N-Mixture Model - Min Detections, Non-normal Covariates

75 views
Skip to first unread message

Mike Boyd

unread,
Sep 19, 2021, 8:07:12 AM9/19/21
to unmarked
Hey all,
Long-time listener, first-time caller...er poster. 
I've been having a heck of a time figuring out a couple of questions that are probably straightforward. I'm hoping that asking here may be better than hours of my poor googling skills.

I'm running n-mixture models for single-season point count data (149 sites x 3 visits). I'm wondering:
1. What is the minimum number of detections to include a species for a model or a straightforward way to determine it?
2. Should I standardize (z-score) covariates that don't have normal distributions? (i.e. temp, Beaufort wind, % of habitat type)?

Any help on this would be appreciated!

Thanks,
Mike

Marc Kery

unread,
Sep 19, 2021, 9:22:14 AM9/19/21
to unmarked
Hi Mike,

there's a first time for everything. So here are some elements of a possible answer:
  1. can't say that in generality. Off the top of my head I would say that if you have detections at fewer than 5-10 sites, things may not turn out so good. But an alternative is simply to fit a model and see whether the results "look good" (for which assessment you may then again need to ask in the group). Be sure to repeat the analysis with different values of the K tuning parameter to see that you get stable estimates (see, e.g., Kery, Ecology, 2018).
  2. Typical regression models (and in the widest sense the Nmix is a regression model) make no assumptions at all about any distribution of the explanatory values (this is a quite widespread mis-conception). Hence, strictly there is no need at all to transform them. However, it may help the numerical likelihood maximisation routines if you do. Plus, it may make comparison among covariates more straightforward if you standardize them all, e.g., you can then more easily say which one has the "strongest influence"  by looking at the absolute magnitude of their estimates.
Best regards  -- Marc


From: unma...@googlegroups.com <unma...@googlegroups.com> on behalf of Mike Boyd <mike....@gmail.com>
Sent: Saturday, September 18, 2021 01:17
To: unmarked <unma...@googlegroups.com>
Subject: [unmarked] N-Mixture Model - Min Detections, Non-normal Covariates
 
--
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/da55a50c-d5de-4699-b31c-14e76dd8f793n%40googlegroups.com.

Mike Boyd

unread,
Sep 20, 2021, 1:35:19 PM9/20/21
to unmarked
Hi Marc,
Thanks for the quick reply! I'll give most species a try and see how the results look as I vary K and good to know about the distribution of variables. I'll try standardizing my covariates (normal/non-normal) for my models to see how the MLEs turn out.
Thanks again for the help and I'm looking forward to interacting more with the group in the future. 

Mike
Reply all
Reply to author
Forward
0 new messages