The dilemma of hierarchical and classical variable selection in polynomial regression and a model of the average January minimum temperature in Turkey

Abstract

The existence of many exponential and/or interaction terms in Polynomial Regression causes some problems in modeling, especially with observed data. One of these is the hierarchy problem. In this study, the nonhierarchical pattern of classical variable selection as opposed to hierarchical ones will be investigated. A dilemma occurs as a consequence of the uncertainty as to which terms have to be included in the polynomial regression. As a consequence, instead of facing this dilemma, the use of both selection procedures together is suggested for prediction purposes. An application concerning the modeling of the average minimum temperature for Turkey is given using meteorological data.