[R-sig-ME] Random effects with data collected at different scales

[Ben Bolker]

  I agree.  Fitting a random effect would presumably help a bit to take
care of the variation in circle radii, but it seems like an odd way to
do it.  If the authors are doing a log-link model or a log-transformed
model, then adding log(radius) as a predictor variable will translate to
fitting a dependence of the form RT ~ radius^b, which also has the nice
property that it can interpolate between b=1 (animals are moving
linearly, so expected time to leave the circle will scale proportionally
to the distance to the edge) and b=2 (animals are random-walking,
expected time scales as the square of the distance).  (If there are
multiple observations per individual, the random effect is needed *in
addition* to a fixed effect of log(r).)

  If the authors are using advanced statistical methods and you're
feeling uncertain of your ability to evaluate them, sometimes editors
will be responsive to a suggestion that the paper should be seen be a
statistical expert in addition to the regular scientific reviewers.

  cheers
    Ben Bolker


On 2019-04-04 6:36 a.m., Vinicius Maia wrote:
> Hi Carla,
> 
> It is a little bit odd, I would suggest to use the circle size as a control variable to deal with the circle size effect on the response (if there is).
> 
> Best
> 
> Obter o Outlook para Android<https://aka.ms/ghei36>
> 
> ________________________________
> From: R-sig-mixed-models <r-sig-mixed-models-bounces using r-project.org> on behalf of Brandt, Carla Freitas via R-sig-mixed-models <r-sig-mixed-models using r-project.org>
> Sent: Wednesday, April 3, 2019 2:36:33 AM
> To: r-sig-mixed-models using r-project.org
> Subject: [R-sig-ME] Random effects with data collected at different scales
> 
> Dear all,
> I'm reviewing a scientific article and have a doubt about an important part of the article. I wonder if any of you could help...
> For confidentiality issues, I will illustrate the question with an example.
> 
> The authors tracked marine animals and measured residence time (RT) inside imaginary circles along the track (at regular 10 km along the track). The problem is that the size of the circles was different for each animal (meaning that RT for some animals is the time spent in 10 km radius areas for example, while for other animals RT is the time spent in 60 km radius areas).
> The authors then fitted a GAM with individual ID as random effect (using s(ID, bs="re") and argue that this will account for the fact that RT was measured at different scales for different individuals. Is this valid? The authors provide this reference:
> https://stat.ethz.ch/R-manual/R-devel/library/mgcv/html/random.effects.html
> 
> Random effects are often used in telemetry studies to take into account repeated measurements from the same individual and to take into account random intrinsic variability between individuals. Is it valid to use random effects to take into account that data from different individuals were collected at different scales?
> 
> Thank you in advance for any help about this.
> Best regards,
> 
> Carla
> 
> 
>         [[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>