1
$\begingroup$

is it possible to fit a GLM over space and time?

I have shrimp densities measured over 8 months, and in each month there are 4 stations sampled along a freshwater-saltwater gradient, so I have 32 observations of shrimp densities in total.

I have Temperature,Salinity,Turbidity,Oxygen,Depth as my environmental variables. I have 32 observations for each variable here as well.

So do I need to block for space or time? Do I have to use a manyglm?

thanks!

Pieter

Improve this question
$\endgroup$
4
  • $\begingroup$ density is expressed by the number of individuals/m² $\endgroup$
    – Pieter Torrez
    Commented Mar 21, 2015 at 15:44
  • 1
    $\begingroup$ You probably need a GLMM. I've voted to migrate this to stat.stackexchange.com since it isn't a programming question. $\endgroup$
    – Roland
    Commented Mar 21, 2015 at 16:01
  • 1
    $\begingroup$ There are multiple ways of accounting for these repeated measurements. What is your objective, though? Are you trying to directly model, or average across, these effects? Is there any experimental or pseudoexperimental condition that you are trying to determine for its impact upon yield rates? $\endgroup$
    – AdamO
    Commented Mar 21, 2015 at 18:10
  • $\begingroup$ You should be able to use GEE for this. Hopefully someone who knows more about GEE will come along and answer confidently $\endgroup$
    – shadowtalker
    Commented Mar 21, 2015 at 18:27

2 Answers 2

Reset to default
1
$\begingroup$

I would look into using Integrated Nested Laplace Approximation, which is specifically for the type of data you have. There is also an R package for it here.

Otherwise I would perhaps use a generalized linear mixed effects model. If your primary question is how shrimp respond to those environmental variables, then you could use the month as a random effect. This is essentially correcting for the fact that your observations from month to month are likely to be related to each other, although this does not account for the spatial autocorrelation that might also be in your data. If the sampling sites are independent of each other then I wouldn't worry about spatial autocorrelation. This is a great guide to using GLMMs for ecological questions.

I'm not sure what the distribution of your shrimp densities are, but that would change the distribution you would use for the GLMM. If it is normally distributed then you could use a linear mixed effect model.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Hi @smccain, just chiming in to say that your answer is still much appreciated. Do you happen to have an updated link to the "great guide to using GLMMs for ecological questions"? That sounds like something I should read! Thanks. $\endgroup$
    – Skaqqs
    Commented Aug 19, 2021 at 17:31
1
$\begingroup$

This, I believe, is the purpose of Generalized Estimating Equations. I'm not personally too familiar with them, but these slides seem to offer a good overview of their application to spatial regression.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.