An eigenvector method for estimating item parameters of the dichotomous and polytomous Rasch models

Area of Science:

• Psychometrics
• Statistical Modeling

Background:

• The Rasch model is a cornerstone in psychometrics for analyzing item response data.
• Accurate estimation of item parameters is crucial for reliable assessments.
• Existing methods like joint maximum likelihood estimation have limitations, especially with missing data.

Purpose of the Study:

• To present a new technique for extracting Rasch model item parameters using matrix eigenvectors.
• To demonstrate the applicability of this method to both dichotomous and polytomous data.
• To highlight the advantages of this novel approach over traditional methods.

Main Methods:

• Item parameters are derived from the eigenvector of a matrix constructed from pairwise item comparisons.
• The technique is applied to a previously published dataset for validation.
• Comparison of results with joint maximum likelihood estimation.

Main Results:

• The eigenvector technique yields item parameter estimates comparable to joint maximum likelihood estimation.
• For highly challenging items, the proposed method demonstrates superior estimation accuracy.
• The technique effectively accommodates and transparently handles missing data.

Conclusions:

• The eigenvector method offers a robust and efficient alternative for Rasch model item parameter estimation.
• Its ability to handle missing data and link to graph theory enhances its utility in social science research.
• This approach provides a unique framework for analyzing assessment networks within the Rasch model context.