Statistical Design of Experiments: Impact of missing and/or changing experiments


I am new to both this forum and the concept of statistical design so please, treat me like the novice I am!

OK, so I am hoping to carry out experimentation on a material. I am testing five different physical parameters of 4 different materials, all of the factors are numerical (i.e. no categorical factors). All of these factors are easy to control. I will be able meet the experimental requirements my software (StatEase version 9) asks me to except for one physical parameter.

Now, the problem is that for one of these physical factors, some of the materials can't have the same value due to a lack of suppliers. So some materials can meet all of the required values, whilst others can't. Inevitably, this means some of the required experiments will need to be either changed slightly or not completed altogether. I currently have it at about 85% of experiments able to be completed. Will a statistical design give me meaningful results or is it entirely worthless?

Below is a table (sort of) outlining a little better what I mean by some of the materials meeting some of the values:

ZSM 15 20 25 40
MOR 15 20 40
FER 20
BEA 15 25 40

The material ZSM has all required variations of this physical parameter.

Any help would be greatly appreciated!!



TS Contributor
The main issues that this will cause are imbalance and non-orthogonality. This may prevent you from analyzing the experiment using the default DOE analysis provided by StatEase, but should not prevent an analysis using regression or GLM. There may be a risk of aliasing (confounding) of higher order interactions caused by the loss of orthogonality.