Immigration and Wages — Digest
TL;DR
The surge literature is consolidating rather than settling: the revised BPEA conference draft now details exactly which specification choices separate the +1.3% and −0.7% native wage estimates, while conceding that the skill-group evidence is inconclusive in both directions. A new NBER working paper adds a long-run growth channel — historical immigration raised income per capita 8.2% by 1940 — that complements rather than contradicts the near-zero short-run wage effects. And the measurement problem deepened: the 2025 ACS remains delayed, and a proposed 2030 census rule would redefine who counts in the apportionment base.
The surge replication literature: converging on employment, still split on the native wage sign
The revision to the Hunt, Orrenius & Zavodny BPEA draft does the field a service by naming the exact specification differences behind the three-way estimate dispute — and by conceding where adjudication has hit its limits. The employment side is now effectively closed: the authors report that "the smaller standard errors and small coefficients for 2021-2024... allow us to rule out moderate longer-run effects" on native employment (post-pandemic-surge-labor-housing-market-impacts). The native wage side is not. Footnote 39 attributes the Clemens, Nice & Rigol +1.3% estimate to a stack of concrete choices — weighting, commuting zones instead of metros, EOIR-based instrument construction, net rather than gross surge immigration — while Wilson & Zhou's −0.7% rests on "an entirely different IV coefficient... statistically significant at the 10% level" (post-pandemic-surge-labor-housing-market-impacts
).
The most important new disclosure is negative: the complementarity hypothesis — that surge migrants substitute for the least-educated natives and complement everyone else — fails its own test. On the education breakdown, the authors write: "due to large standard errors, no coefficient is statistically significant and the test is inconclusive," with the only notable point estimate (+2.1% for high school dropouts) running against the theory (post-pandemic-surge-labor-housing-market-impacts). That matters because the skill-group story is the one most often generalized in public debate; right now, no design in this literature supports citing skill-group effects in either direction. The paper also concedes the aggregate wage puzzle remains open — "the disparity between the imprecise negative effect of the surge migrants on the aggregate wage and our positive effects for natives' wages in the ACS thus remains unresolved" (post-pandemic-surge-labor-housing-market-impacts
).
What to watch: the first response paper or direct replication that isolates which of the footnote-39 choices drives the sign flip — and any published adjudication of the aggregate-wage composition puzzle.
A long-run growth paper reframes what "the effect of immigration" even means
Costas Arkolakis, Sun Kyoung Lee and Michael Peters' NBER working paper 35392 links the universe of historical US patents (~434,000) and over 600,000 immigrants to census records, 1880–1920, and estimates the era's 20+ million immigrants raised US income per capita by 8.2% by 1940 — with income per capita still 9.6% lower in 2000 had the inflow never happened (historical-immigration-growth-innovation). The NBER Digest's summary puts the mechanism plainly:
"Immigration to the US between 1880 and 1920 raised income per capita by an estimated 8.2 percent by 1940. Skills mattered most but the location choices of immigrants were also important." — NBER Digest, October 2026 via historical-immigration-growth-innovation
Two findings deserve attention on their own. Skills dominate: the same immigrants arriving without their skills would have left income per capita 5.9% lower by 1940 — roughly three-quarters of the total effect. And the quota-acts counterfactual is strikingly modest: removing the 1921 and 1924 restrictions would have raised income per capita only ~1.7% by 2000, "because the excluded groups were small relative to the US population and, on average, less skilled than those who continued to arrive" (historical-immigration-growth-innovation). The right way to read this alongside the surge literature is horizon discipline: long-run income-per-capita and innovation effects are a different question from short-run native wage substitution, and neither contradicts the other. One weighing caveat — the 8.2%/9.6% magnitudes come from a calibrated spatial growth model, not a reduced-form natural experiment, so they inherit the model's assumptions; the skill-decomposition and patent-linkage facts are the more direct evidence.
What to watch: whether the skill-selection decomposition gets picked up in the modern surge literature, where skill mix is currently unmeasured well enough to test.
The data infrastructure is being rebuilt under political direction, and 2025 is the casualty
The 2025 ACS 1-year estimates — the data year that would measure the post-surge retrenchment — remain in limbo past their normal mid-September release. As of October 2, Census would only say the release date is "being determined" while it seeks "DAO-compliant solutions to allow the release of the ACS 1-year tables later this year," under the June 2026 Commerce order banning the bureau's long-standing noise-infusion privacy technique (acs-2025-release-delay-disclosure-avoidance). The politicization concern is now explicit in the scientific press:
"The change marks the first time political appointees have dictated the (Census Bureau's) statistical methods, census watchers say." — Science, July 2026 via acs-2025-release-delay-disclosure-avoidance
The escalation is the September 10 proposed rule for the 2030 census: it would exclude non-citizens without a green card from the apportionment base, remove race and ethnicity questions, and narrow "usual residence" to tax-record evidence, with comments closing October 13 and more than 20,000 already submitted (acs-2025-release-delay-disclosure-avoidance). For this topic specifically, the apportionment-base change is the dangerous part: it would create a definitional break in what "the immigrant population" means in the backbone datasets every future study of immigration and native labor markets will use — independent of what the underlying economics is.
What to watch: whether the 2025 ACS tables actually land "later this year" with documented methodology, and whether the 2030 rule is finalized or litigated.
What surprised us
- Construction employment did not rise with the surge. "None of the models suggests that surge migration led to an increase in construction employment," and the buslift IV shows a relative decrease in construction's employment share — though the authors caution payroll data miss off-books work (post-pandemic-surge-labor-housing-market-impacts
). The sector everyone assumed was the pressure valve shows no measurable effect.
- The aggregate wage decline is too big to be composition. Even if surge migrants worked for nothing, a 1pp share increase should cut average wages only ~1%; the estimate is ~1.4% — something real is happening to the average wage that isn't just who's in the pool (post-pandemic-surge-labor-housing-market-impacts
- The quota acts look cheap in the growth counterfactual. Given how central the 1920s restrictions are to the political narrative, a ~1.7% income-per-capita effect by 2000 is a small number — a reminder that the long-run growth lens and the restriction-cost lens (mobility, wages) don't rank policies the same way.
- An enforcement-shock literature is quietly forming — Cox & East, Escobari et al., and Hernandez are now testing the reverse experiment: what ICE activity does to native employment. That's the natural-experiment pipeline to watch next.
Open threads worth a vote
- [watch] 2030 Census Residence-Criteria Proposed Rule: Finalization or Litigation — steer whether to track the rule's finalization, its litigation, or both; either outcome creates the definitional break in census-based labor data flagged above.