[R-sig-ME] Question regarding GLMMs and proportional data

ONKELINX, Thierry Thierry.ONKELINX at inbo.be
Thu Jan 22 09:55:43 CET 2015

Dear Rachel,

IMHO you should choose the most appropriate distribution depending on the nature of the response. If it comes for  a number of Bernoulli trails, then the binomial family makes more sense than the gamma. I presume that the denominator is integer? If the denominator is real, then the data generating process is not a set of Bernoulli trials.

I'm not sure if you can compare AIC among models with a different family.

Best regards,


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
+ 32 2 525 02 51
+ 32 54 43 61 85
Thierry.Onkelinx op inbo.be

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

-----Oorspronkelijk bericht-----
Van: R-sig-mixed-models [mailto:r-sig-mixed-models-bounces op r-project.org] Namens Rachael Eaton
Verzonden: woensdag 21 januari 2015 20:59
Aan: r-sig-mixed-models op r-project.org
Onderwerp: [R-sig-ME] Question regarding GLMMs and proportional data


I am having some trouble identifying the best model structure for my data.
I am trying to construct models to look at a response variable of proportions. These data are highly skewed toward zero, with a max value of 0.2. I also have one random variable and a few fixed effects. Therefore, I decided to use GLMM. I applied a log-transformation of (1 + the response variable). I then modeled these using glmer with a gamma family with an inverse link.

Following concerns from a colleague, I also tried modeling the untransformed data with a binomial family, since I know the denominators for my proportion data. However, the model fit for this approach was not as strong as for the gamma approach. AIC strongly favored the model using the gamma distribution over the model using the binomial distribution.

If you have any recommendations or suggestions on how best to model these data I've described, I would greatly appreciate it.

*Rachael Eaton*
PhD Candidate
Department of Zoology
Graduate Program in Ecology, Evol Biology & Behavior Michigan State University rachaeleaton.weebly.com <http://www.rachaeleaton.weebly.com>

        [[alternative HTML version deleted]]

R-sig-mixed-models op r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

More information about the R-sig-mixed-models mailing list