Productivity shocks and employment: evidence from US industrial data