[R-sig-ME] lmer or glmer?

[ONKELINX, Thierry]

Dear Michael,

I'd rather assess the normality graphically. Non-normality can be caused by missing variables or interactions as well. I tend to avoid transformations as much as possible since they complicate the interpretation of the results. However, a priori knowledge about the relationship should be honored. E.g. if you know that a relationship is exponential, then use this information and transform either the response or the covariate to make things linear again.

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
Belgium
+ 32 2 525 02 51
+ 32 54 43 61 85
Thierry.Onkelinx op inbo.be
www.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 Michael Jackson
Verzonden: zondag 11 januari 2015 21:59
Aan: r-sig-mixed-models op r-project.org
Onderwerp: [R-sig-ME] lmer or glmer?

Hi Ben,

Im wondering if you can help - ive been going round in circles for some weeks now and busily getting more bamboozled. Im basically wanting to use model selection (AICc scores, deltas etc) based on "lme4" outputs to identify the key nutritive drivers of animal consumption. Ive attached my data for you as I think it'll help greatly in showing my issues....My questions are...

1) Should I use REML=FALSE if wanting to compare mixed effects models via their AICc scores, deltas etc. where the random effects are always incorporated, but the fixed effects change between models. Everything ive read suggests yes...do you agree?

2) Ive got 14 different models to run - some additive, some interactive...an example of my code is as follows
m2 = lmer(Consumption ~ Fat + Protein + (1|Product/Trial), data=mydata, REML=FALSE)
m3 = lmer(Consumption ~ Sugars * Fat + (1|Product/Trial), data=mydata, REML=FALSE)

However - no matter which model I run and/or how I scale or transform my variables (both predictor and response) a shaprio wilks test, for example  shapiro.test(resids(m2)) is always significant ....Would using "glmer" be my option here? After reading your paper "Generalised linear mixed models: a practical guide for ecology and evolution" and this article on stackexchange<http://stackoverflow.com/questions/25356633/error-message-when-performing-gamma-glmer-in-r-pirls-step-halvings-failed-to-re>....Would the following example code be OK based on my raw untransformed data?

glmer(Consumption ~ Fat + Protein + (1|Product/Trial), data=mydata, family=Gamma(link=log))

Id value your comment and assistance as I dont want create loads of models and then make incorrect inferences of my AICc outputs based on flawed regression outputs...

Thanks in advance

Mike
..............
PhD Candidate
Centre for Biodiversity and Restoration Ecology Room KK 411 Kirk Building Kelburn Parade Wellington 6012 _______________________________________________
R-sig-mixed-models op r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
Disclaimer<https://www.inbo.be/nl/disclaimer-mailberichten-van-het-inbo>