Model selection for analysis of uniform design and computer experiment

In this paper, a new variable selection procedure is introduced for the analysis of uniform design and computer experiment. The new procedure is distinguished from the traditional ones in such a way that it deletes insignificant variables and estimates the coefficients of significant variables simultaneously. The new procedure has an oracle property (Fan and Li⁸). It is better than the best subset variable selection in terms of computational cost and model stability. It is superior to the stepwise regression because it does not ignore stochastic errors during the course of selecting variables. The proposed procedure is illustrated by two examples, one is a typical example of uniform design, and the other one is a classical example for computer experiment.