A General Unfolding IRT Model for Multiple Response Styles

Area of Science:

• Psychometrics
• Statistical Modeling

Background:

• Respondents exhibit diverse response styles on Likert-type items, such as extreme responding, central responding, acquiescence, and disacquiescence.
• Standard unfolding item response theory (IRT) models assume no response style, leading to poor model fit and biased parameter estimates when these styles are present.
• Existing IRT models for response styles are often limited to specific styles and not applicable to unfolding data.

Purpose of the Study:

• To propose a general unfolding IRT model that accommodates various response styles.
• To integrate a softmax function for flexible handling of different response styles through scoring functions.
• To evaluate the model's performance and the consequences of ignoring response styles.

Main Methods:

• Development of a general unfolding IRT model incorporating a softmax function.
• Estimation of model parameters using Bayesian Markov chain Monte Carlo (MCMC) algorithms.
• Empirical data analysis and simulation studies to assess parameter recovery and bias.

Main Results:

• The proposed general unfolding IRT model demonstrates good parameter recovery.
• Ignoring response styles leads to significantly biased parameter estimates when using standard unfolding IRT models.
• The new model effectively accommodates various response styles, improving estimation accuracy.

Conclusions:

• The developed general unfolding IRT model provides a robust framework for analyzing Likert-type data with response styles.
• Accurate psychometric analysis requires accounting for individual response tendencies.
• The findings highlight the critical impact of response styles on parameter estimation in IRT.