Economist Warns: Diff-in-Diff Pitfalls Distort AI Job-Impact Studies
paulnovosad · x · 2026-09-16
Economist Paul Novosad cautions that papers claiming negative wage/employment effects of AI must be read carefully: the same result can be framed in two very different ways, and he would not conclude that AI-exposed workers are worse off.
He points to a widespread issue in the diff-in-diff literature, where authors make implicit assumptions about which groups trend normally—famously a problem in the China shock papers too. Readers should scrutinize control-group assumptions before accepting headline findings.
Related event: Economists Debate Framing of GenAI Wage Studies(10 posts)→
More from AGI Musings
- Low Internet Penetration May Limit Both AI's Impact on Africa and Our Ability to Measure It — mioana · 2026-09-16
- Only 30% of Africans Regularly Use the Internet, Just 8% Transact Online — mioana · 2026-09-16
- A veteran teacher argues the real AI risk in schools is cognitive offloading — TarzanoftheJungle · 2026-09-16
- DeepMind launches DeepMind Institute, says AGI is now within reach — ShaneLegg · 2026-09-16
- AI floods: Princeton's Narayanan maps 50 institutional processes breaking under AI slop — _akpiper · 2026-09-16
- Open source will pass the frontier: Reddit argument says safety drama is a smokescreen — rasheed106 · 2026-09-16