A reinterpretation of the tests of overidentifying restrictions