The Little Bootstrap and Other Methods for Dimensionality Selection in Regression: X-Fixed Prediction Error

Abstract When a regression problem contains many predictor variables, it is rarely wise to try to fit the data by means of a least squares regression on all of the predictor variables. Usually, a regression equation based on a few variables will be more accurate and certainly simpler. There are various methods for picking "good" subsets of variables, and programs that do such procedures are part of every widely used statistical package. The most common methods are based on stepwise addition or deletion of variables and on "best subsets." The latter refers to a search method that, given the number of variables to be in the equation (say, five), locates that regression equation based on five variables that has the lowest residual sum of squares among all five variable equations. All of these procedures generate a sequence of regression equations, the first based on one variable, the next based on two variables, and so on. Each member of this sequence is called a submodel and the number of variables in the equation is the dimensionality of the submodel. A complex problem is determining which submodel of the generated sequence to select. Statistical packages use various ad hoc selection methods, including F to enter, F to delete, Cp , and t-value cutoffs. Our approach to this problem is through the criterion that a good selection procedure selects dimensionality so as to give low prediction error (PE), where the PE of a regression equation is its expected squared error over the points in the X design. Because the true PE is unknown, the use of this criteria must be based on PE estimates. We introduce a method called the little bootstrap, which gives almost unbiased estimates for submodel PEs and then uses these to do submodel selection. Comparison is made to Cp and other methods by analytic examples and simulations. Little bootstrap does well—Cp and, by implication, all selection methods not based on data reuse give highly biased results and poor subset selection. Key Words: Best subsetsMallows's C p ; Subset selectionVariable selection