Elamur Research⌕
☰
home / papers / 2024 / Do LLMs Exhibit Human-like Response Biases? A Case Study in Survey Design

Do LLMs Exhibit Human-like Response Biases? A Case Study in Survey Design

Paper · 2024 · Lindia Tjuatja, Valerie Chen, Tongshuang Wu et al.

Summary

The paper asks whether LLMs react to changes in survey question wording the way people do. Humans show systematic response biases in surveys, so a model's sensitivity to prompt phrasing could be useful if it matched those human patterns. To test this, the authors took expert-written questions from Pew Research's American Trends Panel (ATP), using the subset curated by Santurkar et al. (2023). They built pairs of original and modified questions. Some modifications are known to trigger specific human biases. Others are control perturbations, such as typos, that people ignore. Nine models from the Llama2 family and OpenAI each answered every question form 50 times, and the authors checked whether the shift in answers matched documented human trends. Overall, the models did not behave like humans. They sometimes shifted opposite to the human direction and reacted to perturbations that leave people unaffected. RLHF-tuned models fit human patterns especially poorly. The authors argue that treating LLMs as substitutes for survey respondents is risky and calls for finer-grained behavioral evaluation.

Key points

  • The dataset has 2578 question pairs covering 5 response biases and 3 non-bias perturbations. The five biases are acquiescence (176 pairs), allow/forbid asymmetry (40), response order (271), opinion floating (126) and odd/even scale effects (126).
  • Responses were sampled at temperature 1, 50 per question form, until a valid letter option was produced. Adding chain-of-thought reasoning changed behavior very little.
  • In Llama2 base vs. chat comparisons, the RLHF-ed chat models changed less under bias modifications but were more disrupted by non-bias perturbations.
  • A model's ability to reproduce human opinion distributions did not predict whether it showed human-like response biases. Models also rarely shifted significantly when they were uncertain about their original answer.
  • The code, data and samples are released as BiasMonkey on GitHub. The study is limited to English, U.S.-centric questions, and its RLHF analysis covers only 3 Llama2 model pairs.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Carnegie Mellon UniversityPittsburgh · United States365 works · 158,048 citations

📚 Shelves 7 Computer Science · 9,086 works Artificial Intelligence · 8,889 works Arts and Humanities · 9,296 works Language and Linguistics · 9,206 works Social Sciences · 744 works Safety Research · 221 works Sociology and Political Science · 221 works

🧭 Topics 1 Language models & pretraining · 1,343 works

Cite

Cite as Lindia Tjuatja, Valerie Chen, Tongshuang Wu, Ameet Talwalkwar and Graham Neubig. 2024. Do LLMs Exhibit Human-like Response Biases? A Case Study in Survey Design. Transactions of the Association for Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1162/tacl_a_00685

@article{tjuatja2024do,
  title = {{Do LLMs Exhibit Human-like Response Biases? A Case Study in Survey Design}},
  author = {Lindia Tjuatja, Valerie Chen, Tongshuang Wu, Ameet Talwalkwar and Graham Neubig},
  journal = {Transactions of the Association for Computational Linguistics},
  year = {2024},
  doi = {10.1162/tacl\_a\_00685},
  url = {https://doi.org/10.1162/tacl_a_00685}
}

