R help - Mar 2006 - Using of LME function in non-replicate data

Hello all R-users!

 In Jun-2005, I find the follow discussion about using
of
 LME function ( in NLME library ) for fitting
non-replicate data

 The thread: ANOVA vs REML approach to variance
component estimation
 

http://tolstoy.newcastle.edu.au/R/help/05/06/6498.html


 Someone expose the follow problem:

 # non-replicate data
 y <- c(2.2, -1.4, -0.5, -0.3, -2.1, 1.5, 1.3, -0.3,
0.5, -1.4, -0.2, 1.8)

 ID <- factor( 1:12 )

 library(ape)
 library(nlme)
 varcomp(lme(y ~ 1, random = ~ 1 | ID))

# RESULTS:

 # ID Within
 # 1.6712661 0.2350218 

 Prof. Dr. Douglas Bates reply this:

 > It's a spurious convergence in lme. There is no
check in lme for the
 > number of observations exceeding the number of
groups. There should
 > be. I'll add this to the bug reports list. 

 Alright!
 But I have one similar problem and one doubt. 

 I have 49 distinctive experiments split in 7 blocks (
split plot design non-replicate )

 I fitting models with ~ 10 or ~ 20 coefficients (
several responses. )

 ( 
   it seems describe the data by 
   experimental versus predicted responses plot and
   residuals plot
 )

 
 My doubt: The components of variance given by LME
function are

 reliable approximate estimates or this variance are
spurious too?

 ... I thinked that this varinces were calculate by
"lack of fit terms".

 In the case of this variances are wrong, even so can
I use the REML coefficients estimates?


 Thanks in advanced! 
 Regards.

 Cleber
 Chemistry student

You did not provide a simple, self-contained replicable example, so I 
can not say for sure.  You say you have 49 observations in 7 blocks. 
The issue described in the post you cite would be a problem if you had 
49 blocks of size 1.  I've tried to fit models like that, only to find 
after getting strange results that 'lme' did not bother to tell me how 
stupid I was:  It just tried to do what I asked.

	  If you've got 7 blocks with 7 treatments and 10-20 fixed effects, I 
do not believe that the issue described in that post would be a problem 
for you.  You may have other problems, like too much noise to find the 
signals you most care about.  However, those are different from the 
issue you cited.

	  hope this helps.
	  spencer graves

Cleber N.Borges wrote:
> 
> 
>  Hello all R-users!
> 
>  In Jun-2005, I find the follow discussion about using
> of
>  LME function ( in NLME library ) for fitting
> non-replicate data
> 
>  The thread: ANOVA vs REML approach to variance
> component estimation
>  
> 
> http://tolstoy.newcastle.edu.au/R/help/05/06/6498.html
> 
> 
>  Someone expose the follow problem:
> 
>  # non-replicate data
>  y <- c(2.2, -1.4, -0.5, -0.3, -2.1, 1.5, 1.3, -0.3,
> 0.5, -1.4, -0.2, 1.8)
> 
>  ID <- factor( 1:12 )
> 
>  library(ape)
>  library(nlme)
>  varcomp(lme(y ~ 1, random = ~ 1 | ID))
> 
> # RESULTS:
> 
>  # ID Within
>  # 1.6712661 0.2350218 
> 
>  Prof. Dr. Douglas Bates reply this:
> 
>  > It's a spurious convergence in lme. There is no
> check in lme for the
>  > number of observations exceeding the number of
> groups. There should
>  > be. I'll add this to the bug reports list. 
> 
>  Alright!
>  But I have one similar problem and one doubt. 
> 
>  I have 49 distinctive experiments split in 7 blocks (
> split plot design non-replicate )
> 
>  I fitting models with ~ 10 or ~ 20 coefficients (
> several responses. )
> 
>  ( 
>    it seems describe the data by 
>    experimental versus predicted responses plot and
>    residuals plot
>  )
> 
>  
>  My doubt: The components of variance given by LME
> function are
> 
>  reliable approximate estimates or this variance are
> spurious too?
> 
>  ... I thinked that this varinces were calculate by
> "lack of fit terms".
> 
>  In the case of this variances are wrong, even so can
> I use the REML coefficients estimates?
> 
> 
>  Thanks in advanced! 
>  Regards.
> 
>  Cleber
>  Chemistry student
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide!
http://www.R-project.org/posting-guide.html

Hello Spencer and all R-users


Many thanks for your reply, Spencer!
I don't put the data because I thought that I would be
inconvenient.

