Programmable automation and the skill question: A reinterpretation of the cross-national evidence