Maximum likelihood missing values estimation in patterns of missing MAR and MCAR in structural models.

In the research's of the applied field is very common to find matrices of data with lost values. The main strategies used in order to fix this problem, are the methods listwise, pairwise and maximum likelihood estimates. This article shows through Monte Carlo simulation in the field of the structural models, that irrespective of the pattern of missing simulated (missing completely at random, monotonic missing or conditional missing) the estimates through the maximum Likelihood algorithm EM throws the better results, concerning the biases in the estimate of the parameters of the models, decrease of the standard errors, and the possibility of finding convergent and adequate solutions in those patterns of missing where the strategies MCAR (listwise and pairwise) are impossible to use.