ApilotstudyofvalueaddedanalysisforBeijingseniorsecondaryschool用增值評(píng)價(jià)技術(shù)對(duì)學(xué)生成績(jī)進(jìn)行分析的初步研究摘要：用多水平分析技術(shù)對(duì)我國(guó)7所高中1999年參加高考的l051名學(xué)生的中招和高招成績(jī)及其他變量，建立7種模型進(jìn)行分析，結(jié)果發(fā)現(xiàn)在校正了學(xué)生先前的學(xué)習(xí)成績(jī)和有關(guān)其他變量的影響后，各學(xué)校對(duì)學(xué)生的數(shù)學(xué)、英語(yǔ)和總成績(jī)的影響(增值)明顯不同，基于多水平分析結(jié)果的比較比傳統(tǒng)的原始分比較更加公平、科學(xué)。對(duì)各變量之間的相互影響也進(jìn)行了分析，并與國(guó)際上的有關(guān)研究結(jié)果進(jìn)行了比較，本文也對(duì)如何把考試分?jǐn)?shù)用于學(xué)校管理進(jìn)行了初步的探討。關(guān)鍵詞：多水平分析；增值評(píng)價(jià)；成績(jī)；學(xué)校管理Abstract:DatafromsevenBeijingseniorsecondaryschoolswasanalyzedusingMultilevelanalysistechniqueforschooleffectivenessintermsofvalueaddedmeasureanditsimplicationforschooladministration．Similartrendsinourstudywerefoundwiththeinternationalstudies．Students’Priorattainmentcanexplainalargepartofoutcomevariance，themodelincludingstudent’spriorattainment，schoolcontextfactors，studentindividualvariablesisoptimalmodel．Schooleffectsdifferintermofsubject，studentgroupcategorizedbygenderandotherindividualfactors．Thisstronglyshowthemultifacetcharacteristicsofthevalueaddedmeasure，andsuggestamultiple，dynamicvalueaddedmeasureisessentialforschoolevaluation．Wefinddifferentschooleffectsintermofsubject，thismeansvalueaddedmeasureisausefultoolforschoolevaluation，butforimplicationinBeijingeducationalpractice，itneedfurtherresearch．Theresearchalsoprovidedanexampleabouthowtomakegooduseofexamscoreforeducationadministrationandothereducationalpractice．Keywords:Multilevelanalysistechnique;valueaddedmeasure;examscore;schooladministration

IntroductionTherehavebeenmanystudiesaboutschooleffectivenessmeasurementbasedonvalueaddedanalysisusingmultilevelstatisticstechnique(Thomasetal．，l994，1996，1997，1999，Sammonsl997，1999，Yang1999，Goldsteinl996，1997)．Generally,thesestudytopicscoverthedefinitionofschooleffectivenessandimprovement，thevalue—addedmethodandtheapplicationofmultilevelanalysis，andsoon，mainlyontheanalysisofstudentacademicattainment．Hereisabriefreviewofthesestudies．SchooleffectivenessandvaluedaddedmethodAsresponsetoColeman’sreport(Colemanetal．，l966)，whichreferredtothe‘equalityofeducationalopportunity’andtoJencks’book(Jencksetal.,l972),whichwasonthe‘Inequality：areassessmentoftheeffectoffamilyandschoolinginAmerica’，andresponsetothePublicationof‘LeagueTable’inUK(Goldsteinl996)，manyresearchesonschoolingeffectonstudentattainmentemergedinUS，UK，andotherareas．Althoughthedefinitionofschooleffectivenesshavesomediscrepanciesfordifferentresearchers(Cheng1996)，somecharacteristicsofeffectiveschoolsuchasprofessionalleadership，maximizationoflearningtime，purposefulteaching，highexpectation，positivereinforcement，etal．，areacceptedbymostpeople(Sammonsl999)．Generally，measureofschooleffectivenessinvolveschoosinganoutcomeandthenstudyingaveragedifferenceamongschoolsafteradjustingforrelevantfactorssuchastheintakeachievementsandsomestudentbackgroundcharacteristics．Althoughtheunadjustedresultsareinformativetosomeextentforeducationevaluation，justanexamresultdoesnotshowaschool’sprogresscompletely．Furthermore．a(chǎn)singlestatisticmaynotbeanadequatesummaryofschool’seffectonstudent’sprogress．schoolmayboosttheprogressofdifferenttypesofstudentatdifferentrateindifferentdepartments／subjects．Forfairerandmoreaccuratewaytomeasureandreportschoolperformance，theValued—AddedMethodisacceptedbymanyeducationresearcherandpractitionernow．