I am studying the effect of forest structure on recruitment. One of the variables is species richness defined as the number of species.

The aim is to quantify the effect of species richness on recruitment by using a glmm. But I want to avoid obtaining a variable which reflects an elevational gradient instead of species richness. The correlation between elevation and species richness makes this difficult.

The model also includes precipitation and temperature-related variables. Degree day sum, water balance and soil information to be more precise. Is this sufficient to account for the elevation? Or how do I know if my precipitation and temperature variables are good enough? I am unsure if I should scale the variable for species richness by elevation.

So far I use z-transformation for every 500 m of elevation.

But I couldn't find any literature on similar problems. Is there a general way to normalize/standardize a variable that is correlated with another variable?

  • $\begingroup$ How strong is the correlation between species richness and elevation? What is the r? You could try calculating variance inflation factors, to investigate how much any collinearity affects the model. $\endgroup$ – rw2 Jun 2 at 10:12
  • $\begingroup$ correlation is -0.57. VIF is below 2 for all variables except degree day sum which has 3.7. $\endgroup$ – Hans Jürgen Jun 2 at 10:19
  • $\begingroup$ Why wouldn't you include both richness and elevation in the model? $\endgroup$ – Lewian Jun 2 at 10:56
  • 1
    $\begingroup$ Putting elevation in the model will give you an estimate for the effect of species richness that takes into account that elevation may also play a role. This is not without its problems (the data cannot perfectly tell apart influences of strongly correlated variables), but seems better to me than leaving elevation out of the model. Whether your way of taking into account elevation is even better is very hard to tell, because it depends on how exactly elevation and richness are related, also from the subject matter perspective. $\endgroup$ – Lewian Jun 2 at 11:14
  • 1
    $\begingroup$ VIFs are not a method to decide which variables to use, however I'd agree that having two variables in a model with correlation 0.96 is not a good idea. Now this suggests that degree day sum and other variables may already pretty much account for the effect of elevation, and your transformation may not be needed. Ultimately it's an interesting question but hard to comment on without seeing the data and knowing all the details (which is one reason why you won't find any "standard solution" in the literature). $\endgroup$ – Lewian Jun 2 at 20:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.