Investigating Heterogeneity in Response Strategies: A Mixture Multidimensional IRTree Approach.

Area of Science:

• Psychometrics
• Statistical Modeling
• Survey Methodology

Background:

• Self-report measures are crucial in research but can be influenced by response style (RS) effects.
• Traditional Item Response Theory (IRT) models, like IRTree, assume uniform RS across respondents.
• Individual differences in the nature and strength of RS effects (e.g., midpoint RS, extreme RS) are often overlooked.

Purpose of the Study:

• To introduce a novel statistical model, the mixture multidimensional IRTree (MM-IRTree), to address the heterogeneity of response strategies.
• To detect and model individual differences in response styles (midpoint and extreme) within a latent class framework.
• To improve the validity of self-report measures by accounting for diverse response strategies.

Main Methods:

• Development of the mixture multidimensional IRTree (MM-IRTree) model, incorporating four latent classes based on response strategy profiles.
• Class-specific strategies include: extreme RS only, midpoint RS only, both RS, and neither RS.
• Simulation studies to evaluate MM-IRTree performance against traditional IRTree models under mixed response strategy conditions.

Main Results:

• The MM-IRTree model demonstrated robust performance in parameter recovery and class membership identification in simulations.
• Traditional IRTree models showed significant performance degradation when response strategies were mixed within the population.
• Empirical data analysis confirmed the presence of distinct latent classes with substantial sizes, supporting the MM-IRTree's utility.

Conclusions:

• The MM-IRTree model effectively captures heterogeneity in response strategies, offering a more valid approach to analyzing self-report data.
• Acknowledging and modeling individual differences in response styles is essential for accurate measurement in psychological and social sciences.
• The proposed model provides a valuable tool for researchers seeking to enhance the validity and interpretability of self-report measures.