适用于二级和多级评分项目组成混合测试加权最大似然潜在特质估计（英文）Abstractinthisarticle,aweightedmaximumlikelihood(WML)latenttraitestimatorisproposedforthemixed-typetestscomposedofbothdichotomousandpolytomousitems・TheN~RalgorithmisusedtoobtaintheWMLestimator,andtherelevantequationsrequiredaregivenindetaiLToevaluatetheperformanceoftheWLE,aSimulationstudyisconducted,andtheobtainedresultsshowthattheWMLestimatorhasbetterpropertiesthanthemaximumlikelihoodestimation(MLE)・Finally,anempiricalstudyisusedtodemonstratethemethodology.Keywordsltemresponsetheory；Mixed一typemodels；Dichotomousitems；Polytomousitems；Maximumlikelihoodestimation；WeightedlikelihoodestimationCLCnumberO211.2DocumentcodeAlintroductionSofar,therearealotofapproachesaboutthebiasoftheabilityestimationreductionhavebeenproposed・Forinstance,Warm(1998)proposedaWMLforapplicationintestsofdichotomousitems・TheWMLestimatorconsistentlydisplayedthesmallerlevelofbiasthantheMLEestimator・ThenPenfieldandBergeron(2005)generalizedtheWMLtothepolytomousIRTmodels(Samejima,1998；Wang,2001；PenfieldandBergeron,2005)・Comparedtodichotomousitems,polytomousitemsprovidesuperiorinformationconcerningtheleveloftheestimatedlatenttrait(Donoghue,1994；EmbretsonandReise,2000,p.95；Jodoin,2003；PenfieldandBergeron,2005)・Theseapproachesmentionedabovefocusedseparatelyontestscomposedofeitherdichotomousitemsorpolytomousitems・However,inpracticethemixed-typetestcomposingofbothdichotomousandpolytomousitemsismorecommonlyused,forinstancetheNationalAssessmentofEducationalProgress(NAEP).Therefore,webelievethatitisofinteresttoproposeaWMLthatappliedtothemixed-typetest.Thisisthemainworkofthisarticle.Thepurposeofthisarticleistwofold：(a)topresentthederivationsoftheWMLestimatorunderamixed—typeitemresponsemodeland(b)tocomparethepropertiesoftheWMLestimatortothatoftheMLestimatorsunderdifferenttestconditions.Tothisend,theremainderofthisarticleisorganizedintofoursections・First，twomodelsusedinthearticlearebrieflysummarizedandpresentthederivationsoftheWMLunderthemixed-typetest・Second,asimulationstudyisconductedtoevaluatetheperformanceoftheproposedWMLbycomparingwiththeusualMLE・Third,arealdatasetfromalarge-scalereadingassessmentisusedtodemonstratethedifferencebetweenthetwoestimationprocedures・Finally,weconcludethearticlewithdiscussion.2.1RASCHandPCMInthispaper,weconsideramixed-typemodelthatisthecombinationofthefollowingRaschmodel(RM)andthepartial-creditmodel(PCM)・Tosimplifythenotation,theexamineesubscriptwillnotbeshowninthefollowingderivations.TheRMisdefinedaswherePij(0)istheprobabilityofselectingjofpolytomousitemiatabilitylevel0,bivdenotesthestepparameterofitemiofcategoryv,andmdenotesthenumberoftheresponsecategory.2.2TheWeightedMaximumLikelihoodEstimatorTofacilitatethepresentationoftheestimationmethod,therelevanttechnicalaspectsoftheIRTabilityestimationmethodsaredescribedbelow.BasedontheaboveRMmodelandPCM,thelikelihoodofresponsecanbewrittenastheproductoftwotypesoflikelihoodfunctions：3.ITheDesignofThisSimulationToevaluatetheperformanceoftheproposedWMLmethod,anintensiveSimulationstudywasconductedtocoverawiderangeofindexvalues,suchasthetotainumberofitemsandtheproportionofdichotomousandpolytomousitemsinamixed-typetest.ExceptfortheWMLmethod,theMLEmethodisalsousedtoestimatedthelatenttrait0.Then,wecomparethetwoestimatorsunderninedifferenttestconditions,the...