Change in JMP 15 REML?

[#1]

Hi,

I'm trying to generate least squares means via REML. The idea is to get a single value for each replicated individual in my study (i.e., I have three reps per individual and I want one number to represent an individual). In the past when I've done this (with the same model on the same data) in JMP 14 the lsmeans have been very close to the arithmetic mean. Now, in JMP 15, the lsmeans for all individuals are very close to each other and very different from the arithmetic mean per individual. Has something changed between versions that would explain this? 

Thanks in advance for any help or suggestions!

[#2]

So you have the same data set analyzed with the same set up in the Fit Model launch dialog in both JMP 14 and JMP 15 but the LSMeans are not the same? I realize that this question seems obvious but it has happened many times in the past that the analysis was not the same in both cases. Perhaps you could save the script for the platform and post it here. That act would go a long way towards reproducibility.

Also, you do not need to use a random effect model and REML estimation if the only variable is individual. The ANOVA model will also produce valid LSMeans estimates.

[#3]

Thank you for your quick reply. Embarrassingly, it seems that the analysis wasn't the same... 

However, regarding using ANOVA to generate lsmeans: when I use ANOVA I get very different results from when I use REML. The trait ("ADI") and the effect ("UniqueID") are the same in both analyses. Any insight as to why that might be the case? (Apologies if this should be a different discussion topic--first time posting here--and also apologies if it ends up being another silly error on my part!)

REML:

Fit Model(
	Y( :ADI ),
	Effects,
	Random Effects( :UniqueID ),
	NoBounds( 1 ),
	Personality( "Standard Least Squares" ),
	Method( "REML" ),
	Emphasis( "Minimal Report" ),
	Run(
		:ADI << {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 )}
	),
	SendToReport(
		Dispatch( {"Response ADI"}, "Effect Details", OutlineBox, {Close( 0 )} ),
		Dispatch(
			{"Response ADI", "Effect Details", "UniqueID"},
			"Least Squares Means Table",
			OutlineBox,
			{Close( 0 )}
		)
	)
);

ANOVA: 

Oneway( Y( :ADI ), X( :UniqueID ), Means( 1 ), Mean Diamonds( 1 ) )

[#4]

My apologies. I meant if you remove the &Random attribute from :UniqueID effect in the Fit Model launch dialog, then Fit Least Squares should give the LSMeans, too. I did not mean that you should use the Oneway platform. 

And as I think about it more, the variation due to the individual should be treated as a random effect if individuals represent a sample a population of subjects. The LSMeans and their standard errors will be different depending on whether you treat :UniqueID as a fixed or random effect.