Robots and Jobs Revisited: A Bit-Exact Replication, Shift-Share Anatomy, and Capital-Stock Extension of Acemoglu and Restrepo (2020)

Version 1.00, 30 July 2026. This working paper revisits Acemoglu and Restrepo (2020), Robots and Jobs: Evidence from US Labor Markets, through a reconstruction of the underlying data pipeline and a detailed analysis of the identifying support of its shift-share instrumental-variable design. First, I reproduce the benchmark long-difference IV estimates reported in Table 7, columns 3 and 6. Starting from the underlying data sources, I reconstruct the industry-level robot-adoption measures, commuting-zone employment outcomes, commuting-zone-by-demographic-group wage outcomes, baseline controls, exposure measures, and estimation specifications. The replicated estimates match the published benchmark: one additional robot per thousand workers is associated with a decline of 0.388 percentage points in the private employment-to-population ratio and a 0.768 percent decline in log hourly wages. Second, I examine conventional robustness. Alternative controls, weights, exposure definitions, outlier treatments, geographic exclusions, and clustering conventions do not overturn the negative employment and wage estimates. Third, I analyze where the shift-share instrument obtains its identifying variation. Frisch–Waugh–Lovell identifying-mass diagnostics, Rotemberg weights, leave-one-industry-out estimates, balance tests, and industry-shock permutations show that the baseline instrument is strongly concentrated in automotive exposure. Automotive accounts for approximately 84 percent of normalized absolute Rotemberg-weight mass. Once the automotive component is removed, the remaining instrument does not sustain the original negative estimate with a strong first stage. A complete timing-window frontier reconstructs all 78 admissible European industry-shock windows between 1993 and 2007 while holding the outcomes, endogenous U.S. robot-exposure measure, controls, weights, fixed effects, clustering convention, and estimation samples fixed. Among the 60 windows with a Kleibergen–Paap F-statistic of at least 10 in both outcome specifications, all produce negative and statistically significant employment and wage estimates. The baseline timing window is therefore not an outlier, although the alternative instruments remain strongly concentrated in automotive exposure. Finally, I construct an age-adjusted perpetual-inventory measure of robot capital. The stock-based specifications retain negative employment estimates and strong first stages, showing that the qualitative result is not specific to measuring automation through short-window robot-adoption flows. The main conclusion is that the original estimates are computationally reproducible, conventionally robust, and stable across relevant alternative instrument windows. Their interpretation nevertheless depends on a narrow source of identifying support concentrated in automotive-intensive U.S. local labor markets. JEL Codes: J23, J24, O33, R23, C26 Keywords: Industrial robots; automation; employment; wages; replication; shift-share IV; Bartik instruments; Rotemberg weights; local labor markets; commuting zones; technological change; robot capital stock