FAQ
Uncontrolled variables
Original question from a viewer of the DoE Crashkurs für Experimentatoren YouTube video:
“Thanks for the video! In the example, only the controllable factors were entered. The uncontrollable factors were mentioned right at the beginning. How can the uncontrollable factors be taken into account when planning the experiment?”
(This highly informative presentation is also available in French—Introduction aux Plans d'Expériences, Portuguese—Aspectos fundamentais de Planejamento de Experimentos, Spanish—Aspectos fundamentais de Planejamento de Experimentos and English—DOE Crash Course for Experimenters.)
Answer:
I recommend that you first gather others on your R&D team who are familiar with your process and ‘brainstorm’ up a comprehensive list of potential variables. I find that a fishbone (aka Ishikawa) diagram works well as a template. See an example of a fishbone and my advice on how to run a brainstorm in the preface for the Stat-Ease Handbook for Experimenters. Next, recompile the resulting list of variables into the following categories:
Varied randomly in the experiment by design (I call these the “factors”)
Grouped due to being hard to change via a split plot
Held fixed throughout
Blocked (e.g., day-by-day)
Recorded if inconvenient to fix, e.g., ambient humidity—making it possible to regress out later as covariate(s) per this Stat-Ease program Help (for statistical details on this technique known as “Analysis of Covariance” (ANCOVA), see Section 15.3 of Analysis of Experiments, 8th Edition by Douglas Montgomery who says that “ANCOVA…is occasionally useful for improving the precision of an experiment”)
I thought about adding at the bottom of this list of variable types those that are “unknown and uncontrolled.” But, by definition, they cannot be identified. The best you can do is randomize the run order to 'wash them out,’ i.e., not bias estimates of main factor effects by unfortunate correlations.
