Outlier Detection Using t-test in Rasch IRT Equating under NEAT Design

Area of Science:

• Psychometrics
• Educational Measurement
• Statistical Analysis

Background:

• Outliers in anchor items can compromise test equating accuracy and score validity.
• Evaluating anchor item performance stability is crucial before equating.
• Existing outlier detection methods may not be sufficiently robust.

Purpose of the Study:

• To investigate the effectiveness of the t-test method for detecting outliers in anchor items.
• To compare the t-test method with other outlier detection techniques.
• To evaluate the impact of sample size, outlier proportion, item difficulty drift, and group differences on outlier detection.

Main Methods:

• Simulation study design.
• Comparison of the t-test method against logit difference and robust z statistic outlier detection methods.
• Analysis of outlier detection performance across varied simulated conditions.

Main Results:

• The t-test method demonstrated superior sensitivity in identifying true outliers.
• The t-test method showed reduced bias in estimating the translation constant.
• The t-test method resulted in lower root mean square error for examinee ability estimates.
• The t-test method consistently outperformed other methods across all investigated factors.

Conclusions:

• The t-test method is a highly effective tool for detecting outliers in anchor items during test equating.
• Utilizing the t-test method enhances equating accuracy and strengthens the validity of test scores.
• The findings support the t-test method as a preferred approach for ensuring anchor item stability in psychometric practice.