Itisseenasbothlessproblematicintermsofhowacceptableprinciplesaretopolicy—makerandschoolseniormanagers，andlessquestionedeveninthemostscepticalofstaffroom．ThismethodhasbecamethemainstreamofBritishschooleffectivenessstudiesnow．‘Therearevariousofdefiningvalueaddedandtheseencompassbothqualitativejudgmentandquantitativemeasures．Valueaddedapproachesalsodifferinthebalanceofemphasisplacedonevaluatingstudentoutcomesdirectly，orindirectly,viathequalityoftheteachingandlearningprocess.…alldefinitionsofvalueaddedhavethecommonaimofassessingthequalityandextentofaschool’seffectivenessinpromotingstudentachievement.(Sammonsetal．1997,P.24)’．Thomasandcollegues(1998)definethe‘Valueadded’asanindicationoftheextenttowhichanygivenschoolhasfosteredtheprogressofallstudentsinarangeofsubjectduringaparticulartimeperiodoroverparticularyearsincomparisonwiththeeffectsofotherschoolsinthesamesample．Thisdefinitionisnotonlyfocusingontheacademyprogress，alsoonthestudents’attitudestolearningandotherimportantoutcomes.However,thisdefinitionrefersspeciallytoacademicattainmentlevelofattainmentincomparisonwithsimilarstudentsinotherschools．Forcalculatingthevalueaddedcomponent，accuratebaselineinformationaboutstudent'spriorattainmentisveryimportant，sothevalueaddedmethodscancomparesoutcomeafteradjustingforvaryingintakeachievement．Theconceptofvaluedaddedisnotonlyanindicatorofschooleffectiveness，alsoatoolforschoolself—evaluation．Thevalueaddedaschoolcontributedtoindividualstudenthasthefollowingpurpose(Thomasl998)：※Itoffersafairerandmoremeaningfulwayofpresentingschoolexaminationresult；※Itisatoolthatcanprovidebothdetailedandsummarydatathataschoolcananalysisaspartofitsself—evaluation；※Itcanbeusedtoexaminetrendsinvalueaddedperformanceovertime.Inrelationtoschoolimprovementinitiatives；※Itprovideperformancemeasuresthatcanbecontrastedagainstothertypesofdataavailableinschoolssuchasmeasuresofpupils’affectiveorvocationaloutcomesorinformationabouttheviewsofkeygroupsobtainedusingteacher，parentandpupilquestionnaires；※Itcanprovideadditionalguidanceinmonitoringandtargetsettingforindividualstudentsandspecificgroupsofstudents．Someresultsfrompreviouslystudies．Comparisonbetweenschoolvalueaddedforoverallattainmentandsubjectsshowschoolsvarydifferentlyacrossdifferentareaofacademicattainment．Thomas(1996)foundcorrelationbetweenvalueaddedscoresforEnglishandmathematicsis0.46．Otherresearch(Thomasl995)foundthesecorrelationbetweendifferentsubjects(English，Englishliterature，math，science，F(xiàn)renchandhistory)rangefrom0.20to0.72，thismeansmanyschoolshavemixedresultsfordifferentdepartments．onlyafewhastheconsistentresultsfordifferentsubjects．Thesefindingssuggeststronglytheneedtolookatschools’valueaddedperformanceindetailnotonlyacrossdifferentacademicsubjectsassignificantsubjectdifferencesmaybemaskedbyoverallattainmentinsomeschool．Studiesshow，onaverage，somedifferentlevelsofacademicperformance／progressbetweenboysandgirls．Yang’s(1999)researchshowgirlsperformbetterthanboysinKS1reading，KS1writing，butworseinMathematicsandSciences；Thomas(1996)findgirlsperformbetterintotalscoreandEnglishlanguage．Goldstein’s(1996)researchalsoshowsasmalldifferenteffectforboysincomparisontogirls．Student’sageseemshowsomeeffectontheirsattainment．Twostudiesshowoldstudentsmakemoreprogressintheirstudies(Thomasetal．