topic Re: JMP Statistical Analysis - Fit Model - Omitting of Results in Discussions

JMP Statistical Analysis - Fit Model - Omitting of Results

Lefkiosvanrooij — Fri, 09 Jun 2023 00:24:16 GMT

During statistical analysis of our data set using JMP, we encounter the following problem that I would like to share with you for to ask whether other users of JMP have encountered this before and know why it occurs and how to solve it.

Fit Model(
	Y( :Measured RP% ),
	Effects,
	Random Effects( :Assay ),
	NoBounds( 0 ),
	Personality( "Standard Least Squares" ),
	Method( "REML" ),
	Emphasis( "Minimal Report" ),
	Run(
		:Measured RP% << {Summary of Fit( 1 ), Analysis of Variance( 0 ),
		Parameter Estimates( 1 ), Scaled Estimates( 0 ),
		Plot Actual by Predicted( 0 ), Plot Regression( 0 ),
		Plot Residual by Predicted( 0 ), Plot Studentized Residuals( 0 ),
		Plot Effect Leverage( 0 ), Plot Residual by Normal Quantiles( 0 )}
	),
	Where( :Sample == "Accuracy 40" ),
	SendToReport(
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40"},
			"Parameter Estimates",
			OutlineBox,
			{Close( 0 )}
		),
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40", "Parameter Estimates"},
			"Std Error",
			NumberColBox,
			{Visibility( "Collapse" )}
		),
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40", "Random Effect Predictions"
			},
			"Std Error",
			NumberColBox,
			{Visibility( "Collapse" )}
		),
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40",
			"REML Variance Component Estimates"},
			"Std Error",
			NumberColBox,
			{Visibility( "Collapse" )}
		),
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40",
			"REML Variance Component Estimates"},
			"95% Lower",
			NumberColBox,
			{Visibility( "Collapse" )}
		),
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40",
			"REML Variance Component Estimates"},
			"Sqrt Variance Component",
			NumberColBox,
			{Visibility( "Visible" )}
		),
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40",
			"REML Variance Component Estimates"},
			"Pct of Total",
			NumberColBox,
			{Visibility( "Collapse" )}
		),
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40",
			"REML Variance Component Estimates"},
			"CV",
			NumberColBox,
			{Visibility( "Visible" )}
		),
		Dispatch(
			{"Response Measured RP% Sample=Accuracy 40", "Effect Details", "Assay",
			"Least Squares Means Table"},
			"Std Error",
			NumberColBox,
			{Visibility( "Collapse" )}
		)
	)
)

Re: JMP Statistical Analysis - Fit Model - Omitting of Results

Lefkiosvanrooij — Tue, 10 Nov 2020 11:53:16 GMT

Note: An explanation of the problem encountered is described in the word document attached

Re: JMP Statistical Analysis - Fit Model - Omitting of Results

SDF1 — Tue, 10 Nov 2020 15:42:58 GMT

Hi @Lefkiosvanrooij ,

I was able to recreate the error that you got using s smaller data table set based on what you included in the DOCX file.

So, I think it has to do with your selection (or rather de-selection) of the "Unbounded Variance Components" in the Fit Model window. If you check it, the CIs are calculated a different way then if they are unchecked, see the JMP help here. It may be that you can't use the REML method in which case you can use EMS method for estimating variances. You can also check the box "Estimate Only Variance Components".

I am not a statistician, so I can't comment on which method is appropriate to use for your analysis, but I believe with the "Unbounded Variance Components" unchecked, then you get the error of the fit reporting 0 for some of the samples, like Accuracy 40.

Hope this helps!,

Re: JMP Statistical Analysis - Fit Model - Omitting of Results

Mark_Bailey — Tue, 10 Nov 2020 16:38:17 GMT

As @SDF1 claimed, the unbounded estimates option determines the result of estimating the variance components with REML. Unbounded estimates can result in negative estimates. Variance cannot be negative, but these results are actually the covariance, which can be negative. This option provides better hypothesis tests and BLUPs. It is important if you include fixed effects in your model. (Your script shows no fixed effects.)

You can bound the estimates and constrain them to be non-negative. Any negative estimate is forced to be zero.