Interventions
No human dataThis page describes a planned study. It has not been run. There is no result to report here, and none is fabricated below.
H5 -- Fact-table disclosure
Hypothesis: showing the structured fact table alongside the Biased AI narrative reduces the H1 choice-share effect relative to Biased AI alone, but does not reduce it to zero.
Design: a separate 2-arm randomized comparison (Biased AI vs Biased AI + Fact Table), run as its own study with its own preregistration -- not part of the primary 3-arm confirmatory test described in research/hypotheses.md. It requires its own recruitment, its own power analysis, and its own frozen preregistration before any result from it could be reported as confirmatory.
Status: not implemented, not scheduled. No arm, route, stimulus set, or analysis model exists for it yet in this repository.
Why this is separate from /results
The primary results dashboard reports on the 3-arm Control / Neutral AI / Biased AI design only. Mitigation and disclosure are explicitly out of that design’s scope (see research/hypotheses.md > Secondary / extension hypotheses). Mixing a not-yet-run intervention result into the primary results page would misrepresent what has actually been tested -- so it lives here instead, clearly labeled as a future study.
Other planned, not-yet-run extensions
- H3 -- Hallucinated attribute. A capability the product does not have, moderated by whether the participant believes the hallucination is real.
- H4 -- Confidence bias. High-certainty vs hedged phrasing of the same AI recommendation, holding content fixed.
Each requires its own arm(s) and its own preregistration before being treated as confirmatory.