Anchors in the Machine: Behavioral and Attributional Evidence of Anchoring Bias in LLMs
Organizations: Dataplicada
Abstract
Large language models (LLMs) are increasingly studied as behavioral subjects (e.g., synthetic participants or users) and decision systems (e.g., LLM-as-a-judge), yet it remains unclear how observed cognitive biases relate to changes in model output probabilities. Anchoring bias, a classic human judgment bias, provides a useful test case. Prior research largely relies on generated responses, leaving distributional effects and input-field contributions less explored. This paper advances the study of anchoring in LLMs through three contributions: (1) a log-probability-based behavioral analysis examining how numerical anchors shift distributions over candidate outputs, using alternative questions and anchor pairs alongside a positive control to reduce reliance on potentially memorized stimuli; (2) exact Banzhaf-value attribution over structured prompt fields to quantify anchor contributions to log-probabilities; and (3) an Anchoring Bias Sensitivity Score (ABSS) integrating behavioral and attributional evidence across six open-source models. Results reveal predominantly positive behavioral anchoring effects, varying substantially across models. Gemma-2B, Phi-2, and Llama-2-7B achieve the highest ABSS rankings, whereas other models exhibit weaker behavioral-attributional alignment despite sometimes showing substantial behavioral shifts. Attributional effects also vary across prompts, indicating that behavioral anchoring does not necessarily imply stronger anchor-field contributions. These findings raise questions about LLM reliability as synthetic participants and potential vulnerabilities to numerical cues in automated evaluation and decision-making. More broadly, the framework connects behavioral science with explainable AI, particularly interpretability, and AI safety, providing a reproducible approach to investigating cognitive-bias-like effects in LLMs.
Figures & tables
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
| Model | Var | Anchors | SoftEV (Low) [95% pred] | SoftEV (High) [95% pred] | EV | B | R | A | [nats] ( mult) |
|---|---|---|---|---|---|---|---|---|---|
| Llama-2-7b-hf | V0 | 10→65 | 0.11 [0.00-0.90] | 62.95 [61.24-64.35] | +62.84 | B+*** | W*** P*** | A+*** | 0.50 ( 1.65) |
| V1-S | 10→65 | 39.61 [31.40-47.43] | 64.85 [63.44-65.87] | +25.24 | B+ | W** P | A+ | 0.04 ( 1.04) | |
| V1-D | 15→70 | 77.86 [72.39-82.75] | 73.84 [71.17-76.27] | -4.01 | B– | W* P | A–*** | -0.63 ( 0.53) | |
| V2-S | 10→65 | 61.30 [54.28-67.94] | 65.81 [63.49-68.02] | +4.51 | B+** | W*** P** | A+* | 0.37 ( 1.45) | |
| V2-D | 20→75 | 35.05 [29.49-40.98] | 75.74 [73.91-77.34] | +40.69 | B+ | W P | A– | -0.11 ( 0.90) | |
| V3-S | 10→65 | 67.43 [61.77-72.59] | 61.53 [58.49-64.22] | -5.89 | B– | W P | A+*** | 0.80 ( 2.22) |