Graded Response Testlet Model’s Application In The Test Of The Elder’s Ability Of Everyday Problem Solving On The Observed Tasks

With the development of modern psychology and education measurement,especially with the development of the item response theory since the1950s, thetesting construction is becoming more scientific. Item response theory is based onstrong assumption of local indepentdent, but with the more widely using of testlets,item response theory to estimate the parameters of testlets will have some deviation,testlet response theory should be applied for test quality analysis at this point. Atcurrent, the researches of the testlet response theory both in domestic and overseas aremostly simulation study, very little empirical research. This study in order toinvestigate the quality of the questionnaire “Performance-based assessment on theChinese elders’ ability of everyday problem solving”, which is based on the observedtask of daily living, surveying on629old people under the age of60to84inempirical study, using the testlet response model and traditional item response modelfor data analysis. The results are as follows:(1) The results of items analysisis showing that the questionnaire“Performance-based assessment on the Chineseelders’ ability of everyday problem solving” has a good quality, the questionnaire hasmoderate difficulty and satisfactory discrimination.(2) The questionnaire meet thecondition of Unidimensionality hypothesis, namely the questionnaire only measuresthe elders’ ability of solving the everyday problems.(3) The questionnaire has breachthe local independent premise, the results of Q3index and (testlet effects var)show that the local item dependence exists within testlets.(4) When the questionnairelength is fixed, whether the classification of the four testlets or finer classification of10testlets, the parameter estimates of testlet response model are basically consistent,especially the ability parameter estimation results are nearly the same.(5) Comparedwith the testlet response model, traditional item response model tends to overestimatethe discrimination and difficulty of the test, obviously overestimate the abilityparameter estimation and underestimate the standard error of ability parameterestimate.