sites without species

[Claudio]

Hi all,

I am fitting Joint species distribution model and Spatial  joint species distribution model without imperfect detection (lfJSDM and sfJSDM). The data are presence/absence of amphibians. Due to the target group and to the environment, I can (reasonably) assume detection=1. Thus, in addition to investigate species correlation and ecological covariates which affects species presence, I am also interested in understand why some sites (about 30%) do not host any amphibian.

I think the proper way to manage this is (i) to include in the dataset also the sites without amphibians, and to analize the covariate effects at community level.

Alternatively, (ii) I could include a further "dummy species" which occurs only in the sites without amphibians. I think such model would not be useful for species correlation, but it could furnish information about why some sites do no host amphibians. 

I even run such datasets, and I got different models as the best fitting, starting from the fact the best for (i) was a spatial factor model and the best for (ii) was a latent factor model.

The questions are: am I right (i) is better than (ii)? How do lfJSDM and sfJSDM handle sites without species? Any further suggestion?

thanks in advance

Claudio

[Jeffrey Doser]

Hi Claudio,

I would suggest using the first approach you outline there. Sites without any species observations will still serve to provide information about the factors driving species and community-level occurrence patterns. If there is a consistent pattern in where locations without any species observations occur, that could be reflected in either (a) covariates in the model or (b) the latent (spatial) factors. You could visualize the latent factors in maps which could help to reveal some interesting community level patterns (see the appendix and github repository in this publication for some examples of how to do this). Including a "dummy species" in the way you outlined would lead to bias in the community-level estimates and as you point out would make the species correlations and latent factors much more difficult to interpret.

Jeff