'One robot destroys 5.6 jobs': what the study really measured

The confusion is between a local effect and a national one. Acemoglu and Restrepo measured what happens in a commuting zone when robots arrive there, using US local labor markets across 19 industries from 1990 to 2007. The 5.6 workers is the drop in a local area per robot added to that area, and it includes spillovers to other local businesses.

That is not the same as saying the national workforce shrinks by 5.6 for every robot installed anywhere. The authors are explicit that the estimate is a local-labor-market effect, and their own paper frames the 5.6 against a comparison figure of 6.2 workers in a scenario without trade, which shows it is a modelled number with assumptions, not a headcount.

For effect size, the cleaner figures are the per-thousand-workers ones: one more robot per thousand workers reduces the employment-to-population ratio by 0.18 to 0.34 percentage points and wages by 0.25 to 0.5 percent. Those are small, precise, and far less dramatic than the 5.6 headline.

If you cite 5.6, say it is a local-labor-market estimate from Acemoglu and Restrepo, not a national per-robot figure. That single qualifier is the difference between using the number correctly and misusing it.

For the size of the effect, prefer the per-thousand-workers figures: 0.18 to 0.34 percentage points on employment and 0.25 to 0.5 percent on wages. They are the study's more careful estimates.

Name the period. The analysis covers 1990 to 2007, before the most recent wave of automation, so it is evidence about a past era, not a measurement of today's robots.

Reading 5.6 as a national count. It is a local-labor-market effect per robot in that area, including local spillovers, not 5.6 jobs erased from the national economy per robot.

Presenting it as the last word. Later work, including by the same authors, revisits the national picture, and other economists dispute the magnitude. The 5.6 is one influential estimate, not a settled fact.

Ignoring the date. The data runs to 2007. Quoting it as a description of automation today skips nearly two decades.

The figures trace to two free primary sources from the National Bureau of Economic Research: the digest summary of Robots and Jobs and the full working paper. We fetched both and confirmed each figure appears in the text, including the 5.6 workers figure, the 0.18 to 0.34 percentage-point range, the 0.25 to 0.5 percent wage effect, and the paper's own 6.2-worker comparison.

The full working paper is a PDF. We read it with a text extractor and confirmed the figures in its actual text, rather than relying on second-hand descriptions, so the local-labor-market framing comes from the source itself.

The published version of this study appeared later in a peer-reviewed journal, which sits behind a paywall. Where the two versions differ in emphasis, we cite only what the free NBER sources state, and we flag that the popular 5.6 line is the working-paper figure.