Components of the item selection algorithm in computerized adaptive testing.

Area of Science:

• Educational Measurement and Psychometrics
• Computer Science and Artificial Intelligence

Background:

• Computerized adaptive testing (CAT) significantly boosts measurement efficiency in high-stakes testing.
• CAT tailors test item difficulty to individual test-takers for precise assessment.

Purpose of the Study:

• To provide a comprehensive overview of conventional CAT item selection algorithms.
• To discuss methodologies for test content balancing, item selection criteria, and item exposure control.
• To explore holistic CAT approaches and factors influencing CAT design.

Main Methods:

• Detailed explanation of three core CAT components: content balancing, selection criteria, and exposure control.
• Review of specific methods within each component, including test script, constrained CAT, Fisher information, b-matching, a-stratification, Kullback-Leibler information, randomesque, and fade-away methods.
• Discussion of automated test assembly methods like shadow tests and weighted deviation models.

Main Results:

• Item usage and exposure counts are influenced by the chosen selection criteria and exposure control methods.
• Factors such as computer resources, item pool size, and test length are critical for CAT design.
• The principles of CAT are increasingly integrated into adaptive learning environments.

Conclusions:

• The reviewed CAT algorithms and technologies can aid medical health educators and test developers in efficient CAT implementation.
• CAT's adaptive nature offers a pathway to more individualized and efficient educational assessment.
• Adoption of CAT principles can enhance both learning and assessment in educational settings.