Hello all,
I am currently carrying out regression models with two versions of
the Box-Cox transformation (with and without a shift parameter):
1) z1 = (y**lambda - 1) / lambda, for lambda ne 0
z1 = log(y) , for lambda=0
2) z2 = ((y+lambda2)**lambda - 1) / lambda, for lambda ne 0
z2 = log(y + lambda2), for lambda=0
After carrying out a regression, I need to then make predictions on
the original scale.
I would like to know how I transform back to the original scale:
a) with and without a shift parameter (lambda2)
b) for lambda=0 (log), and lambda=0.5 (sqrt) ?
In the case of z1 and lambda=0, I gather the equation is:
E(y) = exp( XB ) + 0.5 sigma**2 (where sigma**2 is the residual
variance from regressing z1 on X.
Yours,
David Carr
Dr. von Haunersches Kinderspital
Munich
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|