Hello,
I am implementing the BoxCox transformation for multiple regression
using the function boxcox() from library MASS (this seems to be the
best one). To avoid negative values in the variable to be transformed
however, one must specify a shift parameter. However different
choices for the value of the shift parameter result in different
estimates for the exponent (lambda) and in different values for the
maximum likelihood of the estimate.
I would like to know if a function is available that will
simultaneously estimate both the lambda and the shift in a BoxCox
transformation. Since in general, the general BoxCox transformation
requires specification of both of these parameters (whether or not the
variable takes on negative values) would there be a reason why the
shift parameter has been dropped from the BoxCox estimation? Any
comments or thoughts would be appreciated.
Thanks,
Ernest
