cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Browse apps to extend the software in the new JMP Marketplace
Choose Language Hide Translation Bar
r30363
Level II

how to make cumulative probability plots in JMP?

How can I make cumulative probability plot like the one below? Appreciate your help.

r30363_0-1658152068599.png

 

 

36 REPLIES 36
hogi
Level XII

Re: how to make cumulative probability plots in JMP?


@hogi wrote:

@Mauro Thanks a lot for the Kudo - an also for the other 3.
With 4 Kudos this feature should be kind of save to appear in a future Jmp version  


Hm,Summary Statistics: add Cumulative Probability is still in Acknowledged state.

Seems that more Kudos are needed to trigger the development.

hogi
Level XII

Re: how to make cumulative probability plots in JMP?


@hogi wrote:

 


Hm,Summary Statistics: add Cumulative Probability is still in Acknowledged state.

Seems that more Kudos are needed to trigger the development.


Good news:

there is even an older wish toAdd CDFs to Graph Builder from 2022 - and this wish got in the queue for a future Jmp release

 

Great!

hogi
Level XII

Re: how to make cumulative probability plots in JMP?

Additional advantages of using the built-in Summary Statistics of Graph Builder:

  • For Group X / Group Y / Wrap / Page subplots, the calculation is always correct - no need to split the calculation by manually  adding columns as argmuments of the Col Aggregations:
Col Rank( var, groupbyCol, GroupXCol, GroupYCol, WrapCol, PageCol, excluded());
Col Number( var, groupbyCol, GroupXCol, GroupYCol, WrapCol, PageCol, excluded())

 

  • If a Local Data Filter is activated for the plot, GraphBuilder Summary Statistics restricts the calculation automatically to the filtered rows, the Formula in a Transform Column ignores the LDF:   Transform Column: Bug with excluded rows 
hogi
Level XII

Re: how to make cumulative probability plots in JMP?

Till cumulative probability will be available via Summary Statistics in Graph Builder (Add CDFs to Graph Builder ) ....

Plot quantile by actual can be used as a workaround:

It's similar to CDF - but with access to the data points.

 

disadvantages:
- many black points!

- no split plots

- ...

 

hogi_0-1695893723741.png

 

Names Default to Here(1);
dt = Open( "$SAMPLE_DATA/Big Class.jmp" );	
Oneway(
	Y( :weight ),
	X( :age ),
	All Graphs( 0 ),
	Plot Quantile by Actual( 1 ),
	CDF Plot( 1 ),
	Local Data Filter( Close Outline( 1 ) )
)

 

hogi
Level XII

Re: how to make cumulative probability plots in JMP?

[acually, intended to be a reply to 
https://community.jmp.com/t5/Discussions/how-to-make-cumulative-probability-plots-in-JMP/m-p/527722/...

... hm, at this position it's more or less a replica o the previous post - which triggered me to understand the meaning of


@r30363 wrote:

 Left side  Y-axis with standard deviation, right side Y-axis with cumulative probability


:]

Hi @r30363 , via OneWay, I could offer left side cumulative probability -  and right side standard deviation (not 1,2,3, but 0.67 = 75%, 1.28=90%, ...)

hogi_0-1697057994140.png

 

Names Default to Here(1);
dt = Open( "$SAMPLE_DATA/Big Class.jmp" );
Oneway(
	Y( :weight ),
	X( :sex ),
	All Graphs( 0 ),
	Plot Quantile by Actual( 1 ),
        Normal Quantile Label( 1 ) // by default
)

 

hogi
Level XII

Re: how to make cumulative probability plots in JMP?


@r30363 wrote:

 if I want to do Left side  Y-axis with standard deviation, right side Y-axis with cumulative probability, how to add a right Y in graph builder


The left and right Y axis are not linked. So, after changing one of them one has to click on "reset Normal Quantile Scale" - or is there a Plot Scale Change Handler in Jmp? *)

hogi_0-1697060761062.png

 

 

 

Names Default To Here( 1 );
dt = Open( "$SAMPLE_DATA/Big Class.jmp" );
New Column( "normal Quantiles" );

