As long as you don't have major outliers already or create such by a
unwise transformation -- and clearly I advise against that -- the shape
of the marginal distribution of a predictor is likely to be a much
smaller deal than whether the relationship between the response and the
predictor, conditional on other variables, is linear or not.
Reducing a nonlinear relationship to something more nearly linear is
then the main motive for a transformation. As pointed out more than once
in this thread, powers such as square roots or squares would deal
smoothly with zeros. Which flavour you want depends on your data, which
we can't see.
I am not clear that any kind of folded transformation is natural for
a predictor, bounded or not.
Marck Bulter
I totally agree, conversion is an awfull solution, fitting the data to
the model. But still I have to do something with the heteroskedacity and
the non normal resid's.
As suggested in your transit files, I will give folded transformation a
try.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
