Controlled Sequential Bifurcation: A New Factor-Screening Method for Discrete-Event Simulation

Abstract

Screening experiments are performed to eliminate unimportant factors so that the remaining important factors can be more thoroughly studied in later experiments. Sequential bifurcation (SB) is a recent screening method that is well suited for simulation experiments; the challenge is to prove the "correctness" of the results. This paper proposes controlled sequential bifurcation, a procedure that incorporates a hypothesis-testing approach into SB to control error and power. A detailed algorithm is given, conditions that guarantee performance are provided, and an empirical evaluation is presented.

State-of-the-Art Review: A User’s Guide to the Brave New World of Designing Simulation Experiments

“…also discuss SB for random simulations, including a supply-chain case study. Wan et al (2003) propose a modification, called controlled sequential bifurcation, and provide proof of its performance under heterogeneous-variance assumptions. Other screening techniques with less restrictive metamodels are discussed by Campolongo et al (2000), Dean and Lewis (2004), Holcomb et al (2000a, b), Lin (1995), and Trocine and Malone (2001).…”

Efficient experimental design tools for exploring large simulation models

“…FFCSB is an extension of the CSB group screening procedure (Wan et al 2006) that adds a pre-screening stage to determine the sign of each factor's effect and pre-rank the effects. Rather than requiring the researcher to consolidate large sets of factors into a few composite categories (such as "industrial age" and "21 st century") that can be examined using a full factorial design, FFCSB allows the researcher to efficiently assess the importance of all individual factor effects.…”

“…This Phase 1 activity uses a single replication of an efficient fractional factorial design to sort the factors into two groups according to the sign of their estimated main effects. In Phase 2, one of two sequential group testing procedures is used to further delineate the factors: the procedure of Wan et al (2006) called Controlled Bifurcation (CSB), or an extended version called CSB-X ). In each step of CSB or CSB-X, the cumulative effect of a group of factors is tested for importance.…”

“…CSB is highly efficient for large-scale problems, particularly when a small proportion of the factors are truly important and these factors are clustered, since CSB can eliminate unimportant factors in groups. For example, Wan et al (2006Wan et al ( , 2009 demonstrate that the CSB and CSB-X procedures can save up to 75% of the runs required by one replication of a fractional factorial design, which in turn requires far fewer runs than a full factorial design. Another benefit of CSB and CSB-X is their validity when the error variances differ across different factor settings, as is commonly the case in simulation experiments.…”

“…In this paper, we describe new, stateof-the art experimental designs that allow analysts to overcome such difficulties with guaranteed correctness. For example, Controlled Sequential Bifurcation (CSB) is a new group screening method developed by Wan et al (2006) that produces dramatic efficiency gains in large-scale computational experiments when the important factors are sparse. As another example, the related technique Fractional Factorial Controlled Sequential Bifurcation (FFCSB; see Sanchez et al , 2009Oh 2007) extends the effectiveness of CSB in various, common modeling problems.…”