When modeling data sampled in the field, I often come across the problem of determining the Family of the dependent variable for GLM (or GLMM).
An example: in an ecological study, I have ~ 60 patches. I have sampled territories of an animal species within these patches as well as a set of explanatory variables. The number of territories per patch (NTP) is the dependent variable in a GLM.

NTP is integer with values between 0 and 14:

enter image description here enter image description here

How can I test for the family of NTP? Any help appreciated as this is an issue many ecologists come across each time they conduct statistical analysis.

  • $\begingroup$ I would like to know a little more about what you know about glms. To me this seems like a zero-inflated model for counts, but maybe I am oversimplifying this. $\endgroup$
    – Alex
    Sep 16, 2017 at 18:40
  • $\begingroup$ At any rate, my abbreviated algorithm is as follows. If response is continuous, binary, or count use normal, binomial, Poisson models respectively. The next question concerns the relation of the variance to the mean. Is there overdispersion? If so use the appropriate model such as the gamma for counts. In your case it seems that you have a lot of 0 counts. There is a special model for counts, the zero-inflated Poisson. What you can do is run a zero-inflated Poisson and a regular Poisson and compare the goodness of fit. $\endgroup$
    – Alex
    Sep 16, 2017 at 18:56
  • $\begingroup$ If you initially post to the wrong group, please don't make a new post, flag the old one for migration. Your original post has also been migrated here in any case, so now we have two identical posts here to deal with. Please read the third-last paragraph (the one on cross--posting) at help center $\endgroup$
    – Glen_b
    Sep 17, 2017 at 4:34


Your Answer

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