How to set confidence interval's method to likelihood instead of Wald for the Distribution Profiler in 'Fit Life by X'

[#1]

Hi everyone,

This seems to be a simple question, but I cannot find the solution.

In Analyze toolbar > Reliability and Survival > Fit Life by X, there is a distribution profiler.

My problem is that the confidence intervals (CIs) shown here are using Wald's method, even when I selected the likelihood as the preferred CI method.

Unfortunately, I also cannot find the button to change from Wald to likelihood CI.

So, my question is simply "how to change the CI's method from Wald to likelihood in the distribution profiler?"

Thank you so much in advance for your help.

Note: I am currently using JMP17.

Please find below the picture that summarize my question.

View more...

For reference, I attached my dummy data and the code to generate the plots.

Fit Life by X(
	Distribution( Weibull ),
	Nested Model Tests( Location and Scale ),
	Y( :ResponseVariable ),
	X( :Category ),
	Censor( :Censor ),
	Censor Code( 1 ),
	<<Fit Weibull,
	Use Transformation Scale( 0 ),
	Show Overlay by Levels( 1 ),
	<<Set Scale( Weibull, Simultaneous, Show Nonparametric CI( 0 ) ),
	<<Set Scriptables(
		{Weibull Estimate Probability( {560}, {0, 1, 2, 3} ),
		Scatter Plot Show Distribution(
			Weibull( 0 ),
			Lognormal( 0 ),
			Loglogistic( 0 ),
			Frechet( 0 ),
			Exponential( 0 ),
			SEV( 0 ),
			Normal( 0 ),
			Logistic( 0 ),
			LEV( 0 )
		)}
	),
	Confidence Interval Method( Likelihood ),
	SendToEmbeddedScriptable(
		Dispatch( {"Scatterplot"}, {Use Transformation Scale( 0 )} ),
		Dispatch(
			{"Comparisons", "Distribution Profiler"},
			{1, Confidence Intervals( 1 ), Term Value(
				Category( "A", Lock( 0 ), Show( 1 ) ),
				ResponseVariable( 560, Lock( 0 ), Show( 1 ) )
			)}
		),
		Dispatch(
			{"Comparisons", "Quantile Profiler"},
			{1, Confidence Intervals( 1 ), Term Value(
				Category( "A", Lock( 0 ), Show( 1 ) ),
				Probability( 0.5, Lock( 0 ), Show( 1 ) )
			)}
		),
		Dispatch(
			{"Comparisons", "Hazard Profiler"},
			{1, Confidence Intervals( 0 ), Term Value(
				Category( "A", Lock( 0 ), Show( 1 ) ),
				ResponseVariable( 560, Lock( 0 ), Show( 1 ) )
			)}
		),
		Dispatch(
			{"Comparisons", "Density Profiler"},
			{1, Confidence Intervals( 0 ), Term Value(
				Category( "A", Lock( 0 ), Show( 1 ) ),
				ResponseVariable( 560, Lock( 0 ), Show( 1 ) )
			)}
		),
		Dispatch(
			{"Results", "Weibull Results", "Distribution Profiler"},
			{1, Confidence Intervals( 1 ), Term Value(
				Category( "C", Lock( 0 ), Show( 1 ) ),
				ResponseVariable( 560, Lock( 0 ), Show( 1 ) )
			)}
		),
		Dispatch(
			{"Results", "Weibull Results", "Quantile Profiler"},
			{1, Confidence Intervals( 1 ), Term Value(
				Category( "C", Lock( 0 ), Show( 1 ) ),
				Probability( 0.5, Lock( 0 ), Show( 1 ) )
			)}
		),
		Dispatch(
			{"Results", "Weibull Results", "Hazard Profiler"},
			{1, Confidence Intervals( 0 ), Term Value(
				Category( "C", Lock( 0 ), Show( 1 ) ),
				ResponseVariable( 560, Lock( 0 ), Show( 1 ) )
			)}
		),
		Dispatch(
			{"Results", "Weibull Results", "Density Profiler"},
			{1, Confidence Intervals( 0 ), Term Value(
				Category( "C", Lock( 0 ), Show( 1 ) ),
				ResponseVariable( 560, Lock( 0 ), Show( 1 ) )
			)}
		)
	),
	SendToReport(
		Dispatch(
			{"Comparisons", "Distribution Profiler"},
			"10000",
			ScaleBox,
			{Min( 0 ), Max( 1.08602927047269 ), Inc( 0.2 ), Minor Ticks( 1 )}
		),
		Dispatch(
			{"Results", "Weibull Results"},
			"",
			TabListBox,
			{Set Selected( 6 )}
		),
		Dispatch(
			{"Results", "Weibull Results", "Distribution Profiler"},
			"10000",
			ScaleBox,
			{Min( 0 ), Max( 1.08602927047269 ), Inc( 0.2 ), Minor Ticks( 1 )}
		)
	)
);

[#2 · Accepted solution]

Thank you very much for your effort to elaborate the question!

What you observed is the behavior by design. And you won't be able to see Likelihood Confidence Intervals in profilers.

The reason is due to computation cost and usability. Calculating Likelihood Confidence Intervals is computationally intensive.

It is by design not to show Likelihood confidence intervals in the profiler, or the profiler will be unusable. That is true for all profilers in reliability platforms. And probably true elsewhere. I am glad that you found the places to get Likelihood confidence intervals for predictions. And you shall see when you have enough failure observations, the two types of intervals are rather close. If you have fewer failures, Likelihood confidence intervals are more trustworthy, and you should be alerted by the wide Wald intervals in the profiler first.

[#3]

Hi Peng Liu,

Thank you very much for your answer. It is a bit unfortunate though that we cannot have this feature due to it being the computationally heavy operation.