Multivariate measurement error models using finite mixtures of skew-Student distributions

In regression models, the classical assumption of normal distribution of the random observational errors is often violated, masking some important features of the variability present in the data. Some practical actions to solve the problem, like transformation of variables to achieve normality, are often of doubtful utility. In this work we present a proposal to deal with this issue in the context of the simple linear regression model when both the response and the explanatory variable are observed with error. In such models, the experimenter observes a surrogate variable instead of the covariate of interest. We extend the classical normal model by jointly modeling the unobserved covariate and the random errors by a finite mixture of a skewed version of the Student t distribution. This approach allows us to model data with great flexibility, accommodating skewness, heavy tails and multi-modality. We develop a simple EM-type algorithm to perform maximum likelihood inference of the parameters of the proposed model, and compare the efficiency of our method with some competitors through the analysis of some artificial and real data.