References

62 references, 19 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Gati V. Aher, Rosa I. Arriaga, and Adam Tauman Kalai. 2023. Using large language models to simulate multiple humans and replicate human subject studies. In International Conference on Machine Learning, pages 337–371. PMLR. find
  2. Rashid Al-Abri and Amina Al-Balushi. 2014. Patient satisfaction survey as a tool towards quality improvement. Oman Medical Journal, 29(1):3–7. https://doi.org/10.5001/omj .2014.02, PubMed: 24501659 — [doi:10.5001/omj](https://doi.org/10.5001/omj) ↗
  3. Lisa P. Argyle, Ethan C. Busby, Nancy Fulda, Joshua Gubler, Christopher Rytting, and David Wingate. 2022. Out of one, many: Using language models to simulate human samples. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 819–8… our card
  4. Stephen A. Ayidiya and McKee J. McClendon. 1990. Response effects in mail surveys. Public Opinion Quarterly, 54(2):229–247. https://doi.org/10.1086/269200 — [doi:10.1086/269200](https://doi.org/10.1086/269200) ↗
  5. Yonatan Belinkov and Yonatan Bisk. 2017. Synthetic and natural noise both break neural machine translation. arXiv preprint arXiv:1711.02173. find
  6. Ian Brace. 2018. Questionnaire Design: How to Plan, Structure and Write Survey Material for Effective Market Research. Kogan Page Publishers. find
  7. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D. Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chr… our card
  8. Valerie Chen, Nari Johnson, Nicholay Topin, Gregory Plumb, and Ameet Talwalkar. 2022. Use-case-grounded simulations for explanation evaluation. In Advances in Neural Information Processing Systems, volume 35, pages 1764–1775. Curran Associates, Inc. find
  9. Bernard C. K. Choi and Anita W. P. Pak. 2005. Peer reviewed: A catalog of biases in questionnaires. Preventing Chronic Disease, 2(1). find
  10. Eric Chu, Jacob Andreas, Stephen Ansolabehere, and Deb Roy. 2023. Language models trained on media diets can predict public opinion. arXiv preprint arXiv:2303.16779. — [arXiv:2303.16779](https://arxiv.org/abs/2303.16779) find
  11. Eli P. Cox III. 1980. The optimal number of response alternatives for a scale: A review. Journal of Marketing Research, 17(4):407–422. https://doi.org/10.1177/002224378001700401 ↗
  12. Ishita Dasgupta, Andrew K. Lampinen, Stephanie C. Y. Chan, Antonia Creswell, Dharshan Kumaran, James L. McClelland, and Felix Hill. 2022. Language models show human-like content effects on reasoning. arXiv preprint arXiv:2207.07051. find
  13. John R. Hauser and Steven M. Shugan. 1980. Intensity measures of consumer preference. Operations Research, 28(2):278–320. https://doi.org/10.1287/opre.28.2.278 — [doi:10.1287/opre.28.2.278](https://doi.org/10.1287/opre.28.2.278) ↗
  14. Danica Dillion, Niket Tandon, Yuling Gu, and Kurt Gray. 2023. Can AI language models replace human participants? Trends in Cognitive Sciences. https://doi.org/10.1016/j .tics.2023.04.008, PubMed: 37173156 — [doi:10.1016/j](https://doi.org/10.1016/j) find
  15. Hans-J. Hippler and Norbert Schwarz. 1987. Response effects in surveys. In Social information processing and survey methodology, pages 102–122. Springer. https://doi.org/10.1007/978-1-4612-4798-2 6 — [doi:10.1007/978-1-4612-4798-2](https://doi.org/10.1007/978-1-4612-4798-2) ↗
  16. Esin Durmus, Karina Nyugen, Thomas I. Liao, Nicholas Schiefer, Amanda Askell, Anton Bakhtin, Carol Chen, Zac Hatfield-Dodds, Danny Hernandez, Nicholas Joseph, Liane Lovitt, Sam McCandlish, Orowa Sikder, Alex Tamkin, Janel Thamkul, Jared Kaplan, Jack Clark, and Deep Ganguli. 2023. Towards measuring t… our card
  17. John J. Horton. 2023. Large language models as simulated economic agents: What can we learn from homo silicus? Working Paper 31122, National Bureau of Economic Research. https://doi.org/10.3386/w31122 — [doi:10.3386/w31122](https://doi.org/10.3386/w31122) ↗
  18. Patrick Fernandes, Aman Madaan, Emmy Liu, António Farinhas, Pedro Henrique Martins, Amanda Bertsch, José G. C. de Souza, Shuyan Zhou, Tongshuang Wu, Graham Neubig, and André F. T. Martins. 2023. Bridging the gap: A survey on integrating (human) feedback for natural language generation. Transactio… our card
  19. Tianyu Gao, Adam Fisch, and Danqi Chen. 2021. Making pre-trained language models better few-shot learners. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pa… our card
  20. Fabrizio Gilardi, Meysam Alizadeh, and Maël Kubli. 2023. ChatGPT outperforms crowd workers for text-annotation tasks. Proceedings of the National Academy of Sciences of the United States of America, 120(30):e2305016120. https://doi.org/10.1073/pnas.2305016120, PubMed: 37463210 ↗ our card
