cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Check out the JMP® Marketplace featured Capability Explorer add-in
Choose Language Hide Translation Bar
schilakwad
Level I

How do Space Filling designs automatically generate a run number? Also, can someone please explain the "discrepancy" and "MaxPro" statistics?

Hi, 

 

I’m a biological scientist aiming to physically execute a space filling DoE using Fast Flexible filling to test 11 factors (10 continuous and 1 categorical) to characterise an enzymatic reaction. When I generate the design I see that JMP has populated the run number box with 110 runs. If I add another factor, this goes up to 120 and if I reduce a factor it decreases 100 and so on. Can anyone please explain how this run number is generated? Does it represent a minimum number of runs required to identify significant factor interactions since it changes with each change?

 

Additionally, I have also found it challenging to understand the meaning behind the values of the “discrepancy” and “MaxPro” statistics in determining whether my design is sufficiently powerful. For example, my discrepancy statistic is 0.111634 and my MaxPro statistics are between 45 and 48. Is there rules of thumb I can follow when it comes to interpreting these values?

 

In the past, I’ve executed a large space fill as we weren’t constrained when it comes to throughput. However, for this experiment I’d like to know the least number of experiments I need to characterise my system.

 

Thank you. 

1 REPLY 1

Re: How do Space Filling designs automatically generate a run number? Also, can someone please explain the "discrepancy" and "MaxPro" statistics?

Space-filling designs were developed to provide sufficiently dense data for estimating a flexible interpolator such as a Gaussian process for a highly non-linear response in a computer experiment. These designs have also been successfully used in some cases of formulations to create a mixture design. The search algorithms used by space-filling designs are described here. The algorithm and its options determines the number of runs.

 

We generally use custom design for fitting a linear statistical model. (Note that it might include terms for non-linear effects in the response.)

 

On the other hand, if you want to estimate the parameters for a non-linear function as is often the case with kinetics problems, JMP also has a non-linear design platform. You can read about it here.

 

Why did you choose a space-filling design in the first place?