Because the separation for GISE is an almost singular regression problem, the classic least square method and principal component method are not satisfactorily for solving the problem.

• 由于工具误差分离存在着很强的复共线性，传统的最小二乘方法与主成分方法不能很好的解决这一问题。