Inference on the overlap coefficient: The binormal approach and alternatives