A Pseudo-Random Number Assignment Strategy for Simulation Experiments

Simulation models are often used to select alternatives to system designs. A model of a simulation model, referred to as ametamodel, can be used to explore the functional relationships between the mean response of a simulation model and a set of simulation input variables. The estimated metamodel is then used to predict the system response for speci ed sets of input variables and to identify the optimal operating conditions for the input variables. Previous work in this area has included Schruben and Margolin (1978), Hussey, Myers, and Houck (1987), Law and Kelton (1991), Sargent (1992), Donohue, Houck, and Myers (1992), Song and Su (1996), and Barton (1997). This paper has proposed an e ective pseudo-random number assignment strategy (called Q-MB strategy) for estimating the second-order metamodels for 3-FAC designs. We have shown that the proposed strategy outperforms well-known existing strategies, including the IR, CR, and modi ed-AR strategies in terms of A-optimal and D-optimal criteria. We can summarize the idea behind the Q-MB strategy as follows. First, all design points are partitioned into as many orthogonal blocks as possible such that no metamodel coe cients are confounded. These blocks are then arbitrarily grouped into pairs, each of which contains two blocks. Finally, we assign a common stream set to the design points in the same block, antithetic stream sets to the two blocks in the same pair, and independent stream sets to blocks in di erent pairs or to an unpaired block if there is one.