A novel boosting method for the selection of the variance components in linear mixed models have been recently proposed. The method relies on the directions of negative curvature to deal with a non-convex loss function, which is the negative profile log-likelihood. Differently from previous proposals in the literature, which focus on the selection of the predictors in the fixed part of the model, the procedure provides a selection of the random part. More specifically, starting from a model without random effects, the algorithm sequentially includes them by updating the parameters that determine their covariance matrix. After summarizing the proposal, this paper presents an original application.