The real issue is not the distribution of the responses but the
distribution of errors assumed to be additive and normal. Before I
tried bootstrapping, I'd do other things first:
1. What are the response variable(s)? What do they look like on a
normal probability plot (qqnorm)? Might a transformation make the
hypothesis of additive normal errors more plausible?
2. What do the residuals look like in a normal probability plot?
I do the simple things first. If time and money permit and the
problem seems sufficiently important, then I investigate other
alternatives like bootstrapping.
Venables and Ripley, Modern Applied Statistics with S, discuss
bootstrapping, as does Frank Harrell, Regression Modeling Strategies.
hope this helps.
spencer graves
J.Illian at abertay.ac.uk wrote:> Dear all,
> I have a data set o which I'd like to fit lme model. There are three
factors
> one of whoich is nested. This should be easy to do using lme in R, but the
> problem ist that the data is highly non-normal. I was thinking about
> bootstrapping the distribution but don't have much experience of doing
this
> in R and most references I find don't seem to go beyond the
> "two-sample-t-test" setting.
>
> Any suggestions are very welcome.
>
> Thanks
>
> Janine
>
> ------------------------------------------
> Janine Illian
> lecturer in statistics
> SIMBIOS
> School of Computing and Advanced Technologies
> University of Abertay Dundee
> Bell Street
> Dundee, DD1 1HG
> Scotland, UK
> Tel: +44-(0)1382-308488
> Fax: +44-(0)1382-308537
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help