Abstract

Cognitive diagnostic models (CDMs) are a family of psychometric models designed to provide categorical classifications for multiple latent attributes. CDMs provide more granular evidence than other psychometric models and have potential for guiding teaching and learning decisions in the classroom. However, CDMs have primarily been conducted using large samples. This study examines estimating CDMs in small sample conditions to aid formative learning. Three CDMs were compared across simulated classroom assessment conditions: deterministic input, noisy "and" gate (DINA) model, non-parametric cognitive diagnosis (NPCD), and supervised artificial neural network (SANN). We found all models estimated examinee classifications at the smallest sample size. Accuracy of individual attribute mastery classifications was acceptably high for the models under certain conditions. Effective item discrimination was the most important factor to accurately classify. The DINA and NPCD models were more resilient to measurement error than the SANN. Recommendations for application of CDMs in the classroom are provided.

摘要

认知诊断模型 (CDM) 是一系列心理测量模型，旨在为多个潜在属性提供分类分类。与其他心理测量模型相比，CDM 提供了更细粒度的证据，并具有指导课堂教学决策的潜力。然而，CDM 主要是使用大样本进行的。本研究检查了在小样本条件下估计 CDM 以帮助形成性学习。在模拟课堂评估条件下比较了三种 CDM：确定性输入、噪声“和”门 (DINA) 模型、非参数认知诊断 (NPCD) 和监督人工神经网络 (SANN)。我们发现所有模型都以最小的样本量估计考生分类。在某些条件下，模型的单个属性掌握分类的准确性是可以接受的高。有效的项目区分是准确分类的最重要因素。DINA 和 NPCD 模型比 SANN 更能适应测量误差。提供了在课堂上应用 CDM 的建议。