Dear Jane,
Well it depends what transformation you ended
up with. If it was a log for example, then your model will have
become a multiplicative rather than additive one. The one very
important thing to remember of course is that if the power
transform is <0 then, even though the transform is monotonic,
the direction of effects is reversed. That is a positive coefficient
on an independent variable says that increasing it decreases
the dependent variable.
On useful thing to do is to produce predicted values of the
transformed dependent variable from a grid of the independent
ones, backtransform it to the original scale and then do
plots of predicted vs independent. The shape and nature
of the effect of the independent variables should then
become apparent unless you have a lot of high order interactions
in the model,
Best Wishes,
Philip
On Fri, 2 Aug 2002 16:00:00 +0100, Jane McFerren
<[log in to unmask]> wrote:
>Dear list
>
>
>Having carried out a Box Cox transformation of a dependent variable, I am
having
>trouble interpreting the coefficients in the output of my linear model.
>
>Can anybody point me in the direction of a webpage or paper that explains
how
>coefficients (beta's) in a model with a Box Cox transformed dependent
variable can be
>intepreted.
>
>Thanks in advance
>Jane
>
>______________________________________________________
>Check out all the latest outrageous email attachments on the Outageous
Email Chart! - http://viral.lycos.co.uk
>
>
|