|
If there are few predictors that are highly skewed among a larger set of predictors in case of a linear regression problem, should a BoxCox transformation be applied to only these few predictors or should it be applied to all the predictors present in the set - I know that the remaining of the predictors are actually not skewed, it is just a few of them. If the answer is just few of them, why would that work if we modify just a few of the predictors while fixing the values of the rest of them? |