，l996，1997)，‘0naverage，theoldestl5+pupilsfwithbirthdaysinSeptember／October)attainedapproximately2GCSEpointsmorethantheyoungestl5+pupils(withbirthdaysinJuly／August)(Thomasl996，P．12)’．Studiesonpupilssocial-economicstatusshowstudentsentitledwithfreemeal，whichisacomprehensiveindicatoroffamilyeconomicclass，seemperformworsethanthosewithoutfreemealonaverage(Yanget、al．，l999，ThomasS，etal．，l997，1999)．Thesefindingscouldbeusefullyinterpretedasmeaningthatmostschoolshavedifferentlevelsofeffectivenessforpupilsfrommoreorlesseconomicallyadvantagedbackground．Pupilswithdifferentethnicbackgroundalsoshowdifferentgainsinacademicattainment(Nuttalletal.,l989，Thomasl996，Sammonsl999).‘Incomparisontothewhitegroup，otherethnicgroupsobtainedsignificantlyhighscores，giventheattainmentonentrytoschoolhavebeentakenintoaccount．(Thomasetal.，1996．P．12)’．ButsomestudiesdonotfindevidenceofsignificantdifferenceinprogressinsomesubjectforsomeethnicgroupSammonsl999)．Theseneedfurtherstudyandhighlightagaintheimportanceofdetailedanalysisinexploringtheaffectingfactoronstudents’improvement．Inshort，student'spriorattainmenthasthelargestimpactontheirlaterattainment，theadjustmentusinggenderorsocial—economicstatusaresmallbycomparison.Nuttall’s(1989)studysuggestvariabilityinhighabilitypupilsbetweenschoolsismuchlargerthanthatoflowabilitypupils，Thomas(1993)foundcorrelationbetweenvalueaddedscoreforhighestandlowestabilitystudentrangefrom0.73to0.76fordifferentsubject．Therecentlystudiesusingfinelygradedpriorattainmentmeasuresinbothinnercityareas(Goldsteinetal.，1993)andincountyLEAs(Thomas，etal．，l996)havealsoestablishedsignificantdifferentialsecondaryschooleffects．Theseevidencepointthatsomeschoolarenotequallyeffectiveinpromotingtheattainmentofallpupilswithdifferentpriorattainment，irrespectiveoftheirpreviousstrengthsorweaknesses．StudiesonstabilityandconsistenciesofschooleffectivenessshowschooleffectsaremoststablefortotalGCSEperformancescorebutthedepartmenteffectsarelessstableovertime．Thecorrelationbetweenschooleffectacrosstimeforperformanceindifferentacademicsubjectsisfrom0.38to0.92(Thomasetal．，l997)，andonlyaverylowpercentageofschoolcanmaintainthesameposition，says，alwayseffectiveorineffective，fortotalscoreandsubjectscorethrough3years．Thesesuggestthatweshouldlookresultsortrendsoverseveralyearsinschoolevaluationpractice．Studiesaboutschoolbackgroundvariable’seffectsonstudents’attainmentgainedshowsomeschoollevelvariablehavelittleeffectonchildren’sprogress．Yang’s(1999)researchshowsthecoefficientforvariableofpercentageofstudentsentitledwithfreemealinmodel’sfixedpartisjust-0.007．Givenothervariablecontrolled，somestudiesfindwhenpriorattainmentdataareavailablenoschoolcontextfactorsaresignificantandthefitofthemodelissubstantiallyimproved(Thomasetal．1996)．Therearesomeresearchonschooleffectivenessandschoolimprovementareainchina(Cheng1996，Wang1998)，butmostofthemfocusontherationalthought，fewresearchbasedonevidenceanalysisisfound．ThispilotstudyaimstogiveatentativeanalysisforsomedatafromseveralBeijingseniorsecondaryschools，usingmultilevelanalysistechnique，thencomparetheresultwiththepreviouslystudiesintermofinternationalview．BeingsupportedbyBritishCouncil，forthefeedbackpurposetoschool，thisstudywillexploretheimplicationofvalueaddedmeasurementmethodineducationpracticeinBeijingarea．TheanalysisofvaluedaddedmeasuerDataThedatacomefromseveralseniorsecondaryschoolsinBeijingCity，China，whichrefertothestudentswholeftseniorsecondaryschoolinl998．