Which race-adjustment model is the econometrically correct one for the Mariel wage effect — Borjas's simple adjustment or Clemens's fully-interactive one — and does an independent source adjudicate?

economicslabor-economicsimmigrationmariel-boatliftmeasurement-artifactnatural-experimentscensus-methodologyeconometricsverification

This capture goes directly to question-verify-mariel-race-adjustment-model-correctness, the question raised during the 2026-07-29 promotion that produced claim-mariel-wage-effect-depends-on-race-adjustment-model-choice. That note recorded, from Borjas's own reply paper, that a simple race fixed-effect leaves his Mariel wage-decline estimate intact while a "fully-interactive" race model (race interacted with metropolitan area and education) shrinks it toward zero and insignificance — but the note carried an [unverified-secondary-characterization] flag because the description of the interacted model came only from Borjas's own paraphrase of a Michael Clemens blog post, which had not been read directly. This session located and read that post directly (Michael Clemens, "What Economists Can Learn from the Mariel Boatlift, Part Two," cgdev.org, 2017-05-30), along with an independent contemporaneous assessment of the broader dispute (David Roodman, davidroodman.com, 2017-05-25).

Claim: Clemens's own account grounds the case for a race-interacted model in a specific empirical objection — the assumption that the black/non-black wage gap is identical across cities is contradicted by the sample's own composition

Claim type: specific technical-mechanism claim, with a load-bearing quantitative figure. Tier 1–2 required — cleared (Clemens's own primary venue, own words, archived capture read directly).

Clemens writes, in his own reply to Borjas, that the simplest race adjustment (a single, non-interacted black indicator, matching what Borjas had done) "impose[s] the strong assumption that the wage gap between blacks and non-blacks is identical in every city, at all levels of education. This is unlikely given that 63 percent of the sample is black in Miami, but under 10 percent in the Borjas control cities. In our paper we show that the difference between black and non-black wages is indeed different in Miami than in other cities." His argument for the more flexible specification is therefore not a general preference for more parameters — it is a specific claim that one of the simple model's identifying assumptions is empirically false for this particular comparison. Allowing the black wage effect to vary by city (rather than holding it fixed everywhere) is, on his account, what a compositional shift this large in one city and not the others requires.

Provenance:

• source_url: https://www.cgdev.org/blog/what-economists-can-learn-from-mariel-boatlift-part-two (fetched via Wayback Machine capture, see source_archived_via in frontmatter, because the live URL 403'd automated fetch)
• source_sha: 36d50d92d1122f478dbe685e4a3333b34ba1950d71c2aa6f6cd7859d8089574f
• source_author: Michael Clemens
• source_date: 2017-05-30
• source_tier: 1
• exact quote: "unlike the analysis in our original paper, but following Borjas' blog post, columns 3 and 4 impose the strong assumption that the wage gap between blacks and non-blacks is identical in every city, at all levels of education. This is unlikely given that 63 percent of the sample is black in Miami, but under 10 percent in the Borjas control cities. In our paper we show that the difference between black and non-black wages is indeed different in Miami than in other cities."

Claim: On Clemens's own reported results, the effect shrinks progressively as the race adjustment is loosened — roughly a third under the simple adjustment, over half once the black effect is allowed to vary by city, and to nothing (relative to Card's control cities) once it is also restricted to the less-than-high-school subsample where the compositional shift actually occurred

Claim type: specific technical-mechanism claim, with quantitative figures as stated by the source (exact regression coefficients appear only in a table graphic not captured by text extraction, so the figures below are Clemens's own rounded characterizations, not exact numbers). Tier 1–2 required — cleared on the rounded figures as stated; exact coefficients not independently confirmed this session.

