglishimmersionprograms,wereAsian,whereasall students ntheSpanishbilingualprogram32%of thesample)were,of course,Latino.Similarly,Legarreta1979) comparedSpanish-dominanthildren n bilin-gualinstructionwithmainlyEnglish-dominanthildren aught n English.Finally,CarlisleandBeeman(2000) comparedSpanish-dominanthildrentaught 80% in Spanish and 20% in English with those taught 80% inEnglish and 20% in Spanish, so there was no English-only comparisongroup.12. Brief studies. A few studies cited by Rossell and Baker involved treat-mentdurations ess than1year.Forthe reasonsdiscussedearlier,studiesof bilingualeducationlastingonly 10 weeks (Layden, 1972) or 4 months(Balasubramonian,Seelye, & de Weffer, 1973) are clearly not relevant.All but one of these brief studies failed to meet inclusion standardsonothercriteriaas well (e.g., they lackedpretestsor had outcomesotherthanreading).

    ThePresent ReviewThisreview carriesoutabest-evidencesynthesisof studiescomparingbilingualandEnglish approaches o readingin the elementaryand secondarygradesthatmeet theinclusioncriteriaoutlinedabove. These includethemethodologicallyade-quate studies cited in the Willig (1985), Rossell and Baker(1996), and Greene(1997) reviews,as well as otherstudieslocatedin an exhaustive searchof the lit-erature,as describedpreviously.The characteristics ndfindingsof these studiesaresummarizedn Table 1 (p. 260).258

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    14/39

    Studies of Beginning Reading for Spanish-Dominant StudentsThelargestnumberof studiesfocusedonteachingreading o Spanish-dominantstudents n theearlyelementarygrades.Thirteen tudiesof thiskindmettheinclu-sion criteria.Threecategoriesof bilingualprogramsweredistinguished.The most commonamongthequalifyingstudieswere studiesof pairedbilingualstrategies, n whichstudentsweretaught oread nEnglishand n Spanishatdifferent imes of theday,beginning in kindergartenor firstgrade and continuingthroughthe end of thestudy.Pairingmay nothave begunon the firstday of the school year,but if chil-drenwerebeing taughtto read in bothSpanishandEnglishduring heir firstyearof reading nstruction, heprogramwas considereda pairedmodel.A secondcat-egory involvedevaluationsof programsn whichchildrenweretaughtreading nSpanishfor 1 yearbeforea transition o pairedbilingualinstruction Englishand

    Spanish).A thirdcategoryconsisted of a single study by Saldateet al. (1985),which did not describe the treatments well enough to permit categorization,although t seemed to evaluatea transitionalmodel.In Table 1, the elementarystudies of Spanish-dominantchildren are listedaccording o thesetreatment ategories,with thehighest-quality tudies isted first.That s, randomizedmultiyear tudies are istedfirst, hen matchedmultiyear tud-ies, thenmatched1-yearstudies. The studieswill be discussedin the same order.Studiesof PairedBilingualPrograms

    TenqualifyingstudiescomparedpairedbilingualandEnglishimmersionpro-grams.Plante(1976) randomlyassigned Spanish-dominant, uertoRican childrenin a New Haven,Connecticut,elementaryschoolwitha pairedbilingualmodelorwithEnglish-only nstruction. nitially,72 childrenwererandomlyassigned,45 tothepairedbilingualgroupand 27 to anEnglish-onlycontrolgroup.By the endofthe study,31 childrenremained n the pairedbilingualgroupand 22 remained nthecontrolgroup.Thechildrenbeganin kindergartenrfirstgrade.The treatmentinvolveda team-teachingarrangementwith one nativeSpanish-speakingeacherand one English-speaking eacher.It was describedas follows:

    [Spanish-dominanthildren]aretaught heirbasicskills,i.e., reading,writ-ing,arithmetic,ocialstudies,andscience, n Spanish.At thesametime,theEnglish-speaking ngloteachernitiates heteachingof English,beginningwithan aural-oralpproach.WhenanEnglishoralvocabularys sufficientlydevelopedn individual hildren, heinitiatesnstructionn thereadingandwriting fEnglish.Thekeypremisenthis nstructionalrganizations a con-cept of diagnostic-prescriptivenstructionwithboth SpanishandEnglishresourcesbeingavailable.Plante,1976,p. 40)

    Analysesof pretestscores for the finalsamplefoundthat the two groupsweresimilar on measuresof SpanishandEnglish oral vocabulary.The controlgroupwas nonsignificantlyhigheron bothmeasures.Two yearslater,allchildrenweregiventheEnglishform of the Inter-AmericanTest of ReadingandtheEnglishMetropolitanAchievementTest. Secondgradersin the pairedbilingualtreatment coredsignificantlyhigherthan control secondgraders effect size (ES) = +0.78). Theeffect size for thirdgraderswas also pos-itive but nonsignificant ES = +0.26). Totaleffects were nonsignificant,with a259

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    15/39

    N TABLE 1clL Languageof readinginstruction:Descriptive informationand effectsizesfor qualifyingstudiesIntervention Design, N, grade Sample EvidStudy description duration levels characteristics initial

