[R-sig-ME] Lmer and variance-covariance matrix
Jarrod Hadfield
j.hadfield at ed.ac.uk
Fri Mar 11 14:56:29 CET 2011
Hi,
In addition, each trait is only measured once for each id (correct?)
which means that the likelihood could not be optimised even if the
data-set was massive. If you could fix the residual variance to some
value (preferably zero) then the problem has a unique solution given
enough data, but I'm not sure this can be done in lmer? . Since the
structure of the residuals is at the moment quite inflexible you
probably can't use lmer to fit multi-response models, unless the
responses are non-Gaussian and non-binary.
Cheers,
Jarrod
On 11 Mar 2011, at 13:18, Douglas Bates wrote:
> On Thu, Mar 3, 2011 at 7:03 AM, Antoine Paccard
> <antoine.paccard at unine.ch> wrote:
>> Dear modelers,
>> I have been trying in the past few months to obtain a variance-
>> covariance matrix using lmer. I failed multiple times until I decided
>> to do it under SAS. Now, I am going back on it and would like to run
>> it with R.
>> I measured 15 different traits and my data is organized this way:
>> fam id trait score
>> 57 1 1 0.047207645
>> 57 2 1 1.420790311
>> 57 3 1 -0.290782077
>> 57 1 2 -0.585655473
>> 57 2 2 -0.986483343
>> 57 3 2 0.290187057
>> 57 1 3 0.741162271
>> 57 2 3 1.59448736
>> 57 3 3 .
>> . . . .
>> . . . .
>> . . . .
>> 57 1 15 .
>> 57 2 15 .
>> 57 3 15 .
>> 58 1 1 .
>> 58 2 1 .
>> 58 3 1 .
>> . . . .
>> . . . .
>> . . . .
>> 58 1 15 .
>> 58 2 15
>> 58 3 15
>> .
>> .
>> .
>> 100 1 15
>> 100 2 15
>> 100 3 15
>> with “fam” being the family, “id” the individual, “trait”
>> the traits measured and “score” the result of each measures.
>> So far I have been trying to fit this model:
>> d <- data
>> d$fam1 <- factor(d$fam)
>> d$id1 <- factor(d$id)
>> d$trait1 <- factor(d$trait)
>> w.family <- lmer(score ~ (trait1 | fam1/id1 ), data=d)
>
>> But I was getting some “stack overflow” error messages so I ran
>> the
>> model this way:
>> w.family <- lmer(score ~ (trait1 | fam1/id1 ), data=d,control = list
>> (maxIter = 500))
>
> If there are 15 levels of trait you are trying to estimate 240
> variance-covariance parameters (120 for fam1 and 120 for fam1:id1).
> That is a very large optimization problem, I'm not surprised that
> there is difficulty in finding the optimum.
>> Although it still didn’t work and I am now wondering what is wrong
>> in
>> this model. The reason why I have put “trait1” in the random
>> effect
>> is because it was the only way for me to obtain a variance-covariance
>> matrix on all my traits.
>> I am currently trying to put “trait1” also as fixed effect:
>> w.family <- lmer(score ~ trait1 + (trait1 | fam1/id1 ),
>> data=d,control
>> = list (maxIter = 500))
>> I would be curious to know what you think about such a model and why
>> it actually doesn’t work.
>> Moreover, I am trying to write done the equation for the model:
>> w.family <- lmer(score ~ (trait1 | fam1/id1 ), data=d)
>> but can’t figure it out. I know I should get something as:
>> Scoreij = µ + fam(i) + Ij(i) + ε
>> But I am not sure if that’s right.
>> What do you think? Thanks a lot for your help,
>> Best,
>> Antoine Paccard
>> -----------
>> Antoine Paccard,
>> Laboratory of Evolutionary Botany,
>> Institut of biology, Faculté des Sciences,
>> University of Neuchâtel, Unimail,
>> Rue Émile-Argand 11,
>> 2000, Neuchâtel Switzerland
>> Office: (0041) (0)32 718 23 49
>> Fax: (0041) (0)32 718 30 01
>> www.unine.ch/members/antoine.paccard/
>> http://www2.unine.ch/evobot
>>
>>
>> [[alternative HTML version deleted]]
>>
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>>
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
More information about the R-sig-mixed-models
mailing list