cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Choose Language Hide Translation Bar
CB21C68
Level I

GroupBy in Variability Charts

Is there a way to group by in Variability charts?

 

If i have several samples I want to group by the sample and draw a cell means and bars for each sample.

 

Here's an example that JMP generates:

 

Capture1.PNG

 

It's just plotting one cell mean, but I want to draw separate cell means grouped by each sample individually.

 

Seems like it's impossible to do that right?

 

This is what I want to be able to do:

 

Capture2.PNG

 

*EDIT:

 

I know I can do something like this using the bivariate chart:

 

Capture5.PNG

 

But what if I want  to do something like this:

Capture4.PNG

 

Seems like JMP is only allowing me plot this:

 

Capture3.PNG

4 REPLIES 4
statman
Super User

Re: GroupBy in Variability Charts

Welcome to the community.  It would be helpful if you attached your data table (this way we can play around to find what you're looking for.) and let us know what version of JMP you are using.  In any case, have you tried Graph Builder?  This platform is extremely flexible.  If you use the By option in Variability plots, you will get multiple charts.  Also, right click is your friend.  Often right clicking on the graph or axis provide a list of many possible customizations.

"All models are wrong, some are useful" G.E.P. Box
CB21C68
Level I

Re: GroupBy in Variability Charts

Hello Statman.

 

I'm on jmp 13.2 

I don't think graph builder is flexible enough to do what I want.

I've attached a file that contains some example data I'm working with, with all of the details sanitized. 

 

The script for plotting is below:

 

Variability Chart(
	Y( :Measurement % ),
	X( :StressTest, :StressContition, :Subsample, :Days ),
	Max Iter( 100 ),
	Conv Limit( 0.00000001 ),
	Number Integration Abscissas( 128 ),
	Number Function Evals( 65536 ),
	Analysis Type( "Choose best analysis (EMS REML Bayesian)" ),
	Connect Cell Means( 1 ),
	Std Dev Chart( 0 ),
	Points Jittered( 1 ),
	SendToReport(
		Dispatch(
			{"Variability Chart for Measurement %"},
			"Variability Chart",
			FrameBox,
			{Grid Line Order( 5 ), Reference Line Order( 6 ),
			Row Legend(
				Sample,
				Color( 1 ),
				Color Theme( "JMP Default" ),
				Marker( 0 ),
				Marker Theme( "" ),
				Continuous Scale( 0 ),
				Reverse Scale( 0 ),
				Excluded Rows( 0 )
			)}
		)
	)
)

 

Capture6.PNG

 

The graph it's generating is exactly what I'm looking for in terms of the visual aesthetics, except that I want it to be able to draw the range bars and the connected cell means for each sample individually (i.e. for sample 1 & 2).

 

If it's not possible to do what I want using the variability gauge chart, then how do I replicate the same visual aesthetic of multiple charts side by side with one axis on the left side with all of the same labels on the bottom using something else?

 

I'm open to any suggestions.

 

jthi
Super User

Re: GroupBy in Variability Charts

One example what could be done in Graph Builder:

jthi_0-1639329518463.png

 

-Jarmo
statman
Super User

Re: GroupBy in Variability Charts

Still not sure what you mean by range bars?  The way you have displayed the data is as individual data points.  There are no ranges.  If you want the range of one of your components, you need to leave it out of the plot (if you want the range of subsamples, leave it out and select the box plots option for variability charts, see below)  If you leave sample out of the variability plot and do the plots by sample, then you can get the cell means for each sample (I added those scripts).  Also, the arrangement on your variability plot does not look hierarchically correct?  Days can't be nested in sample?  But I don't know the points you are trying to make with your plots.

example_data - Variability Chart of Measurement %.png

 

 

 I don't think JMP 13 has the capabilities of graph builder that are available with JMP 16.  Here is an example of JMP 16 graph builder output:

example_data - Graph Builder.png

"All models are wrong, some are useful" G.E.P. Box