I am confused. First i have no experience with Rstudio and second I can't imagine that the syntax there and as in R are as different as you make it seem: I would start with some references: R for everyone: advanced analytics and graphics | Lander, Jared P | download (b-ok.cc) and Using R and RStudio for Data Management, Statistical Analysis and Graphics | Nicholas J. Horton, Ken Kleinman | download (b-ok.cc) and make sure your syntax is correct e.g. how are these R variables: (note the strange spelling as well)

interacion item type, family=poisson)

Response variable (461 interactions): count data

Predictor variables ('species' (n = 10), 'individuals' (n = 39), 'interaction item type' (n = 3)): categorial data

Does Rstudio allow 3 word variable names? Note that you seem to be confusing GLM and glm. try this: write a good research question for this study. What do wish to accomplish? Note ANOVA is not an R function name. Further,

Model: myGLM

I completely agree with David Eugene Booth in his view, that it is very hard to understand, what exactly you are aiming for in your models and hence what causes the problem.

I think one issue that hampers understanding and might also be responsible for your problem is the way your models include, presumably all observations in factors. From what I understand your dataset consists of different species with 39 individuals each. Thus you have (at least) 390 observations whereas species is a factor that can be treated as a predictor, however you can not treat individuals as a factor because these are unique individuals. Each Individual corresponds to only one species (or am I missing something here?). You can however incorporate the nested structure of individuals in species but that would require a multilevel approach and I would support David Eugene Booth s advice to provide research questions, a plan for analysis and simpler models first.

David Eugene Booth and Sven Greving:

Thank you very much for the answers!

As RStudio is just a user interface for R I agree it was redundant to report the version (using R x64 4.0.3 btw).

The function glm() also originates from the package 'stats' - not from 'lme4'. My fault - I confused it with my GLMM functions! Sorry for that!

I tried to condense my problem. Obviously, I was not successful with that – so I try it again.

My research question here is simply if the interaction with an item (response variable) is statistically significant affected by the species and/or individual and/or interaction item type (the three predictor variables).

Since the response variable is n = 461 (interactions), this is the total data point count for my model. Every interaction (data point) is therefore categorised in species AND individual AND item (e.g. individual4 from species2 had an interaction with item1). Within the species I had different sample sizes of individuals. Individuals showed different interaction rates. And the interaction items were used in different intensities (more interactions with item1 than with item3). Every interaction is one data point and done by a certain individual (in total 39 individuals) as member a certain species (in total 10 species).

Sven Greving: That means I have 461 observations not 390. The amount of interactions per individual, species and item varies. The amount of individuals per species varies as well. Sorry if I was not clear here as well.

The multilevel approach is given due to utilising GLM with additive predictor variables (I do not use interaction terms).

Regarding the syntax I'm familiar with the function itself. All following bold variables/models are just named before in my R script. For simplification here I renamed them disregarding any syntax rules in R to clarify my model itself.

myGLMChisq)

Species 0

Individual 1365.6 29 < 2.2e-16 ***

item 5082.8 2 < 2.2e-16 ***

Thanx again. I hope I could formulate myself better this time.

I'm looking forward for any suggestions! :)

PS.: the response variable 'interaction' contains count data which means e.g. 20 interactions per observation unit in the experiment.

David Eugene Booth: Thanx for the literature - I will looking for solutions there as well. :)

Why don't you just show your R code and output? I think one (of perhaps several, I can't tell from your text) is treating n=39 individuals as fixed effects, and this is a lot. But I can't tell.

Thanks also @ Daniel Wright and Henning Teickner

I attached is the associated code.

Maybe it makes thinks more clear.

Henning Teickner

Henning Teickner

Your code:

length(unique(paste0(species, "_", individuals, "_", 'interaction item type')) == length(interaction)

results in 39 as expected.