Abstract
When two-level designs are split into blocks of size two, the traditional method is to use mirror-image (foldover) pairs of runs, in which the signs of the factors are completely changed in the second run of the pair. This enables the estimation of all main effects clear of blocks, but of nothing else. We point out that by employing more runs, but still using blocks of size two, all the available effects can be estimated free of blocks. This is useful to know in situations where runs are cheap but the response varies over time, batch, or some other characteristic external to the experimental factors. In general, for a basic 2k-p design (p ≥ 0), (k - p) 2k-p runs are needed, with equal replications for all treatment combinations.
Additional information
Notes on contributors
Norman R. Draper
Dr. Draper is a Professor of Statistics. He is a Fellow of ASQC.
Irwin Guttman
Dr. Guttman is a Professor of Statistics.