Errors in the research of Emmanuel Saez and Gabriel Zucman
AI issues its judgment: 138 papers audited 2,035 problems found 147 serious enough to change a headline number 69% of papers have at least one 97% of major and moderate errors with a clear direction overstate the paper’s claim And on the method: Two independent audits of each paper. One by Claude Opus 5.5, one […] The post Errors in the research of Emmanuel Saez and Gabriel Zucman appeared first…
The research of Emmanuel Saez and Gabriel Zucman has been scrutinized by AI, revealing 2,035 problems across 138 papers. Of those, 147 were deemed serious enough to alter headline numbers, accounting for 69% of papers with at least one error. A staggering 97% of major and moderate errors displayed a clear direction that overstated the paper's claim.
Two independent audits of each paper were conducted, with neither AI system seeing the other's work. Each AI system obtained the original papers, appendices, and replication data, recomputed the headline numbers from the authors' own files and original sources, and checked cited sources for accuracy. Afterward, a third agent re-verified every claimed problem and attempted to refute each one, discarding those that did not hold up and searching for additional problems missed by the auditors.
The remaining issues were graded by their impact on the paper's headline claim, with major changes affecting core numbers materially, moderate issues impacting secondary results or a core result modestly, and minor errors being presentation or arithmetic slips. The survey is extensive and thorough, as detailed on the linked source.
Written by urgent.news from Marginal Revolution's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
