Improving the assessment of measurement invariance: Using regularization to select anchor items and identify...

Area of Science:

• Behavioral Sciences
• Psychometrics
• Statistics

Background:

• Evaluating measurement invariance is crucial in behavioral sciences.
• Measurement invariance fails when differential item functioning (DIF) exists, where item responses differ across groups.
• Traditional DIF detection methods iteratively test items, risking incorrect anchor selection and DIF identification.

Purpose of the Study:

• To propose and evaluate an alternative method for selecting anchors and identifying DIF using regularization.
• Specifically, to apply a lasso penalty within the two-parameter logistic item response theory model.
• To compare the performance of lasso regularization against the likelihood ratio test method for DIF analysis.

Main Methods:

• Utilized regularization, a machine learning technique, specifically a lasso penalty.
• Applied the method to group differences in item parameters within the two-parameter logistic item response theory model.
• Compared lasso regularization with the likelihood ratio test method in a 2-group DIF analysis using simulations and empirical data.

Main Results:

• Lasso regularization demonstrated superior control of Type I error compared to the likelihood ratio test method.
• This advantage was particularly evident when substantial DIF was present and sample sizes were large.
• The power of lasso regularization showed only a minor decrement, indicating robust performance.

Conclusions:

• Lasso regularization is a promising alternative for testing DIF and selecting anchor items.
• The method offers improved accuracy and reliability in measurement invariance assessment.
• It provides a more robust approach, especially in complex scenarios with widespread DIF.