Detecting Bad Science with Data