Graph Builder(
	Transform Column( "Cumulative Probability", Formula( Col Rank( :height, :sex ) / (Col Number( :height, :sex ) + 1) ) ),
	Show Control Panel( 0 ),
	Variables( X( :height ), Y( :normalQuantiles ), Y( :Cumulative Probability, Position( 1 ), Side( "Right" ) ), Overlay( :sex ) ),
	Elements( Points( X, Y( 2 ) ), Smoother( X, Y( 2 ) ) ),
	SendToReport(
		Dispatch(
			{},
			"Cumulative Probability",
			ScaleBox,
			{Scale( "Normal Probability" ), Format( "Best", 12 ), Min( 0.01 ), Max( 0.99 ), Inc( 0.1 ), Minor Ticks( 1 )}
		)
	)
);
New Namespace(
	"reset"
);
reset:NormalQuantileScale = Function( {},
	myScript = (Current Report() << XPath( "//AxisBox" ))[3] << get script();
	myMin = Normal Quantile( myScript["min"] );
	myMax = Normal Quantile( myScript["max"] );
	(Current Report() << XPath( "//AxisBox" ))[2] << Min( myMin ) << Max( myMax ) << inc( 0.5 );
);
Current Report()[List Box( 2 )] << append( Button Box( "reset Normal Quantile Scale", reset:NormalQuantileScale() ) );
reset:NormalQuantileScale()
hogi
Level XII

Re: how to make cumulative probability plots in JMP?

asking for a friend/colleague:

 

With the current version of Jmp, is there an easy way to generate a split grouped probability plot?


I can generate such a plot via Graph Builder, but I have to implement a hand-made functionality which re-generates the Transform column once the user selects different columns or different entries in the Local Data Filter - and it's extremely slow! 

 

hogi_4-1697259519716.png

 

Accepting that the graph won't be as interactive as what a Jmp user is used from Graph Builder, one can use Live Distribution/Compare Groups - with a By variable. It can be much faster than Graph Builder - but there are cases where you have to wait literally "hours" till the plot shows up [see below].
The original report doesn't look like expected - but with several commands to the report (*) the plot gets almost as nice as in Minitab.

The benefit compared to the Graph Builder approach: more Distributions/scales to select from

The disadvantages compared to Minitab:
- not stable
- very slow

 

hogi_3-1697259294008.png

 

Open( "$SAMPLE_DATA/Cars.jmp" );Graph Builder(
	Transform Column(
		"CDF",
		Formula(
			Col Rank(
				:Wt,
				:Doors,
				:Year
			) / (Col Number(
				:Wt,
				:Doors,
				:Year
			) + 1)
		)
	),
	Variables( X( :Wt ), Y( :CDF ), Group Y( :Year ), Overlay( :Doors ) ),
	Elements( Line Of Fit( X, Y, ), Points( X, Y, Legend( 14 ) ) )
);

myld = Life Distribution(
	Perspective( Compare Groups ),
	Y( :Wt ),
	Grouping( :Doors ),
	Confidence Interval Method( Wald ),
	Select Distribution( Distribution, Normal ),
	Select Scale( Normal ),
	<< Show Statistics( 0 ),
	By( :Year)
);
(myld  << XPath("(//OutlineBox[text()='Statistics'])")) << Visibility( "Collapse" );
(myld << xpath("//OutlineBox[text()='Individual Group']")) << Visibility( "Collapse" );
((myld << xpath ("//RadioBox")) << parent) << Visibility( "Collapse" );
(myld << xpath("//OutlineBox[text()='Compare Distribution']//IfBox"))<< Visibility( "Collapse" );
//(myld << xpath("//OutlineBox[text()='Compare Distribution']")) <<Set Title( "" );
// current Report()[ListBox(1)] << Set Horizontal( 1 );
(myld << xpath("//OutlineBox")) <<Set Title( "" );
myABS = (myld << xpath ("//OutlineBox//AxisBox"));
Transform Each({ABList},myABs[1::Nitems(myABs)-1],ABList[2]) << Label Row(
				{Automatic Tick Marks( 0 ), Show Major Ticks( 0 ),
				Show Minor Ticks( 0 ), Show Major Labels( 0 )} );
myTEBs = (myld << xpath ("//OutlineBox//TextEditBox"));
Transform Each({TEBList},myTEBs[1::Nitems(myTEBs)-1],TEBList[2])<<Set Text( "" ) ;

 

 

hogi
Level XII

Re: how to make cumulative probability plots in JMP?

At the moment, Life Distribution is definitely the most versatile/elegant way in Jmp to create a probability plot.

The confidence intervals look very nice.

 

On the other hand, it takes a while for Life Distribution to open - depending on the number of data points, it can take hours:
TS-00122043

hogi_1-1705614388557.png

BHarris
Level VI

Re: how to make cumulative probability plots in JMP?

With a little work you can almost reproduce what you get from the Fit Y by X CDFs -- you just have to add a line and switch its connection to "Step" and Summary Statistic to "Max".  However unlike a real CDF you don't get the vertical lines at the endpoints going to 0 and 1.0.  I'm hoping this will make it into Graph Builder as a first-class citizen someday.

hogi
Level XII

Re: how to make cumulative probability plots in JMP?

With a little work?