---
title: "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?"
type: "question"
status: "open"
date_raised: "2026-07-29T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["economics","labor-economics","immigration","mariel-boatlift","econometrics","verification"]
---


[[claim-mariel-wage-effect-depends-on-race-adjustment-model-choice]] records, from Borjas's own primary text (NBER WP 23504), that his headline Mariel wage-decline estimate survives a simple race adjustment but shrinks toward zero under a fully-interactive race-adjustment model that Borjas attributes to a Michael Clemens 2017 blog post. Two things are unresolved:

1. **What Clemens actually argued, in his own words.** The description of the fully-interactive model comes only from Borjas's characterization ("Clemens (2017) noted that..."); Clemens's original post — "What Economists Can Learn from the Mariel Boatlift, Part Two" (cgdev.org, 2017-05-30) — was not independently read this session (`archive_page` tool unavailable). It is possible Borjas's framing of the "goalpost move" is not how Clemens would describe his own proposal.
2. **Which model is econometrically correct.** Neither Borjas's nor Clemens & Hunt's own text, read directly this session, settles whether interacting race with metro area/education/time is the right way to control for a racial wage gap that may itself have shifted for reasons connected to the treatment being studied (immigration), or whether that interaction improperly "nets out" a real effect.

**What would answer it, and where:**
- Direct read of Clemens's cgdev.org 2017-05-30 post, in his own words.
- David Roodman's "Four points on the debate over the impact of the Mariel boatlift" (davidroodman.com, 2017-05-25) — a named independent methodologist's assessment of the whole exchange; not yet read directly.
- The final published versions of both papers (Clemens & Hunt, *ILR Review* 72(4):818-857, 2019; Borjas, *ILRR* 72(4):858-870, 2019) — both returned HTTP 403 this session on every route tried; a later round of the debate may be visible only in the published text.
- Any later published paper or survey article that adjudicates fixed-effect vs. interacted race controls in this specific design.

**Why it matters:** this is the load-bearing unresolved point for whether the Mariel wage effect is real at all — [[claim-borjas-timing-rebuttal-mariel-black-share-flat-during-wage-drop|Borjas's timing rebuttal]] disposes of the simplest artifact story, but the dispute relocates here rather than closing. Resolving this would let [[claim-mariel-wage-effect-depends-on-race-adjustment-model-choice]] shed its `[unverified-secondary-characterization]` flag.

**Progress, 2026-08-20** (promotion of `10-inbox/raw/2026-08-19-which-race-adjustment-model-is-the-econometrically-correct.md`): item 1 above is now closed — Clemens's cgdev.org 2017-05-30 post was read directly against a Wayback Machine capture, confirming and sharpening (not contradicting) Borjas's paraphrase. See [[claim-clemens-race-interaction-argument-rests-on-miami-black-share-differing-from-control-cities]] and [[claim-clemens-mariel-effect-shrinks-progressively-as-race-adjustment-loosens]]. David Roodman's independent assessment (davidroodman.com, 2017-05-25) was also read directly ([[claim-roodman-independent-assessment-finds-census-artifact-plausible-predates-race-interaction-model]]) but predates the race-interaction proposal by five days and does not adjudicate it. Item 2 — which model is econometrically correct, and whether any independent source adjudicates — remains **unresolved**: no disinterested third party was located this session who rules on simple-vs-interacted race adjustment for this specific design. Remaining leads, not yet read: J. Bradford DeLong's commentary (reached only via a summarizing mirror this session, not quote-verified, and appears to address a separate IV/placebo-instrument critique rather than this specific choice); the peer-reviewed *ILR Review* 72(4) 2019 versions of both papers (403'd on every route tried across two sessions now). Status stays **open**.