Investigating heterogeneity in IRTree models for multiple response processes with score-based partitioning

Area of Science:

• Psychometrics
• Statistical modeling

Background:

• Item response tree (IRT) models measure latent traits while accounting for response processes.
• A key assumption of IRT models is the homogeneity of response processes across all respondents.
• Detecting heterogeneity in these processes is crucial for accurate measurement.

Purpose of the Study:

• To propose a novel method for detecting parameter heterogeneity in IRT models.
• To develop a model-based partitioning algorithm to identify sources of differing response behaviors.
• To address the limitation of assumed homogeneity in IRT models.

Main Methods:

• Utilizing score-based tests to detect violations of parameter homogeneity.
• Applying extraneous person covariates to identify heterogeneity sources.
• Employing a partitioning algorithm for subgroup analysis.

Main Results:

• Simulation studies confirmed accurate Type I error rates and sufficient power.
• The method effectively differentiates between various types of parameter heterogeneity.
• The approach demonstrated utility with metric, ordinal, and categorical person covariates.

Conclusions:

• The proposed score-based partitioning approach effectively detects parameter heterogeneity in IRT models.
• This method allows for the identification of subgroups with distinct response behaviors.
• Empirical application confirmed the practical utility of the approach in analyzing latent response processes.