q94 · askamerica

How much did employment change in the places most exposed to automation, compared with everywhere else?

Answer: Counties whose 2008 employment was most concentrated in manufacturing (NAICS 31-33) — the industries where essentially all industrial-robot deployment occurs, per Acemoglu & Restrepo (2020) — grew total employment only +5.0% cumulatively from 2008-2023, versus +14.4% in the rest of the country. The median county in the most-exposed group actually shrank (-0.6%), while the median county elsewhere grew (+2.2%). The gap is statistically significant (Welch's t=7.43, p=1.5e-13), survives controlling for county size (OLS coefficient -38.2, p=6.3e-13), and is not driven by any single state (leave-one-state-out sensitivity: coefficient stays between -35.2 and -41.0 across all 53 state/territory omissions, no sign flip).

Full report (dashboard, methodology, sources, caveats): published via AskAmerica publish_report as a self-contained local HTML page, title "Automation-exposed counties grew employment +5.0% (2008-2023) vs +14.4% elsewhere — a robust, statistically significant gap" (link is a 127.0.0.1-local URL, not fetchable from here — share the report link the tool returned with the human reader).

Method

No table in this corpus carries a direct robots-per-worker or IFR robot-shipment measure, so this analysis used the standard literature proxy: a county's 2008 employment share in manufacturing (NAICS 31-33), the sector where industrial robot deployment concentrates (motor vehicles, electronics, machinery, metals, plastics), per Acemoglu & Restrepo (2020, Journal of Political Economy, "Robots and Jobs: Evidence from US Labor Markets"). Counties were split at the 75th percentile of 2008 manufacturing share (18.2%) into "most exposed" (799 counties, avg 27.1% manufacturing) vs. "rest" (2,397 counties, avg 7.6%).

Data: econ.county_qcew (BLS QCEW), county grain, 2008-2024 (the table's declared schema window is 2019-2024, but it actually holds complete data back to 2008, confirmed via data_coverage).

Results

Group	Counties	Avg. 2008 mfg. share	Employment 2008	Employment 2023	Cumulative change	Median county change
Most exposed (top quartile)	799	27.1%	14.48M	15.20M	+5.0%	-0.6%
Rest of counties	2,397	7.6%	121.33M	138.83M	+14.4%	+2.2%

Indexed to 2008=100, "rest" counties reached 114.6 by 2024 while the most-exposed group reached only 105.6 — a persistent, widening divergence, not a one-time shock (both groups fell together in the 2008-2010 recession and again in 2020).

Statistical tests

Welch's t-test on county-level percent employment change by group: t=7.43, p=1.5e-13 (n=2,397 vs 799).
OLS: pct_employment_change ~ mfg_share_2008 + log(baseline_employment) — coefficient on manufacturing share = -38.2 (SE=5.28, p=6.3e-13, n=3,196). R²=0.036 — a real, robust partial effect, not the dominant driver of most counties' growth.
Leave-one-state-out sensitivity: coefficient ranges -35.2 to -41.0 across 53 omissions, no sign flip, no loss of significance anywhere.

What this adds vs. the literature

Acemoglu & Restrepo (2020) and Rust Belt regional accounts (Brookings, EIG, policybynumbers.com) already establish the direction and mechanism over 1993-2014 and 2000-2010 respectively. This analysis reproduces the finding independently using this warehouse's own county-level BLS QCEW data, extends the window through 2008-2024 (including the pandemic), and adds a leave-one-state-out robustness check not typically reported at this grain in the regional-account literature.

Caveats

Manufacturing employment share is a composition-based proxy for automation exposure, not a direct robot-adoption count (this corpus has no IFR robot-shipment data). It misses non-manufacturing automation exposure (warehouse/logistics, AI-exposed office work) and treats all manufacturing counties as equally exposed regardless of actual plant-level automation.
R²=0.036: manufacturing exposure explains only a small share of variance in county employment growth; many other forces (population shifts, energy booms, remote-work shifts) dominate most individual counties.
This is a robust statistical association, not an independently identified causal estimate the way Acemoglu-Restrepo's robot-shipment instrument is.

Sources

Acemoglu, D. & Restrepo, P. (2020). "Robots and Jobs: Evidence from US Labor Markets." Journal of Political Economy 128(6). https://www.journals.uchicago.edu/doi/abs/10.1086/705716
NBER Working Paper w23285 (same paper, ungated). https://www.nber.org/papers/w23285
CEPR VoxEU summary. https://cepr.org/voxeu/columns/robots-and-jobs-evidence-us
IZA Discussion Paper 12293. https://docs.iza.org/dp12293.pdf
Economic Innovation Group, "Manufacturing jobs have recovered, but not everywhere." https://eig.org/manufacturing-rebound/
Brookings, "Rebuilding the employment security system for the Rust Belt that created it." https://www.brookings.edu/articles/rebuilding-the-employment-security-system-for-the-rust-belt-that-created-it/
policybynumbers.com, "Manufacturing Employment in New York and the Rust Belt since 2010: A Recovery?" https://policybynumbers.com/manufacturing-employment-in-new-york-and-the-rust-belt-since-2010-a-recovery
policybynumbers.com, "The Decline of Manufacturing in New York and the Rust Belt." https://policybynumbers.com/the-decline-of-manufacturing-in-new-york-and-the-rust-belt
AskAmerica econ.county_qcew (BLS QCEW county employment, 2008-2024) — primary computed data source.

Visuals

dashboard.png — the published report's inlined dashboard (headline stats + indexed employment chart)
employment_index_chart.png — standalone employment index line chart (2008=100), most-exposed vs. rest of counties