Originallythereare7schools’dataavailable，butsinceoneschooldidnotprovideitsstudents’priorattainmentinformation，datafromthisschoolisonlyincludedinthebeginninganalysis．Theother6schooldataareincludedinthefinalmultilevelanalysis．Thecasehavingmissingvaluecannotbeincludedintheanalysisautomaticallywhenmultilevelmodelarefitted，so，only617casecanenterthefinalanalysisinourstudy，contrastwiththeoriginall051caseinthesample，thatmeansthesamplesizeisabitsmall．Descriptiveanalysisisavailableinotherpaper．Thevariablesinsampledatabaserefertoeachstudent’sage，gender，ethnicity，parents’occupation，student’stype，major，andtheirdifferentattainmentofdifferentsubjectsindifferenttime，etal．．Sincedifferentstudentsfromdifferentdistricthavedifferentsubjectpriorattainment，onlythreecoresubjects，Chines，Math，Englishandthetotalscoresfornationalexamareincludedinthecurrentanalysis．Studentsareclassifiedtolowband(bottom25％)middleband(Middle50％)andhighband(upper25％)foranalysisaccordingtothesumoftheirChinese，mathandEnglishscorewhenleavingJuniorsecondaryschool．Schoolbackgroundvariableincludespercentageoflowbandstudents，ratioofteacher／pupil，schooltype，whichisatermformerlycalledandnotusednowasacrudeindicatorofcomprehensiveschoolinputandteachinglevel．Nootherbackgroundvariablesavailablethistime.Age，ratioofteacher／pupilandattainmentscoresarenormalized(Mean0，standarddeviationl)fortheanalysispurpose．(Goldsteinetal．，l999)．AnalysismethodAdvancedMultilevelmodelingisusedasthemethodofanalysis(Goldstein，1995，1999)．Theadvantageofmultilevelmodelingisnotonlythatitcancapitalizeonthehierarchicalstructureofthedata，alsothatthistechniquecanbeusedtolookatpotentiallyinterestingdifferences，suchasthosebetweentheperformanceofdifferentstudenttypehavingtakenintoaccountoftheirpriorattainment．Theanalysiscanprovideanestimateofresidual(valueaddedscore)foreachschoolafterestablishedeveryvariable’simpact．a(chǎn)measureofresidualplotwith95%confidenceintervalforeachschoolcanbeattachedforcomparinggraphically(Goldsteinl995)．Fourmeasuresofstudents’performanceinnationalexamasoutputvariableareanalysed：thetotalscore，theChinesescore，theEnglishscore，themathscore，theircorrespondingsubjectsscoresinleavingJuniorsecondaryschoolarefittedaspriorattainment．Sincethereareonly6school，whichmeansnumberof2ndlevelisinsufficient，therearemanycomputingproblemsinourmultilevelanalysis，soallvariablesenteringthemodelareonlysetinfixedpartsexceptconstantvariable．Variablescontributingnosignificantchangeofgoodness—fittothemodelorhavingnosignificantfixedpartcoefficientwillnotbeincludedinmostequationformostcase．Differentmodelsfordifferentacademicperformancearefittedfirstly，andthenamultivariatemodelfortotalscoreandmathisfitted．Themodelsareasfollow：1．Interceptonlymodel：thisisthebasicmodelthatprovidesbasicanalysisandcomparingbaselineforotheranalysis．Onlytheconstantvariableisincludedintheequationandsetrandomatbothschoolandstudentlevel．2．Priorattainmentonlymodel：onlysinglepriorattainmentorcombinedpriorattainmentisfittedinthemodel，allpriorattainmentissetonlyinthefixedpart．3．Studentbackgroundonlymodel：onlyStudentbackgroundvariableisfittedinthemodel，allaresetonlyinthefixedpart．4．Schoolbackgroundonlymodel：onlyschoolbackgroundvariableisfittedinthemodel，allaresetonlyinthefixedpart．5．Priorattainment+studentbackgroundmodel：thecombinedpriorattainment+studentbackgroundvariablesarefittedinthemodel,allaresetonlyinthefixedpart．6．