I want to perform a fractional factorial design and I have 5 factors one of them is Hard to change. Using Custom design I got a table with 18 experiments with 4 whole plots. I can not do all the experiments in one day and I am not sure if I will have minor changes from day to day experiments.
I am wondering if it is possible to do a whole run/ Day?
In my previous Designs I often use Blocking since I did not have any Hard to change factors but in this situation, I do not know what to do
Thank you in advance
You have the possibility to use blocking when you are defining your factors.
When none of the factors are hard or very hard to change, the blocking can be made in the "Design generation" part. In your case, this option won't appear here (only the number of whole plot), but in the options of "Add Factor", you have the possibility to add a blocking factor, with a defined number of runs per block (to take into account day-to-day variation in your example).
Having this blocking factor possibility and the Hard to change factor, JMP Custom design platform should be able to show you a suitable design.
Hope this answer will help you,
Questions/Comments for you:
1. Are all of the factors continuous?
2. What is the model you are predicting? ( What terms do you want to estimate?)
3. JMP will suggest you do replicates of the WP so you can get estimate of the WP error and thus a statistical test. You need to think about this and the implications. This will mean you have to change the hard to change factor multiple times (which is what you are trying to prevent). I have some philosophical thoughts here...If the factor is very hard to change, you wouldn't want to manage a process with that variable. You would still be interested in knowing how large an effect it has, but that would be ultimately to set it at some constant "improved value". Also varying a hard to change factor in an experiment may introduce sources of variation well above the normal random variation in the process, so be very cautious of statistical tests as this may bias the MSE.
4. Now the day-to-day variation requires some additional thought. First, if you know (or are suspicious) you have day-to-day variation, do you know specifically why (what is actually changing that has an effect on your response variable(s))? Directed sampling is very useful in these situations. Are you concerned about robustness of your model to day-to-day variation?
Some follow=up thoughts:
1. If you are screening, then why do you have a discrete variable with more than 2 levels? Pick the extremes of the three levels and test those first.
2. If you don't know whether you have day-to-day variation, perhaps you should do directed sampling PRIOR to dong any experimentation. If you don't understand what the noise is doing, it will be very difficult to use the conclusions from your experiment to extrapolate into the future.
3. I already commented on the hard-to-change factor.
Can you upload your data table so we can look at what you've got and how you did your analysis?