[R-sig-ME] error question

[Ben Bolker]

Alejandro Martínez Abraín <a.abrain at ...> writes:

> 
> Hi everyone,
> 
> I am running Poisson mixed models in R and I get this error message:
> 
> Mensajes de aviso perdidos
> In aic(y, n, mu, weights, dev) : non-integer x = 2.500000
> 
> Any clues on what it means and how to solve it???
> 
> Thanks
> Alejandro


  To be precise, this is a warning rather than an error message, which
means that R should have given you an answer; it is just telling you there
might be something wrong with it.

  As a general rule of thumb, Poisson models only make sense with integer
responses (in particular, they assume that the variance is equal to the
mean: if your response variables are measured in units of (e.g.) individuals
**per m^2** rather than counts, you could nonsensically change the answers
simply by changing units to per cm^2 or per km^2.  There are situations
where running Poisson models with non-integer response values is OK, but
if you're not sure then you should assume it's not.
 
  If you have counts divided by some fixed area (or time), analyze the
raw counts rather than the densities.  If you have counts divided by
different fixed areas for different observations, then analyze the
counts and use a term
+offset(log(area)) in your model to account for the variation.