On the Subbagging Estimation for Massive Data