Design of Experiments with hard-to-change factor and blocking

[#1]

Dear JMP community,

I am using the DoE JMP Custom Design Platform to build the following design:

* 4 continuous factors (2 hard to change and 2 easy to change)

* 1 blocking factor

* 18 experiments should be conducted in two days (9 in the first day and 9 in the second)

A possible effect between both days should be modelled with the blocking factor by constructing two blocks with 9 runs per block.

Now, if I click through the menu I got a design with 18 experiments separated in two blocks. However, my problem is that the block column is also randomized. Since the first 9 experiments must be done in day 1 and the second in day 2, the blocking column must be:
1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2 

How can I tell JMP to get this blocking column?

Thanks for your help in advance!

Best,

Mike

[#2 · Accepted solution]

@mike2 

The hard to change factors are going to take precedence over the easy to change factors, of which the blocking factor is one. A workaround is to

1. Start with the two hard and two easy to change factors.
2. Add day as a Very Hard to change categorical factor with 2 levels.
3. Build your model. Remove all interactions involving day.
4. Set Number of Whole Plots to 2.
5. Create your design

You'll get a warning message saying you should add another Whole Plot. Unless you want to test for statistical differences between days, you don't have to. If you do, the problem of design and analysis becomes more difficult since day as a factor would most likely have to be treated as a random effect (fixed effects, like blocks, have a finite, typically small, number of levels).

You'll have to make a few changes when you go to analyze the data.

1. Remove Whole Plots&Random from the model. It's confounded with day.
2. Ignore any statistical test involving day. It will be using the incorrect error component (denominator) for the F test.

Hope this makes sense. A few years ago, I recorded a Mastering JMP session that touch on blocking and split plot designs. There are a number of additional links at the bottom of that page as well.

[#3]

Hi Don,

thanks for your quick answer and the nice work around. However, I still have an open question:

I would like to analyze the resulting design using the DoE / Design Diagnostics / Evaluate Design menu. Following your advice, I should also drop the Whole Plots factor for doing this, since its confounded with the day factor. But, unfortunately JMP forces me to include also the very hard to change factor. Do you also know here a workaround?

Thanks in advance!

Mike

[#4]

If I understand correctly, day is used as the very hard to change factor and left in the model if you want to estimate the difference between the two days. It's just that you don't want to have both Whole Plots&Random and day in the model and you can't perform the proper statistical test for the day difference with just two observations (i.e., two days).