    Studiesofpaired bilingualeducationPaired Random

    bilingual assignment,2 years

    Paired Randombilingual assignment,2 & 3 years

    Paired Matchedbilingual control,2 years

    Paired Matchedbilingual control,2 years

    55 students,Grades1-2, 2-3

    160 students,Grades1-2, 1-3

    171 students,GradesK-l

    153 studentsin fourschools,GradesK-l

    Spanish-dominantPuertoRicanstudents n NewHaven,CT

    DisadvantagedPuertoRicanstudents nPerthAmboy,NJ

    Spanish-dominantstudents nCorpusChristi,TX

    Spanish-dominantLEP students

    Well mon Soralvbut EEngli

    Well mon Iandiachie

    MatcheEngliSpanprete

    Very wmatcSESback

    Plante(1976)

    Huzar(1973)

    Campeauetal.(1975),CorpusChristiRamirezetal.(1991)

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    16/39

    Campeau Pairedet al. bilingual(1975),HoustonJ. R. PairedMaldo- bilingualnado(1977)

    Matchedcontrol,3 years

    Matchedcontrol,5 years

    Alvarez Paired Matched(1975) bilingual control,2 years

    Cohen(1975)

    Paired Matchedbilingual control,2-3 years

    Campeau Pairedet al. bilingual(1975),Kingsville,, TX

    Matchedcontrol,1 year

    206 students, Spanish-dominant MatcheGrades students n langK-2 Houston,TX andaachie126 students, Spanish-dominant MatcheGrades students n six and n1-5 elementary yearschools in schoCorpusChristi,TX147 students, Spanish-dominant MatcheGrade2 children n two andischools in langAustin,TX profi

    90 students, Spanish-dominant MatcheGrades students n and iK-I, 1-2, RedwoodCity, lang1-3 CA profi89 students, Spanish-dominant MatcheGradeK students n ande

    Kingsville,TX

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    17/39

    TABLE 1 (Continued)Languageof readinginstruction:Descriptiveinformationand effectsizesfor qualifyingstudiesIntervention Design, N, grade Sample EvidStudy description duration levels characteristics initial

    Studiesofpaired bilingualeducationCampeau Paired Matched 77 students, Hispanicstudents Similaret al. bilingual control, GradeI in SantaFe, prete(1975), 1year NM PBESantaFeStudiesof I-year transitionalbilingualeducation

    J. A. Bilingual, Random 20 students, Spanish-dominant Well mMaldo- 1-year assignment, Grades special disabnado transition 3 years 2-4, 3-5 education lang(1994) students n profiHouston,TX andfbackCampeau Bilingual, Matched 125students, Spanish-dominant Similaret al. 1-year control, Grades students n Engli(1975), transition 2 years K-l Alice, TX butTAlice, TX Span

    Studyof bilingualeducation(unspecified)Saldate Unspecified Matched 38 students, Spanish-dominant Well met al. control, Grades students n prete

    (1985) 3 years 1-3 Douglas,AZStudiesof heritage languages

    Paired Matchedbilingual control,1year

    193 students, French-dominantGrade 1 students nLafayette,LA,

    Well minitiaabilit

    Morgan(1971)

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    18/39

    CatholicSchools

    Paired Matchedbilingual control,1year

    63 students, Choctaw studentsGrade2 in MS

    MR

    Well minitiaprofi

    SecondarystudiesCovey(1973)

    Paired Randombilingual assignment,

    1yearKaufman Paired Random

    (1968) bilingual assignment,1 and2 years

    200 students, Spanish-dominantGrade9 students nPhoenix,AZ

    139 students, Spanish-dominantGrade7 students nNew YorkCity

    Well mprete

    Initialvocacompscorelangnon-lIQ,aHoffbilinschewerecova

    t) Note. PBE = Paired Bilingual Education; TBE = Transitional Bilingual Education; El = English Imma Basic Skills;MAT = MetropolitanAchievementTest;SRAAS = Science ResearchAssociatesAssa Effect size estimated;data forexactcomputationwere not available.

    Doebler&Mardis(1980)

