More Thinking, More Bias: Length-Driven Position Bias in Reasoning Models

Xiao Wang

arXiv:2605.06672 · 2026-05-11 공개 · arXiv · PDF

chain-of-thought reasoning-models position-bias deepseek-r1 mmlu gpqa trajectory-length mcq-evaluation

Abstract

Chain-of-thought (CoT) reasoning and reasoning-tuned models such as DeepSeek-R1 are commonly assumed to reduce shallow heuristic biases by thinking carefully. We test this on position bias in multiple-choice QA and find a different story: within any reasoning-capable model, per-question position bias scales with the length of the reasoning trajectory. Across thirteen reasoning-mode configurations (two R1-distilled 7-8B models, two base models prompted with CoT, and DeepSeek-R1 at 671B) on MMLU, ARC-Challenge, and GPQA, twelve show a positive partial correlation between trajectory length and Position Bias Score (PBS) after controlling for accuracy, ranging from 0.11 to 0.41 (all p < 0.05). All twelve open-weight reasoning-mode configurations show monotonically increasing PBS across length quartiles. A truncation intervention provides causal evidence: continuations resumed from later points in the trajectory are increasingly likely to shift toward position-preferred options (16% to 32% for R1-Qwen-7B across absolute-position buckets). At 671B, aggregate PBS collapses to 0.019, but the length effect still manifests in the longest quartile (PBS = 0.071), suggesting that accuracy gates the expression of length-driven bias rather than eliminating the underlying mechanism. We additionally find that direct-answer position bias is a distinct phenomenon with a different footprint (strong in Llama-Instruct-direct, weak in Qwen-Instruct-direct, and uncorrelated with trajectory length): CoT reasoning replaces this baseline bias with length-accumulated bias. Our results argue that reasoning-capable models should not be treated as order-robust by default in MCQ evaluation pipelines, and offer a diagnostic toolkit (PBS, commitment change point, effective switching, truncation probes) for auditing position bias in reasoning models.

한국어 요약

📋 한 줄 요약

**[LLM 평가/추론 모델]** 사슬 사고(CoT)와 추론 튜닝 모델이 길게 생각할수록 객관식 QA에서 위치 편향이 오히려 누적된다는 사실을 13개 설정에서 정량적으로 입증.

🎯 핵심 기여도

💡 핵심 아이디어

"더 많이 생각하면 더 객관적이 될 것"이라는 통념과 달리, 추론 토큰이 길어질수록 모델은 정답 정렬보다 위치 정렬에 더 끌리는 편향이 강해진다. 이는 정확도가 떨어진다는 의미보다, 추론 경로 자체가 "옵션 위치"라는 표면 신호를 누적적으로 강화한다는 메커니즘적 발견이다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: "추론 모델은 평가 입력 순서에 강건하다"는 암묵적 가정에 정면으로 반박하며, MCQ 기반 LLM 벤치마크 전반의 신뢰성 재검토를 요구한다. **한계**: 분석이 객관식 QA에 집중되어 개방형 생성·도구 사용 시 동일 메커니즘이 어떻게 나타나는지는 별도 연구가 필요하다.

🚀 실용적 활용