R help - May 2024 - Using intervals() function for nlme model - Statistics Lab ETHZ

Dear members of the R-Help Team,


we are reaching out to you with a question regarding using
intervals()<https://www.rdocumentation.org/packages/nlme/versions/3.1-163/topics/intervals>
to create inference for
nlme()<https://www.rdocumentation.org/packages/nlme/versions/3.1-164/topics/nlme>
function.

The maintainer for the nlme package is listed as the R-Core Team and we were
advised by moderators of R-Core to write to R-Help to receive help in this case.


We are students of Statistics at ETH Zurich currently collaborating with ETH
Crop Science lab on a project focused on soybean growth. We are modelling the
growth using the nlme function and are facing the issue of creating a valid
inference for the nlme model (we are especially interested in confidence
intervals for the fixed effects).


Based on the documentation of nlme package, the function interval is supported
for gls, lme, and lmList classes. Although nlme is not mentioned, we still
receive intervals without any errors or warnings.


We would be therefore very grateful if you could offer your insights on the
following questions:

  *   Is the intervals() function appropriate for nlme to create confidence
intervals?

  *   If not, what do the outputted values represent?

  *   Do you have any other recommendations/ ideas on how should we proceed to
calculate intervals in R (e.g. more suitable packages)?


Your help is greatly appreciated.


We are looking forward to hearing back from you!



Yours sincerely,

Tatiana, Marvin & Paul

from

Statistics Lab


	[[alternative HTML version deleted]]

Dear Tatiana, Marvin & Paul,

If you don't get a better answer here, try
r-sig-mixed-models at r-project.org.

On Fri, 3 May 2024 11:29:19 +0000
"Bielakova  Tatiana" <tbielakova at student.ethz.ch> wrote:
> Based on the documentation of nlme package, the function interval is
> supported for gls, lme, and lmList classes. Although nlme is not
> mentioned, we still receive intervals without any errors or warnings.
You're right to question this. Programming-language-level inheritance
ends up being backwards for statistical methods:
<https://notstatschat.rbind.io/2023/06/07/blank-cheque-inheritance-and-statistical-objects/>.
Linear mixed-effect models are a subset of the more general mixed-effect
models, but in R, "nlme" objects inherit from "lme", not the
other way
around.
>   *   Do you have any other recommendations/ ideas on how should we
> proceed to calculate intervals in R (e.g. more suitable packages)?
Here's a vignette from the nlraa package where bootstrap is used to get
a confidence interval for an implicit parameter:
https://cran.r-project.org/package=nlraa/vignettes/Bootstrapping.html

The author seems to be using intervals() on an nlme model without any
problems. It may help to see if the intervals() output agrees with the
bootstrap results from nlraa::boot_nlme().

-- 
Best regards,
Ivan