Title: Approximating the asymptotic covariance matrix when using the EM algorithm

Abstract: This talk discusses an efficient and accurate numerical approximation strategy useful for obtaining the asymptotic covariance matrix of the parameter estimates (ACOV) when fitting models with the Expectation-Maximization (EM) algorithm. A short background on the EM is included, and current methods for obtaining the ACOV are briefly discussed. Next, a numerical approximation approach is presented and contrasted with previous approaches to highlight the computational benefits of this new approach. Finally, instructive and real-world examples are presented to demonstrate the methodology concretely, and a Monte Carlo simulation study is included to demonstrate the ACOV estimator's behaviour for a set of item response theory models.