Variable Selections

Abstract

In data analysis, there may be many variables involved in the study of interest. Hence, the data analyst must decide how to select the most important variables from all those available. Selecting the best variables not only increases forecast accuracy, but also reduces the cost of data collection. In this talk, we will consider variable selection for parametric, nonparametric, semiparametric, single-index, wavelet and time series models, as well as univariate and multivariate response structures. While our focus assumes normal random errors, we also discuss quasi-likelihood and robust model selection. Finally, we investigate the use of marginal likelihood in Box-Coxtransform model selections.