A General Approach for Estimating Projective IRT Models

Area of Science:

• Psychometrics
• Statistical Modeling
• Educational Measurement

Background:

• Projective Item Response Theory (PIRT) models simplify multidimensional item response models using lower-dimensional proxies.
• Current PIRT methods use logistic approximations, limiting them to specific response functions and requiring intensive computation for variability estimates.
• These limitations hinder PIRT's practical use, particularly with smaller sample sizes.

Purpose of the Study:

• Introduce a general maximum marginal likelihood PIRT (MML-PIRT) approach.
• Overcome limitations of existing PIRT methods, including restricted response functions and computational demands.
• Enhance the flexibility and statistical efficacy of PIRT models.

Main Methods:

• Leverage components of the expectation-maximization (EM) algorithm for MML estimation.
• Utilize expected count information from EM-MML to fit proxy response functions.
• Apply to a broader class of multidimensional IRT models.

Main Results:

• The MML-PIRT approach fits proxy response functions for any focal trait.
• It accommodates a wider range of multidimensional IRT models compared to existing methods.
• Provides accurate and efficient large-sample variability estimates for PIRT models.

Conclusions:

• The proposed MML-PIRT method significantly enhances the flexibility and statistical efficacy of PIRT applications.
• It addresses limitations of current PIRT techniques, making them more practical for empirical research, especially with moderate sample sizes.
• This advancement expands the utility of PIRT in psychometrics and related fields.