Add CDFs to Graph Builder - Page 2

BHarris · ‎10-01-2022

Recommend adding the ability to create CDF plots within Graph Builder.

TorP · ‎07-15-2024

That's a pity. Distribution graphs are the main reason we still use Minitab in our department.

Currently it still produces such graphs faster and easier than jmp.

(Minitab has much less evaluation options in the reports, but that's secondary if you first need the distribution graphs.)

Example w/ data from Big class.jmp

Minitab: Automatic settings, 3 clicks in dialogue window

jmp (no JSL, just std. options)

1. Graph Builder

Automatic settings

y-axis manually set to Probability Normal and adjusted to 0.01 and 0.99

2. Lifetime Distribution

Automatic settings. Nice graphics, but

- No panel option

- No info about distribution type (Normal, Weibull,...) and CI

- Legend missing when copied w/ "Copy Graph" (available in original report)

Summary: Such graphs in the graph builder w/ the above issues fixed would really be appreciated.

And it seems like most pieces are already available, but in diff. places/platforms?

hogi · ‎07-15-2024

It's even worse than that.
For Graph Builder, the cumulative probability has to be calculated manually (for every subplot!)
Cumulative Percentage is no option (!)

See: airline Delay (with some negative values)

Open( "$SAMPLE_DATA/Airline Delays.jmp" );
Graph Builder(	Variables( X( :Arrival Delay ), Wrap( :Month ), Overlay( :Airline ) ));

Minitab as reference - this is how the plot should look like:

JMP / Graph Builder via Cumulative Percent "trick":

pro: months are in the right order

con: WHOAH! COMPLETELY WRONG PLOT !!!

Under the line:
CUMULATIVE PERCENTAGE
IS NOT
CUMULATIVE PROBABILTY

hogi

"Add CDF/Cumulative Probability to Graph Builder Statistics":
tiny effort compared to e.g. 🙏 X group: restrict the values on the axis to the respective group

maybe we can get @XanGregg 's support to get the functionality into JMP19?

Sarah-Sylvestre

Hi everyone, thank you for the discussion and feedback! We are still looking into adding this in a future release, but our developers are currently focused on other features and functionality improvements in JMP 19, including other wish list requests. In the meantime, you can achieve similar results using Platform Presets (see steps below), however we understand this will not be sufficient in all cases.

Steps: Put a numeric variable on the X-Axis, select the line element, then change the settings in the line panel to what is shown in the image. Then make any other changes you see fit, save as a preset, and the next time you open graph builder, add a numeric variable to the X and select the preset you created.

hogi

@PatrickGiuliano

BHarris

@Sarah-Sylvestre : Is that behavior strictly equivalent to running a Fit Y-by-X where GB X = FYX Y and GB Overlay = FYX X? I just did a couple of spot checks and it looks the same, but I'm worried there might be some important differences, e.g. Add CDFs to Graph Builder - JMP User Community .

PatrickGiuliano

@BHarris, I would be skeptical that it's the same in every case even if appears the same in the cases that you have tested. My suggestion would be to use either Fit Oneway or Life Distribution if you want classically "correct" CDF plots, at least for the time being. See @hogi's commentary in that other post you referenced. If you have a specific example that you'd like us to look at more carefully we'd be happy to look at that, you can email us at support@jmp.com.

@hogi Thanks for making me aware. You may consider discussing this one (again if not already) with the R&D team next time you have a chance to make it to JMP Discovery. In most decisions to 'table' an idea for possible future implementation there are both prioritization considerations and technical constraints (which also affect prioritization).

Cheers, Patrick (JMP Technical Support)

hogi

@PatrickGiuliano , I agree: use Fit Oneway or Life Distribution if you want classically "correct" CDF plots.

DON'T USE Graph Builder / Cumulative Percent ! *)

The difference:
Cumulative PROBABILITY counts each row with THE SAME weight.
Cumulative PERCENT (Sum %) counts each row with the corresponding value (negative values count negative -> easy to show the difference)

*) Cumulative Percent is OK for some rare cases (1+2):
1) ALL VALES ARE POSITIVE
2) ALL VALUES ARE VERY CLOSE TO EACH OTHER (approximation holds: all values ~ THE SAME)

Examples with

Open( "$SAMPLE_DATA/Airline Delays.jmp" );

Life Distribution:

50% → 0

Analysis/Distribution:

Fit Y By X:

Graph Builder / Cumulative Percent:

50% → 100

Sarah-Sylvestre

Thanks Patrick! My suggestion was going to be the same but Hogi has provided some great explanation on the differences. I have passed all of that feedback on to the developers.