[R-sig-ME] family with Julian dates

[Ben Bolker]

Douglas Bates <bates at ...> writes:

> 
> On Sun, Mar 27, 2011 at 1:50 PM, Iker Vaquero Alba <karraspito at ...> wrote:
> 
> >    Dear list members:>
> >    I am fitting a model with lmer where the response 
> variable is the laying date of several swallow pairs,
> measured as Julian dates (absolute number of days of the year
> elapsed, where 1st Jan is 1).
> 
> >    I was wondering wether I should use family=gaussian 
> or poisson when writing the model, as I am not sure
> wether this Julian dates can be considered as count or continuous data.
> 
> >    What family would you reccommend me to use, and how can I test the
suitability of each of them?
> 
> Definitely not Poisson.  I would suggest a Gaussian family.

  I agree that you should consider the laying date as continuous
rather than discrete.
  If the data (within 'treatment' groups), or alternatively the
residuals, are not plausibly normally distributed (e.g. skewed),
then I would consider transformations of the response
variable.  (This is definitely easier
if all of your predictor variables are categorical.)  See
boxcox() in the MASS package ...