The data follow below.


The model that I am using is the following one:

( Scheffe model in Z variable ) * ( Scheffe model in X
variable )

obs.: I make one test for fit this model in SAS and I
obtained the same results.

Thanks
Regards


Cleber
Chemistry' student




####### Model


r.lme <-lme( NP ~ -1 +

 x3:z2 + x1:x2:z2 + x3:z1 + x2:x3:z3 + x1:z1 +
x1:x2:x3:z1:z3 +
 x1:x3:z2 + x2:x3:z2 + x3:z1:z2:z3 + x1:z3 + x3:z1:z2
+
 x1:x3:z1 + x2:z1 + x1:x2:z1:z3 + x2:z3 + x2:x3:z2:z3
+
 x1:x2:x3:z3 + x2:x3:z1:z3 + x1:x2:x3:z1 + x1:x2:x3:z2
+
 x1:x2:x3:z1:z2:z3 + x2:z2 + x1:z2 + x3:z3 +
x1:x3:z1:z2 +
 x1:x2:z1:z2:z3 + x1:x2:x3:z2:z3 + x1:x3:z1:z2:z3 +
 x1:x3:z2:z3 + x2:z1:z2:z3

, data=ret,random=~1|wp )



####### data

       z1       z2       z3       x1       x2       x3
wp     y NP
 1.000000 0.000000 0.000000 1.000000 0.000000 0.000000
 1  11.6  6
 1.000000 0.000000 0.000000 0.000000 1.000000 0.000000
 1   5.8  3
 1.000000 0.000000 0.000000 0.000000 0.000000 1.000000
 1   5.9 12
 1.000000 0.000000 0.000000 0.500000 0.500000 0.000000
 1  12.4  5
 1.000000 0.000000 0.000000 0.500000 0.000000 0.500000
 1   4.6  5
 1.000000 0.000000 0.000000 0.000000 0.500000 0.500000
 1   7.8  7
 1.000000 0.000000 0.000000 0.333333 0.333333 0.333333
 1   8.5  8
 0.000000 1.000000 0.000000 1.000000 0.000000 0.000000
 2  11.6  2
 0.000000 1.000000 0.000000 0.000000 1.000000 0.000000
 2  14.9  2
 0.000000 1.000000 0.000000 0.000000 0.000000 1.000000
 2 114.0 17
 0.000000 1.000000 0.000000 0.500000 0.500000 0.000000
 2  19.2 17
 0.000000 1.000000 0.000000 0.500000 0.000000 0.500000
 2  10.2  3
 0.000000 1.000000 0.000000 0.000000 0.500000 0.500000
 2  11.2 16
 0.000000 1.000000 0.000000 0.333333 0.333333 0.333333
 2  18.5 11
 0.000000 0.000000 1.000000 1.000000 0.000000 0.000000
 3   5.1  4
 0.000000 0.000000 1.000000 0.000000 1.000000 0.000000
 3   4.7  3
 0.000000 0.000000 1.000000 0.000000 0.000000 1.000000
 3   8.5  2
 0.000000 0.000000 1.000000 0.500000 0.500000 0.000000
 3  12.8  3
 0.000000 0.000000 1.000000 0.500000 0.000000 0.500000
 3   5.4  2
 0.000000 0.000000 1.000000 0.000000 0.500000 0.500000
 3  20.8 10
 0.000000 0.000000 1.000000 0.333333 0.333333 0.333333
 3   7.0  9
 0.500000 0.500000 0.000000 1.000000 0.000000 0.000000
 4  11.6  5
 0.500000 0.500000 0.000000 0.000000 1.000000 0.000000
 4  13.3  3
 0.500000 0.500000 0.000000 0.000000 0.000000 1.000000
 4 114.0 19
 0.500000 0.500000 0.000000 0.500000 0.500000 0.000000
 4  19.8 11
 0.500000 0.500000 0.000000 0.500000 0.000000 0.500000
 4  10.2  4
 0.500000 0.500000 0.000000 0.000000 0.500000 0.500000
 4  10.3 15
 0.500000 0.500000 0.000000 0.333333 0.333333 0.333333
 4   8.3 11
 0.500000 0.000000 0.500000 1.000000 0.000000 0.000000
 5  11.6  5
 0.500000 0.000000 0.500000 0.000000 1.000000 0.000000
 5  13.8  4
 0.500000 0.000000 0.500000 0.000000 0.000000 1.000000
 5  19.4  8
 0.500000 0.000000 0.500000 0.500000 0.500000 0.000000
 5  11.3  9
 0.500000 0.000000 0.500000 0.500000 0.000000 0.500000
 5  10.2  4
 0.500000 0.000000 0.500000 0.000000 0.500000 0.500000
 5  11.2  6
 0.500000 0.000000 0.500000 0.333333 0.333333 0.333333
 5  14.7 18
 0.000000 0.500000 0.500000 1.000000 0.000000 0.000000
 6   3.1  2
 0.000000 0.500000 0.500000 0.000000 1.000000 0.000000
 6  13.7  3
 0.000000 0.500000 0.500000 0.000000 0.000000 1.000000
 6  19.5  9
 0.000000 0.500000 0.500000 0.500000 0.500000 0.000000
 6  24.0 10
 0.000000 0.500000 0.500000 0.500000 0.000000 0.500000
 6  10.2  4
 0.000000 0.500000 0.500000 0.000000 0.500000 0.500000
 6  22.2 17
 0.000000 0.500000 0.500000 0.333333 0.333333 0.333333
 6   7.0 10
 0.333333 0.333333 0.333333 1.000000 0.000000 0.000000
 7  11.5  5
 0.333333 0.333333 0.333333 0.000000 1.000000 0.000000
 7  13.7  2
 0.333333 0.333333 0.333333 0.000000 0.000000 1.000000
 7 114.0 18
 0.333333 0.333333 0.333333 0.500000 0.500000 0.000000
 7  19.8 13
 0.333333 0.333333 0.333333 0.500000 0.000000 0.500000
 7  10.2  5
 0.333333 0.333333 0.333333 0.000000 0.500000 0.500000
 7  11.0 16
 0.333333 0.333333 0.333333 0.333333 0.333333 0.333333
 7  18.3 13




