Usually
the loss of information from omission of true predictors is
larger than the gain by deleting noise variables. The correct
answer is (b), the aim of stepwise selection is variable reduction.
This reduction may be rather unstable, as explained in material
that follows.