Unverified50% confidenceFactExact time
The paired comparison design separates the two factors of modality preference and evidence position, allowing changes in answers after swapping the order to be directly attributed to position itself.
1
Sources
50%
Confidence
Long-term
Relevance
9/24/2026
First Seen
Sources
Related Entities
Related Claims
Unverified'Position bias' is a systematic bias documented in multiple studies: LLMs in two-option comparisons tend to favor the option listed first or last76% similarUnverified位置偏差是偏好数据集中一个被广泛研究的系统性问题，人类标注者评估并排回答时会受展示顺序影响75% similarUnverifiedThe distributional shift is driven by two key factors: the recipient's prior weight and the candidate evidence skew.75% similarUnverified配对设计通过在同一个体上观测两种处理的差值将个体差异从误差项中剥离，等价于将组间方差转化为组内比较以提高统计功效74% similarUnverifiedThere is a systematic bias in annotator preferences, with higher scores given to polite, clearly structured, and risk-averse responses.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/946096API
curl https://kongchang.com/api/v1/knowledge/claims/946096MCP
get_claim(id=946096)