FAQ
Advisability of minimal-run small central-composite design for optimizing a process
Original question from a College of Pharmacy Professor:
“Thank you for your very informative DOE FAQ Alerts. I am a research scholar in pharmaceutical sciences using Design-Expert software for my design and analysis of experiments. I have tried different response surface designs offered by your program. To save time, I am now considering a 15-run small central-composite design (SCD) for a three-factor optimization experiment. What are your thoughts on ‘cutting corners’ via SCDs?”
Answer:
As Design-Expert (DX) software warns upon selection of SCDs, these near minimal-run templates do not withstand outliers nearly as well as the far stouter conventional CCDs. In your case for three factors, the extra 5 runs for the full 20-run CCD will be a solid investment in reduced leverages and variance inflation factors (VIFs). See this for yourself by doing design evaluations with the outstanding tools provided by DX.
As I once heard George Box, the co-inventor of CCDs, say: “When in doubt, build it stout.” As you know being an experienced DOE practitioner, runs often go awry due to so many factor changes and response measurements. Therefore, we advise using SCDs only as a last resort.
By the way, for 6 or more factors, Design-Expert offers CCDs with minimum-run resolution V cores. This unique response-surface design option reduces the number of design points substantially without putting the experimenter in jeopardy of outliers. Nice!
- Mark
|