I'm looking for a dataset which I can use in R modelled by the glm function with family=poisson. I need there to be at least four predictor variables and the count has to be the explanatory variable.
There are a couple of packages which provide quite a few count regression examples. See for example package
countreg on R-Forge (https://R-Forge.R-project.org/R/?group_id=522), or
AER (https://CRAN.R-project.org/package=AER) or
COUNT (https://CRAN.R-project.org/package=COUNT) on CRAN. The
AER package is more general but count data sets include
For most of these datasets/models there are overdispersion and/or excess zeros present so that a more general model fits better, e.g., negative binomial, zero inflation or hurdle model.
If you need a well-fitting Poisson model, then
AER would be worth a look. If it is ok to discuss the limitations of the Poisson model and show some generalizations, then I like the
CrabSatellites data in the
countreg package. See the examples on the corresponding manual pages for concrete models and illustrations.
Based on your requirements, the British Doctor's Smoking and Lung Cancer dataset is ideal.
Regarding GLM with a Poisson family function, most Poisson regression packages can also handle GLM's log and linear link functions. There are also geometric mixture models which can be handled by some packages. The Explorer (free) package will identify what power-link function is best for a geometric mixture model between additive and multiplicative. In summary, you commonly don't need GLM for many flavors of Poisson regression.