0

Is there any way to compare the fit of quasi-Poission and negative binomial models? If so, can it be done in R?

chl
  • 50,972
  • 18
  • 205
  • 364
JKO
  • 673
  • 2
  • 8
  • 18
  • 2
    What is your motivation for comparing these models? Have a look at [Verhoef and Boveng, 2007, Ecology, 88(11)](https://esajournals.onlinelibrary.wiley.com/doi/10.1890/07-0043.1). See also [this post](https://stats.stackexchange.com/questions/66412/comparison-negative-binomial-model-and-quasi-poisson) on comparing quasi-Poisson and negbin models (in R). – Jessica Burnett Nov 26 '18 at 22:36
  • 1
    @Jessica B It is precisely the issues highlighted by Verhoef & Boveng 2007 that motivate me to compare these models. Without some way to compare them (such as through a simulation) there is not a good way to know which one is the best choice for your data, but, critically, they are likely to yield different results. So - they are not equivalent, but it is not clear how to choose between the two of them without running some kind of simulation. I have seen the post by Elena Spark and Gavin Simpson, but did not find that it answered my question. It may be that the answer is simply "no". – JKO Nov 27 '18 at 16:01
  • In your OP you ask for an answer to, "is this possible in R," rather than "is this a good idea and if so, why?". Perhaps the answer to the latter is already posted in stats exchange? If you find something, please share as I am curious the answer(s). – Jessica Burnett Nov 27 '18 at 16:17
  • 1
    @Jessica Burnett It is not only a good idea to compare to find the best match for your data, but it is irresponsible not to (per Verhoef & Boveng 2007). Ironically, it seems (to me) like there is no good or clear way to do this. Hopefully, someone will correct me on that last point! – JKO Nov 27 '18 at 17:31
  • 1
    To be clear--my previous comment was to question whether your original question should include the term "in R"? – Jessica Burnett Nov 27 '18 at 18:03
  • @JessicaBurnett Good point. I suppose that I do want to know that more generally, but, if the answer is yes, it would be ideal (for my purposes) to know whether it is a problem that someone has solved using R. – JKO Nov 28 '18 at 16:35

0 Answers0