B0523
Title: General Stratum Orthogonal Arrays (GSOAs): Constructions and properties
Authors: Ulrike Groemping - Berliner Hochschule fuer Technik (Germany) [presenting]
Abstract: Stratum Orthogonal Arrays (SOAs) and their generalization to General Stratum Orthogonal Arrays (GSOAs) were proposed for experimenting with quantitative variables in computer experiments. A review of SOA constructions, using a unifying notation in the form of simple equations, has been recently given. (G)SOAs can be latin hypercube designs or have much fewer distinct values for each experimental variable. Designs for computer experiments are commonly requested to have good space-filling properties. Many different metrics for assessing space-filling can be found in the literature. An entirely different approach has been presented recently: a stratification pattern derived from a generalized word length pattern captures the stratification properties of (G)SOAs, which can also be considered an aspect of space-filling. The purpose is to present general ideas behind (G)SOAs and their constructions and explain a recent stratification pattern. The R package SOAs allows us to apply all the constructions and assessments. Note that SOAs were initially named ``strong orthogonal arrays'' although they usually have very low orthogonal array strength; this contradiction is the reason why this author has repurposed the acronym to ``stratum orthogonal arrays''.