|
Abstract:
|
Large data sets have become increasingly frequent in behavioral research, and this presentation focuses on a statistical learning method designed to explore and analyze large data sets when knowledge concerning the structural relations between the variables is limited or absent. The emphasis is on gaining new insights, rather than confirmation. Most statistical learning algorithms were developed for topics that involved the prediction of a univariate outcome. However, in the behavioral sciences, constructs of interest are often unobservable and multiple items are used to measure these constructs adequately. In this roundtable discussion, I will present our recent extension of an existing boosting algorithm designed for multivariate outcomes. Details of the algorithm are described at http://arxiv.org/abs/1511.02025.
|
