FAQ
Design evaluation reveals high leverage points—should I be worried?
Original question from a Group Quality Control Trainer:
“I am conducting a three-factor, central-composite-design (CCD) response-surface experiment to determine the optimal level of addition and efficacy of an enzyme cocktail for my client. The enzyme supplier has recommended a specific level of addition for the cocktail they are supplying. My client wants to assess whether to follow the supplier's recommendations or find an optimal level independently. I am utilizing Design-Expert® software for this purpose, as indicated by the attached design file.
“I have noticed that the leverage for some of the runs in the design is quite high, ranging from 0.81 to 0.88. According to your Program Help, high-leverage runs should be replicated to mitigate their impact on the final prediction and optimization.
“Could you please review my design and provide your advice? Is my concern valid? If so, how can I reduce the influence of these high-leverage runs without increasing the design size—I am already at my limit. Any additional advice you can offer would be greatly appreciated.”
Answer from Stat-Ease Consultant Joe Carriere:
“You mentioned you can't add any more runs, so it is a bit of a moot point, but I don't think those leverages necessarily mean you need to replicate them if you run them with care. They are not that close to 1. Also, those leverage values are for the full quadratic model. It may be that some of those model terms are not required, so the actual leverages may decrease.”
I totally agree with Joe. By the way, this user blocked his design into 5 days, which increased the leverage. With the same number of runs in an unblocked design, the highest leverage drops to 0.6720. (However, I definitely prefer blocking over not doing so due to concerns about leverage.)
For more information on leverage, see this Stat-Ease program Help page. I am working on a blog post that provides a case study with one alarmingly large leverage (exceeding 0.99!). Stay tuned to the Stat-Ease home page for this post.
(Learn more about leverage by enrolling in the next Modern DOE for Process Optimization public workshop.)
