Explaining Variability in Response Style Traits: A Covariate-Adjusted IRTree.

Area of Science:

• Psychometrics
• Survey Methodology
• Statistical Modeling

Background:

• Response styles like extreme or midpoint responding can bias survey results and threaten the validity of score interpretations.
• Existing models may not adequately account for individual differences in response styles.

Purpose of the Study:

• To incorporate person-level covariates into a multidimensional item response tree model to explain heterogeneity in response styles.
• To assess the model's parameter recovery and performance using simulation studies and an empirical example.

Main Methods:

• Developed and applied a multidimensional item response tree model incorporating person-level covariates.
• Utilized Markov chain Monte Carlo (MCMC) estimation, specifically the Gibbs sampler, for parameter estimation.
• Conducted two simulation studies and analyzed an empirical dataset (National Longitudinal Study of Adolescent to Adult Health).

Main Results:

• The model demonstrated small mean bias and root mean square error for item intercepts and discrimination parameters, especially with larger sample sizes (>1,000).
• Covariates significantly predicted extreme and midpoint response styles; for instance, non-White race, male gender, and high parental support were linked to extreme responding.
• Item and regression parameters were estimated with sufficient accuracy, indicating the model's practical utility.

Conclusions:

• The multidimensional item response tree model with covariates effectively explains individual differences in response styles, enhancing survey data validity.
• The model provides accurate parameter estimates, particularly with sample sizes over 1,000 and MCMC estimation.
• Identifying predictors of response styles offers valuable insights for survey design and interpretation.