Robots at work? pitfalls of industry level data

In a seminal paper Graetz and Michaels (2018) find that robots increase labor productivity and TFP, lower output prices and adversely affect the employment share of low-skilled labor. We show that these effects hold only, when comparing hardly-robotizing with highly-robotizing sectors and collapse, when only the latter are analyzed. Controlling for demographic workforce variables reestablishes the productivity effects, but still rejects positive wage effects and skill-biased technological change. Additionally, we find no effects, when the investigation period is extended to the most recent data (2008-2015) and document non-monotonicity in one of the instruments, which calls the respective results into question.