Priorattainment+schoolbackgroundmodel：thecombinedpriorattainment+schoolbackgroundvariablesarefittedinthemodel．a(chǎn)llaresetonlyinthefixedpart．7．Fullvariablemodel：allvariablesarefittedinthemodel．a(chǎn)llaresetonlyinthefixedpart．ThenewMLWin(visionl.1)softwareisusedforourstudy．Onefeatureofthisnewvisionsoftwareistoprovidedetailedsimulationresultfortheanalysis(Goldsteinetal．，l999)．Thesimulationtechniquecanmakeaccurateinferencesonthebasisofsimulatedparameterestimate,thisisusefultohavemethodsforproducingaccurateintervalestimateswithsmallsamples．Aftereachmodelwasfitted，threesimulationmethodareused：Gibbssampling(MLWindefaultoption)，MetropolisHastingSampling(univariateMHforfixedandrandomeffectsparameterisset，theotheroptionisMLWindefault)，Bootstrapestimation(MLWindefaultoption)．AssaidintheMLWinmanual．a(chǎn)llsimulationmethodsarenotusedformodelexploration，justforobtainingunbiasedestimatesandaccurateintervalestimatesatthefinalstagesofanalysis．Itisneedtotrydifferentsimulationmethodstoensurestableconclusion(Goldsteinetal．，l999)．Inouranalysis，resultsfromthreesimulationmethodshowsomekindconsistencyforfixedpartparameters，soonlyresultsfromMHmethodsarereported(seeappendixl8—21)togetherwithMlwindefaultIGLSanalysisresults．Sinceweonlyhavesixschooldataavailable,theanalysisshowschoollevelparameterhavinglargestandarderror,whichmeansthesamplingerrorislarge．Thesimulationresultbasedontheseschoollevelparameterswithlargesamplingerrorisuncertaintyforfurtheranalysis．SothecomparisonandanalysisarebasedonIGLSresult.ThefindingDescriptiveanalysesofeachvariableshowstudent’sintakeacademicattainmentsexceptChinesesubjectaredifferentindifferentschool.Afterthedescriptiveanalysis,eachvariablementionedaboveisfittedforthemodelseparately．Ethnic，majorandratioofteacher／pupildonotshowsignificanteffectonmodelgoodness-fitchange，sothesevariablesaren’tincludedinthelatermultilevelanalysis．Student’stype，parents’Jobhavemanymissingvalue．theyarenotincludedinthelatermultilevelanalysisformakinggooduseofthepriorattainmentinformation．Schooltypehavesomeeffectformodelgoodness—fitchange，butwhenincludedwithothervariable，therewillnosignificanteffect，evencausecomputingerrorprobablyfromthesmallsamplesize．wedonotincludeitinthelatermodeltoo．Finallyonlystudentsacademicscore，age，genderandpercentageoflowbandstudentareincludedinthelastmultilevelmodelsetting．Themultilevelanalysisresultsareshownintableltotable4．(P128～P131)Fortotalscore(tablel)，withoutincludingpriorattainmentinthemodel，bothpupilbackgroundandschoolbackgroundvariablecanexplainsomepercentageforthemodelgoodness—fit．Pupilbackgroundvariablesseemcontributemoretothemodelgoodness—fitchangethanschoolcontextvariable，causing16％decreaseingoodness—fitcomparingwiththeinterceptonlymodel．Thepriorattainmentcanexplainthemajorityofvarianceofmodel，itcause22.77％decreaseinmodelgoodness—fit．Sinceonlypercentageoflowbandstudentbeingschoollevelvariableisincludedinthefullmodel，anditscoefficientinfixedpartshowsnosignificance．Thismeansfortotalscore，schoollevelvariableinourstudymakeslesssignificanteffectonpupilattainment，afterpriorattainmentandpupillevelvariablearetakenintoaccount.Fromtheintra—schoolcorrelation(0.1240—0.2863fordifferentmodel)wecanconcludethereisconsiderablevariationbetweenschool’sperformanceintotalscore．Despitethereductionsinvarianceinschool1evelinsomemodel．comparingwithinterceptonlymodel，thereisstillaschooleffecteventhemostvariableshavebeentakenintoaccount，forthefullmodel．