DoE center points problem

Question

Hello,

I'm currently designing a DoE with three different factors, two continuous (temperature, time) and one categorical (batch number with 4 different batches). I used the Custom design, D-optimal to design it and with some disallowed combinations. However, I would like to add some runs to my design that are center points and not only the suggested extremities, but it seems like it's not possible to do so. Is it because of the categorical factor? Is there a way to go around this problem? I’ve also tried to do an augmented design on my DoE but that didn't work either.

What happens if I just manually add a run myself in the data table, will the model use the additional run?

P_Bartell · Answer

The only suggestion I have in addition to everything @Byron_JMP mentioned (of which I concur) is wrt to adding the center points once you've decided the exact method and how many actual treatment combinations is to compare the original design with no center points to the design with center points in the Evaluate Design platform to see what the new design buys you in terms of power and other characteristics...might not be worth the trouble? Especially if you are in factor screening mode.

Mytom · Answer

Hi,

Thank you for your quick reply.

I further investigated the problem myself and observed that as soon as I choose the "Disallowed combinations" option, the center points are no longer available to pick. It doesn't matter if the actual combinations falls within the center point.

According to your reply my best option seems to be to manually add them myself (with the information you got).

Byron_JMP · Answer

An opinion based on less than a complete understanding of the underlying problem.

Just to make sure I'm on the same page as you, a center point run is the middle of each continuous factor.  In typical DOE coding (-1, 0, 1) a center point is where all the continuous factors are set to "0".  The problem is that categorical levels are coded 1, 2, 3..., there is no "0" center. When there are categorical factors, you'll have to decide which level you want to consider the "center", or you might want a center point run for each level of the categorical factor.

Since you have disallowed combinations, it's possible that the center points for all the continuous factors fall into the disallowed range?

Will the model work if extra runs are added? (...or,  Will the data still be analyzeable if I add runs that affect the orthogonality of the design?)

1. Adding, manually, center point runs will probably be OK.

2. The Fit Model platform that you'll use to do the analysis works great on both well-structured experiments and completely ad-hoc happenstance data.

Hope this helps.

Cheers,

B