A note on scheffé's method of deriving confidence sets for directions and ratios of normals