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    19/39

    Slavin& Cheungmeanweightedeffect size of +0.50. However,these differencesdid not controlfor the controlgroup's pretestadvantage,so pretest-adjusteddifferenceswouldhave further avoredthe experimentalgroup.On theMetropolitanAchievementTest,totalreadingscoresfavoredtheexper-imental group by 0.4 grade equivalents among second gradersand 0.5 gradeequivalentsamong thirdgraders.No standarddeviationsor analyses were pro-vided, however.Finally,therewere substantialdifferences n retentionrates.Onlyone of the 31experimental hildren(3%)was retained n grade,as comparedwith 13 of the 22controlchildren 59%).Retentionratesaredeterminedby theteachers nvolvedona subjectivebasis,but this is neverthelessanimportantinding.Not surprisingly, hildren n thebilingualgroupscoredsubstantiallybetteron aSpanishreading estthandid theEnglish-onlycontrolgroup overallES = + 1.02).The Plante(1976) study is small, and with only one class in each treatment,teachereffects werecompletelyconfoundedwith treatment ffects. Yet its use ofrandomassignmentanda 2-yearlongitudinaldesign with modestattritionmakesthis studyanimportantpartof the researchbaseon bilingualeducation.Huzar (1973) carriedout a randomizedexperimentinvolving 160 Spanish-dominantPuertoRican childrenin PerthAmboy, New Jersey.On entryto firstgrade, hechildrenwereassignedto one of two treatments: airedbilingualor con-trol. The pairedbilingual treatmentwas described as follows: One bilingualteachergave reading nstruction o the class in Spanishfor 45 minutes each day,while the monolingualteachergave readinginstruction n English for the sameperiodof timeeachday (Huzar,1973, p. 34).Inthecontrolgroup,studentsweretaughtonly in Englishfor 45 minutesa day.Allteachingprocedures,qualityof materials,andtimeperiods orreading nstruc-tion were the same, with the exception that the experimentalclasses receivedinstruction n bothSpanishandEnglish,withcorrespondingextbooks.Therearetwopotentialconfounds n thisstudy.First, t is unclearwhatthe con-trolgroupwas doing while the experimentalgroupreceived45 minutes of dailySpanishreading nstruction. tmaybe thatthe experimental tudentswere receiv-ing more totaltimein reading EnglishplusSpanish).Second,theEnglish readingtexts used inthe two programswere different.Theexperimental lasses usedapho-netic program, he MiamiLinguisticReaders.Control studentsused ScottFores-man'sOpenHighwaysseries.The 160 study subjectswereassignedatrandom o fourexperimentalandfourcontrolclasses. Two classes of eachtreatmentwere at thesecond-and third-gradelevels, respectively.Thirdgradershad been in theirrespectivetreatments or 3years,andsecondgradershadbeen in theirsfor 2 years.MetropolitanReadinessTestfirst-gradecores werecollected fromschoolrecords;Lorge-ThordikeIntel-ligence Test scores were obtained orthirdgradersand showedno significantdif-ferencesbetween treatment roupsat eithergrade evel.Theposttestwas theEnglish reading est of theInter-American eries.For sec-ond graders, here were no differences (ES = +0.01). Differences at the third-gradelevel directionally avoredthe experimentalgroup(ES = +0.31) but werestatisticallysignificant orboys (ES = +0.44) but notgirls (ES = -0.06).The Huzar 1973) andPlante(1976) studiesareparticularlymportant,despitetakingplacemorethana quarter enturyago.Botharemultiyear xperiments hat,264

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    20/39

    Researchon Language of ReadingInstructionforELLsbecause of use of randomassignment,canruleout selectionbias as an alternativeexplanationfor the findings.Both startedwith childrenin the early elementarygradesand followed them for 2 to 3 years. Both used pairedbilingual readinginstructionby differentteachers in Spanishand English, with transitionto all-Englishinstructionby second orthirdgrade.Theuse of bothSpanishandEnglishreading nstruction achdaymoreresemblestheexperienceof Spanish-dominantstudentsin many two-way bilingual programs(see Calder6n & Minaya-Rowe,2003) than t does transitional ilingualmodels,whichdelayEnglish readinguntilsecondor thirdgrade.Oneof the mostwidelycitedstudiesof bilingualeducation s alongitudinal tudyby Ramirezet al. (1991), which comparedSpanish-dominanttudents n Englishimmersionschoolswith those in two forms of bilingualeducation: arlyexit (tran-sitionto English n Grades2-4) and lateexit (transitiono Englishin Grades5-6).Schools in severaldistrictswerefollowed over 4 years.Immersionandearly-exitstudents n four two-treatment chools were well matched,but a groupof onetreatment chools that mplementedonly bilingualor immersion reatmentswerepoorlymatched,according o the authors.Late-exitstudentswere lower than theircomparisongroupsin socioeconomic status(SES), and their schools had muchlowerproportions f nativeEnglishspeakers.Forthesereasons,no directcompar-isons weremadeby the authorsbetweenlate-exitand other schools.The two-treatment omparisonof early-exittransitionalbilingualeducationandEnglishimmersion s the important ontributionof the Ramirezet al. (1991)study.In the immersionprogram,almost all teacherspeech was in Englishat allgradelevels. In the early-exitclasses, teacherspeechwas 31%Spanish n kinder-garten,29%in firstgrade,24% in second grade, 17% in thirdgrade,and 2%infourthgrade.Theearly-exitprogramwas describedas follows:

    In an early-exitprogram hereis some instructionn the child's primarylanguage, 0to60 minutesperday.This s usually imited otheintroductionof initial eading kills.Allothernstructions inEnglish,withthechild'spri-mary anguageusedonlyasa support,or clarification.However,nstructionintheprimaryanguages quicklyphasedout overthe nexttwoyearsso thatbygrade wo,virtually llinstructions inEnglish. Ramirez tal., 1991,p.2)Althoughthe Ramirezet al. studyis invariablycited as a studyof transitionalbilingualeducation,it is in fact a studyof paired bilingualeducation,using thedefinitionappliedin the presentreview. The authorsnoted that in the early-exitkindergartens, 5.1% of instructionalimewas spenton Englishlanguageartsand29.9% on Spanishlanguagearts. In firstgradethe correspondingnumberswere33.5% (English) and 24.2% (Spanish). In contrast, in the English immersionclasses, 63.6%of instructionalime was spenton Englishlanguagearts n kinder-garten,and 60.2%in firstgrade.Percentagesf timespentonmath, ocialstudies,andscience,allinEnglish nbothtreatments, erenearlydenticalnthetwo treatmentonditions.Overallnstructionaltimewas equivalent n the two conditions,so the immersionandearly-exitclasseswerespendingabout hesame amountof timeonlanguagearts,asEnglishandSpan-ish time ntheearly-exit rogramwassimilar oEnglish ime n the mmersion lasses.Thelongitudinal xperimenthad two stages.Childrenwerepretested nkinder-gartenand thenposttestedatthe end of firstgradeon theEnglish Comprehensive

    265

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    21/39

    Slavin & CheungTest of Basic Skills (CTBS). The experimentaland control students were wellmatchedon pretests,SES, preschoolexperience,andother factors. On first-gradeCTBSreading,students n thebilingualgroupscoredsignificantlyhigherthantheimmersionstudents ES = +0.53).

