Research opportunity · Aggregate null or distributional shift? Employment
Why do Nordic registers and US data disagree about young workers in AI-exposed jobs?
Earlier version (v1)
This is RO-LD02-01 as it was published at v1, kept unchanged. The current version is v2; its changes are listed there.
Question
When the same exposure measure, age bands, occupation coding and event-study specification are applied to US data and to Nordic population registers, how much of the gap between the large US young-worker divergence and the near-zero Nordic estimates is explained by occupation mix, education composition, sensitivity to interest-rate rises and cohort size, and how much remains?
Untested beyond the original settingOne PhD chapter or paperLow costPartly addressed
Why it matters
The strongest displacement-like signal in this debate comes from US young workers, while Norwegian and Finnish registers find little or nothing. Each paper uses its own exposure index, age cut-offs and model, so readers cannot tell whether the countries really differ or the studies simply measure different things. A harmonised comparison would tell us whether the US result is a general early warning or a US-specific pattern, which is the question supervisors, policymakers and graduates are actually asking.
The debate lists this among what we still do not know: Why are the broad young-worker effects near zero in Norway while US datasets show large divergences?
Evidence that motivates it
- By June 2026, US workers aged 22–25 in highly AI-exposed occupations were about 19% below the employment path of similarly aged workers in less-exposed occupations, with the divergence driven mainly by reduced hiring rather than separations.
The US payroll evidence shows a 19% employment gap for 22–25-year-olds in highly exposed occupations, driven mainly by hiring; the claim itself does not license reading this as a causal AI effect, which is why the cross-country contrast is informative. - Norwegian private-sector registers show near-zero employment differences between the most- and least-AI-exposed occupations since late 2022, and the young-worker gap across full exposure quintiles is an imprecise zero.
Norwegian private-sector registers show near-zero gaps between most- and least-exposed occupations and an imprecise zero for young workers; the claim does not say exposure equals use or that no occupation is affected. - In US Census administrative data, regression-adjusted employment of workers aged 22–24 in the most AI-exposed industry-state cells fell 12% over 10 quarters after ChatGPT, with reduced hiring the primary mechanism.
US Census administrative data independently show a 12% relative decline for 22–24-year-olds in the most exposed industry-state cells, so the US pattern is not an artefact of one private payroll source; clean causal attribution to AI is not licensed.
A possible design
Write one pre-registered specification (one exposure index plus one alternative, the same age bands, occupation crosswalk to ISCO-08, monthly or quarterly cell-level Poisson event study with occupation and period fixed effects) and run it on US public data (CPS basic monthly files, ACS) and on at least one Nordic register through an existing collaborator or remote-access system (for example microdata.no in Norway). Then decompose the US–Nordic difference by re-weighting the US occupation and education mix to the Nordic distribution, adding each occupation's sensitivity to the 2022 interest-rate rises, and adjusting for the size of entering cohorts. Report what share of the gap each step removes and what remains unexplained.
Data needed
US: CPS and ACS public microdata (free). Nordic: one population register with occupation and age, via remote access (Norway's microdata.no, used by the Norwegian study below, is open to researchers at approved institutions; eligibility needs checking) or a co-author with Statistics Sweden, Finland or Denmark access. Published exposure crosswalks (Eloundou et al. and one alternative). An occupation-level interest-rate sensitivity measure, built from pre-2022 industry composition.
Feasibility
One PhD chapter or paper. Low: public US data are free; Nordic register access needs an application, and fees and eligibility rules vary by country and must be checked. The main hurdle is register access and making occupation codes comparable (US SOC versus ISCO). CPS cells for 22–25-year-olds in narrow occupations are small, so the US side should report precision honestly and may need to pool months.
Has this been done?
Partly done already: related work covers some of this question (first-pass screen of 12 October 2026; not yet checked in depth).
Separate single-country studies exist for Norway (two papers), Finland and Sweden, and the US paper runs many robustness checks, including education and interest-rate controls. The Norwegian and Finnish authors offer institutional explanations (unions, employment protection, demographics) for why their results differ from the US but do not test them. No study was found that runs one identical specification across the US and several Nordic registers and decomposes why the estimates differ. The Swedish paper also finds a widening 22–25 gap from 2024 with failed pre-trend tests, the Norwegian 22–25 replication shows an emerging gap from spring 2025, and a Banque de France analysis reports a large relative fall in permanent hiring of 21–29-year-olds in the most exposed French occupations, so the European picture is itself not uniformly null.
