METAMODEL ESTIMATION UNDER CORRELATION METHODS FOR SIMULATION EXPERIMENTS
Abstract
The objectives of this research are threefold. First, a means of statistically validating the use of the Schruben-Margolin correlation induction strategy together with the follow-up analysis of Nozari, Arnold, and Pegden (1986) in a simulation experiment designed to estimate a linear metamodel for the response is established. This validation test consists of a three-stage statistical procedure. Each stage of the procedure tests a key assumption about the behavior of the response across all points in the design. The first stage tests for multivariate normality, the second stage tests for the induced covariance structure postulated by Schruben and Margolin, and the third stage tests for the adequacy of the proposed linear model. Because the test in each stage presupposes the properties tested in previous stages, these diagnostic checks on the experimental design and analysis must be performed in the indicated order. Second, an alternative to the analysis of Nozari, Arnold, and Pegden (1986) is developed for the case when the use of the Schruben-Margolin correlation induction strategy is inappropriate. Third, the use of the Schruben-Margolin strategy and the use of control variates are combined for a simulation experiment designed to estimate a linear metamodel for the response. Under certain conditions, this combined strategy is shown to be superior to the Schruben-Margolin strategy for a large class of experimental designs.
Degree
Ph.D.
Subject Area
Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.