On Fri, 30 Mar 2012, Sven Schreiber wrote:
On 03/30/2012 08:52 AM, Talha Yalta wrote:
> R^2 and F stats are different whether one uses WLS or does a manual
> weighing. I noticed this after first dividing all variables in a model
> by X1 and then doing a wls with a weight variable = 1/X1^2. Everyting
> was exactly the same except R^2, adjR^2, F and p(F).
without really looking into this, this doesn't sound like a bug. Those
statistics change with transformations, and so you have to be really
careful how to calculate them if you do it manually, to make them
comparable. But maybe you checked that already?
Yes, how to calculate these statistics in the context of WLS
is debatable. Gretl calculates them in the same way that R
does. See the help for "wls" for an account of how R^2 is