Projection-Based Stopping Rules for Computerized Adaptive Testing in Licensure Testing.

Area of Science:

• Psychometrics
• Educational Measurement
• Computerized Adaptive Testing (CAT)

Background:

• The confidence interval (CI) stopping rule is standard in computerized adaptive testing (CAT) for classification decisions in licensure settings.
• CI rules exhibit reduced efficiency in near-cut regions of the ability (θ) scale, often failing to achieve early test termination.
• This inefficiency can lead to unnecessarily long tests, impacting candidate experience and operational costs.

Purpose of the Study:

• To introduce and evaluate novel projection-based stopping rules for CAT.
• To address the limitations of CI stopping rules, particularly in near-cut regions of the ability scale.
• To enhance the efficiency of CAT by reducing test length without compromising psychometric integrity.

Main Methods:

• Developed projection-based stopping rules that estimate the final ability (θ) score range at hypothetical test completion.
• Conducted a simulation study to compare the performance of projection-based rules against the traditional CI rule.
• Performed an empirical study to validate the findings in a real-world testing context.

Main Results:

• Projection-based stopping rules significantly reduced test length compared to the CI rule.
• The reduction in test length was achieved without compromising critical psychometric qualities, including ability (θ) estimation precision and classification accuracy.
• These rules are operationally simple, requiring no additional parameters beyond the existing CAT framework.

Conclusions:

• Projection-based stopping rules offer a more efficient alternative to CI rules in CAT, especially in challenging ability regions.
• The proposed rules provide a practical method to shorten tests while maintaining high standards of measurement quality.
• It is recommended to use projection-based rules in conjunction with CI rules for optimal performance across the entire ability scale in CAT.