A comparison of parameter covariance estimation methods for item response models in an expectation-maximization framework

The Expectation-Maximization (EM) algorithm is a method for finding the maximum likelihood estimate of a model in the presence of missing data. Unfortunately, EM does not produce a parameter covariance matrix for standard errors. Both Oakes and Supplemented EM are methods for obtaining the parameter covariance matrix. SEM was discovered in 1991 and is implemented in both open-source and commercial item response model estimation software. Oakes, a more recent method discovered in 1999, had not been implemented in item response model software until now. Convergence properties, accuracy, and elapsed time of Oakes and Supplemental EM family algorithms are compared for a diverse selection IFA models. Oakes exhibits the best accuracy and elapsed time among algorithms compared. We recommend that Oakes be made available in item response model estimation software.