mariana fernandez olalla
2009-Jul-30 10:26 UTC
[R] temporal and spatial pseudorreplication with lmer
Dear all,
I am trying to fit a generalized linear mixed model to deal with both temporal
and spatial pseudorreplication.
I have repeated seasonal measurements (3 seasons during 3 years and 2 season
during the last year, factor named percod, with 11 levels) of a bird
presence/absence (pres.f) on 14 artificial ponds (charca.f). The ponds are
integrated in 4 greater spatial units (zepa). I have 11 explanatory variables,
some of them are continuous
(prof1,prof2,prof4,capac,perim_veget,div_veg,div_herp) and some are factors
(ungul.f,ganado.f,turb.f,peces.f). Not all artificial ponds have been monitored
in all seasons, so I have some missing values in the dependent variable.
I have thought that the correct model should be, ignoring the spatial
pseudorreplication:
m1<-lmer(pres.f~ganado.f+ungul.f+turb.f+prof1+prof2+prof4+capac+perim_veget+peces.f+div_veg+div_herp+(percod|charca.f),family=binomial)
AIC BIC logLik deviance
124.4 184.2 -42.18 84.35
Random effects:
Groups Name Variance Std.Dev. Corr
charca.f (Intercept) 4.9533e-11 7.0380e-06
percod 5.3095e-13 7.2866e-07 -0.995
Number of obs: 147, groups: charca.f, 13
The standard errors of the random effects are really small. I am not sure how to
interpret it, but I suspect it is not a good thing. Can someone help me?
Besides, to take into account the spatial pseudorreplication I have thought to
add a random effect more: (1|zepa/charca.f). When I run the model it appears
Warning message:
In mer_finalize(ans) : singular convergence (7)
I have read that it means that the data does not support such a great number of
parameters estimation. Am I correct? What should I do?
I hope someone could help me. Thanks a lot!
Mariana Fernandez-Olalla
Ph.D Student
ETSI Montes. UPM
Madrid (Spain)
marianaolalla@hotmail.com
_________________________________________________________________
[[alternative HTML version deleted]]
