topic Re: "Animals" Split Plot DOE in Discussions

"Animals" Split Plot DOE

frankderuyck — Thu, 08 Jun 2023 21:11:44 GMT

Hi, in JMP Sample Data there is a small data set "Animals" for illustrating a split plot design, see attachment.

I understand that nesting subject into species generates six whole plots for estimating random error. However it is strange to see that the 4 levels of the factor Season, to my opinion a "hard to change" factor is treated as a subplot effect? How is this Animals DOE created using Custom design? I can do this when using Species and Subject both as hard to change factors and Season as easy to change? Technically this works but I have problems with the time effect "Season" easy to change..

Re: "Animals" Split Plot DOE

ih — Mon, 19 Sep 2022 18:58:38 GMT

My take: Consider the relative difficulty of changing each column. I presume this was developed with some sort of tracking device attached to each of three foxes and three coyotes. Adding another tracking device on another animal is probably more expensive than just waiting a year to get information from all six animals, and adding a tracking device to another species of animal seems harder still. Here 6 total animals were involved. Waiting a year to capture data from those six subjects seems easy compared to capturing new subjects each season.

Re: "Animals" Split Plot DOE

frankderuyck — Mon, 19 Sep 2022 20:46:14 GMT

I also assume that animals are tracked. Season levels can't be changed from run to run so this is a hard to change factor and consequently each season is a separate plot which is not the case in this DOE; looks like all three factors here are hard to change, right?

Re: "Animals" Split Plot DOE

ih — Tue, 20 Sep 2022 12:30:45 GMT

I think season is easy to change in this case, just wait 3 months and it is changes. Perhaps the month the tracker was installed could be assigned, but that might mean bringing the team together multiples times to tag animals which could be cost prohibitive.

It is possible this design was not created with the DOE platform at all. The goal might have been to estimate variation which they could then be used in the power analysis of a future DOE, or it might have been a budget-constrained experiment.

I believe this mixed model is the same technique you would use to analyze a split plot design, but the same technique can be used other places where an individual or unit is tested with more than 2 treatments.

Re: "Animals" Split Plot DOE

statman — Tue, 20 Sep 2022 17:30:44 GMT

My take on the data set. IMHO, this isn't an experiment. This is a sampling plan that includes subjects nested in species and systematic sampling over seasonal time periods. Experimentation is the purposeful manipulation of factors to see their effect on response variables. I don't see any manipulation. Instead, this is directed sampling of the variation that occurs in nature. Also, I believe the numbering is wrong in the data table. Subjects should be 1-6. The same subject can't be both a fox and a coyote.

Split-plots are the family of experiments and are appropriate anytime there is a restriction on randomization in experimentation. There can be multiple splits and therefore multiple sub-plots.

Re: "Animals" Split Plot DOE

frankderuyck — Tue, 20 Sep 2022 17:28:26 GMT

OK Thanks a lot for these interesting comments, yes I believe this is an illstrating samling plan. Suppose we want to create an experimenta plan on this case, how to do this with Custom DOE?

Re: "Animals" Split Plot DOE

statman — Tue, 20 Sep 2022 17:36:35 GMT

Sorry, the question is non-sensical to me? Are you asking how do we create a data table to match how the data was collected? I would draw a sampling tree where the layers represent to components of variation and the numbering depicts the relationship(s) between the layers (e.g., nested, systematic or crossed). Then I would create the data table to match the tree.

Re: "Animals" Split Plot DOE

frankderuyck — Tue, 20 Sep 2022 18:40:57 GMT

In the JMP DOE tutorial the animal case is to illustrate a split-plot DOE, I understand that the data set rather is a sampling plan and does not really represent a split plot scheme. So there are three factors that influence the miles response variation (1) Season, (2) Species and (3) Subject. My question is how to generate a correct split-plot DOE using Custom design; what are the hard and easy to change factors? Goal is to get an interaction model with these 3 factors.

