Re: st: st)why log linear model is better?

[Richard Goldstein <richgold@ix.netcom.com>]

If you really want to use things like R-squared when the dependent
variables differ, look at Patrick Royston's -brsq- command, originally
in STB-14

Rich

woong-tae chung wrote:

> Thanks for the comments. Both are helpful
> WT 
> ----- Original Message ----- 
> From: "Nick Cox" <n.j.cox@durham.ac.uk>
> To: <statalist@hsphsun2.harvard.edu>
> Sent: Tuesday, August 29, 2006 6:10 AM
> Subject: RE: st: st)why log linear model is better?
>
>
>
>
> > The same applies, even more strongly, to 
> > interpretation of standard errors, which 
> > are no longer on the same scale. 
> > In addition, look at diagnostic plots
> > such as residual vs fitted, in each case. 
> > Nick 
> > n.j.cox@durham.ac.uk 
> > David Greenberg
> >
> >
> >
> > > YOu have to be careful in comparing R-square for a regression in which
> > > the dependent variable has been transformed with one in which 
> > > it has not
> > > been transformed. The dependent variables are not measured on the same
> > > scale, and this can throw off the comparison. IF it does turn out that
> > > the equation with transformed variables provides a better fit, the
> > > explanation will not be a statistical one, but a substantive one. The
> > > equation with transformed variables better describes the processes at
> > > work. Only someone with a knowledge of those processes could offer an
> > > explanation as to why that is. 
> > Woong.Chung@colorado.edu
> >
> >
> > > > I need a help to find out reasonable explanation for my model 
> > > > specification.After running simple linear regression using OLS, 
> > > > ROBUST standard errors(due to
> > > > heteroskadasticty) and SUR, it turned out that log linear 
> > > regression:
> > > > log(y)=a1+a2log(x1)+a3log(x2)+a4log(x3)+...e
> > > > seems to be fit so well in any cases rather than level or other 
> > > > transformationregressions:
> > > > y=a1+a2x1+a3x2+a4x3+.....+e
> > > >
> > > > in terms of lower standard errors and higher R squares.
> > > >
> > > > I am looking some explanations why this happens and also want to 
> > > > know how tell
> > > > whether the log linear regression method is my best specification
> > > >
> > > > Mostly y x2 x3 are ratio and x1 is level( but x1 is not a 
> > > > denominator of other
> > > > ratios)
> > > > Within my knowledge, the log transformation would be helpful for 
> > > > multiplicativedata set. I don't know it would be applied to my case

*
*   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/