Review of recent developments of parameter estimation methods in item response theory models with focus on studies reported in Psychometrika

Abstract

This research reviewed articles that developed parameter estimation methods in item response theory models, with special focus on recent articles published mainly in Psychometrika. As a result, various new methods to manage the computational burden problem with respect to the item factor analysis and multidimensional item response models, which have high dimensional factors, were introduced. Monte Carlo integral methods, approximation methods for marginal likelihood, new optimization methods, and techniques used in the machine learning field were employed for the estimation methods. Theoretically, a new type of asymptotical setting, that assumes infinite number of sample sizes and items, was considered. Several methods were classified apart from the maximum likelihood method or Bayesian method. Theoretical development of interval estimation methods for a latent trait parameter were also proposed and they provided highly accurate confidence intervals.