Yoshiro - In answer to your questions:
1. Is your "y" a binomial proportion? The arcsine(sqrt(y)) transformation
was designed to make biniomial proportions closer to being normally
distributed with equal variance. With glm and a binomial family you can
indeed do regression analysis on you original variable.
2. Nonlinear least squares applies to the way parameters occur in the
regression model. The non-linearity has nothing to do with the distribution
of the dependent variable. (Any kind of)least squares is equivalent to
maximumum likelihood when you have a homoscedastic normal dependent
variable. So if you don't have a normally distributed variable, noinlinear
least squares won't help you.
3. There is no built-in Stata command for doing stepwise glm, etc.This
question has been asked before - there may be an FAQ on it.
Al Feiveson
I am applying regression analysis
to a dependent variable 'y', which has non-normal
distribution.
arcsine(square root( y )) considerably
transformed 'y' to a normal distribution.
(1) However, are there any better method (such as glm ?)
to apply regression analysis to this y as it is?
(2) Are there advantages in using non-linear least square method (nl)
for this case?
(3) In addition, if "regress" command is to be used,
are there any stepwise command, equivalent to "sw"
for maximum likelihood estimate?
Thank you for your assistance in advance.
_________________________________________________________________
Help STOP SPAM with the new MSN 8 and get 2 months FREE*
http://join.msn.com/?page=features/junkmail
*
* 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/
*
* 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/
