R help - Aug 2021 - Issue regarding specifying pdBlocked matrix for random effects vcov in nlme package

Dear Dr. David;

Re; List rejection

I'm unable to explain why this posting was refused. It does have 4 image 
files attached. That is more than I typically see on Rhelp. I'm only a 
volunteer moderator, and not one of the owners of the list. I speculate 
that if you were to repost to rhelp (and NOT to R-Core where this 
message does not belong) and you were to substitute copied text from 
your console for all except the first image, it might go through (at 
least to the moderation queue where it could be reviewed by human eyes 
and wetware. (I suspect the number of images and links caused an 
automatic spam rejection.)

I also wonder if your modified re-submission should instead be directed 
to the R-SIG-mixed-models list since that appears to be the question 
topic. Some of the experts there are not regular contributor of viewers 
of Rhelp which is focused on the maechanics of the R language itself.

-- 

David Winsemius

Volunteer moderator.

On 8/2/21 1:43 PM, Benjamin Davis wrote:
>
> Hello,
>
> ??????????????? I have come across an issue regarding specifying the 
> vcov of the random effects while using the /medrc/ package, which I 
> believe is very likely to originate from the /nlme/ package. I have 
> posted the question to StackOverflow if it is easier to read there:
>
>
https://stackoverflow.com/questions/68626894/specifying-the-variance-covariance-matrix-for-random-effects-for-medrc-or-nlme
>
<https://urldefense.com/v3/__https://stackoverflow.com/questions/68626894/specifying-the-variance-covariance-matrix-for-random-effects-for-medrc-or-nlme__;!!HGYKHdhaPg!Ghq6dEJafKNBdT7h64jc9_H5JJWEnEQ891_QRsUZClnGXCeYntzRiXDmmq2XMTQ$>.
>
> I am using a 3-parameter log-logistic non-linear function
>
> and am including a species indicator variable based on whether my data 
> originates from humans (H) or rats (R). I am looking to specify a 
> blocked variance-covariance matrix for the random effects where 
> covariances are estimated for the off-diagonals of function parameters 
> within the same species, but not between species.
>
> I have attempted to estimate this vcov using the ?medrm? function from 
> the /medrc/ package. While the notation is slightly different from the 
> ?nlme? function, the estimation is performed in the same manner. I 
> have attempted to specify the random effect vcov using the 
> ?'pdBlocked' function.
>
> M3b <- medrm(inhibition ~ concentration, curveid=b + d + e ~ species,? 
> data=OP,
>
> ?????????????????????????????random= 
> list(subject=pdBlocked(list(b~species, d~species, e~species))),
>
> ?????????????????????fct=LL.3(), control=c(drmc(method="CG"), 
> nlmeControl(msMaxIter = 150)))
>
> However, this results in almost the opposite of what I intended, 
> providing covariances within parameter type but across species (which 
> is non-sensical given that no subject can be both a human and a rat).
>
> Do you have any recommendations on how to fix this, ideally using one 
> of the existing pd-matrix functions?
>
> Thanks in advance for your reply.
>
> *Benjamin Davis, Ph.D.*
> Senior Scientist
> *Exponent*
> Direct +1-202-772-4942
> Email davisb at exponent.com <mailto:davisb at exponent.com>
>
> *Benjamin Davis, Ph.D.*
> Senior Scientist
> *Exponent*
> Direct +1-202-772-4942
> Email davisb at exponent.com <mailto:davisb at exponent.com>
>
	[[alternative HTML version deleted]]