A decade of state data shows no negative within-state relationship between immigration and non-college wages

Summary

No. Over 2013-2023, the AskAmerica state panel shows no evidence that rising immigration pushed down wages for non-college workers within states over time. Nominal median earnings for workers with less than a high-school diploma rose from about $20,400 to $32,700 (real terms: +22.5% after inflation) even as the foreign-born share of the adult population rose modestly nationally (11.1% to 12.2%). In a two-way fixed-effects panel regression (51 states+DC x 11 years, controlling for each state's own trend and each year's national shock, plus log per-capita income), a one-percentage-point rise in a state's foreign-born adult share is associated with a statistically significant increase of about $515/year in median earnings for less-than-high-school workers (p=0.02), and a positive but not statistically significant association (+$346/yr, p=0.16) for high-school-graduate-only workers. This matches the recent mainstream academic consensus (Caiumi & Peri, NBER 2024) that immigration since 2000 has had a small positive, not negative, effect on non-college native wages, largely due to native-immigrant task complementarity and the rising skill mix of recent immigrant arrivals. A structurally different, national skill-cell approach associated with George Borjas continues to find negative effects (3-9%) for the least-educated natives specifically (high-school dropouts) under different modeling assumptions (no capital adjustment, closed local labor markets) — that dispute is not resolved by this state-panel evidence, and is described below.

Step 1 — What the research literature says

This question has been the subject of a three-decade economist dispute, most associated with David Card (UC Berkeley) on one side and George Borjas (Harvard) on the other, with Giovanni Peri and coauthors producing the most current syntheses.

• Card's spatial-correlation approach (e.g. the 1980 Mariel Boatlift study, and Card 2001/2009) compares local labor markets that received different numbers of immigrants and finds little to no negative wage effect on low-skill natives on average, though Card (2001) found some evidence of lower occupation-specific wages in high-immigration gateway cities (Miami, LA) in the 1980s.
• Borjas's national skill-cell approach (Borjas 2003 and later) argues that local labor markets equilibrate via native migration and capital flows, so cross-city comparisons understate the true national effect; using a national supply-and-demand accounting framework instead, Borjas estimated immigration reduced wages of native high-school dropouts by 7.4-9% and native workers overall by 3-4% between 1980 and 2000.
• Ottaviano & Peri's rebuttal showed that once native and immigrant workers within an education-experience cell are treated as imperfect substitutes (not perfect substitutes, as Borjas's baseline model assumed) and capital is allowed to adjust, the estimated wage effect for high-school dropouts flips from Borjas's -1.7% to roughly +1.1%, and the overall native wage effect becomes modestly positive (about +0.5-0.6%).
• The most current update, Caiumi & Peri (NBER Working Paper 32389, April 2024) — confirmed directly from the NBER abstract — extends this literature through 2022 with an improved instrument (a skill-based shift-share) and finds immigration had "a positive and significant effect between +1.7% to +2.6% on wages of less-educated native workers" over 2000-2019, with no significant employment crowd-out, and "small positive effects on wages of non-college natives" even across the high-immigration 2019-2022 period.
• The Center for Immigration Studies (a restrictionist-oriented think tank) continues to circulate Borjas's older estimates in current advocacy and testimony (Camarota, Borjas testimony), arguing low-wage/low-education immigration specifically has expanded the low-skilled workforce by roughly 25% over two decades and depressed wages for natives without a diploma.
• Mainstream reviews (Cato Institute, American Enterprise Institute) and pro-immigration advocacy groups (American Business Immigration Coalition) both cite the Card/Peri complementarity finding as the current weight of evidence, while acknowledging Borjas's estimates as the standing minority position within the profession.

In short: the dispute is not about the data so much as about the identifying assumption — whether local labor markets are open (workers, capital, and trade adjust across regions, favoring near-zero/positive estimates) or effectively closed (favoring Borjas's more negative national estimates). No single dataset available here resolves that structural disagreement; this analysis instead reports which pattern the actual, decade-long US state data show, and where that lines up with each side.

Step 2 — What the state data show (2013-2023)

Using Census ACS 5-Year data (tables B20004 median earnings by education, and B06009 nativity-by-education), a panel of 51 states+DC x 11 years (2013-2023, n=561) was built. Two outcomes were tested: median earnings of workers with less than a high-school diploma, and of high-school-graduate-only workers (no college) — the two groups a 'no college degree' question implies, tested separately rather than pooled, since the two skill groups can diverge.

Naive cross-state comparison is confounded. Comparing states to each other in a given year (pooled OLS, controlling for each state's log per-capita income), a higher foreign-born adult share is associated with statistically significantly lower median earnings for less-than-HS workers (-$155/yr per +1 percentage point of foreign-born share, p<0.001). This raw pattern is what a purely cross-sectional read of the data would report as "proof" immigration lowers wages. But it is exactly the confound the literature warns about: high-immigration states (CA, NJ, NY, FL, TX) differ from low-immigration states (WV, MS, MT, WY) in industry mix, cost of living, minimum-wage regimes and urbanization — all of which independently affect low-skill nominal pay, and none of which this simple comparison holds fixed.

Within-state change over time tells a different story. Adding state and year fixed effects — so the estimate compares a given state's own foreign-born share and earnings against its own trend, net of any nationwide shock in a given year — flips the sign: a 1-point rise in a state's foreign-born adult share is associated with roughly +$515/year higher median earnings for less-than-HS workers (t=2.37, p=0.02, cluster-robust SE by state, n=561) and a statistically insignificant +$346/year for HS-graduate-only workers (p=0.16). Leave-one-state-out sensitivity checks on the (less rigorous) pooled specification confirmed that correlation's sign never flips when any single state is dropped — the negative pooled correlation is a broad pattern, not one state's artifact — which makes the FE reversal a genuine finding about specification, not noise in either direction.

Over the same decade, nominal median earnings for less-than-HS workers rose from $20,383 (2013) to $32,652 (2023) nationally; in constant 2023 dollars the 2013 figure is $26,660, so real earnings for this group rose about 22.5% over the decade (CPI-U deflated), even as the national foreign-born adult share ticked up from 11.1% to 12.2%.

Step 3 — Interpretation and what this can and cannot claim

This state-year panel is a spatial-correlation design — the same family Card, Peri and Ottaviano-Peri use, and the same family Borjas critiques for not accounting for cross-state equilibration (natives and capital moving toward or away from high-immigration states in ways correlated with local wages). It is not a national skill-cell model, so it cannot arbitrate the Card-vs-Borjas dispute on its own terms. What it does show, directly and robustly in this data: the answer to "has immigration pushed down non-college wages" depends entirely on whether the comparison controls for fixed state differences. Once it does (state+year fixed effects, the more credible design for causal interpretation per the standard panel-econometrics logic), there is no negative relationship in the last decade of US state data — if anything, a positive and marginally significant one for the least-educated group, consistent with the direction (though larger in magnitude here) of Caiumi & Peri's 2024 national estimate of +1.7-2.6%.

Caveats: (1) this is an association, not a randomized experiment — states with rising immigrant shares also had specific local economic conditions (partially, but not fully, captured by the per-capita-income control) that could independently raise wages, so reverse causality (immigrants move toward booming low-skill labor markets) remains a live concern the state fixed effects only partially address. (2) The earnings series is nominal-population median earnings, not a fixed cohort's wage growth, so composition changes (e.g., which workers remain in the less-than-HS category) are not separately controlled. (3) econ.state_personal_income's per-capita-income series was found to be loaded only for 1929-1957 in this corpus (a data-ingestion gap, reported separately), so the income control used here comes from the Census ACS median-per-capita series instead, not BEA.