    The second phaseof the studyfollowed studentsfromspringof firstgradetothirdgrade.It is not clear why the kindergarten ohortwas not simply followed,althoughtherewere seriousproblemswith attritionover the longitudinalstudy.However,the second-phasestudydoes notqualifyfor inclusionin this review,asit uses a pretestgiven long aftertreatmentshadbegun.TheRamirezet al. studywas so importantn its timethat heNationalResearchCouncil convened a panel in 1991 to review it and a study by Burkheimer t al.(1989). The panel's report(Meyer& Fienberg,1992) supported he conclusionsof the Ramirezet al. comparisonof the early-exit and immersionprograms nGradesK-l. It expressedconcernabout the second phaseon the same criterionas thatappliedin the presentreview, that the pretestsfor the second phaseweregiven after thestudentshadalreadybeen in bilingualor immersion reatments or2 years.In the mid-1970s,theAmericanInstitutesof Research AIR)produceda seriesof reportsonbilingualprogramsaround he U.S. (Campeau,Roberts,Oscar,Bow-ers, Austin,& Roberts,1975).These are of some interest,with one majorcaveat:The AIRresearcherswerelookingforexemplarybilingualprograms.Theybeganwith 96 candidatesandultimatelywinnowedthe list down to 8. Programswereexcludedif datawereunavailable,notbecausetheyfailedto show positiveeffectsof bilingualprograms.Nevertheless,these sites were chosen on theirreputationsforexcellence,and a sitewouldclearlybe less likelyto submitdata f thedatawerenotsupportiveof bilingualeducation.Also, theCampeauet al. (1975) evaluationswere organizedas successive 1-yearstudies,meaningthatpretestsafterthe firsttreatmentyear (K or 1) are of little value. For reasons described earlier, weincludedonlycohorts hatwerepretestedbeforetreatments egan.Withthese cau-tions in mind,theCampeauet al. (1975) studiesare describedbelow.A studyinCorpusChristi,Texas,evaluateda pairedbilingualprogram n threeschools.Inkindergarten,hepairedbilingual reatmentnvolveda 2-hour anguageperiodalternatingEnglishandSpanish. Inbothlanguages, hepatterned racticesstressbasicsentencepatterns nd llustrate hanges n wordforms andwordorder(Campeau t al., 1975,p. D-59). Over the course of kindergartenheproportion fSpanish nstructiondecreased rom90%to50%.Infirstgrade, hepairedbilingualclasses had a daily 2-hourEnglishreadingandlanguageartsperiodanda 1-hourSpanishreadingandlanguageartsperiod:Phoneticanalysis killsare aught irst nSpanishbecauseof thehighlypho-neticnature f the anguage.... English eadiness kills... are aughthroughtheHarcourt raceJovanovicheries.Phonetic nalysis kills are ntroducedafter hechildhas earned hem houghhisSpanish eadingessons.... Thereis a heavyphonicsemphasis,withcomprehensionkills receivingheavieremphasisasdecoding killsdevelop. Campeaut al., 1975,p. D-60)Classes in the controlgroupswere taughtmonolinguallywithoutregard or thelanguagedominanceof thechildren p. D-66), butotherwise nstructional trate-gies were similar.