themeanofschoollevelvarianceis0.146．Someaveragedifferenteffectsbetweendifferentgroupsarefound．Boysperformancebetterthangirls，thefixedpartcoefficientinfullmodelis0.232．Elderstudentsseemperformanceworsethantheyounger，thefixedpartcoefficientinfullmodelforageis-0.175．FormathandEnglishsubject(table24)，thesimilartrendisfound．Pupilbackgroundvariablesandschoolbackgroundvariablehavesomekindeffectsonstudentattainment，thepriorattainmentcancausesignificantdecreaseinmodelgoodness—fit，especiallyinthefullmodel．AnalysisforChineseattainmentshowstheschoollevelvarianceisoforfullmodelandpupilbackground+priorattainmentmodel，alsoextremelysmallinothermodel，especialinthemodelsincludingpriorattainment．Thismeanswedonotfindtheseschoolshavedifferenteffectinstudent’sChinesesubjectstudy．FromthedescriptiveanalysiswedonotfinddifferenceforpriorandlaterChineseattainmentbetweenschools，thiswouldbeonereason，andalsonoschooleffectsarefoundinChineseattainmentthistime．thereasonneedfurtherstudy．Fromtheresultswecansayschoolsstillhavesomekinddifferenteffectonstudent’ssubjectacademicattainment．Astotheeffectsfrompriorattainment，mathseemshavelittleimpactonpupil’sChineseandEnglishattainment．ChineseandEnglish，especialEnglishpriorattainment，havesignificanteffectonstudent’slateracademicoutcome．Ageseemtobeanegativevariableaffectingstudent’sattainmenthere．Boyseemsdobetterinmathsciencethangirldo，butworseinEnglish．Thecurrentschoolbackgroundvariableseemstohavenoeffectonstudent’sacademicattainmentduringthe3-yearstudy，althoughthecoefficientislarge，theestimatestandarderrorislargeinallfullmodelscomparingwithotherexplanatoryvariables．WetrytosetstudentbandvariableinstudentlevelinallfullmodelandfindhighbandvariablevarianceinstudentlevelisnegativeforallandsignificantlynegativeforEnglishandChinese，thismeansstudentswithhighpriorattainmenttrendtovarylessintheirlaterattainment．Herewetrytoanalysisthisdatausingmultivariatemethod，becauseofthecomputingerrorcausedbythesmallsamplesize，onlyamultivariatemodelincludingtotalscoreandmathmodelisfittedatlast．Therearesomesimilartrendswithpreviouslyanalysis，ageseemhavenegativeeffectonstudent’sstudy，especiallyfortotalscore，boydoesbetterthangirlhere．Schoolsstillhavedifferenteffectonstudentsinthesetwoacademicattainment．thevarianceatschoollevelis0.260and0.178respectively，varianceatstudentlevelisstilllarge，hereitis0.159and0.110formathandtotalscorerespectively．ThecorrelationbetweenthesetwoacademicscoreiS0．978and0．814atschoollevelandstudentlevelrespectively．Forthefeedbacktoeachschool，differentresidualplotwith95％confidentialintervalispresented．AccordingtoGoldstein，formoreaccuratecomparisonbetweenanytwoschools，weusel.39asSDmultiplier(Goldsteinetal．，l995)．Residualscatterplotsforcomparisonbetweentotalscore，mathandEnglishareavailabletoo．Fordetailedanalysis，therawscorescatterplotsfordifferentattainmentareprovidedtoeachschoo1．Eachschool’sresultsarehighlightedforidentification．ForChinese，sincetheschoollevelvarianceis0inthisstudy，noresidualplotforChinesesubjectisavailablehere．Fromtheplot，wefindschoolshavedifferentresidualpositionacrossdifferentacademicattainmentexceptChinese．Fromtherawscorescatterplot，wecanfoundsomeschoolhaveoutliers，sayextremelyvalue．Theseoutliersmustaffectschool’sresidualposition，wedon’tknowhowmuchthiseffectis，sincewedonotdothediagnosticexploreshere．Theanalysisfortheseoutliersshouldbeveryusefulforeducationpractice，especiallyforsomecasestudy．Althoughwedidnotcomputetheresidualcorrelationbetweendifferentmodelfordifferentsubject，theresidualplotsherestronglyshowthedifferentschoolshavedifferenteffectsfordifferentsubjects，notincludingChinese．Thesecanbeusedasschoolevaluationtoolineducationpractice．

