ABSTRACT
The expectation maximization (EM) algorithm (Dempster et al. 1977) has had an enormous effect on statistical modeling and analysis. In this chapter, we discuss only its direct application to item response models, including multilevel and mixture models. In this model family, it has been fundamental in providing computationally feasible methods for the maximum-likelihood analysis of large-scale psychometric tests with item models more complex than the Rasch model. What is even more striking is that nearly all, if not all, of these extensions have required or used this algorithm.
