The value of imperfect sample separation information in switching regression models