What means the dot in Effect Summary

[#1]

Hello all,

I have performed a doe and would now like to evaluate the data.

I have an idea what the effect summary is.

I'm just confused because for three effects there is only a dot instead of the logworth value.

I also had to manually add the two effects "tool" and "source" (and also their interaction) via "add".

Is this due to the properties of the column? Have I defined something wrong?

Thanks for your answers in advance.

[#2]

Hi @fizzy,

Could you provide a little more details and context about your DoE ? 
What was the number of runs and model assumed ?

The dots for some of the effects in your model (with no value for Logworth or p-value) means that these effects can't be determined, for different possible reasons, because you don't have sufficient degree of freedom for their estimations.

You might not have sufficient data to estimate these effects, and/or not have data collected at the right locations of your experimental space to be able to determine them.

Another user faced a similar problem and you can find the responses given here : https://community.jmp.com/t5/Discussions/Missing-P-Values-in-Effect-Summary-OF-Fit-Model/td-p/271031

If you need more detailed/personalized help, sharing some of the info for the construction of the DoE (factors used, model, number of runs, ...) as well as the Analysis of Variance, Parameter Estimates, and Effect Tests tables could help us guide you.

I hope this first answer will help you,

[#3]

Hello Victor,

Thanks for your quick reply.

About my DoE:
I am investigating a traffic demand generation model.
In total there are:
5 routes
2 networks
4 vehicle models and
4 traffic generation tools.

In two tools, the traffic is generated stochastically. For these tools I did only one repetition (In a first test I determined the standard deviation and the relevant effect).

The other two tools are data based. I realized the test with the data based tools two times.

To make it more complicated I have 2 data sources, but I had to delete some factor level combinations due to insufficient data.

The difficult thing for me in the evaluation is that the tools have different settings, which makes the settings not comparable with each other.

In total, I have a trial size of 992.

[#4 · Accepted solution]

Hi @fizzy,

From the look on your tables, it seems you may have enough quantity of data, but not enough quality of data. For example, most of the interaction terms are zeroed or biased. Some further considerations :

• Have you taken in consideration all these interactions during the DoE creation ?
  If not, that may explain that you're not able to estimate correctly these terms, as some parts of the experimental space have not been explored.
• Are some factors covariate, or dependent of each others ? Do you have a "singularity" panel at the top of your model ?
  If yes, that may explain that some interactions parameters are biased or zeroed, as they can be estimated independently and precisely from other effect terms.

I hope this answer will help you,

[#5]

Hi Victor,

I have tried to create a full factorial experimental design, however, I had to delete some factor levels due external conditions.

Yes, the singularity details are displayed to me.

If I understood correctly, zeroed means that this parameter is redundant. How do I find out which parameter it is redundant with?

[#6]

Hi @fizzy,

Ok, this may explain why you're not able to estimate precisely or entirely some effects.

In order to better understand the "blind spots" in the design and parameters estimates, you can :

• Check the "Evaluate design" script (or create one by clicking on DoE, Design diagnostics, Evaluate Design, and from there specify your factors, response and model), in order to visualize the correlations between your effects though the "Color Map on Correlation", as well as looking at which terms are aliased (meaning correlated entirely or partially) through the "Alias terms" or "Alias Matrix" panels. You can also check the prediction variance profile, to know if there is a particular area in your experimental space where you have few points/bigger prediction uncertainty. The Estimation Efficiency may also provide interesting information related to parameters estimation uncertainty (through the fractional increase of confidence interval length compared to an ideal design).
• Check the "Singularity details" panel, as it will provide you some detailed informations about how some terms are aliased/correlated to each others, and providing the equation linking them.

“Zeroed” appears for terms involved in a linear dependency whose parameters cannot be estimated. "Biased" appears for terms where there are many estimates solutions that satisfy the least squares criterion. More info here : https://www.jmp.com/support/help/en/17.1/index.shtml#page/jmp/parameter-estimates-report-2.shtml# 

I hope this answer will help you,

[#7]

Thank you very much.

You helped me a lot!