Selecting variables in orthogonal linear models

Abstract

Reducing the number of variables of an orthogonal problem may lead to a more practical predicting equation. Based on the standardized total squared error, or the Cp statistic, this can be attained through several methods. A direct search in the sums of squares due to each variable may save computational work in the combinatory trials required by the Cp statistic method. The result is a reduced equation yet as unbiased as possible.