Ideal Point or Dominance Process? Unfolding Tree Approaches to Likert Scale Data with Multi-Process Models

Area of Science:

• Psychometrics
• Statistical Modeling
• Behavioral Science

Background:

• Likert scale data analysis often relies on assumptions about response processes.
• Existing Item Response Tree (IRTree) models may not fully capture ideal point response behaviors.
• Developing robust analytical frameworks is crucial for understanding latent traits.

Purpose of the Study:

• Introduce and evaluate three novel Unfolding Tree (UTree) models for Likert scale data.
• Compare the performance of UTree models against established Item Response Tree (IRTree) models.
• Investigate respondents' decision-making processes and latent trait structures.

Main Methods:

• Developed three new Unfolding Tree (UTree) models based on the ideal point assumption.
• Conducted simulation studies to assess model performance under various conditions.
• Applied UTree and IRTree models to empirical data for comparative analysis.

Main Results:

• Fit indices effectively distinguished between correct and incorrect models.
• Both UTree and IRTree models accurately recovered parameters when correctly specified, with improved precision for larger sample sizes and more items.
• Mis-specified models yielded biased individual parameter estimates, particularly for ideal point response processes.
• Empirical data supported the ideal point response process over the dominance process.
• Respondents' extreme response choices were primarily driven by target traits, not extreme response style.
• Identified two distinct, moderately correlated target traits influencing decisions across stages.

Conclusions:

• The proposed Unfolding Tree (UTree) models provide a valid framework for analyzing Likert scale data, especially when ideal point processes are present.
• The ideal point assumption better explains respondents' decision-making and response patterns than the dominance process.
• Findings underscore the importance of selecting appropriate models to accurately estimate latent traits and understand response behaviors.