    266

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    22/39

    Research on Language of ReadingInstructionor ELLsExperimental ndcontrolclasses werepre-andposttestedeachyearfor2 years.For reasonsdiscussedearlier,we did not include comparisons n which pretestswere given after treatmentswere alreadyunderway. For this reason, we onlyfocusedon the childrenwho werepretested n kindergartenn the firststudyyearandthenposttestedat the end of firstgradein the secondstudy year.This cohortwas well matchedon theStanfordEarlySchoolAchievementTest.Therewerenodifferences at the end of kindergarten,but the paired bilingual studentsscoredsignificantlyhigher thancontrols on the Inter-AmericanEnglish Reading Test(ES = +0.45) at the end of firstgrade.Therewere also substantialdifferencesonthe StanfordReadingAssessment.Experimental tudentsaverageda gradeequiv-alentof 2.3, controls 1.8. However,no statisticalcomparisonsweremade on theStanford.The paired bilingual group also scored substantiallybetter than thecontrolgroup n Spanishreading,of course.A study in Houston also reportedby Campeauet al. (1975) followed threecohortsof students n seven pairedbilingualand two Englishimmersionschools.The pairedbilingualprogrambegan with a Spanishdecodingprogramand thenSpanishLaidlawbasals,duringa regularSpanisharts block. Studentswere thentaughtEnglishreading,startingwith a transitionalprogramandthencontinuingwiththe HarcourtBraceJovanovichbasal series. Each class hada teacherandanaide. Two controlschools, in which Spanish-dominanttudentsweretaughtonlyin English,wereselected on the basis of similarity o seven experimental choolsin language,SES, andpriorachievement.Experimental ndcontrolkindergartentudentswerewell matchedonthe Inter-AmericanEnglish Ability Test. At the end of firstgradeand second grade,thepairedbilingualstudents coredsubstantially igher han hecontrolstudents,witheffect sizes of +0.69 and +0.54, respectively.Cohen (1975) comparedtwo schools serving many Mexican AmericansinRedwoodCity,California.Oneschool wasusingwhat amounts oatwo-waybilin-gual program, n thatSpanish-dominanttudentsandEnglish-dominant tudentswere taughtin both SpanishandEnglish. However,from the perspectiveof theELLs,the treatmentwas the same as a pairedbilingualmodel. Spanish-dominantstudentsweretaughtSpanishreadingusingreaders uch asPepinen PrimerGradoandEnglishreadingusingthephoneticMiamiLinguisticSeries.Spanish-dominantandEnglish-dominanttudentsweregrouped ogether orcontent area nstructionbutnot for reading.Threesuccessive cohorts were comparedat the two schools:GradesK-l, 1-2, and 1-3. Ineachcase, studentswerepretestedandposttestedona broadrange of English readingmeasures. In all cohorts, Mexican-Americanstudentswere well matchedon English and Spanish pretests.At posttest, therewere no significantdifferences,adjusting or pretests.The data did not allow forcomputationof effect sizes, so zeros were entered n Table 1.Maldonado 1977) conducteda 5-yearlongitudinalstudyon a groupof Mexi-can Americanchildrenin six elementaryschools in CorpusChristi,Texas. Themainpurposeof this studywas to examine how well the bilingualstudentswereable to succeed in the regulareducationprogramof the school districtaftertheyhad left the bilingualprogram.The experimentalgroupconsisted of 47 childrenwho hadparticipatedn the bilingual program or 4 consecutiveyears,from firstto fourthgrades.The controlgroupwascomprisedof 79 childrenwho hadbeen ina regularEnglish-only programfor the same 4 years and grades. The study

    267

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    23/39

    Slavin& Cheungfollowed thetwo groupsuntiltheyreached ifthgrade,1yearaftertheexperimen-tal groupleft the bilingualprogram.Students n the experimentalgroupenrolledin the Title VIIprogram itled ApprendemosEn Dos Idioma We Learn n TwoLanguages),in which they received a minimumof 2 hoursper day in Spanishlanguageinstruction n the areas of languagearts,reading,and mathematicsandsocial studies (p. 103). No specific descriptive informationabout the controlgroupwas provided n the study.No statistically ignificantdifferenceswerefound atanygradeaftercontrollingfor first-gradepretests.It is important o note thatteachersin both the bilingualgroupand the controlgroupwere bilingual.However,it is not statedhow muchthese bilingualteachers in the controlgroupused Spanishin their classroomstohelpchildrenwho were inneed forbilingualexplanations.As the researchertated,It s highlypossiblethatthecontrolgroupbilingualteachersmighthaveused theSpanishlanguagefor clarificationof some concepts.This in turnwould not onlyassist those students n the comprehensionof those conceptsbut at the same timelowerthedifferencebetween the groups n theareasof mathematicsandreading(p. 104).A studyby Alvarez (1975) followed 147 Mexican-Americanchildrenin twoschools in Austin, Texas, fromfirst to second grades.Studentsin the bilingualclassrooms and the control classrooms in each school were well matched onSES and initial languageproficiency. The instructionprogram n the bilingualclassrooms was describedas a balanced combinationof Spanish and English(50 percent/50 percent) (p. 73). However, children who had very limitedEnglishproficiency in these bilingualclassroomswere initially taughtin Span-ish. Reading and language arts were taught in both languages each day for2 hoursby the English-speakingandSpanish-speakingteachers.Orallanguagein English was taughtfor one-half hourdaily to the Spanish-dominant hildren,and oral language in Spanish was taught for one-half hour to the English-dominantchildren.Students n thecontrol classroomshad a curriculum imilar o thatof thebilin-gual classrooms. The only difference was thatthe subject matterwas taughtcompletelyin English;usingthe sametextbooksfor theEnglishcomponentof thebilingualclasses, but using twice as much time as the balancedcombinationofSpanishandEnglishcurriculumdesignedfor Mexicanstudents n bilingualclass-rooms (Alvarez,1975, p. 75). Thebilingualstudentsscoredsomewhathigheronthe English readingvocabularytest, but the controlgroupscoredhigheron theEnglish reading comprehensiontest than the bilingual classes. None of thesedifferenceswerestatisticallysignificant.Two of the studiescarriedout by Campeauet al. (1975) had 1-yeardurations.The firststudywas conducted n two low-SES elementaryschools in Kingsville,Texas. Five gradeswerecomparedat the two schools:K-4. Onlythe resultsfromthekindergarten roups-48 in the bilingual groupand41 in the controlgroup-areinterpretable ecause thepretestsfor othergradeswereadministered fterthetreatmentsbegan.All teachers in kindergartenwere bilingual. Instructiontime was equallydivided between Spanish(50%) and English (50%). Kindergartenerswith verylimitedEnglish were taughtprimarily n Spanishuntil theirEnglishproficiencyreacheda pointwheretheycouldcope withbilingualinstruction.268

