Who Can Automate—and What Happens to Work?

This chapter moves from the average border estimate to the firms and workers behind it. The key question is why a higher minimum wage can make adoption more likely for some manufacturing firms without implying that every firm will automate or that employment must fall. The evidence supports a model of selective adoption and changing job composition; it does not yet provide a complete account of who gains, who loses, or how workers move through the adjustment.

Why an average effect does not describe a typical firm

Robot adoption is rare in the study's long panel. The full sample contains approximately 240,000 single-unit manufacturing firms, but only about 2,600 ever import a robot. The mean adoption indicator is 0.00341. In the border sample, about 87,000 firms are represented and roughly 900 ever import a robot, with a mean adoption indicator of 0.00304.

Those numbers are not a reason to dismiss the relationship. They establish what kind of relationship is being estimated. A policy can change the probability of a rare event at the margin without producing a mass transition across the population. If a wage increase moves some firms across an investment threshold, the average adoption probability can rise even while nearly all firms continue without an observed import event.

The low rate has two explanations in the paper. The panel begins in 1992, when industrial robots were much less common, so many early firm-years occur before adoption was a realistic choice. The import measure also misses domestic production and purchases made through third parties. The observed rate is therefore not a complete prevalence measure. It combines genuinely rare adoption, a long historical window, and incomplete capture.

Selection matters as much as rarity. Firms that adopt may be larger, better financed, more productive, more exposed to competition, or more capable of integrating equipment. Prior U.S. evidence summarized by the paper finds that robot adopters tend to be larger and geographically concentrated in “robot hubs.” These characteristics can make a firm both more likely to adopt and more able to respond to a wage increase. The average estimate should not be applied as if policy exposure were the only relevant input.

The evidence also distinguishes two claims. The current study estimates how minimum-wage exposure relates to the likelihood of an observed adoption event in its defined sample. Its background literature reports that responses in other countries or settings are stronger among firms with routine or manual work, higher productivity, R&D intensity, competitive pressure, or sufficient scale, and weaker in settings with institutions that protect employment. Those findings help explain possible heterogeneity, but they are not a complete subgroup analysis of the U.S. minimum-wage effect.

A small manufacturer can therefore face the same wage rule as a larger neighbor and make a different choice.