Examination of a parsimonious factor screening technique:screening K = 2^{N - 1}- 1 factors in n runs

Abstract

In many experimental situations (particularly in computer simulation studies) a large number of potentially important factors exist. Because of time and budget limitations, it is imperative to screen these factors in order to identify a subset that should be subjected to more detailed' examination. This paper examines the performance of a factor screening technique when it is known that there is at most one active factor (i.e., a factor which has an effect on the response of interest). Although the technique works perfectly in the deterministic case, care must be taken when random error is present.

Keywords:

• variable selection
• supersaturated experiments
• detection of an important factor