测验Q矩阵的修正方法及其比较研究

在认知诊断评估中,构建正确测验Q矩阵十分关键,但比较困难.该文将确定性输入噪音与门模型下3种在线标定方法(极大似然估计方法,边际极大似然估计方法和交差方法)用于测验Q矩阵修正,并与δ方法,γ方法和最小残差平方和方法进行比较.采用模拟研究验证和比较各方法的表现,研究结果显示:边际极大似然估计方法表现良好,交差方法次之; 项目所考查的属性数目是影响δ方法和γ方法的表现.

The Q-Matrix Validation Methods and Comparison to Three Existing Methods

The Q-matrix plays an important role in establishing the relation between latent attribute patterns and ideal response patterns.In practice,the Q-matrix is difficult to specify correctly in cognitive diagnostic assessment and misspecification of the Q-matrix can seriously affect the accuracy of both item parameter estimates and the classification of examinees.In the study,three on-line calibration methods have been extended to validate Q-matrix,and three related methods including the δ method,the γ method,and the Q-matrix refinement method(denoted by RSS)have been compared.A simulation study was conducted to investigate the sensitivity of validation methods to four factors(the distribution of attribute patterns,sample size,the quality of items,and the error rate of q-entries)under the deterministic inputs,noisy “and” gate(DINA)model.Results show that marginal maximum likelihood method performs best,both in terms of accuracy and robustness.