thurstonianIRT

thurstonianIRT fits Thurstonian Item Response Theory (IRT) models to forced-choice questionnaire data to estimate latent trait scores and inter-trait correlations while addressing ipsativity and response biases.

Key Features:

• Model Estimation: Supports model estimation using Mplus, lavaan, and Stan for fitting Thurstonian IRT models.
• Estimation Paradigms: Employs both frequentist and Bayesian estimation methods as part of its modeling capabilities.
• Forced-Choice Questionnaire Modeling: Implements Thurstonian IRT specifically for forced-choice formats to mitigate common response biases inherent to rating scales.
• Trait Score Accuracy: Simulation studies show that trait score and inter-trait correlation accuracy are insufficient when measuring up to five traits but improve substantially when measuring around 30 traits.
• Ipsativity Handling: Uses equal-keying within item blocks to reduce ipsativity and enable valid comparisons of individual trait scores and resistance to faking.
• Simulation Validation: Relies on simulation studies to evaluate measurement accuracy and the impact of test design choices on ipsativity.

Scientific Applications:

• High-stakes testing: Application in contexts where examinees may have incentives to provide deceptive responses and robust trait estimation is required.
• Forced-choice questionnaire design: Use in developing questionnaires that balance social desirability and reduce faking through equal-keyed item blocks and multi-trait measurement.

Methodology:

Model estimation via Mplus, lavaan, and Stan using both frequentist and Bayesian approaches, implementation of equal-keyed item blocks, and validation through simulation studies assessing ipsativity and trait-score accuracy.