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    24/39

    Researchon Language of ReadingInstructionor ELLsAt the endof the study,students n bothgroupsweregiven the sameposttests.The mean gains between the two groups were compared. Students in all sixkindergartenclasses gained significantly more on the English version of theInter-AmericanSeriesthantheircounterpartsn the controlgroup,with an effectsize of +0.42.Another1-yearstudy n SantaFe, New Mexico, comparedpairedbilingualandimmersion programsfor Spanish-dominantstudents. Pre- and posttests werereported or eachyear,butonly firstgradewas interpretable, s pretestsfor otheryears had alreadybeen affected by the treatments.Parents chose to place theirchildren in bilingual or English programs,and apparently parents of higher-achievingchildrenchose thebilingualgroup,as pretestscoreswerehigherin thatgroup.Students n the bilingualclassrooms received abilingual presentationofall the topicsof studyin the normalcurriculum p. D-16). Forexample,studentsweretaughtcertainconcepts n Spanish nthemorningand wereretaughthesamelesson inEnglishin theafternoon. naddition,students n thebilingualclassroomswere groupedby ability duringlanguagearts and reading periods. No specificdescription nformationwas providedabout the controlclassrooms.At the end ofthe study, the bilingual group gained slightly more than the control group inEnglish reading, with an effect size of +0.03. However, the bilingual groupalso gained more in English readingthan the English-only group. No standarddeviationsweregiven, so effect sizes forpretestdifferencesandgainscould not becomputed.

    Studiesof 1-YearTransitionalBilingualEducationJ. A. Maldonado(1994) carried out a small, randomizedstudy involvingEnglish language learnerswho were in special education classes in Houston.Twenty students in second and third grades with learning disabilities wererandomlyassignedto one of two groups.A bilingual groupwas taughtmostly inSpanish ora year,witha45-minuteESLperiod.Duringa secondyear,half of theinstructionwas in English,half in Spanish.Ina thirdyear,instructionwas only inEnglish. The controlgroupwas taughtin English all 3 years. Studentsin bothgroupsreceived the same amount of readingandlanguageinstructionand weretaughtby similarteachers.Childrenwerepretestedon theCTBS and thenposttestedon the CTBS 3 yearslater.Atpretest, hecontrolgroupscorednonsignificantlyhigher han hebilingualgroup,but atposttestthe bilingualgroupscored farhigher.Using the means andstandarddeviationspresented n the article,the effect size would be +8.33, butusingthe given valuesof t andn, the effect size is + 1.67, a more credibleresult.One of the Campeauet al. (1975) studies,in Alice, Texas, comparedSpanish-dominant tudents n bilingualandEnglish mmersionprograms tartingn kinder-garten, ora 2-yearexperiment.The treatment nvolvedteachingkindergartnersnSpanish. n firstgrade,childrenwere transitionedo Englishreadingand were thentaught qualamountsof time neach anguage.Matched ontrol tudentsweretaught

    onlyinEnglish.Whilekindergartnerserecomparabletpretest nEnglishmeasuresof generalability,bilingualstudents coredsubstantially igheron a Spanishabilitytest.Atposttest controllingorpretests), ilingual tudents coredsubstantiallyetteron theInter-Americannglishreading est at the end of firstgrade,after2 yearsofbilingualeducationES = +0.49).269

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    25/39

  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    26/39

    Research on Language of ReadingInstructionforELLsKaufman 1968) evaluatedaprogramnwhichlow-achievingSpanish-speakingseventhgraderswererandomlyassignedto bilingualorEnglish-onlyconditions ntwo New Yorkjuniorhigh schools. One school participatedn the program or ayear and the otherfor 2 years.In the bilingualclasses, studentsreceived threeorfourperiodsof Spanishreading nstruction ach week, while controlswerein art,music, or health education. On standardized tests of reading, there werenonsignificant ifferences avoring hebilingualclassesin the2-yearschool andthe1-yearschool(ES = +0.23 forboth).Thesecondary tudiespointto thepossibilitythatprovidingnative languageinstruction o low-achievingELLs in secondaryschoolmayhelpthemwithEnglishreading.Thisapplications worthyof additionalresearch also see Klingner& Vaughn,2004).

    CanadianStudiesof French ImmersionThere are severalCanadian tudies(e.g., Lambert& Tucker,1972) thathaveplayedanimportant ole in debatesaboutbilingualeducation.These arestudiesofFrench mmersionprograms,n whichEnglish-speakinghildrenaretaughtentirelyorprimarilynFrench ntheearlyelementary ears.RossellandBaker 1996)empha-sized these studiesas examplesof structured nglishimmersion, he approachfavored n theirreview.However,Willig(1985)andotherreviewershave excludedthem.TheCanadian tudiesdonot meettheinclusion tandardsf thisreviewbecausetheAnglophone hildrenwere earning usefulsecond anguage,notthelanguageorwhichtheywould be heldaccountablentheir aterschooling.Althoughmanyof thestudies ookplaceinMontr6al,he childrenivedinEnglish-speakingeighborhoodsandattended chools in an English system.Thefocusof thisreview is on bilingualeducationused to helpchildren ucceed n thelanguagen whichtheywill be taughtin the latergrades;but the French mmersionchildren n Canadawere headedtoEnglish secondary schools. Furthermore,hese studies all involved voluntaryprograms,n whichparentswanted heirchildreno lear French,andthechildrennthese studiesweregenerallyuppermiddleclass,notdisadvantaged.Because Frenchimmersionprogramswere voluntary,children who did notthrive in them could be androutinelywere returned o English-onlyinstruction.This means that the children who completed French immersion programsin Canada were self-selected relatively high achievers. Most important,the

