Skip to content

Binomial glmer() with zero-inflated data

2 messages · Jessie Barker, Mollie Brooks

#
Dear mixed-model enthusiasts,

I have a question about how to model zero-inflated data (I have several 1's
too, but more 0's).

The data are observations of bees visiting flowers to collect nectar. A
visit can either be "pollination" or "nectar robbing" (the bee collects
nectar from a hole through the side of the flower). Each bee made a
sequence of visits, and the response variable that I am interested in is
how many times a bee robbed nectar. (In many cases a bee did not rob nectar
at all - hence the zero-inflation - and in a few cases it only robbed
nectar.) I observed each bee only once.

The predictor variable I am interested in is the proportion of flowers on
the plant that have holes in them (thus enabling nectar robbing). I have
multiple measurements of the response variable on each plant, but only one
measurement of the predictor variable. So I will include plant as a random
effect.

I think I need something like this:

model1 <- glmer ( cbind(rob,pollinate) ~ prop.holes + (1|plant),
data=mydata, family=binomial)

But I don't think that can handle the zero-inflation. Does anyone on this
listserv have any advice?

Thanks in advance - I really appreciate any suggestions!

Best wishes,

Jessie Barker

Junior Fellow
Aarhus Institute of Advanced Studies, Denmark
http://aias.au.dk/aias-fellows/jessica-barker/
#
Dear Jesse,

You probably don?t need to worry about zero-inflation. The binomial distribution should be able to handle the 0s. Your model seems reasonable to me. 

Best regards,
Mollie

???????????
Mollie E. Brooks, Ph.D.
Research Scientist
National Institute of Aquatic Resources
Technical University of Denmark