DiscussionandconclusionFromaboveanalysis，somesimilartrendsarefound，comparingwithotherstudies(Thomasetal．，l994，1997，1997，1999，Yangetal．，l999，Sammonsl999，Goldsteinetal．，l996)．Onaverage，genderhavesomeeffectonstudent’sstudying；priorattainmentcanexplainthelargepartofthemodelvariance；schoolcontexteffectdecreasewhenpupilbackgroundvariableorpriorattainmentaretakenintoaccount．Hertheoptimalmodelistakingallvariables，includingpriorattainment，schoolcontextvariables．studentbackgroundvariables，intoaccountotherthantakingonlyonepartoftheseaffectingvariable，becauseitfitsthemodelbest．Theanalysisresultsstronglysupportthatschoolshavedifferenteffectondifferentstudentgroupfordifferentsubject，exceptChinese．Thesestrengthentheneedstousearangofdifferentvalueaddedmeasureforparticulargroupofstudents，forindividualdepartmentinthemonitoringandevaluatingprocess，evenfordifferentpupilcohortsandpointsintime(Thomasetal．,l999)．Thevalue—addedmeasureforschoolevaluationshouldbemulti—facetanddynamic．ThisresearchisdifferentformmoststudiesinUkinsomefacets．Intheirstudies，student’spriorattainmentissomecomposite／crudeindicator．suchastheLondonreadingtest，thesocialeconomicstatusindicator，suchasfreemealentitlement，isnotaaccurateindicatoreither．‘Theusingofacrudegroupedmeasureratherthanafinelydifferentiatedmeasureofpriorattainmentsmayaffectfindingsaboutthenatureandextendofdifferentialschooleffectiveness’(Thomasetal．1997，p.454)．ChinesesecondaryeducationsystemisdifferentfromUK．StudentsinJuniorsecondaryschoolstudythesamekindsubjectsasinseniorsecondaryschool，sosubjectscoreinJuniorsecondaryschoolshouldbeamoreaccuratemeasureofpriorattainmentinpredicatingthefutureattainmentinsamesubjectthanthecomposite／substituteindicator．Thisistheadvantageofourstudy．Somestudiesfoundolderpupildobetterthanyoungerintermsofbothrawandprogressmeasuresofattainment(Thomasetal．，l996，Sammonsl999),ourstudyfoundthecontrastresult，isthisbecausethedifferentculturebackground，Orresultsfromdifferenteducationsystem?Theunderlyingreasonneedfurtherstudy．GendereffectinourstudyalsoseemdifferentfromsomeUKstudyinsomesubject，fortotalscoreandmath，boydobetterthangirl；forEnglish，girldobetterthanboy，thelaterisinlinewithUK’sstudies．Itisalmostnostrongevidencethatparents’occupationandstudenttypehavesignificanteffectonstudentstudying．UnlikethestudyinUK，nodifferencewasfoundbetweendifferentethnicgroup(Sammonsl999，Thomasetal．，1996)．Thisshouldbetheresultofculturebackgrounddifferenceorotherunknownreason，probablyeducationpolicy．Schoolcontextvariableshavesomeeffectsonstudents’studywhenfittedinthemodelonly，butwhenthepriorattainmentarefittedinthemodel，theseeffectbecomenosignificant，beinginlinewithotherstudies．‘Thisindicatethatschoolcontextfactorsmayonlybesignificantinpredicatingpupiloutcomewhenrichandwiderangingpupilleveldataarelacking(Thomasetal．，l996，pl83)’．HerewedonotfoundschooleffectsonChinesestudy．isthisbecauseoursmallsamplesizeorthereisreallynoschooleffectonChinesestudyorotherunknownreason?Thereasonneedfurtherstudy．However，formthispilotstudy，wecansayvalueadded