Optimization of large simulations using statistical software

Many applications utilize time and memory intensive simulations. We demonstrate a method of decreasing necessary space to 5% or less without losing time, compared to conventional ways, applicable to any statistical software. The idea is to perform simulations in portions. Statistics are calculated for each portion and stored with the current seed(s). After running all portions, the full set of statistics is analyzed. For the same starting seed(s) we obtain the same results for any number of portions. We give examples of programs for SAS and S-PLUS. This approach is important for performing large simulations on multi-user systems and small computers. © 2006 Elsevier B.V. All rights reserved.