Clemens reports, replicating Borjas's own regression framework: applying the simple, city-invariant race adjustment Borjas used, "about one third of the treatment effect disappears—but, as Borjas noted, not the entire effect." Allowing the black effect to differ by city (but not yet by education), "more than half the treatment effect is gone now, and the treatment effects are statistically insignificant relative to the Card control cities." Restricting the black-effect adjustment specifically to workers with less than a high-school education — the subgroup in which "the large shift in the black fraction of Miami's sample occurs" — "there is no wage effect relative to the Card control cities," though a smaller effect (-5 to -8 percent) versus Borjas's own control cities cannot be statistically rejected either way. A parallel check using the larger CPS Outgoing Rotation Group extract shows the treatment effect disappearing even without interacting by education, when the black effect is allowed to vary only by city.

Provenance:

• source_url: https://www.cgdev.org/blog/what-economists-can-learn-from-mariel-boatlift-part-two (Wayback capture, as above)
• source_sha: 36d50d92d1122f478dbe685e4a3333b34ba1950d71c2aa6f6cd7859d8089574f
• source_author: Michael Clemens
• source_date: 2017-05-30
• source_tier: 1
• exact quote: "About one third of the treatment effect disappears... Columns 5 and 6 therefore remove the black effect from wages allowing the black effect to differ between cities, but not between education levels. More than half the treatment effect is gone now, and the treatment effects are statistically insignificant relative to the Card control cities... in columns 7 and 8 of the table, we remove the black effect specifically for workers with less than high school. There is no wage effect relative to the Card control cities."

Claim: An independent, named methodologist (David Roodman) assessed the broader census-composition critique and found it plausible and case-weakening, but his assessment predates the fully-interactive model and does not itself adjudicate simple-versus-interacted race adjustment

Claim type: specific technical-mechanism claim, with quantitative figures (p-values). Tier 1–2 required — cleared (named analyst, own venue, direct read via archive_page).

David Roodman — previously a CGD colleague of Clemens's, now at Open Philanthropy which funds CGD and Clemens's migration work specifically (a disclosure he states himself) — published an assessment of the dispute five days before Clemens's fully-interactive-model post. Roodman's own independent check of Borjas's data found "no sudden drop or trend break in 1980": a formal test of a constant-decline-rate hypothesis against the adjusted wage series "returns high p values: 0.82 for the first graph, 0.37 for the second," meaning deviations from a steady long-term decline "cannot be distinguished from the play of chance." On the compositional-artifact question specifically, he wrote: "Clemens and Hunt point up an interesting discovery. If that discovery can explain the long-term decline in wages in the Borjas sample—and they convince me that it can—it further weakens the case for the Mariel boatlift as a wage driver." Because this post is dated 2017-05-25 and Clemens's fully-interactive-model reply is dated 2017-05-30, Roodman's assessment addresses the general census-coverage mechanism, not the specific choice between a simple race fixed effect and a race-by-city(-by-education) interaction that is the subject of this capture's core question.

Provenance:

• source_url: https://davidroodman.com/blog/2017/05/25/four-points-on-the-debate-over-the-impact-of-mariel-boatlift/
• source_sha: fa973116de8a060e558ffc57f775112e245a6f67baaef219e4e82c158c998645
• source_author: David Roodman
• source_date: 2017-05-25
• source_tier: 2
• exact quote: "A formal test of that hypothesis, for the adjusted averages (in orange), returns high p values: 0.82 for the first graph, 0.37 for the second... Nevertheless, Clemens and Hunt point up an interesting discovery. If that discovery can explain the long-term decline in wages in the Borjas sample—and they convince me that it can—it further weakens the case for the Mariel boatlift as a wage driver." / disclosure: "I used to work for the Center for Global Development, where I was a colleague of Michael Clemens. Now I work for the Open Philanthropy Project, which provides general support to CGD and specific support for Michael's work on migration."

