Scale Alignment in the Between-Item Multidimensional Partial Credit Model.

Area of Science:

• Educational Measurement and Psychometrics
• Statistics
• Developmental Psychology

Background:

• Comparing latent trait estimates across dimensions in multidimensional item response models requires relative scaling.
• Traditional scaling methods like standardization do not guarantee Rasch model properties, specifically item ordering across dimensions.
• Previous work introduced scale alignment to enforce cross-dimensional item ordering by transforming item parameters.

Purpose of the Study:

• To extend the scale alignment methodology to the between-item multidimensional partial credit model.
• To adapt scale alignment for item response models estimated with incomplete data.
• To validate the extended scale alignment method using simulation studies and real-world data.

Main Methods:

• Linear transformation of item parameters within dimensions to achieve scale alignment.
• Application of scale alignment to the multidimensional partial credit model.
• Estimation of models using incomplete data, incorporating scale alignment.
• Illustrative analysis using the Kindergarten Individual Development Survey (KIDS).

Main Results:

• The extended scale alignment method effectively enforces item ordering across dimensions for polytomous models.
• Scale alignment performs well even with incomplete data, demonstrating robustness.
• The method provides a valid approach for scaling dimensions in complex item response models.

Conclusions:

• Scale alignment is a valuable technique for ensuring comparability of latent traits across dimensions in multidimensional item response models.
• The extension to partial credit models and incomplete data broadens the applicability of scale alignment.
• This method enhances the psychometric rigor of multidimensional assessments, particularly for developmental data like the KIDS survey.