[R-sig-eco] Zero inflated data on some levels of a random factor in mixed models

[Mollie Brooks]

Dear Krista,

glmmADMB will only model a zero inflation constant that equally applies to all species (i.e. no predictors for number of zeros). It sounds like in your case, zero inflation varies by species. The easiest thing to do is to model the zero/non-zero parts separately as you suggest.

If n is nectar and dat is your data frame, then in lme4, this might look something like

m0=glmer((n>0) ~ species + (1|plant), family=binomial, data=dat)

m1=lmer(n ~ species + (1|plant),  data=subset(dat, n>0))

Note, with only 4 species, it should be included as a fixed rather than random effect. 
Do you have repeated measures of individual flowers? If not, then there’s no need to include it as a random effect.

Cheers,
Mollie
------------------------
Mollie Brooks, PhD
Postdoctoral Researcher, Population Ecology Research Group
Institute of Evolutionary Biology & Environmental Studies, University of Zürich
http://www.popecol.org/team/mollie-brooks/


On 15Sep 2014, at 9:41, Krista Takkis <krista.takkis at gmail.com> wrote:

> Dear all,
> 
> 
> 
> I have a set of data on nectar volumes from four plant species. Two
> species have ample zeroes in the data (for one species almost 1/3 of
> the flowers had no nectar), but two species don’t have excessive
> zeroes in the data and have a normal distribution. I am trying to find
> out, what would be the correct way to model the trait responses in
> this situation. I would like to analyse all four species in one mixed
> model, but should I try to account for the zero inflated data, if the
> problem is only with half of the species? And if so, then how could I
> do it properly?
> 
>   An answer to an earlier question on the topic of zero inflated data
> (https://stat.ethz.ch/pipermail/r-help/2014-May/374444.html) suggested
> to model the zero and non-zero data separately. With not too many
> zeroes in case of two species and wishing to combaine all four
> species, I probably cannot use this method in this case or could it be
> possible somehow? Till now I have used function glmmPQL (MASS) to
> model this trait with species/plant/flower as a random factor.
> However, as far as I know, this function does not allow to account for
> the zero inflated data. I found that MCMCglmm and glmmADMB would allow
> to account for zero inflated data, but before learning to use a new
> package I wanted to ask, whether this would be the correct way to
> approach this kind of data in the first place and whether there might
> be a way to do this using glmmPQL function?
> 
>   Could you give me some suggestions, what might be the best way to
> deal with this kind of data?
> 
> 
> 
> Thank you in advance,
> 
> 
> 
> Krista Takkis
> 
> Department of Geography
> 
> University of the Aegean
> 
> _______________________________________________
> R-sig-ecology mailing list
> R-sig-ecology at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology


	[[alternative HTML version deleted]]