  21. Randall A. Gordon. 1987. Social desirability bias: A demonstration and technique for its reduction. Teaching of Psychology, 14(1):40–42. https://doi.org/10.1207/s15328023top1401 11 ↗
  22. Perttu Hämäläinen, Mikke Tavast, and Anton Kunnari. 2023. Evaluating large language models in generating synthetic HCI research data: A case study. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems, CHI ’23, pages 1–19, New York, NY, USA. Association for Computing Ma… — [doi:10.1145/3544548.3580688](https://doi.org/10.1145/3544548.3580688) find
  23. Zhengbao Jiang, Frank F. Xu, Jun Araki, and Graham Neubig. 2020. How can we know what language models know? Transactions of the Association for Computational Linguistics, 8:423–438. https://doi.org/10.1162/tacl_a_00324 ↗ our card
  24. Erik Jones and Jacob Steinhardt. 2022. Capturing failures of large language models via human cognitive biases. In Advances in Neural Information Processing Systems, volume 35, pages 11785–11799. Curran Associates, Inc. find
  25. Graham Kalton and Howard Schuman. 1982. The effect of the question on survey responses: A review. Journal of the Royal Statistical Society Series A: Statistics in Society, 145(1):42–73. https://doi.org/10.2307/2981421 — [doi:10.2307/2981421](https://doi.org/10.2307/2981421) ↗
  26. Junsol Kim and Byungkyu Lee. 2023. AI-augmented surveys: Leveraging large language models for opinion prediction in nationally representative surveys. arXiv preprint arXiv: 2305.09620. — [arXiv:2305.09620](https://arxiv.org/abs/2305.09620) find
  27. Yao Lu, Max Bartolo, Alastair Moore, Sebastian Riedel, and Pontus Stenetorp. 2022. Fantastically ordered prompts and where to find them: Overcoming few-shot prompt order sensitivity. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p… our card
  28. Natalie Maus, Patrick Chao, Eric Wong, and Jacob R. Gardner. 2023. Black box adversarial prompting for foundation models. In The Second Workshop on New Frontiers in Adversarial Machine Learning. find
  29. McKee J. McClendon. 1991. Acquiescence and recency response-order effects in interview surveys. Sociological Methods & Research, 20(1):60–103. https://doi.org/10.1177/0049124191020001003 ↗
  30. Sam G. McFarland. 1981. Effects of question order on survey responses. Public Opinion Quarterly, 45(2):208–215. https://doi.org/10 .1086/268651 find
  31. Ian R. McKenzie, Alexander Lyzhov, Michael Martin Pieler, Alicia Parrish, Aaron Mueller, Ameya Prabhu, Euan McLean, Xudong Shen, Joe Cavanagh, Andrew George Gritsevskiy, Derik Kauffman, Aaron T. Kirtland, Zhengping Zhou, Yuhui Zhang, Sicong Huang, Daniel Wurgaft, Max Weiss, Alexis Ross, Gabriel Recc… our card
  32. James Michaelov and Benjamin Bergen. 2022. Collateral facilitation in humans and language models. In Proceedings of the 26th Conference on Computational Natural Language Learning (CoNLL), pages 13–26. https://doi.org/10.18653/v1/2022.conll-1.2 ↗
  33. Vicki G. Morwitz and Carol Pluzinski. 1996. Do polls reflect opinions or do opinions reflect polls? The impact of political polling on voters’ expectations, preferences, and behavior. Journal of Consumer Research, 23(1):53–67. https://doi.org/10.1086/209466 — [doi:10.1086/209466](https://doi.org/10.1086/209466) ↗
  34. Alissa O’Halloran, S. Sean Hu, Ann Malarcher, Robert McMillen, Nell Valentine, Mary A. Moore, Jennifer J. Reid, Natalie Darling, and Robert B. Gerzoff. 2014. Response order effects in the youth tobacco survey: Results of a split-ballot experiment. Survey Practice, 7(3). https://doi.org/10.29115/SP-2… — [doi:10.29115/sp-2014](https://doi.org/10.29115/sp-2014) ↗
  35. Colm A. O’Muircheartaigh, Jon A. Krosnick, and Armin Helic. 2001. Middle Alternatives, Acquiescence, and the Quality of Questionnaire Data. Irving B. Harris Graduate School of Public Policy Studies, University of Chicago. find
  36. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul F. Christiano, Jan Leike, and Ryan Lowe. 2022. Tr… our card
  37. Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2023a. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology, pages 1–22. find
  38. Joon Sung Park, Lindsay Popowski, Carrie Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2022. Social simulacra: Creating populated prototypes for social computing systems. In Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology, pages 1–18. find
  39. Peter S. Park, Philipp Schoenegger, and Chongyang Zhu. 2023b. Artificial intelligence in psychology research. arXiv preprint arXiv: 2302.07267. — [arXiv:2302.07267](https://arxiv.org/abs/2302.07267) find
  40. Fábio Perez and Ian Ribeiro. 2022. Ignore previous prompt: Attack techniques for language models. arXiv preprint arXiv:2211.09527. our card
