On quantitative uniqueness for elliptic equations