[R-sig-ME] Doubts on zero-inflated model (glmmTMB)

André Pardal @ndre@p@rd@|@@ouz@ @end|ng |rom gm@||@com
Wed May 8 10:53:30 CEST 2019


I am trying to run mixed effect zero-inflated models in order to
investigate spatial variability in density of a intertidal barnacle and
effect of environmental variables.

First, I am trying to determinate the best random model. I have 3 scales of
variability: *region* (2 levels), *sub-regions* (3 levels nested in each
region = 6) and *locations* (number of levels *unbalanced* nested in each
sub-region and region, total = 62). I am using negative binomial family
(nbinom1: best fit than other possibilities).

After selecting such model, I intend to compare it with most parsimonious
mixed model considering random effects plus the effect of a specific

For example:
Best random model: density ~ subregion + location
Best mixed model:   density ~ temperature + subregion + location

My doubts are:

1) Following recommended approaches for regular GLMM's, should I use REML
instead of ML for estimation of parameters?

2) For making models comparable, how should I set zero-inflation component?
I am using ziformula = 1. Is that the best approach? When I try to consider
all terms in zero-inflation component (ziformula = ~.) it usually leads to
errors. Overparameterisation I guess.

3) Once I have my best random and mixed models, should I refit them using
ML method and then compare them (as recommended for regular GLMM's)?

4) (Finally) How to validate model? I am trying to use DHARMa package for
checking simulated residuals vs fitted values, but it doesn't work well for
models with strong random effects (my case).

Thank you so very much.

My best,


Visiting PhD student
School of Ocean Sciences
Bangor University
Menai Bridge, Anglesey, UK

	[[alternative HTML version deleted]]

More information about the R-sig-mixed-models mailing list