1
$\begingroup$

I was wondering whether you could help me with the following issue:

I am running a negative binomial model (overdispersed) on count data of sampled insects at 22 sites. Here, I would like to select the best predicting explanatory variables out of five potential variables (i.e. I am not testing a hypothesis, but rather explore the data). To do so, I used a model selection approach based on AICc (Akaike Information Criterion). Now, I learned that a model selection based on AICc (following Burnham & Anderson) is quite heavily discussed.

My question here is: since when is the model selection approach based on AICc not acceptable anymore and why? It seems that I have missed a discussion on this issue. If you could point me in any direction here, it would be very much appreciated.

Many thanks, Tanja

| improve this question | | | | |
$\endgroup$
  • $\begingroup$ I would just try AIC and then AICc and see if there are differences. If AIC peaks more than AIC, then the AICc correction for small sample size is doing its job. If both pick the same number of variables, then it doesn't matter and you can choose either. You could also try BIC (en.wikipedia.org/wiki/Bayesian_information_criterion) that also includes a correction and see what makes sense. I would assume BIC and AICc should give similar results. $\endgroup$ – arkaia May 22 at 1:58
  • $\begingroup$ Please spell out your acronyms on first use. $\endgroup$ – David Hammen May 22 at 8:55
  • $\begingroup$ Where's the link to earth science ? $\endgroup$ – a_donda May 22 at 18:33
  • $\begingroup$ you could follow the guidance in: towardsdatascience.com/… and stats.stackexchange.com/questions/319769/… $\endgroup$ – arkaia May 23 at 13:31
  • $\begingroup$ This is a common technique use in many Earth Sciences analysis, but you could also see if you get more action in: stats.stackexchange.com $\endgroup$ – arkaia May 23 at 13:32

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.