Penalized Beta Regression and Adaptive Random Lasso

• Activity Number: 87
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Youran Fan* and Xiao-Feng Wang
• Companies: Cleveland Clinic and Cleveland Clinic
• Keywords: Adaptive Lasso ; variable selection ; beta regression ; bootstrap

Abstract

In a variety of studies from different disciplines, the response variables of interest are proportions, or percentages, or fractions. The types of variables are continuous but naturally bounded on the open unit interval (0, 1). The class of beta regression models has been commonly used by practitioners to model them. In this paper, we investigate high-dimensional beta regression. A computationally intensive method, the adaptive random lasso method, is proposed for variable selection in the beta regression. We apply an adaptive lasso with a two-step bootstrap: a measure of importance is yielded from the first step for each covariate; the final selected variables as long as their estimates and confidence intervals are obtained in the second step. The benefit of the proposed method is illustrated through extensive simulation studies. The method is also applied to a real data study in neuroscience.