Prediction and Tolerance Interval for the Profiler Tool

[Marco_]

The Profiler is a powerfull tool in JMP and widely used for the analysis and interpretation of Design of Experiment models.

To make prediction about future runs, the uncertainity of a model has to be considered. However, JMP is currently limited to confidence intervals, which only allow to make predictions about the range where the true mean falls into. Prediction and tolerance intervals on the other side would extend this so single runs or even for a poportion of a population.

Other software tools such as DesignExpert offer these options already for years and I guess many other JMP users would appreacitae such an option in one of the future JMP versions.

[Marco_]

Quick note for other JMP users. There is a workaround to visualize at least prediction intervals in the Prediction Profiler. 

Save the PredSE Formula and add the MSE to that formula (see picture below with the example of 9.1816). Select both columns (PredFormula and PredSE) in the Profiler and you get an approximate 95% PI.

You have to manual adapt each PredSE formula and that can be time-consuming in cases where you have multiple responses. 

[Marco_]

With the help of @ih it was possible to find an automated way to adapt all PredSE Formula columns at once.

Simply run the code and go to the Profiler where you can combine the PredFormula with the adapted PredSE formula in order to visualize the PI for each model instead of a CI.

Use the following code:

// Select all PredSE columns
dt = currentdatatable();
Col_List = dt << Get Column Names("String");

for(i = 1, i <= N Items(Col_List),i++,
If(Contains(Col_List[i], "PredSE"),
Column(Col_List[i]) << Set Selected(1);
);
);


dt << Get Selected Columns;



//'fix' each selected column formula by manipulating the formula as a string
for each( {col, c}, dt << Get Column References( dt << Get Selected Columns ),
		newf = regex(
			char(col << Get Formula),
			"(.* \* (\d+\.?\d*))",
			"\1 + \2"
		);
		//show(Parse(newf));
		Eval( Eval Expr(
			col << Set Formula( Expr( Parse(newf) ) )
		) );
);

[ih]

@Marco_ I believe you can export this formula directly from some platforms such as fit model (see below screenshot), but I agree it would be nice to see that as shaded regions right in the profiler.

View more...

[Marco_]

@ih 
Thanks for the info. What you need to export is the StdErr Pred Formula. However, both export tools (StdErr Pred Formula and Indiv Confidence Limit Formula) only gives you the Confidence Interval instead of the Prediction Interval. Thus, the manual adaption is necessary until JMP implements this.

[jszarka]

It would be good to have a save formula for tolerance intervals to mirror what we can already get with confidence and prediction intervals.

[Sarah-Sylvestre]

@Marco_ - Thank you for your suggestion! We have captured your request and will take it under consideration.

[Marco_]

@Sarah-Sylvestre 
Thanks a lot! I guess it would be highly appreciated by the cummunity if a save CI/PI/TI formula button is available in all model regression plattforms (including GLM, mixed model etc.)

[Sarah-Sylvestre]

@Marco_ Hello, this is a duplicate of a feature request that we are already reviewing.  We have documented your feedback and will be considering this for inclusion in a future release. 

[Marco_]

@Sarah-Sylvestre 
Thanks for the information. Could you send me a link to the feature request here in the community which address exactly my request?

[Sarah-Sylvestre]

@Marco_ The request that this duplicates is not from the community wishlist, but came in through our technical support team. The request was almost the same, as it asked for PI in the prediction profiler. We amended that request to include that you asked for TI as well, which is being considered.