Spencer Graves wrote:
>       You did not provide a simple, self-contained
replicable example, so I can not say for sure.  You
say you have 49 observations in 7 blocks. The issue
described in the post you cite would be a problem if
you had 49 blocks of size 1.  I've tried to fit models
like that, only to find after getting strange results
that 'lme' did not bother to tell me how stupid I was:
 It just tried to do what I asked.
>
>       If you've got 7 blocks with 7 treatments and
10-20 fixed effects, I do not believe that the issue
described in that post would be a problem for you. 
You may have other problems, like too much noise to
find the signals you most care about.  However, those
are different from the issue you cited.
>
>       hope this helps.
>       spencer graves
>
> Cleber N.Borges wrote:
>
>>
>>
>>  Hello all R-users!
>>
>>  In Jun-2005, I find the follow discussion about
using
>> of
>>  LME function ( in NLME library ) for fitting
>> non-replicate data
>>
>>  The thread: ANOVA vs REML approach to variance
>> component estimation
>>  
>>
>>
http://tolstoy.newcastle.edu.au/R/help/05/06/6498.html
>>
>>
>>  Someone expose the follow problem:
>>
>>  # non-replicate data
>>  y <- c(2.2, -1.4, -0.5, -0.3, -2.1, 1.5, 1.3,
-0.3,
>> 0.5, -1.4, -0.2, 1.8)
>>
>>  ID <- factor( 1:12 )
>>
>>  library(ape)
>>  library(nlme)
>>  varcomp(lme(y ~ 1, random = ~ 1 | ID))
>>
>> # RESULTS:
>>
>>  # ID Within
>>  # 1.6712661 0.2350218
>>  Prof. Dr. Douglas Bates reply this:
>>
>>  > It's a spurious convergence in lme. There is no
>> check in lme for the
>>  > number of observations exceeding the number of
>> groups. There should
>>  > be. I'll add this to the bug reports list.
>>  Alright!
>>  But I have one similar problem and one doubt.
>>  I have 49 distinctive experiments split in 7
blocks (
>> split plot design non-replicate )
>>
>>  I fitting models with ~ 10 or ~ 20 coefficients (
>> several responses. )
>>
>>  (    it seems describe the data by    experimental
versus predicted responses plot and
>>    residuals plot
>>  )
>>
>>  
>>  My doubt: The components of variance given by LME
>> function are
>>
>>  reliable approximate estimates or this variance
are
>> spurious too?
>>
>>  ... I thinked that this varinces were calculate by
>> "lack of fit terms".
>>
>>  In the case of this variances are wrong, even so
can
>> I use the REML coefficients estimates?
>>
>>
>>  Thanks in advanced!  Regards.
>>
>>  Cleber
>>  Chemistry student
>