Re: "Animals" Split Plot DOE

statman — Tue, 20 Sep 2022 20:46:35 GMT

Sorry, I'm not trying to be difficult and just an FYI, I have no affiliation with JMP other than as a user. It will be impossible to get a crossed study with subject and species. These are hierarchical (nested). You can possibly get an estimate of the interaction between season and species.

Y= Season + Species + Season*Species + Subject[Species]

Re: "Animals" Split Plot DOE

frankderuyck — Wed, 21 Sep 2022 06:19:47 GMT

This is a very interesting discussion on a special case that is to my opinion not the usual way to set-up a spit-plot DOE. A split plot usually kicks off by specifying hard & easy to change factors and whole & sub plots; it is still not clear to me why Season is easy? One can't change a time effect level from run to run; wait 3 monthst to change? On the other hand one can change from fox to coyote in each run so rather easy to change? Confusing..

This case is one that shows up frequently in industrial studies i.e. where in treatments different product or supplier batches are used that can't be changed from run to run. Also in medical studies with groups of patients and healthy people, in each group there are different participants ("subjects"). Usually these studies take a long time where time blocking is necessary.

So how to tackle these studies with Custom design that can generate any DOE? Row - Column experiments with covariates?

Sincerely thanks for input, I appreciate!

Re: "Animals" Split Plot DOE

frankderuyck — Wed, 21 Sep 2022 12:07:24 GMT

IY= Season + Species + Season*Species + Subject[Species] is the required interaction model. How do you specify this in Custom design?

Re: "Animals" Split Plot DOE

statman — Wed, 21 Sep 2022 14:07:38 GMT

I'm sorry, this may not be the answer you want, but I would not use custom design to design the sampling plan. The customer design platform is meant to design experiments not sampling plans.

Also, I will add not all split-plots are created to handle difficult to change factors (these are just convenience split-plots. There are other times when you want to restrict randomization to improve the precision of the design (efficiency split-plots). I create these manually in JMP.

For more on this see

Box, G.E.P., Stephen Jones (1992), “Split-plot designs for robust product experimentation”, Journal of Applied Statistics, Vol. 19, No. 1

Re: "Animals" Split Plot DOE

frankderuyck — Wed, 21 Sep 2022 14:41:16 GMT

Is a DOE not meant to gather data, create a sampling plan to meet an objective? In this case assess whether there is a difference in run difference between a fox & coyote.

Re: "Animals" Split Plot DOE

frankderuyck — Wed, 21 Sep 2022 15:17:01 GMT

Suppose we're also interested in the effect of age & gender of both animals & all interaction effects and that #runs is restricted, some DOE support will be a help.

Re: "Animals" Split Plot DOE

statman — Wed, 21 Sep 2022 16:25:28 GMT

Yes all of these components can be assessed with sampling. DOE is when you manipulate factors in an “organized” fashion. There is no manipulation in the study you suggest.

Re: "Animals" Split Plot DOE

frankderuyck — Wed, 21 Sep 2022 17:27:26 GMT

Strange.. in the JMP DOE tutorial, this case study is an example for a split plot design?

Re: "Animals" Split Plot DOE

ih — Wed, 21 Sep 2022 18:53:11 GMT

Which DOE tutorial do you refer to? Could you include a link?

Re: "Animals" Split Plot DOE

frankderuyck — Wed, 21 Sep 2022 19:50:18 GMT

Here's the link

https://www.jmp.com/support/help/en/16.1/?os=win&source=application#page/jmp/simple-split-plot-or-repeated-measures-model.shtml

Re: "Animals" Split Plot DOE

frankderuyck — Wed, 21 Sep 2022 19:54:56 GMT

Also

https://www.jmp.com/support/help/en/16.1/?os=win&source=application#page/jmp/split-plot-design-example.shtml

Re: "Animals" Split Plot DOE

ih — Wed, 21 Sep 2022 20:20:36 GMT

Okay this makes sense, both of those links refer to analyzing data containing split plots, not designing the experiment itself. Split plots can show up in all sorts of places besides tables created with the custom DOE platform, and the animals data table is just such an example. To learn about designing experiments where different treatments are applied to the same experimental units I suggest setting the Animals data table aside and watching this talk by @DonMcCormack: