A note on rates of convergence in least squares estimation