Central question status

[unverified — could not confirm or deny after search, on the specific "which model is econometrically correct" question]. Clemens's own words, now read directly, confirm and sharpen (rather than contradict) Borjas's characterization of the dispute: Clemens argues the interacted model is required because the simple model's implicit assumption — one shared black/non-black wage gap across all cities and education levels — does not hold in this sample. Borjas argues, in the text already in the vault, that the interacted model is conceptually flawed because it may "net out" real wage effects that immigration itself drove through the racial wage gap. Both positions are now sourced directly to their authors. No independent, disinterested source was located this session that rules on which specification is econometrically correct for this design. Roodman comes closest as a named, credentialed independent voice, but his read predates the interacted-model proposal and engages a different, broader question (whether there was any 1980 trend break at all). J. Bradford DeLong's blog post appears to weigh in on the pro-Clemens/Card side of the wider dispute but could not be quote-verified this session (reached only via a summarizing mirror) and, per available snippets, addresses a separate IV critique rather than this specific race-model choice. The final, peer-reviewed ILR Review 2019 versions of both papers — which might contain a settled or refereed-commentary round — returned HTTP 403 on every route tried and were not read. The dispute, as far as this session can confirm, remains a live modeling disagreement between the two principals, not one closed by a third party.

Further leads

• J. Bradford DeLong, blog post dated 2018-08-08 (title not directly confirmed; reached via a Goodreads mirror), sides with Clemens/Hunt/Card but was not quote-verified this session — worth locating on DeLong's own venue (likely "Grasping Reality," bradford-delong.com or its Substack successor) and archiving directly.
• Borjas, "The Wage Impact of the Marielitos: The Role of Race," ILR Review 72(4), 2019 (published, peer-reviewed version) — https://gborjas.scholars.harvard.edu/sites/g/files/omnuum4696/files/gborjas/files/ilrr2019.pdf — 403'd on every route tried across two sessions now; may contain a referee-driven resolution not present in the 2017 working paper.
• Clemens & Hunt, "The Labor Market Effects of Refugee Waves: Reconciling Conflicting Results," ILR Review 72(4), 2019 (published version) — also not accessed; the NBER/CGD working-paper text has been used as the primary throughout the vault's Mariel coverage instead.
• A Cato Institute blog post and a Center for Immigration Studies report both surfaced in search results as taking sides in this debate but were not read directly this session (the former already flagged as not-promoted in the 2026-07-29 capture; the latter, cis.org, is a restrictionist advocacy venue and would need independent-of-cause framing before citing).
• Peri & Yasenov's synthetic-control critique of Borjas's sample construction is a separate, already-captured line of attack (claim-mariel-boatlift-wage-effect-flips-with-sample-construction) — not re-litigated here, but worth checking whether their published replication (JHR 2019) touches the race-adjustment question at all.

Entity candidates

• David Card — person — the 1990 null-result study both Borjas and Clemens/Hunt explicitly measure their control-city comparisons against ("relative to the Card control cities" is the benchmark in nearly every quoted figure above); already an entity (entity-david-card), flagged first per the standing instruction to catch the older foundational figure a dispute argues against, not just its immediate combatants.
• David Roodman — person — independent, named, credentialed methodologist; a claim now rests on his own directly-read words for the first time (in the 2026-07-29 capture he was judged "not yet read directly... not promoted"; that gap is now closed for this narrow point). Worth a hub given his recurring role as the field's go-to independent referee on immigration-economics disputes.
• J. Bradford DeLong — person — independent economist (UC Berkeley) whose commentary sides with Clemens/Hunt/Card in this dispute; not yet read on his own primary venue this session, candidate for a future capture rather than a hub today.
• Michael Clemens — person — already an entity (entity-michael-clemens); this capture is the first time a claim rests on his own words rather than Borjas's paraphrase of them.
• George Borjas — person — already an entity (entity-george-borjas); referenced but not re-quoted here to respect the single-source concentration cap on NBER WP 23504.
• Jennifer Hunt — person — Clemens's co-author on the underlying paper; previously judged "UNSURE" for a standalone hub (2026-07-29 capture) since no claim is independently attributed to her apart from the joint paper; that remains true here.