Breeding programs are activities developed during several years and then must be flexible to adjust new situations created by changes in market trends, economic context and those caused by increase of amount and quality of data and scientific discoveries. The adjust to the latter must be done essentially by means of change and chose of the most suitable model to describe the phenomenon at specific scenario. Data of swine breeding program involving Duroc, Landrace and Large White races were analyzed through Bayesian theory to two candidate models. Three levels of à priori information were simulated: informative, little informative and non informative. The behavior of the à posteriori distributions curves and estimates associated of each level of information à priori were analyzed and compared. The results indicate that in the simplest model the samples of the three races are sufficient to yield estimates that aren't altered by à priori information. Duroc's estimates, with regard to complex model, are altered by à priori knowledge before data collecting and then should be tried the best representation of the à priori distribution to obtain suitable estimates given the actual knowledge of the breeder at that moment.
Likelihood function; à priori information; Pig; Bayesian theory
