[R-sig-Geo] testing autocorrelation on residuals of a binomial mixed-effect model

[Henri Vallès]

Dear all,

I have multiple geo-referenced sites and at each site I have multiple
sampling units where I record the presence/absence of my organism of
interest along with environmental descriptors. I am using
presence/absence as my dependent variable in a binomial mixed-effects
model (glmmML function) with substrate descriptors as fixed effects
and sites as random effects. I am using the "average presence" at a
site, i.e. "proportion of sampling units where the organism was found"
and the site coordinates to examine spatial auto-correlation across
site-level values using the moran.mc function of the "spdep" package
(after creating a spatial weighting object) and I do find a
significant auto-correlation.

I would be interested in finding out whether the Pearson residuals of
the binomial mixed-effect model can be used in the same manner
(averaging residuals at a site and using moran.mc function for a test
across sites) to assess whether the spatial correlation was linked to
environmental features.

Would this make any sense?

Best regards

Henri