Data strongly influenced by factors not accounted for by the statistical model can bias estimates of genetic parameters and values. Moreover, several traits of economic importance do not follow a normal distribution or have censored data. The objective of this study is to describe and illustrate the application of hierarchical Bayesian models for the detection and muting of outliers and for the analysis of censored data. First, the traditional specification of the animal model in hierarchical stages is presented under the Bayesian approach for normally distributed uncensored data. Then, this model is extended by introducing an independent weighting variable, which allows for the specification of thick tail residual densities from the Normal/independent distribution family. Finally, to cover censored data analysis, the basic model is extended by the inclusion of a variable with truncated normal distribution based on the lower limit in the observed value of the trait at the evaluation time, for those animals that have not yet completed their reproductive life at the evaluation time.
Bayesian inference; censored data; outliers; robust analysis
