DOE question

[#1]

I am hoping that someone here can help me with this question about the use of blocking variable in DOE custom design. I have six factors, three at 3 levels, two at 2 levels, and one blocking variable in 2 levels. I am trying to figure out how to enter the information on the blocking variable in the custom design framework. When I try to add a factor as a block, it asks for the number of runs. It is not clear to me exactly what it means. If I choose the number of runs as 108 (3x3x3x2x2) then I get two blocks, but very large number of runs; but by choosing smaller number of runs, it give me more than two blocks. There is also a check box for grouping runs in blocks, and I was wondering if that is what I should be using.

I would appreciate any help in this matter. Thanks in advance.

[#2]

I'm not sure of your experimental question, but I have a suggestion:

Enter your factors into the Custom Design as usual and note the number of runs when you get to the Design Generation step. If you want half your runs in one block and half in the other, click the "Group runs into random blocks of size:" and enter half the desired number of runs.

For example, if your design has 16 runs, group runs into blocks of size 8. the generated table will have a blocking factor that divides the runs into two groups.

Your mileage may vary.

[#3]

Thanks for the suggestion. I have total of 48 runs; so when I chose blocks of size 24, the runs are now divided into two blocks. However, I run into the following problem: I have three response variables, and I am trying to optimize the factor levels that maximize overall desirability. Without the blocks, I get a prediction profiler that allows me to jointly optimize all responses. However, with blocks, I can only choose factors levels for one response variable at a time, which is not good for me.

Message was edited by: AlexG

[#4]

Hi Alex,

When you include the blocking factor in your design and your model, JMP changes the fitting personality in Fit Model to fit a REML model. In JMP 8, this results in fitting each of the responses separately, so you can't get a joint profiler.

You can save the prediction formula for each of your responses and then use the Profiler under the Graph menu with all three prediction columns and you'll have the joint profiler you're looking for.

JMP 9 makes this much better and offers the joint profiler in the initial report.

Jeff

[#5]

Hi Jeff,

Thank you very much. This is exactly what I need.

Very much appreciate your help.

[#6]

Hi, Jeff

I'm now using JMP8.
So, I often connct Profile using "Profiler under the Graph menu".
At that time Profiler of the Fit Model displays the confidence intervalat the same time, but "Profiler under the Graph menu" can't displays this. Is it possible for "Profiler under the Graph menu" to display confidence interval?

Thanks in advance.

KAI

[#7]

Hi Kai,

Yes, you can save the StdErr Pred Formula along with the Prediction Formula. Then use both of these new columns in the Profiler from the Graph menu and JMP will recognize the StdErr formula and offer to use it for confidence intervals.

Jeff

[#8]

Hi Jeff,

Thank you for your quick response.

I confirmed your way.
When the model doesn't include Random effect, this way work.
But when the model include Random effect, I can't save the StdErr Pred Formula.
Is it the limitation of JMP8?

Any way, version upgrade to the JMP9 is one of the good solution.

Thank you again.

Kai