    bilingual programs o which French mmersion s comparedwere nothinglikebilingualeducation n theUnited States.At most, childrenreceived 30 to 40 min-utes daily of Frenchas a second language,with far less time in Frenchreadinginstruction han a U.S. studentin a bilingualprogramwould receive in Englishduringandaftertransition see Genesee & Lambert,1983). Yet in manystudies,English comparisongroupswere not learningFrenchat all. In the widely citedstudyby LambertandTucker(1972), Anglophonesin French mmersionclasseswerecomparedwithAnglophones taughtonly in EnglishandwithFrancophonestaughtonly in French.Ironically,studiesof this kind,cited by Rossell andBaker(1996) as comparisonsof immersionandbilingualeducation,are in factcompar-isons of immersionandmonolingualeducation.If they existed, Canadian tudiesof, say, Spanish speakers earningFrenchin Francophone chools in Quebec orEnglish in Anglophone schools in the rest of Canadawould be relevant to thisreview. But studies of voluntaryimmersionprogramsas a means to acquiringFrenchas a secondlanguageareonly tangentiallyrelevant.271

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    27/39

    Slavin& CheungWhilethe Canadianmmersionstudies are notdirectlyrelevant o thequestionof theeffectiveness of bilingualprogramsorELLslearning hesocietallanguage,they arenevertheless nteresting n gaininga broaderunderstanding f theroleofnative languagein foreign languageinstruction.As a group,these are matchedstudies of highmethodologicalquality.Quitein contrast o U.S. studies,however,the focus of the Canadianstudies is on whether French immersionharms the

    Englishlanguagedevelopmentof nativeEnglish speakers.It is taken as obviousthatFrenchalldaywill producemorefacilityinFrench han30 to 40 minutesdailyin second-languageclasses.Overall,the Canadianstudies painta consistentpicture(Lambert& Tucker,1972; Lambert & Tucker, 1977; Barik & Swain, 1975; Barik, Swain, &Nwanunobi, 1977;Genessee & Lambert,1985; Day & Shapson, 1988). At leastfortheoverwhelminglymiddle-classstudents nvolved,French mmersionhadnonegativeeffect on Englishreadingachievement,andit gave studentsfacility in asecondlanguage.The relevance to the U.S. situation s in suggestingthat similarsecond-language immersion programs, perhaps including two-way bilingualprograms or English-proficient hildren,are not likely to harmEnglishreadingdevelopment.However,the relevanceof these studies to anycontextin whichthechildren of immigrantsareexpectedto learn the languagethatthey will need tosucceed in theirschool and in the largersociety is minimal.Comparisonof PairedBilingualand TransitionalBilingual Programs

    As notedearlier,manyof the programswith the strongestpositive effects forEnglish anguage earnersusedapairedbilingualapproach,n which childrenweretaughtreading nbothEnglishand theirnative anguageat different imeseachdayfrom the beginningof theirschooling. This approachcontrastswith transitionalbilingual education (TBE) models, in which children are first taught to readprimarilyn theirnativelanguageandonly then transitioned radually o English-only instruction.Only one study has compared readingoutcomes of these twobilingualapproaches.A longitudinal tudy by GerstenandWoodward 1995) initiallyfavoredpairedbilingual nstruction verTBEbutlaterfound thetwo to be equivalent.Thisstudywascarriedout withSpanish-dominant LLsin 10elementary chools in El Paso.Five schools usedaprogramn which all subjectsweretaught nEnglishbutSpan-ish instructionwas also provided, or 90 minutesdailyin firstgradeanddecliningto 30 minutesdaily in fourthgrade.The transitionalbilingualprogram nvolvedmostly Spanish nstruction,with 1 hourperday forESL instruction,withgradualtransition o Englishcompleted n thefourthor fifthgrade.Thechildrenwerewellmatcheddemographicallyon entryto firstgradeand scorednear zero on a mea-sureof Englishlanguageproficiency.InGrades4, 5, 6, and7, IowaTests of BasicSkills were compared or the two groups.OnTotalReading,the pairedbilingualstudents coredsignificantlyhigher han hetransitional ilingualstudents nfourthgrade(ES = +0.31), butthe effects diminished n fifthgrade(ES = +0.18) andwereverysmall inthesixth(ES = +0.06) andseventhgrades ES = +0.08). Testsof languageandvocabularyshowedsimilarpatterns.Thispatterns probablydueto the factthatthetransitional ilingualstudentshadnotcompletedtheir ransitionto Englishin fourthandfifthgrades.Whentheyhad done so, by sixthgrade,theirreadingperformancewas nearlyidenticalto thatof the othergroup.272

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    28/39

    Researchon Language of ReadingInstructionor ELLsResearchcomparingalternativebilingualmodels is far from conclusive, butnothing suggeststhat t is harmful o children'sreadingperformanceo introducebothnativelanguageandEnglishreading nstructionat different imes eachday.

