nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Use of Real instead of target factor values in DoE

JMPdiscoverer

Hey Jmp users,

I have one question regarding DoE. I have heard that it is better to use real instead of target values in DoE, but I still didn't understand how jmp can recognize the particular applied design. Some values might be far away from the target and if you have a small experimental domain, it will not be that easy to differentiate the matrix levels.

Any clarification ?

Thanks a lot!

Accepted answers

Dan_Obermiller

I will also try to take a shot at what you mean.

Suppose a design had two variables A and B and the runs look like this:

A B

100 1

100 2

200 1

200 2

150 1.5

Now in conducting these trials, on the first run, you did not achieve A=100. Instead, it was A=95. So, typically the recommendation is that you change the 100 to a 95 before analyzing the data. Why? Because you want to analyze what the actual data were, not what you intended.

If you are very far away from the design point, it will likely affect the analysis. It should! Your first question would be why are you so far away from the intended design point?

Note that @statman is correct on the coding, and if JMP created the design it adds a column property to automatically code to the -1, +1 scale. However, the coding is based on the design levels. If you are quite a ways off, the default JMP coding may not be appropriate and should be modified by editing that Coding column property.

All comments

P_Bartell

Can you please articulate how you are defining 'real' and 'target values?

statman

Not sure I understand your question, but here goes...

Actually it is better to use coded levels for factors in an experiment (the coding should be equipped-distant centered on zero, e.g., -1, 1 for 2-level designs). This "normalizes" the beta coefficients and allows for comparisons of the coefficients in subsequent analysis (otherwise the coefficients have to account for the actual units which may be quite different across factors). If you use "real" values for the levels, JMP is still going to code those for analysis (behind the scenes).

Dan_Obermiller

I will also try to take a shot at what you mean.

Suppose a design had two variables A and B and the runs look like this:

A B

100 1

100 2

200 1

200 2

150 1.5

If you are very far away from the design point, it will likely affect the analysis. It should! Your first question would be why are you so far away from the intended design point?

JMPdiscoverer

@Dan_Obermiller : This is what I meant.

More details:

DoE: Faced centered central composite design, 3 factors, 3 levels, 2 center points

Heat treatment :

x1 pH

x2 Temperature 58 60 62

x3 Treatment time 105 135 165

In one experiment, we needed to apply heat treatment at 62°C for 165 min. However, something went wrong, the target value of 62°C was not reached and we needed to adjust that manually which results in increased incubation time (182 instead of 165 min).

The treatment time was slightly different for all the runs (7-9 minutes difference) but for that particular point the difference was 17 minutes.

Should we adjust the coding column property ?

Thanks !

JMPdiscoverer

If you want to heat a solution for 30 minutes but you did it during 34 minutes.
Target = 30
Real = 34

Mark_Bailey

Update the Time value to 34 after the run is finished but before the regression analysis.

Dan_Obermiller

The answer to your question could be determined by leaving the coding property alone and looking at the VIF numbers, then changing the coding property and looking at them again.

In my opinion, I would leave the coding property alone and just perform the analysis. You have one observation that is far from target, but I would not adjust the coding for that one outlier. I would adjust it if you were consistently hitting that 182 mark instead of 165.