Definition
Plain language
A test where you show people the exact same proposal but change only who supposedly backs it, to reveal hidden bias toward that group.
As stated in the literature
A survey-methodology design that holds content fixed and varies only the named sponsor across conditions, isolating the endorser's effect on ratings; applied here to measure geopolitical bias in LLM scoring.
Also called: endorsement experiments
Why it matters: It isolates bias by holding everything constant except the endorser, revealing prejudice that would otherwise hide behind differences in content.
For example, you send reviewers identical project proposals but tell one group it is backed by Country A and another that it is backed by Country B, then compare their scores.
Heard on the show
“The author reaches for a move political scientists have used for years — the endorsement experiment.”Episode 215 — The Same Policy Scored 85 for the US and 36 for Russia