Randomizing an experiment completely is often either impossible or prohibitively expensive. That's where split-plot designs can be valuable. Split-plot designs allow you to fix certain factors for several runs in a row. Within each block of runs (or whole plot), the factors that are hard to change remain fixed while the others vary at random from run to run. This makes the logistics of running a design simpler.
If you’ve had the opportunity to see JMP R&D Director Bradley Jones demonstrate how JMP Custom Designer handles split-plot designs elegantly and efficiently, you will want to read the paper he recently co-authored.
In the May 2007 issue of Journal of the Royal Statistical Society: Series C (Applied Statistics), Brad and Professor Peter Goos from Universiteit Antwerpen introduce a new method for generating optimal split-plot designs. In the paper, they demonstrate the usefulness of this flexibility with a 100-run polypropylene experiment involving 11 factors. In the experiment, they found a design that is substantially more efficient than designs produced using other approaches.