摘要
认知诊断模型能否拟合测验数据,直接决定诊断结果的准确性。目前国内鲜有研究涉及认知诊断测验下的模型-资料拟合检验。文章将模型整体拟合指标及基于PPMC的项目拟合指标应用于认知诊断模型-资料拟合检验。模拟研究基于DINA,R-DINA和R-RUM三个诊断模型检验各拟合指标的表现。结果显示整体和项目拟合指标在识别数据产生模型时皆有较高准确率。采用整体和项目拟合指标比较了三个竞争模型与Tatsuoka带分数减法数据的拟合情况,显示R-RUM拟合最好。
One key issue in cognitive diagnostic assessement（CDA） is to select a suitable diagnostic model for a specific test. Mismatch between diagnostic model and test data world lead to decreasing classification accuracy, At present, few studies have addressed model - data fit criterion for CDA. Based on posterior predictive model checking（PPMC）, this study introduces several global fit indices and item - fit indices for model evaluations in CDA. The global fit indexes may provide information for answering the question as to the utility of the data for analysis by the model. The item - fit indexes are used to determine the interaction between the item responses and skills that each item is designed to measure. Simulation and real - data studies are conducted to examine the performance of these indices on three CDMs. The simulation results indicate that： （ 1 ） global fit indices are almost able to identify the simulation models and detect poor - fitting models ; （ 2 ） the item fit indices were able to identify fitting items and detect poor - fitting items. The results from real - data analysis indicate that：（ 1 ）according to BIC and DIC4and global G2, the R- RUM performed best followed by R- DINA model, and DINA model worst ; （2） for the number of item fit, the R - RUM and the R - DINA model also outperform the DINA model. s ：
出处
《心理学探新》
CSSCI
北大核心
2016年第1期79-83,共5页
Psychological Exploration
基金
全国教育科学规划教育部重点课题(DHA150285)