    DiscussionThemostimportant onclusion fromresearchcomparing he relativeeffects ofbilingualandimmersionprograms or English languagelearners s that there arefar too few high-qualitystudies of this question. Willig (1985) and Rossell andBaker(1996) agree on very little, but both of these reviews call for randomized,longitudinal valuations o producea satisfyinganswer o this criticalquestion.Ofcourse,manywouldargue hatrandomized valuationsare neededon mostimpor-tantquestionsof educationalpractice see, forexample,Mosteller&Boruch,2002;Slavin,2003);however,inbilingualeducation, heyareespeciallycrucialbecauseof the manyinherentproblemsof selection bias in this field. Furthermore,his isan areain which longitudinal,multiyearstudiesarevirtuallymandatory,o trackchildren nitially taught ntheirnative anguage hrough heir ransitionoEnglish.Finally,althoughrandomized,ongitudinal tudiesof thistopicaresorelyneeded,there are simply too few experimentalstudies of all kinds, includingones withmatchedexperimentalandcontrolgroups.Despitetheseconcerns,however,theexistingresearchon languageof instruc-tionyields someimportantessonsat leastworthyof further tudy.Across 17qual-ifying studies of all types of programs, 12 found effects favoring bilingualeducationand5 foundno differences. None of the studies found resultsfavoringEnglishimmersion.The largest group of studies focused on elementary reading instructionforSpanish-dominanttudents.Nine of 13 studiesin this categoryfavoredbilingualapproaches,and fourfound no differences.The medianeffect size for all 13 stud-ies was +0.45. This effect size is higher than the estimate of +0.21 given byGreene(1997), but Greenedid not locate the Campeauet al. (1975) studies thatadded severalpositive effect sizes. Also, manyof the largest positive effect sizeswerefrom studieswithverysmallsamplesizes. The mean effect size weightedbysamplesize for the 13 studiesof elementary eading orSpanish-dominanttudentswas +0.33. Using proceduresdescribedby Lipsey and Wilson (2001), this effectwas found to be significantlydifferent romzero,Q = 29.6, p < .05, df = 12.Theweightedmean for the threerandomized tudies was +0.62, Q = 8.53, p < .05.It was surprising hatmost of the methodologicallyadequatestudiesevaluatedformsof bilingualeducationquitedifferentfrom those commonlyused in recentyears.Thesearepairedbilingualprograms,n which childrenaretaught o read inEnglishand ntheirnative anguageatdifferent imes eachdayfromthebeginningof their time in school. Anothercategory of programsprovided just 1 year ofnative-language nstructionbefore the transition o English-onlyreading.Pairedbilingual strategieswere used in two of the randomized studies (Huzar, 1973;Plante, 1976), andin a studyof a 1-yeartransitionalprogram J. A. Maldonado,1994). These practicescontrastwithpractices n transitionalbilingualeducation,in whichchildrenaretypicallytaught o read n theirnativelanguagefromkinder-garten o Grades2, 3, or4, andthentransitioned o reading.There areseveralpossible explanations or the prevalenceof the pairedbilin-gualinterventions mongthestudiesreviewed.First,mostof thestudiestookplace

    273

    This content downloaded from 128.223.174.86 on Fri, 14 Feb 2014 21:26:11 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/12/2019 (2005) a Synthesis of Research on Language of Reading - Slavin & Cheung

    29/39

    Slavin& Cheungin the 1970s, whenTitle VII was new. At thattime,pairedbilingualmodelswerepopular.Second,forreasonsdiscussed earlier n thisreview,studiesof transitionalbilingualeducationarevery difficult to perform,as they shouldbegin in kinder-gartenandcontinuepastthepointof transition.A 4-year longitudinal tudywouldbe requiredto follow childrenfrom kindergarten o thirdgrade. Allowing forstudentmobility,such a studymust startwith a largesampleto end up withsuffi-cientnumbersof students.The U.S. Department f Educationhasrecentlyfundedtwo matchedandone randomizedongitudinal tudyto evaluatetransitionalbilin-gual education,but before these only the Ramirez et al. (1991) study had theresourcesto carryout an investigationof this kind,and it did not follow a consis-tentsamplefromkindergarteno thirdgrade.It is importanto note that most of the studiesthat did notqualifyfor inclusionalso usedpairedbilingualmodels,nottransitionalbilingualmodels.A key excep-tion was a series of studiesby Thomasand Collier (2002) that followed childrenwho hadbeen intransitional rogramsbut lackedpretestmeasures rom before thestartof the TBE interventions.Becauseof thedearthof studies of TBE, it is notcurrentlypossibleto say withconfidencewhetherpairedbilingualmodels are more effective thantransitionalmodels.Onlyone study,by Gerstenand Woodward 1995), madethiscomparison.It founddifferencesfavoringpairedbilingualstrategies n Grades4 and 5 but notin Grades6 and 7. However,given the support or pairedbilingualmethods seenin this review, it is worthwhileto speculateaboutwhy pairedmethodsmightbebeneficial.

    Teaching a Spanish-speakingEnglish language learner in Spanish can beexpectedto establishthe alphabeticprinciple,the idea thatwordsarecomposedof distinct sounds represented by letters (see National Reading Panel, 2000).Earlyin theirreadinginstruction,children learnto combine lettersand soundsinto wordsthey know. This process is verydifficult if childrenmustform lettersand sounds into words they don't