认知诊断模型参数估计算法比较

The Comparison of Parameter Estimation of Cognitive Diagnosis Models

该文在不同条件的组合下考查了EM算法和MCMC算法对3种常用的认知诊断模型(DINA模型、DINO模型和G-DINA模型)的参数估计返真性问题.借助项目参数或作答概率分布的偏差、均方根误差、平均绝对离差以及被试的平均属性判准率等指标,评价这2类算法的表现.模拟研究结果表明:MCMC算法更适用于低质量题目、小样本、测验短的条件,而在其他条件下EM算法的表现与MCMC算法的表现相当.

The performance of the EM algorithm and MCMC algorithm under different combinations of conditions is investigated.Three cognitive diagnosis models are considered,such as the DINA(deterministic inputs,noisy"and"gate),DINO(deterministic inputs,noisy"or"gate),and the generalized DINA(G-DINA)model.The bias,root mean square error,average absolute deviation and average attribute correct classification rate are used to evaluate the performance of two parameter estimation methods applicable to the three models under different conditions.The simulation results show that the MCMC algorithm performs better than the EM algorithm for the condition combination of low-quality and short test with small sample size,while these two methods performs similarly in the other condidions.