A comment on "What Happens to Workers at Firms that Automate?"

James Bessen et al. (2025) investigated the effect of company automation events on employee outcomes. The authors found that the automation spike increases the hazard of incumbent workers separating from their employers by 1.7% and the five-year cumulative annual wage income loss of 9.3%. They also claimed that this annual wage income loss was a consequence of working day decrease instead of daily wage. In this report, we computationally reproduced the aforementioned main claims of the paper successfully under the original specification and implementation. However, a coding error led to the inconsistency between the original implementation and the way the authors described in their paper. One sample restriction criterion that excluding workers earning less than EUR 10 per day or EUR 5,000 annually, was not implemented in the original analyses. After adding this restriction, the sample size reduced by one-eighth. The DID estimates on the restricted sample remains stable, but the observed heterogeneity of the effects on workers of different education level and younger ages attenuates. We additionally test the robustness of the main claims by shifting the baseline period from t = -1 to t = 0. The resulting estimates remain the same direction and similar in magnitude, although the wage-income losses become marginally significant under the restricted sample. We also extend the heterogeneity analysis to the workers' fields of study. We find that workers trained in hard sciences experience less negative earnings effects and separation hazards than those in other fields, both in the original and restricted samples. Overall, our replication roughly converges with the study's main findings, while warning that the magnitudes of effects supporting main conclusions may attenuate under alternative sample definitions.