From Chris Gotwalt at SAS/JMP:
"If a 'variance component' estimate is negative it really can no longer
interpreted as a variance, since variances are strictly non-negative. It
can be interpreted as meaning that some of the levels of the random effects
are negatively correlated with one another. When the confidence interval
for a variance component contains zero, that is evidence that there is no
variance associated with that effect, and you might consider dropping that
effect from the model."
Martin Kane
Manager, Non-Clinical Statistics
Human Genome Sciences, Inc.
240-314-4400 x3822
"Davern, Sean"
<sdavern@amgen.com> To:
"'jmp-l@lists.biostat.wustl.edu'"
Sent by:
<jmp-l@lists.biostat.wustl.edu>
jmp-l-owner@lists.biosta cc:
t.wustl.edu Subject: Re: [jmp-l]
Simple Variance Component Analysis
08/30/2005 04:13 PM
Please respond to jmp-l
So what does it mean when the Lot&Random component of variance is -22% of
the total?
How is it possible for the Var Comp Est to be negative?
Sean
-----Original Message-----
From: jmp-l-owner@lists.biostat.wustl.edu
[mailto:jmp-l-owner@lists.biostat.wustl.edu]On Behalf Of
Martin_Kane@hgsi.com
Sent: Tuesday, August 30, 2005 12:51 PM
To: jmp-l@lists.biostat.wustl.edu
Subject: Re: [jmp-l] Simple Variance Component Analysis
The correct way to do this analysis is to use the "Fit Model" analysis.
Put "Response" in the Y role and put "Lot" in the Effects role. Now,
highlight "Lot" in the effects box and select the red "Attributes" arrow.
With this "Attributes" menu, select "Random Effect". This will change the
effect from "Lot" to "Lot& Random".
Doing this will allow you to get variance components out.
Unfortunately, your data set seems to have much more variability within lot
than lot-to-lot.
Sincerely,
Martin Kane
Manager, Non-Clinical Statistics
Human Genome Sciences, Inc.
240-314-4400 x3822
"Davern, Sean"
<sdavern@amgen.com> To:
"'jmp-l@lists.biostat.wustl.edu'"
Sent by:
<jmp-l@lists.biostat.wustl.edu>
jmp-l-owner@lists.biosta cc:
t.wustl.edu Subject: [jmp-l]
Simple Variance Component Analysis
08/30/2005 03:42 PM
Please respond to jmp-l
Can anyone tell me how to get JMP to give a very simply variance component
analysis?
I'm trying to quantify lot-to-lot variability of a raw material. I've made
two independent measurements of 9 lots. I'd like to segregate the overall
variance into within lot variation (coming from run-to-run variation) and
between lot (lot-to-lot) variation. I'd like the total variance to be
reported as the two components.
Attached is the example data set.
<<Example Data.JMP>>
It seems like this should be very simple but I can't seem to get the
answer without doing the calculations by hand.
Thanks,
Sean
Sean Davern
Engineer III
Cell Sciences Process Development
Mail Stop AW2/D2152
Ext. 57074
(See attached file: Example Data.JMP)
|