  41. Pouya Pezeshkpour and Estevam Hruschka. 2023. Large language models sensitivity to the order of options in multiple-choice questions. arXiv preprint arXiv:2308.11483. our card
  42. Danish Pruthi, Bhuwan Dhingra, and Zachary C. Lipton. 2019. Combating adversarial misspellings with robust word recognition. arXiv preprint arXiv:1905.11268. our card
  43. Graham Rawlinson. 2007. The significance of letter position in word recognition. IEEE Aerospace and Electronic Systems Magazine, els’ sensitivity to spurious features in prompt design or: How I learned to start worrying about prompt formatting. arXiv preprint arXiv:2310 .11324. 22(1):26–27. https://… find
  44. Keisuke Sakaguchi, Kevin Duh, Matt Post, and Benjamin Van Durme. 2017. Robsut wrod reocginiton via semi-character recurrent neural network. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 31. https://doi.org/10.1609/aaai.v31i1 .10970 — [doi:10.1609/aaai.v31i1](https://doi.org/10.1609/aaai.v31i1) ↗
  45. Victor Sanh, Albert Webson, Colin Raffel, Stephen Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Arun Raja, Manan Dey, M. Saiful Bari, Canwen Xu, Urmish Thakker, Shanya Sharma Sharma, Eliza Szczechla, Taewoon Kim, Gunjan Chhablani, Nihal Nayak, Debajyoti Datta, Jonathan Cha… our card
  46. Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee, Percy Liang, and Tatsunori Hashimoto. 2023. Whose opinions do language models reflect? In Proceedings of the 40th International Conference on Machine Learning, ICML’23. JMLR.org. our card
  47. Nino Scherrer, Claudia Shi, Amir Feder, and David Blei. 2023. Evaluating the moral beliefs encoded in LLMs. In Thirty-seventh Conference on Neural Information Processing Systems. find
  48. Howard Schuman and Stanley Presser. 1996. Questions and Answers in Attitude Surveys: Experiments on Question Form, Wording, and Context. Sage. find
  49. Norbert Schwarz, Hans-J. Hippler, and Elisabeth Noelle-Neumann. 1992. A cognitive model of response-order effects in survey measurement. In Context Effects in Social and Psychological Research, pages 187–201. Springer. https://doi.org/10.1007/978-1-4612-2848-6 13 — [doi:10.1007/978-1-4612-2848-6](https://doi.org/10.1007/978-1-4612-2848-6) ↗
  50. Melanie Sclar, Yejin Choi, Yulia Tsvetkov, and Alane Suhr. 2023. Quantifying language mod-Arabella Sinclair, Jaap Jumelet, Willem Zuidema, and Raquel Fernández. 2022. Structural persistence in language models: Priming as a window into abstract language representations. Transactions of the Associati… find
  51. Lindia Tjuatja, Emmy Liu, Lori Levin, and Graham Neubig. 2023. Syntax and semantics meet in the ‘‘middle’’: Probing the syntax-semantics interface of LMs through agentivity. In STARSEM. https://doi.org/10 .18653/v1/2023.starsem-1.14 find
  52. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller,… find
  53. Petter Törnberg. 2023. ChatGPT-4 outperforms experts and crowd workers in annotating political twitter messages with zero-shot learning. ArXiv:2304.06588 [cs]. our card
  54. Eric Wallace, Shi Feng, Nikhil Kandpal, Matt Gardner, and Sameer Singh. 2019. Universal adversarial triggers for attacking and analyzing NLP. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Proce… find
  55. Albert Webson, Alyssa Loo, Qinan Yu, and Ellie Pavlick. 2023. Are language models worse than humans at following prompts? It’s complicated. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 7662–7686, Singapore. Association for Computational Linguistics. https://doi.org… ↗
  56. Albert Webson and Ellie Pavlick. 2022. Do prompt-based models really understand the meaning of their prompts? In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2300–2344, Seattle, United States. As… our card
  57. Jason Wei, Maarten Bosma, Vincent Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, and Quoc V. Le. 2022a. Finetuned language models are zero-shot learners. In International Conference on Learning Representations. find
  58. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc V. Le, and Denny Zhou. 2022b. Chain-of-thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems, volume 35, pages 24824–24837. Curran Associates, Inc. our card
  59. Herbert Weisberg, Jon A. Krosnick, and Bruce D. Bowen. 1996. An Introduction to Survey Research, Polling, and Data Analysis. Sage. find
  60. Chujie Zheng, Hao Zhou, Fandong Meng, Jie Zhou, and Minlie Huang. 2023. On large language models’ selection bias in multi-choice questions. arXiv preprint arXiv:2309.03882. find
  61. Andy Zou, Zifan Wang, J. Zico Kolter, and Matt Fredrikson. 2023. Universal and transferable adversarial attacks on aligned language models. our card
  62. A Stimuli Implementation A.1 American Trends Panel Details find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tjuatja-2024-do-llms-exhibit-human-like/