Investigating the Normativity of Trait Estimates from Multidimensional Forced-Choice Data

Area of Science:

• Psychometrics
• Quantitative Psychology

Background:

• Multidimensional forced-choice (MFC) data requires specialized modeling for accurate trait estimation.
• Classical scoring methods for MFC data can be problematic due to their ipsative nature.

Purpose of the Study:

• To evaluate the normativity of Thurstonian IRT trait estimates derived from MFC data.
• To compare Thurstonian IRT estimates with classical ipsative scoring and other data formats (dichotomous, rating scale).

Main Methods:

• Simulation study comparing Thurstonian IRT, classical ipsative scoring, dichotomous true-false (TF), and rating scale data.
• Empirical study assessing the differentiation of behaviors using MFC versus TF responses.

Main Results:

• Thurstonian IRT trait estimates demonstrated normativity when using blocks of opposite keyed items, unlike classical ipsative estimates.
• Measurement precision of Thurstonian IRT was lower than rating scale data but robust to factors like item imbalance or correlated traits.
• Empirical findings showed equal criterion validity but lower construct validity for MFC compared to TF data.

Conclusions:

• Thurstonian IRT modeling effectively addresses limitations of classical scoring for MFC data.
• Enhanced validity in MFC data may require mitigating common method biases in comparative analyses.