topic Which experimental design is most appropriate when one of the factors is hard to change? in Discussions

Which experimental design is most appropriate when one of the factors is hard to change?

Zhangyue — Fri, 25 Apr 2025 13:58:04 GMT

Hello everyone,

I'm relatively new to DoE. Currently, I’m planning to run a screening design for 8 continuous factors, one of which is hard to change due to technical constraints.

Initially, I considered using a Definitive Screening Design (DSD) with a sorted run order. However, I’ve realized that this approach may not be appropriate when a split-plot structure is required.

Given this limitation, I was wondering whether optimal (custom) designs would be more suitable, and if they can accommodate split-plot structures? Are there any other design types you would recommend in this context?

Thanks in advance for your insights!

Re: Which experimental design is most appropriate when one of the factors is hard to change?

Mark_Bailey — Fri, 25 Apr 2025 14:18:47 GMT

A custom design is most appropriate in your case. You can change the optimality criterion through the platform red triangle menu to alias-optimal to better address this screening situation.

Re: Which experimental design is most appropriate when one of the factors is hard to change?

statman — Sun, 27 Apr 2025 16:01:36 GMT

Your situation is certainly a split-plot situation (restrictions on randomization). You will lose precision in detecting the effects of the whole plot factor (the hard to change factor), but you will increase precision for the remaining factors. Typically it is suggested to replicate so you can compare the effect of the whole plot factor against the whole plot by replicate for statistical significance. Ironically, the replication requires the whole plot factor be changed more than once, so that is a challenge. Often plotting the effect of the whole plot factor may provide insight into the whole plot factor's practical significance.