Boosted Beta Regression

摘要: Regression analysis with a bounded outcome is common problem in applied statistics. Typical examples include regression models for percentage outcomes and the of ratings that are measured on scale. In this paper, we consider beta regression, which generalization logit to situations where response continuous interval (0,1). Consequently, convenient tool analyzing responses. The classical approach fit model use maximum likelihood estimation subsequent AIC-based variable selection. As an alternative established - yet unstable approach, propose new technique called boosted regression. With selection can be carried out simultaneously highly efficient way. Additionally, both mean variance modeled using flexible nonlinear covariate effects. consequence, method accounts problems such as overdispersion non-binomial structures.