jarioksa
commented
3 years ago How do you construct your data? Usually the Hmsc data sets are species/taxa/OTUs by sampling units, and in this kind of data there is no way of identifying individuals. Can you construct your data set so that your individuals can be identified (that is, they have names)? In general, in data of species by sampling units, co-occurrence preference, or aggregation, appears as large variances for species. Basically, if your individuals have uniform random distribution over space, the number of individuals per sampling unit follows Poisson distribution. If the distribution is non-uniform and the individuals aggregate in some sampling units (and conversely, some sampling units are often empty), then the variance is higher than in Poisson. In Hmsc distr="lognormal poisson" is a model that can estimate the scale of extra-Poisson variance, but this model has poor mixing, and you should inspect the convergence carefully. If you have spatially dense sampling pattern ('dense' in the scale of movement of your organism), the strength and scale of spatial scale can also be called by intra-species attraction. You should also be aware that there are many other processes than intraspecies attraction that can cause extra-Poisson variance (beginning from a wrong or poor model you fit), and you cannot infer the process from a pattern.
I’m seeking advice on how to quantify intraspecies associations using HMSC, i.e., do individuals of the same species co-occur. Is this something that is already quantified and I just overlooked it in the lit.? @gtikhonov @jarioksa