Our World Statistics Day conversations have been a great reminder of how much statistics can inform our lives. Do you have an example of how statistics has made a difference in your life? Share your story with the Community!
Choose Language Hide Translation Bar
Level I

custom design

i need help in analysis of this model


Re: custom design

You have given us no information with which to help you.


  • What is your experience with the design and analysis of experiments?
  • What is your experience with JMP, JMP custom design, and super-saturated designs?
  • What is the purpose of your experiment? What are your goals?
  • What have you tried already? Did anything work for you?
  • What specific problems did you encounter? What answers are you looking for?


I examined your design. It is a small design (12 runs) so it will likely have low power. It will find real effects only if the effects are very large compared to the random effect of the responses (4-fold to 5-fold larger). Is that result what you expect? Your factor ranges should be wide to produce a large effect. Is that what you did? It is a super-saturated design. You have 10 terms with estimability set to if possible (all possible two-factor interactions). You have 8 terms that are necessary to estimate (include a three-factor interaction). You do not have any terms to account for a potential non-linear response to changing these factors. This appears to be a  factor screening experiment. Is that your purpose?


I examined your analysis. You saved an analysis in which you attempted to estimate all 18 terms with 12 runs, so you have many singularities. Is that analysis what you intended? The methodology for the analysis of super-saturated designs would not recommend such an analysis.


I used stepwise regression for SLOPE. I manually chose terms to add, one term at a time. I arrived at this tentative model.


Screen Shot 2019-04-28 at 7.51.01 AM.png


Notice the large uncertainty in the model prediction. Your CV for SLOPE is about 23%. The largest, most significant effects, are 2 two-factor interactions. Is that reasonable? Does that surprise you?


I found no significant effects for LOD. Does that surprise you? Do you expect the LOD to be normally distributed as expected by this modeling method?


I found PVC:TCP to be the only significant effect for LOQ. Does that surprise you?


I found no significant effects for R.


I found the same effects for RESPONSE TIME as I did for SLOPE.


Other methods are available to select different models in such a case if these models are unsatisfactory. It is obvious that any of the models that you select for this data would need to be empirically confirmed.

Learn it once, use it forever!
Level I

Re: custom design

thank you very much for your help

1- Actually, i have a little experiance with JMP and DOE

2- I designed this model in order to inspect the main effects and ( any interaction if possible) of the three continuous factors PT,PVC:TCP and MT and one categorical factor ionophore of three levels (CX-BCD-none) on the mentioned five responses (slope,LOD,LOQ,R and RT)

3-the designed model is to reach the optimum factors combinationfor max desirability.

4- i set values of 0.6,0.05, 0.15, 0.05, 0.15 importance for the responses slope , LOD,LOQ,r and RT

5- is this model truly designed? and how can i usefully analyze this data especially i already did the twelve experiments??

6- really, i did not use a wide range for each factor.

finally i need your in details of using this model to examine

1- the main effects of each factor

2- interaction terms 

3- prediction formulae for each response

4- optimum factos combination for max desirability




Re: custom design

  1. Your answer is not much, but thank you anyway.
  2. What is your experiment about? What are your factors and responses about?
  3. So this experiment is meant to optimize the factor levels to obtain the desired responses? You do not seem to know much about the effects of these factors, so optimization seems like a leap. Your very economical screening design is unlikely to support the detailed model necessary for optimization.
  4. So LOQ and RT are 3 times more important than the slope, LOD, and r. Got it.
  5. I have no idea what you mean by asking if "this model is truly designed."
    1. Which model? The one that you specified for the custom design? The one that you fit? The one that I fit?
    2. What does "truly designed" mean?
    3. I assumed that the response data are real, so I also assumed that you already did the 12 runs.
    4. Your analysis begins with selecting the best model for the data that you have. My previous reply suggests some approaches that you might take for this part. The next step is to exploit the selected model and find the best factor levels, which would be a good use of the Predicition Profiler as I already demonstrated.
  6. That answer is unfortunate. It might be the reason that few effects were discovered. You missed an opportunity to make the most of an economical design (power).
    1. Your RMSE is about 17 for slope with a mean slope of about 45, for example. Does that sigma seem normal for your slopes?
    2. That sigma means that changing your factor levels must really change the slope by 4-fold or 5-fold to have a chance of detection. Would you expect that much of a change (effect) over these factor ranges?


Your four specific requests at the end of your reply can be answered quite well by the excellent documentation. Please see Help > Books > Design of Experiments Guide. This guide contains both general and specific information about the design and analysis of a variety of experiments. It illustrates each aspect with examples. The sample data sets are provided so that you can practice before analyzing your own data.


(Note that there is another guide for Fitting Linear Models that covers the details about the kind of model that you are using.)

Learn it once, use it forever!

Re: custom design

I just examined your data table again and found it quite messed up. That is, some of the meta-data about the factors and responses was inconsistent or wrong and these problems affect the modeling and inference. That result should not happen if you design the experiment with JMP. You should not change these properties unless you know what you are doing and have a good reason.


I fixed the problems. I re-ran my tentative analyses.* I saved the prediction formula for each tentative model. I profiled these models. I had to exclude LOQ because it never achieved the desired level.


I saved scripts for each step to the data table. The updated data table is attached.


* I manually selected tentative models using the stepwise platform and a generous alpha level of 0.15 to hopefully increase the low power, at the risk that some of these effects will not be confirmed.

Learn it once, use it forever!
Level I

Re: custom design

thank you 

i really need a tutorial videos to understand how to use JMP for DOE 

Level VI

Re: custom design

There are several DOE planning and analysis oriented On Demand and Live webinars available at no charge on jmp.com. These videos are largely focused on 'how to' in JMP...and NOT focused on teaching someone with little knowledge of the DOE problem solving process. So if you are someone looking for a 'how to' in JMP...then I suggest perusing the many offerings scattered around the Mastering JMP webinar series linked here:




If you are looking for a more comprehensive learning experience that teaches the DOE problem solving process AND 'how to' in JMP, SAS offers a variety of training experiences as well. You can peruse these offerings here:





Article Labels

    There are no labels assigned to this post.