Parameter Recovery in Multidimensional Item Response Theory Models Under Complexity and Nonnormality.

Area of Science:

• Psychometrics
• Educational Measurement
• Psychological Statistics

Background:

• Item parameter recovery is crucial for robust assessment development, particularly in item response theory (IRT).
• Existing research extensively covers unidimensional IRT parameter recovery, but less is known about multidimensional IRT (MIRT), especially with complex test structures or non-normal latent traits.
• Understanding parameter recovery in MIRT is vital for advanced applications like computerized adaptive testing.

Purpose of the Study:

• To investigate the impact of complex item structures and non-normal latent trait distributions on item parameter recovery in compensatory MIRT models.
• To evaluate the effects on bias and root mean square error of item parameter estimates.
• To provide insights into the reliability of MIRT parameter estimation under challenging conditions.

Main Methods:

• A simulation study was conducted using compensatory MIRT models for dichotomous items.
• Complex item structures and skewed latent trait distributions were systematically manipulated.
• Item parameter recovery was assessed using bias and root mean square error as outcome variables.

Main Results:

• Skewed latent trait distributions generally led to adverse impacts on item parameter recovery.
• Increased test complexity further decreased the precision of parameter recovery.
• Discrimination parameters, particularly along one dimension, were notably affected when at least one latent trait was skewed.

Conclusions:

• The findings highlight the sensitivity of MIRT parameter recovery to the shape of latent trait distributions and test complexity.
• Assessment developers should exercise caution when interpreting MIRT parameters derived from data with non-normal latent traits or complex structures.
• Further research is needed to develop robust methods for parameter recovery in MIRT under these conditions.