[R-sig-ME] zero one inflated beta mixed model (Ben Bolker)
Highland Statistics Ltd
h|gh@t@t @end|ng |rom h|gh@t@t@com
Wed Feb 12 19:08:53 CET 2020
At present glmmTMB doesn't do zero-one-inflated betas, only
zero-inflated betas. As far as I know your options are (1) use brms,
(2) squish your 1 values to something slightly less than 1, or (3) do
the hurdle model manually (i.e. fit two separate models, one for the
probability that the response== 1, and another (conditional) model for
the zero-inflated beta distribution applied only to the responses <1).
Others on the list may have other suggestions ... (e.g. does INLA
does zero-one-inflated betas?)
In INLA, you would have to do that manually. First, a 0-1 model, and
then a beta model (without the zeros..and 'converted' ones).
Actually....you can also fit a model in which the first column of the
response variable contains the 0-1 data and the second column the
remaining values of the response variable. This is nice for spatial
data; you can test whether the binary part of the model and the
non-binary part of the model have the same spatial correlation, or
whether different spatial correlation terms are needed. Or whether they
share spatial correlation. And you can even deal with barriers (e.g. an
island for coral reef coverage data).
So..to summarize....if it is a 'simple' (zero-inflated) beta GLM or
GLMM, then use glmmTMB (in two steps). If there is spatial correlation,
then use INLA. Once you have fitted both parts, then you can use the
expression in the following snapshot to re-assemble the model:
Dr. Alain F. Zuur
Highland Statistics Ltd.
9 St Clair Wynd
AB41 6DZ Newburgh, UK
Email: highstat using highstat.com
More information about the R-sig-mixed-models