Hello all,
I work with a floral system that present three floral morphs genetically
determined. In isopleth equilibrium, is expected an equivalent floral morph
ratio in a population (1:1:1). However, deviation from this expected are
common and has some ecological and evolutive implications.
I have floral morph frequency of 73 populations which I want to know if
there is any morph less frequent. My main interest concern relative terms,
however smaller populations will have higher chances of deviation by
stochastic process.
So, I want to test it weighting sample or population size. To evaluate it,
I did a GLMM considering population nested to macro region as random effect
and morph as fixed effect I used the weights argument with the absolute
value of each morph per population and use binomial distribution. However,
I am not sure if I can use binomial distribution this way. Also, some
morphs do not appear in the population, so I summed a lowermost value to
the weights column to "trick" the analysis.
Here goes an example of my data and model:
#######################################################################################################################################################
# data.frame
cmr <- read.table ("http://pastebin.com/raw.php?i=iW41BDUE";, header = T)
# package
library(lme4)
# GLMM weighted
set.seed(2714)
m_bin <- lmer ( perc ~ -1 + morph + ( 1 | pop / region), weights = abs +
0.00000001, data = cmr, link = "logit", family = "binomial")
summary (m_bin)
#######################################################################################################################################################
# I also did a model with inflated beta distribution, however its results
has difficult biological interpretation
#######################################################################################################################################################
# Beta inflated model
library(gamlss)
m_beta <- gamlss(perc ~ -1 + morph + random (pop), weights = abs, family =
c("BEINF"), data = cmr, contrasts = cmr$morph )
plot(m_beta)
summary(m_beta)
#######################################################################################################################################################
Is there a better/proper way to test this?
Do I need to perform any prior data transformation to achieve binomial
premisses?
Any additional help and suggestions are much appreciated!
Thanks in advance
Nicolay
# []'s
--
Nicolay Leme da Cunha
Biólogo, Mestre, Doutorando em Ecologia e Conservação
Universidade Federal de Mato Grosso do Sul, 79070-900
Campo Grande, MS, Brasil
E-mail: [email protected]
lattes.cnpq.br/5916316648872099
[[alternative HTML version deleted]]
_______________________________________________
R-sig-ecology mailing list
[email protected]
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
