ACL Findings2026
On the Step Length Confounding in LLM Reasoning Data Selection
Bing Wang, Rui Miao, Chen Shen, Shaotian Yan, Kaiyuan Liu, Ximing Li, Xiaosong Yuan, Sinan Fan, Jun Zhang, Jieping Ye
Abstract
Large reasoning models have recently demonstrated strong performance on complex tasks that require long chain-of-thought reasoning, through supervised fine-tuning on large-scale and high-quality datasets. To construct such datasets, existing pipelines generate long reasoning data from more capable Large Language Models (LLMs) and apply manually heuristic or naturalness-based selection methods to filter high-quality samples. Despite the proven effectiveness of naturalness-based data selection, which ranks data by the average log probability assigned by LLMs, our analysis shows that, when applied to LLM reasoning datasets, it systematically prefers samples with longer reasoning steps (i.e., more tokens per step) rather than higher-quality ones, a phenomenon we term step length confounding. Through quantitative analysis, we attribute this phenomenon to low-probability first tokens in reasoning steps; longer steps dilute their influence, thereby inflating the average log probabilities. To address this issue, we propose two variant methods: ASLEC-DROP, which drops first-token probabilities when computing average log probability, and ASLEC-CASL, which applies a causal debiasing regression to remove the first tokens' confounding effect. Experiments across four LLMs and five evaluation benchmarks demonstrate the effectiveness of our approach in mitigating the step length confounding problem.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- Reasoning with Exploration: An Entropy Perspective cites Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning
- Scaling Data-Constrained Language Models cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Let's Verify Step by Step cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Builds on19
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
- Let's Verify Step by StepHunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards et al.ICLR 2024 · 3,045 citations
- Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM ReasoningShenzhi Wang, Le Yu, Chang Gao, Chujie Zheng et al.NeurIPS 2025 · 592 citations
- Scaling Data-Constrained Language ModelsNiklas Muennighoff, Alexander M. Rush, Boaz Barak, Teven Le Scao et al.NeurIPS 2023 · 475 citations
- Reasoning with Exploration: An Entropy PerspectiveDaixuan Cheng, Shaohan Huang, Xuekai Zhu, Bo Dai et al.AAAI 2026 · 216 citations
Related papers
- Learning to Focus: Causal Attention Distillation via Gradient-Guided Token PruningYiju Guo, Wenkai Yang, Zexu Sun, Ning Ding et al.NeurIPS 2025 · 14 citations
- TokenSqueeze: Performance-Preserving Compression for Reasoning LLMsYuxiang Zhang, Zhengxu Yu, Weihang Pan, Zhongming Jin et al.NeurIPS 2025 · 6 citations
- Selection-Inference: Exploiting Large Language Models for Interpretable Logical ReasoningAntonia Creswell, Murray Shanahan, Irina HigginsICLR 2023 · 110 citations
- The Impact of Reasoning Step Length on Large Language ModelsMingyu Jin, Qinkai Yu, Dong Shu, Haiyan Zhao et al.ACL Findings 2024 · 39 citations
- Causal Sufficiency and Necessity Improves Chain-of-Thought ReasoningXiangning Yu, Zhuohan Wang, Linyi Yang, Haoxuan Li et al.NeurIPS 2025 · 19 citations
