Studies on simulation were carried out aiming to achieve a comparative analysis between the following variance component estimation methods: Restrict Maximum Likelihood (REML), Maximum Likelihood (ML) and Method III of Henderson. All of them were used as a reproducer model, and the REML method was also used as a animal model. The main objective was to verify the effect of the unbalanced data on the estimation of the variance components. A genomic, made of a single quantitative characteristic, ruled by 200 loci was simulated. The genomic was used in the construction of three base-populations with heritability of 0.60, 0.30 and 0.10, constituted by 1000 animals (500 males and 500 females). From each base-population, it was constructed the initial populations. Each initial population was submitted to the mating at random, that producing four populations for each heritabilities studied. For a comparative study of the methods, different kinds and levels of unbalanced were introduced in the populations. For the population with high heritability and fully balanced data, the three methods, used as a reproducer models, presented similar results among themselves. For populations with low heritability, the REML method, used as an animal model, presented results closer to the real value for the additive genetic variance. The induced unbalanced data did not affect the estimation of the variance components by the methods. The differences observed were a consequence of the heritability levels. The REML method, as an animal model, could be considered the most appropriate to estimate the variance components for traits of low heritability.
variance components; methods; animal breeding; simulation
