Title: Variable selection in non-linear regression models: A parsimony-utility approach
Authors: Carlos Carvalho - The University of Texas at Austin (United States) [presenting]
Richard Hahn - University of Chicago (United States)
Robert McCulloch - University of Chicago (United States)
Abstract: A procedure is described for principled and pragmatic Bayesian variable selection for nonlinear mean regression models. The method is motivated from a decision theoretic perspective and uses posterior samples to gauge the adequacy of submodels relative to the unknown true model. It is shown that computationally efficient surrogate prediction models can be used to prescreen submodels, greatly easing the variable search. The procedure is various data sets from the applied regression literature.