This article generalizes a multivariate skew-normal distribution and describes its many interesting properties. The univariate version of the new distribution is compared with two other currently used distributions. The use of the new distribution is illustrated with a real data example suitable for regression modeling. The new model provides a better model fit than its two rivals as evaluated by some suitable Bayesian model selection criteria.
Keywords: Bayesian inference; Gibbs sampler; kurtosis; skewness; skew normal distributions.