How strong this evidence is
- Who searched and checked: an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: none of the 6 nearest works below is recorded as judged from an abstract only. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web search on the date shown; Google Scholar, SSRN and paywalled journal pages were not systematically searched. Absence of a hit is not proof that a harmonised comparison does not exist or is not under way.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- Hernæs & Kostøl, "Has AI Widened Employment Gaps? Tracking Early-Career Employment by Occupational Exposure in Norway," IZA Discussion Paper No. 18767 (June 2026)
Norwegian private-sector register (via microdata.no), Eloundou et al. exposure quintiles, cell-level Poisson event study: a relatively precise null for ages 21–30; a firm-fixed-effects replication of the US 22–25 specification shows the most-exposed quintile falling below the least-exposed from spring 2025, which the authors treat as an emerging pattern, not an identified effect, noting wide intervals and a rejected pre-trend test. One country only; no decomposition of the difference from the US. - Facius & Iacono, "Labor Market Consequences of Generative AI: Early Evidence from Norway," CESifo Working Paper No. 12752 (June 2026)
Norwegian registers January 2015 to March 2025, exposure mainly from the Massenkoff and McCrory index (Eloundou et al. as a check); within-firm, synthetic difference-in-differences and shift-share designs give negative but never individually significant estimates for ages 22–25, and backdating tests suggest part of the post-2022 pattern reflects earlier trends. Discusses Norwegian institutions as a likely reason for the contrast with the US but does not test it or harmonise specifications. - Kauhanen & Rouvinen, "AI Has Not Impacted the Youth Labor Market in Finland," ETLA Working Papers No. 135 (27 January 2026)
Finnish Incomes Register (private sector, January 2019 to September 2025) with Eloundou et al. and Anthropic usage-based exposure measures; follows several of the Brynjolfsson et al. analyses and finds no clear exposure gradient for ages 22–25 and no persistent wage differences. Offers demographic and employment-protection explanations for the difference from the US as hypotheses, without a formal decomposition. - Lodefalk, Löthman, Koch & Engberg, "Same Storm, Different Boats: Generative AI and the Age Gradient in Hiring," Örebro University School of Business Working Paper 2/2026 (16 March 2026)
Swedish postings and full-population employer register: postings decline lines up with monetary tightening; within employers, 22–25-year-olds in the most exposed occupations fall about 5.5% relative by 2025H1, with pre-trend tests rejected. Single country, own exposure index (DAIOE). - Brynjolfsson, Chandar & Chen, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," Stanford Digital Economy Lab (August 2026)
US ADP payroll data: adding college share halves the 22–25 estimate (−0.18 to −0.09, the latter significant only at 10%); interest-rate exposure controls barely change it; pre-trends checked against a 2018 baseline. US only; no international comparison. - Palladino, Bergeaud, Bertheau, Bunel & Degos, "Generative AI: the first changes in the labour market are affecting young people," Banque de France blog post No. 463 (8 October 2026)
French administrative hiring records (DMMO, 2019 to March 2026) with a language-model exposure index: permanent hires of 21–29-year-olds in the most exposed quintile about 28% below November 2022 levels by March 2026, against about 8% in the least exposed; firms reporting intensive AI use show weaker young hiring. Descriptive, single country; contrasts with US, Norwegian and Danish results but uses no common specification.
Search queries used
generative AI young workers employment register data Sweden OR Finland OR Netherlands exposed occupations 2025 2026Norway register AI exposure young workers employment near zero working papercross-country comparison AI exposure entry-level employment young workers administrative data harmonisedharmonized multi-country replication "AI exposure" young workers employment Denmark Norway Sweden Finland United States same specificationwhy AI young worker employment effects differ US versus Europe institutions explanation decomposition Brynjolfsson canaries replicationgenerative AI young workers employment multi-country comparison United States Europe same exposure measure 2026 Denmark Norway Sweden Finland Netherlands
Provenance
RO-LD02-01 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-02 debate on 2026-10-12. Checked by an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.