Termination Criteria for Grid Multiclassification Adaptive Testing With Multidimensional Polytomous Items

Area of Science:

• Educational Measurement
• Psychometrics
• Computerized Adaptive Testing

Background:

• Adaptive classification testing (ACT) is a variation of computerized adaptive testing (CAT).
• Multidimensional multiclassification involves classifying examinees into multiple categories across several dimensions.
• Grid classification is proposed for multidimensional multiclassification to provide clearer examinee standing and facilitate interventions.

Purpose of the Study:

• To implement the sequential probability ratio test (SPRT) and confidence interval method in grid multiclassification ACT.
• To propose two new termination criteria: grid classification generalized likelihood ratio (GGLR) and simplified GGLR for grid multiclassification ACT.
• To evaluate the efficiency of grid multiclassification ACT compared to measurement CAT.

Main Methods:

• Implementation of SPRT and confidence interval method within the ACT framework.
• Development and proposal of GGLR and simplified GGLR as novel termination criteria.
• Simulation studies using both simulated and real item banks with polytomous multidimensional items.

Main Results:

• Grid multiclassification ACT demonstrates higher efficiency than CAT focused on trait estimate precision.
• The GGLR criterion was most effective in terminating grid multiclassification ACT and classifying examinees.
• Both simulated and real item bank studies confirmed the efficiency of the proposed methods.

Conclusions:

• Grid multiclassification ACT offers a more efficient approach for classifying examinees into multiple groups compared to traditional CAT.
• The GGLR criterion is a highly efficient method for terminating grid multiclassification ACT, especially with high-quality item banks.
• The developed methods enhance the precision and efficiency of examinee classification in multidimensional settings.