glmmADMB error messages with percentage cover data (Elwyn Sharps)

Hi Ben, Thank you very much for your reply. The plant cover data
was assessed by eye so I should probably be trying a
(zero-inflated) beta distribution as you suggested. Do you have any
tips for which R package to use? I've seen that this is possible 
using gamlss. However, I was wondering if it is straightforward to
add a random effect when using these models? Thanks again

Elwyn

Date: Wed, 19 Nov 2014 16:04:03 -0500 From: bbolker at gmail.com To:
r-sig-mixed-models at r-project.org Subject: Re: [R-sig-ME] glmmADMB
error messages with percentage cover data (Elwyn Sharps)

On 14-11-19 12:16 PM, Elwyn Sharps wrote:

Hi Alain, Thanks for your reply. Sorry I hadn't noticed this
section of the glmmADMB help file and as some of the models
were running fine, I assumed there wasn't a problem with the
family. Have you got any other ideas on how I can deal with the
very large number of zeros in the data? As it's percentage
cover rather than count data, I don't think I can use Poisson
or Neg Binomial. Thanks again Elwyn

A couple of thoughts:

+ I believe glmmADMB *can* handle zero-inflated binomials (I
would try testing with a simple made-up data set to check that
the results are sensible).

+ does your plant cover variable have a meaningful denominator?
(e.g., if you had point-count data with 100 points, or with a
known number of points)  If you're just taking a percentage
that's assessed from a visual impression, or e.g. a measure of
area from remote sensing, then you probably don't have a variable
that's actually binomial (or zero-inflated binomial).  A beta
distribution might be more appropriate, but it too would have to
be zero-inflated.

+ errors of the sort you're seeing are due to some sort of an 
instability in the model -- might be fixable with a more recent
version of the glmmADMB binary files, which are available but I
haven't gotten around to packaging in a new version yet.

Date: Wed, 19 Nov 2014 17:38:40 +0100 From:
highstat at highstat.com To: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] glmmADMB error messages with
percentage cover data	(Elwyn Sharps)

I have collected data on % cover of a variety of plant
species on heathland (in relation to nesting birds) and am
interested in testing the effect of density of grazing
ponies, the proximity to nest site (%cover measured at 1
and 10m from the nest) and the age of the nest (as a
control for vegetation being measured at nests of different
ages). The response is percentage cover and contains many
zeros (see attached data for an example).


I have tried using glmmADMB, with family= binomial and
zero inflation=T. The response was created using cbind e.g.
 y<-(cbind(Data$Plant_cover, Data$Plant_empty)), where
Plant cover = % cover and Plant empty = 100-Plant cover.


Example model:

Fit1<-glmmadmb(y~Grazing*proximity+Age+(1|site),data=Heath,zeroInflation=TRUE,family="binomial")

zeroInflation:   whether a zero-inflated model should be
fitted (only "poisson" and "nbinom" families).

With emphasis on the last part in this sentence.

Kind regards,


Alain

The majority of the models for Species 1 have been running
fine, however I have been getting the occasional error
message and for Species 2, all models give an error
message.


Example error messages:

Parameters were estimated, but not standard errors were
not: the most likely problem is that the curvature at MLE
was zero or negative

Error in glmmadmb(w ~ Grazing * proximity + Age + (1 |
site), data = heathland,  :

The function maximizer failed (couldn't find STD file) 
Troubleshooting steps include (1) run with 'save.dir' set
and inspect output files; (2) change run parameters: see 
'?admbControl'

Warning message:

In glmmadmb(w ~ 1 + (1 | site), data = heathland,
zeroInflation = T,  :

Convergence failed:log-likelihood of gradient= -0.269018



I just wanted to check that this is a sensible way to deal
with these data and if there is anything that I should be
doing to remedy the error messages? I've tried some other
options, e.g. Arc sine transformation couldn't deal with
the excess zeros. I've also considered quasibinomial in
lme4, but have seen Doug Bates comments on the
unreliability of the model output.



Any advice would be very much appreciated. All the best
Elwyn


------------------------------

-- Dr. Alain F. Zuur

First author of: 1. Beginner's Guide to GAMM with R (2014).
2. Beginner's Guide to GLM and GLMM with R (2013). 3.
Beginner's Guide to GAM with R (2012). 4. Zero Inflated
Models and GLMM with R (2012). 5. A Beginner's Guide to R
(2009). 6. Mixed effects models and extensions in ecology
with R (2009). 7. Analysing Ecological Data (2007).

Highland Statistics Ltd. 9 St Clair Wynd UK - AB41 6DZ
Newburgh Tel:   0044 1358 788177 Email: highstat at highstat.com
URL: www.highstat.com

[[alternative HTML version deleted]]