topic Re: Difference between two Cpk analysis tool in Discussions

Difference between two Cpk analysis tool

mkhoddam — Fri, 08 May 2020 18:57:31 GMT

Hi all,

I have generated random data from Johnson Sl distr. and saved it in a column. Now, when I use these two Cpk analysis tolls, I get diff. results:

1. Analyze> Quality and Process> Process capability

2. Analyze> Fit > Capability analysis (I have tried all of the options for long term, moving range, ...)

For both method I chose Johnson Sl distr. and the same USL.

So, how come I get diff. results? Which one is better?

Thanks,

Mona

Re: Difference between two Cpk analysis tool

Georg — Sat, 09 May 2020 07:55:51 GMT

Dear Mona,

I could not reproduce your Problem, please see the following script, that generates that table,

and the two scripts generate the capability Reports (one via Process Capability, the other via Distribution).

But when running the script more times, all cpk's are different, because the data is newly generated by the Formula behind.

Maybe that's the Problem? You could simply delete the Formula to "fix" the data.

Best regards,

Georg

New Table( "Johnson SI Capability",
	Add Rows( 100 ),
	Compress File When Saved( 1 ),
	New Script(
		"Distribution",
		Distribution( Continuous Distribution( Column( :Johnson SI ) ) )
	),
	New Script(
		"Process Capability",
		Process Capability(
			Process Variables( :Johnson SI ),
			Moving Range Method( Average of Moving Ranges ),
			Individual Detail Reports( 1 ),
			Capability Box Plots( 1 ),
			Goal Plot( 1 ),
			Capability Index Plot( 1 ),
			Process Performance Plot( 0 )
		)
	),
	New Script(
		"Distribution of Johnson SI",
		Distribution(
			Continuous Distribution(
				Column( :Johnson SI ),
				Process Capability( Use Column Property Specs )
			)
		)
	),
	New Column( "Johnson SI",
		Numeric,
		"Continuous",
		Format( "Best", 12 ),
		Formula( Random Johnson Sl( 1, 1, 1 ) ),
		Set Property(
			"Spec Limits",
			{LSL( 1 ), USL( 3 ), Target( 2 ), Show Limits( 1 )}
		),
		Set Selected
	)
);

Re: Difference between two Cpk analysis tool

tonya_mauldin — Mon, 11 May 2020 12:38:01 GMT

You did not state what version you are using. Using JMP Version 15 or 16, the results should be the same. In version 14, I expect them to be different. In Version 14, Process Capability was updated and is using a different Johnson fitter. This Johnson fitter uses quantile matching rather than maximum likelihood by default. The distribution platform in 14 uses maximum likelihood. Beginning with Version 15, the distribution platform was updated and the Johnson fitter uses quantile matching so I would expect them to match.

Re: Difference between two Cpk analysis tool

Georg — Mon, 11 May 2020 13:08:20 GMT

Using JMP15 on Win10 (sorry, I forgot)

Re: Difference between two Cpk analysis tool

tonya_mauldin — Mon, 11 May 2020 14:21:33 GMT

If you want to post an example either here or to technical support, this issue can be researched further. Otherwise, we are just guessing at possible scenarios.

Re: Difference between two Cpk analysis tool

mkhoddam — Mon, 11 May 2020 16:20:57 GMT

@tonya_mauldin I am using JMP 14.3.0

Re: Difference between two Cpk analysis tool

mkhoddam — Mon, 11 May 2020 16:22:15 GMT

@Georg
No, the data is fixed. The difference is still there and the same all the time.

Re: Difference between two Cpk analysis tool

tonya_mauldin — Mon, 11 May 2020 16:37:03 GMT

I expect them to be different in 14.3. Distribution is using maximum likelihood to fit the Johnson distribution. Process Capability is using quantile matching to fit the Johnson distribution. Check your distribution parameters, I doubt they are the same! If the distributional fits are not the same, the capability statistics will be different.

Re: Difference between two Cpk analysis tool

mkhoddam — Tue, 12 May 2020 17:12:55 GMT

Thanks for your reply. I tried maximum likelihood method for Process capability but still getting diff. results.

Re: Difference between two Cpk analysis tool

tonya_mauldin — Tue, 12 May 2020 17:27:36 GMT

Yes, they will be different in all flavors of Version 14. In Version 15 and 16 they will be same as Distribution was updated to the newer method used in Process Capability.