Uma abordagem via mistura finita para modelos de regressão linear com erro nas variáveis

Abstract

In linear regression models with measurement errors it is usually common that the assumption of symmetric normal distribution for measurement error is not the most adequate for the data at hand. This can be evidenced in cases where the measurement error presents have behavior that does not coincide with those of different population subgroups. This work proposes a finite mixture distribution of skew-normal with a mass point at zero. This distribution allows flexibility in errors, accommodating both symmetry and asymmetry in the same. To carry out Bayesian inference, an algorithm of the type Gibbs with Metropolis-Hasting step is developed. To evaluate the performance of the estimates, a simulation study is presented with different symmetries and asymmetries in the measurement error and applied to a real data set.