[R] comparing GLM coefficients & repeatability

[RCulloch]

Many thanks for taking the time to read this! 

I am looking at the repeatability of behaviour between re-sighted
individuals across discrete time periods (annual breeding seasons). My
approach was to run a GLM (with a logit link - the data are proportional,
presence v. absence of behaviour) for each breeding season. I included the
re-sighted individuals as a factor (categorical variable) (i.e. the models
only contained individuals that were seen in all of the breeding seasons).
Inevitably the variables that are retained in the best models are not the
same for each breeding season and in one (out of 3 cases) individual is not
retained within the best model (although I suspect that is a product of a
considerably smaller sample size for that breeding season). I use the best
model that has retained individual id and extract the coefficients of the
individuals. I then use the ICC command in the package psych to test for
repeatability in these values over the three breeding seasons. The results
are in fact repeatable, which does support the basic analyses using just the
behaviour (without trying to account for potential covariates), which is
encouraging. However, I have had a look on nabble and other forms to see if
this is at all statistically sound or if I am making a fundamental error in
how I am treating the coefficients. I have found a couple of posts, but I
don't think that they relate directly to my question. 

I appreciate that some may suggest using mixed-effects modelling with
individual as a random effect. My issue is that the behaviours I am
interested in are very rare and are best suited for a beta-binomial
distribution (tested using Ben Bolker's script/e.g. in his book). And such a
distribution is not available in lme4. Therefore, I'm trying to find another
approach to assess whether individual is important in predicting a
behaviour, and whether individuals are repeatable/consistent in this
respect. 

Any advice would be most appreciated, 

Best wishes, 

Ross



--
View this message in context: http://r.789695.n4.nabble.com/comparing-GLM-coefficients-repeatability-tp3772844p3772844.html
Sent from the R help mailing list archive at Nabble.com.