Robust estimation of the latent trait in graded response models

Area of Science:

• Psychometrics
• Psychological Measurement
• Statistical Modeling

Background:

• Aberrant responses, such as careless or miskeyed answers, can compromise the validity of psychological assessments.
• Existing robust estimators are effective for dichotomous item response theory (IRT) models but not for Likert-type items with multiple response categories.
• A need exists for robust estimation methods applicable to widely used Likert-type scales.

Purpose of the Study:

• To propose and evaluate a novel robust estimator for the graded response model (GRM).
• To address the challenge of aberrant responses in psychological surveys using Likert-type items.
• To enhance the accuracy of latent trait estimation in the presence of response disturbances.

Main Methods:

• Development of a robust estimator for the graded response model (GRM).
• Implementation of two weighting functions (Huber and bisquare) to downweight suspicious responses.
• Conducting simulation studies with varying test lengths, numbers of response categories, and disturbance types.

Main Results:

• The proposed robust estimator significantly reduces bias in latent trait estimates across different simulation conditions.
• Stable standard errors were observed, indicating the estimator's reliability.
• The method demonstrated effectiveness in mitigating the impact of response disturbances on assessment scores.

Conclusions:

• The robust estimator for the GRM is effective in improving the accuracy of psychological assessments with Likert-type items.
• This method provides a valuable tool for researchers and practitioners dealing with response disturbances.
• The findings suggest broader applications in survey methodology and psychometric analysis.