Thierry, thanks a lot for your quick response.
We sampled each site (small water points) only once and we noted the
different PE (human disturbance types). However for our analyses we have
done as if each site was sampled more than once. That is we repeated each
row as many times as different perturbance types were recorded in each
site.
Site
PA
AL
PE
1
0
38
1
1
0
38
3
2
0
138
1
3
0
382
1
3
0
382
3
4
0
382
1
4
0
382
3
Our final aim is to obtain probabilities of presence/absence of each
amphibian species in a site in relation to the different types of
disturbance and altitude, using the "invlogit" function. I think the
"cbind" function is not useful in this case because we are not modelling
proportions.
What do you think?
Best regards,
Silvia
2015-04-30 15:25 GMT+02:00 Thierry Onkelinx <thierry.onkelinx at inbo.be>:
Dear Silvia,
I presume that the values of AL and PE are constant within the site. Did
you sample different locations within each site simultaneous ? Or did you
sample the same location at each site but at different dates?
In case of different locations per site you can simplify your model to.
glm(cbind(n.present, n.absent) ~ AL + PE, family = binomial) With n.present
the number of present locations per site.
Best regards,
Thierry
ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature
and Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium
To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data. ~ Roger Brinner
The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey
2015-04-30 14:50 GMT+02:00 Silvia Rodr?guez Fern?ndez <sileiris at gmail.com
Dear list members,
I?m a PhD student in trouble. I?m running a mix effects model with a
dependent variable (PA: presence/absence, 0/1), one fixed explanatory and
continuous variable (AL: altitude), one fixed factor (PE: initially 16
levels, but reduced to 4 to reduce complexity) and one random term (2421
sites). Basically, the structure of a logistic regression but with a
random
term to prevent temporal pseudoreplication.
model1<-glmmadmb(PA~PE+AL+(1|site), family="binomial")
My data are quite unbalanced becouse I?ve many more zeros than ones. I?ve
tried making a random selection of absences but I get similar problems
than
when using the whole dataset.
I?m getting an output of results in R, but also getting a warning of lack
of convergence, such as:
Convergence failed:log-likelihood of gradient= -0.0195034
Can I trust my results in spite of the warning?
What other alternatives do you suggest?
I?ve tried with the classical lmer and glmer, and I also get convergence
problems as expected.
I?ve also tried with the MCMCglmm package, but I?ve problems with the
specification of the priors.
Any help is welcomed.
Silvia
[[alternative HTML version deleted]]