Metamodel estimation using integrated correlation methods

ABSTRACT

This paper develops a generalized approach for combining the use of the Schruben-Margolin correlation induction strategy and control variates in a simulation experiment designed to estimate a metamodel that is linear in the unknown parameters relating the response variable of interest to selected exogenous decision variables. This generalized approach is based on standard techniques of regression analysis. Under certain broad assumptions, the combined use of the Schruben-Margolin correlation induction strategy and control variates is shown to give a more efficient estimator of the metamodel coefficients than each of the following conventional correlation-based variance reduction techniques: independent streams, common random numbers, control variates, and the Schruben-Margolin strategy.

REFERENCES

• 1: Cramer, H. (1970). Random Variables and Probability Dis t ribu ~ions. Cambridge Univrsity Press, Cambridge.
• 2: Kleijnen, J.P.C. (1974). $=a~istical Techniques i__nn Simulation, Part I. Marcel Dekker, New York.
• 3: Myers, R.H, (1976), Res~ Surface Methodology, Allyn and Bacon, Boston, Massachusetts.
• 4: Nozari, A., Arnold, S.F., and Pegden, C.D. (1984). Control variates for mult ipopulatlon simulation experiments, lie Transactions 16, 159-169.
• 5: A. Nozari , S. F. Arnold , C. D. Pegden, Statistical analysis for use with the Schruben and Margolin correlation induction strategy, Operations Research, v.35 n.1, p.127-139, Jan.-Feb. 1987 [doi>10.1051/ro:2001107]
• 6: Schruben, L.W. andMargo lln, B .B. (1978). Pseudorandom number assignment in statistically de signed s Imula t ion and dis t ribu tion sampling experiments. Journal of American Statistical Association 73, 504-525.
• 7: Schruben, L.W 1979. Designing correlation induction strategies for simulation experiments, Chapter 16 in Current Issues i_n_n Simulation. Academic Press, New York, 235-255.
• 8: Seber, G.A.F. (1977). Linear Regression Anal~sis. John Wiley & Sons, New York.
• 9: Tew, J.D.(1986).Me tamode I estimation under correlation methods for simulation experiments ~ Unpublished Ph.D. dissertation, School of Industrial Engineering, Purdue Unlverslty, West Lafayette, l~dla~a.
• 10: Tew, J. D. and Wilson, J.R. (1987).V~lidation of correlation-induction strategies for simulation experiments. Technical Report VTR 8701, Department o~ Industrial Engineering and Operations Research, Virginia Polytechnic Institute and State University, Blacksburg, Virginia.
• 11: Sekhar Venkatraman , James R. Wilson, Using path control variates in activity network simulation, Proceedings of the 17th conference on Winter simulation, p.217-222, December 11-13, 1985, San Francisco, California, United States [doi>10.1145/21850.253156]
• 12: Wilson, J. R. and Pritsker, A. A. B. (1984}. Variance reduction in queueing simulation using generalized concomitant variables. Journal of Statistical Computation and Simulation 19, 129-153.

INDEX TERMS

Classification:
  I. Computing Methodologies
  I.6 SIMULATION AND MODELING

General Terms:
Design, Experimentation, Measurement, Performance, Theory