Person-fit Index And Its Performance In Cognitive Diagnostic Assessment

Cognitive diagnostic evaluation is an evaluation of individual attributes based on cognitive diagnostic model.Whether the cognitive diagnosis model(CDM)can accurately and reasonably reflect the test data is very important.In recent years,most of the indicators used to verify the rationality of CDM have focused on test-fit and item-fit,but lack of research on person-fit which is the identification and analysis of abnormal reaction subjects.If misfitting responses are not detected out,it will lead to the misinterpretation of students'attribute profiles,which may result in faulty remediation decisions.This paper incorporates ~lzstatistics,developed under the framework of item response theory,into the cognitive diagnosis theory.This paper also introduce a new person-fit statistic,generalized distance index(GDI),developed for detecting misfitting response patterns for cognitive diagnostic assessments.This paper conduct both simulation studies and one empirical research to discuss the stability and accuracy of l_z and our new statistic.This study includes three parts.The first study developed a new person-fit statistic GDI.The feasibility and reconnaissance effect of GDI and ~lz are discussed in the field of cognitive diagnosis through simulation study.The second study,using simulation study,compared the feasibility and effectiveness of the person-fit statistics in the different factors like the discrimination power of test items,the test length,the number of attributes.The third study identified and analyzed the abnormal responding subjects in the real data to illustrate the applicability of person-fit index in educational practice.Three studies have drawn the following conclusions:1.The construction of GDI is feasible and reasonable in this study.GDI and l_z can detect abnormal responding subjects in different misfit proportion and type,so they can be used for person-fit evaluation.2.The sensitivity of GDI and l_z to abnormal reactions is different.The discrimination of item,the test length and the number of attributes interact with each other to influence the detection rate of GDI and l_z.When the discrimination of item is low,GDI had high detection rate at detecting creative reaction and random reaction,and l_z had high power at detecting cheating responding.When the discrimination of item is high,GDI had high power at random responding.At this time,with the increase of test length,the detection rate of GDI and l_z for creative responding and cheating responding reached 100%.In addition,the increase in the number of attributes will also improve the detection rate of GDI and l_z.3.In the empirical study,the critical value of GDI fluctuates in a small range and is more stable.The detection results of GDI and l_z for abnormal response are consistent.And when the abnormal responding is deleted,the goodness of fit of the whole data-model is improved.