The method or technique for producing the sequence of parameter estimates that, under the mild regularity conditions, converges to maximum likelihood estimator. Of particular significance in the context of the incomplete data problems. The algorithm comprises of two steps, called as the E, or
Expectation step and the M, or the Maximization step. In the previous, the expected value of log-likelihood conditional on the observed data and the current estimates of parameters are found. In the M-step, the function is maximized to provide the updated parameter estimates which increase the likelihood. The two steps are alternated until the convergence is attained. The algorithm might, in some cases, becoms very slow to converge.


This is acronym for the Epidemiological, Graphics, Estimation and Testing of the program developed for the analysis of the data from studies in epidemiology. It can be made in use for logistic regression and models might include random effects to permit over dispersion to be modelled. The beta- binomial distribution can be fitted.