GLMM for repeated measures in space and time series

Hi Barbara,

The SD explained by your random effects is very small, it actually
looks to be 0 for station! So you may not need them at all, or at
least not the station one.

Also although your raw seed eaten data does follow a binomial
distribution you aren't modelling the counts ie number of seeds
predated. You are modelling the percentage predated, which is why you
are getting the warning message saying your response isn't an integer.
I'm not entirely sure how glmer handles this, it may be rounding your
percentages to the nearest integer, or may not be.

You may want to
consider modelling the actual percentages using some other model
better suited to this? Possible a beta distribution (although I don't
think they can handle 0 and 100?s, but that can be solved by
adding/subtracting a very small amount). Or maybe even modelling the
weight consumed? Or even splitting it into 3 (or more) categories ie
all eaten, non eaten, some eaten.

Chris Howden
Founding Partner
Tricky Solutions
Tricky Solutions 4 Tricky Problems
Evidence Based Strategic Development, IP Commercialisation and
Innovation, Data Analysis, Modelling and Training

(mobile) 0410 689 945
(fax / office)
chris at trickysolutions.com.au

On 11 Sep 2014, at 0:42, "B?rbara Baraibar Padr?" <

barbara.baraibar at udl.cat> wrote:

Hello,

I'm trying to choose the correct model to analyze my data and I need

some help. I'm measuring seed predation (I leave 1 gram of seeds for 48 in
the field and after 48 hours I take what is left and weigh it again). I do
this in the same 50 petri dishes (stations), 25 of which have one weed
species and the other 25 have another and repeat the same in 3 different
fields during 3 months. So, I have a nested design with:

Fixed effects: Weed_species, Date (time)

Random effects: Station, Field

My results are a bit weird in the sense that I have a lot of dishes with

100% seeds predated and some with 0% predated and few in the middle.

My boss says that my response variable follows a binomial distribution

because each seed can be either predated or not, so I have constructed a
response variable with a success column (seeds_predated/initial_seedweight)
and a failure column (initial_seedweight-seeds_predated)/initial_seedweight

I have tried a GLMM like the one below and I would like to know if the

model is ok for this kind of data (repeated measures in space and in
different times) and how I can validate the model. I have done a Binned
residuals plot and almost all my residuals fit within the intervals, do I
need to do something else?

Thank you very much!!!

success<- seeds_predated/initial_seedweight

failure <- (initial_seedweight-seeds_predated)/initial_seedweight

resposta<- cbind (success, failure)

GLMM1<-glmer(resposta ~ Weed_species + Data + (1|Station/Field),

family=binomial)

Warning message:In eval(expr, envir, enclos) : non-integer counts in a

binomial glm!

Generalized linear mixed model fit by maximum likelihood

['glmerMod']Family: binomial ( logit )Formula: Depredacio ~ Especie + Data
+ (1 | Station/Camp)

AICBIClogLikdeviance
479.6651501.8878 -233.8326467.6651

Random effects:
GroupsNameVarianceStd.Dev.
Camp:Station (Intercept) 5.036e-10 2.244e-05
Station(Intercept) 0.000e+00 0.000e+00
Number of obs: 300, groups: Camp:Station, 100; Station, 50

Fixed effects:
    Estimate Std. Error z value Pr(>|z|)
(Intercept)-0.41330.1787-2.3130.02072 *
EspecieLolium0.50080.21002.3850.01709 *
Data2-0.84620.2681-3.1560.00160 **
Data30.74700.25412.9400.00328 **
---Signif. codes:0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' '1

Correlation of Fixed Effects:
(Intr) EspcLl Data2
EspecieLolm -0.600
Data2-0.404 -0.037
Data3-0.4730.0380.299

--
Barbara Baraibar Padro
ETSEA- Universitat de Lleida
Dep. Hortofruticultura, Botanica i Jardineria
Av. Rovira Roure 191
25198 Lleida (Spain)
Telf: +34 973 702912