Skip to content

data distribution for lme

6 messages · peyman, Andrew Robinson, Rolf Turner +2 more

#
Hi folks,

I am using the lme package of R, and am wondering if it is assumed that
the dependent factor (what we fit for; y in many relevant texts) has to
have a normal Gaussian distribution? Is there any margins where some
skewness in the data is accepted and how within R itself one could check
distribution of the data?

Thanks,
Peyman
#
This is not really an R question -- it is statistics.
In any case, you should do better posting this on the
R-Sig-Mixed-Models list, which concerns itself with matters like this.

However, I'll hazard a guess at an answer: maybe.  (Vague questions
elicit vague answers).

Cheers,
Bert
On Tue, Dec 10, 2013 at 6:55 AM, peyman <zirak.p at gmail.com> wrote:

  
    
#
See inline below.
On 12/11/13 11:28, Bert Gunter wrote:
No! Nay! Never!  Well, hardly ever.   The ***y*** values will rarely be 
Gaussian.
(Think about a simple one-way anova, with 3 levels, and N(0,sigma^2) errors.
The y values will have a distribution which is a mixture of 3 
independent Gaussian
distributions.)

You *may* wish to worry about whether the ***errors*** have a Gaussian
distribution.  Some inferential results depend on this, but in many cases
these results are quite robust to non-Gaussianity.

There.  I have exhausted my knowledge of the subject.

     cheers,

     Rolf
#
Thanks Rolf and Andrew. I was entirely too careless and should take a
trip to the woodshed (google "David Stockman woodshed"  for the
reference).

The correct answer therefore is: maybe for the residuals, for the
"right" model, of course.

But I still think the crowd on r-sig-mixed-models is the right place
to hash it out, if anything meaningful can indeed be made of it.

Cheers,
Bert
On Tue, Dec 10, 2013 at 5:33 PM, Rolf Turner <r.turner at auckland.ac.nz> wrote: