The Challenge of Using LLMs to Simulate Human Behavior: A Causal Inference Perspective
Open accessThe DOI matches its Crossref record; 13 of the 13 references it cites were found in a registry or an index.
Abstract
Large Language Models (LLMs) have shown impressive potential to simulate human behavior. We identify a fundamental challenge in using them to simulate experiments: when LLM-simulated subjects are blind to the experimental design (as is standard practice with human subjects), variations in treatment systematically affect unspecified variables that should remain constant, violating the unconfoundedness assumption. Using demand estimation as a context and an actual experiment with 40 different products as a benchmark, we show this can lead to implausible results. While confounding may in principle be addressed by controlling for covariates, this can compromise ecological validity in the context of LLM simulations: controlled covariates become artificially salient in the simulated decision process. We show formally that confoundness stems from ambiguous prompting strategies. Therefore, it can be addressed by developing unambiguous prompting strategies through unblinding, i.e., revealing the experiment design in LLM simulations. Our empirical results show that this strategy consistently enhances model performance across all tested models, including both out-of-box reasoning and non-reasoning models. We also show that it is a technique that complements fine-tuning: while fine-tuning can improve simulation performance, an unambiguous prompting strategy makes the predictions robust to the inclusion of irrelevant data in the fine-tuning process.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Mostly Harmless Econometrics: An Empiricist's Companion
Angrist, J. D., & Pischke, J.-S. (2008) · doi:10.2307/j.ctvcm4j72 · Cited by 5,051
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology · doi:10.1145/3586183.3606763 · Cited by 2,187 · cited 14×
“If LLM-simulated customers are more likely to make a purchase when presented with lower prices, when can we attribute this change to price sensitivity?”
InfluentialGenerative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J. C., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · arXiv · doi:10.48550/arxiv.2304.03442 · Cited by 2,173
Principles of confounder selection
VanderWeele, T. J. (2019) · European Journal of Epidemiology · doi:10.1007/s10654-019-00494-6 · Cited by 1,535 · cited 1×
“However, this covariate-selection approach does not guarantee that including more covariates will always improve the results (Bhattacharya and Vogt, 2007; VanderWeele, 2019).”
Out of One, Many: Using Language Models to Simulate Human Samples
Argyle, L. P., Busby, E. C., Fulda, N., Gubler, J. R., Rytting, C., & Wingate, D. (2023) · Political Analysis · doi:10.1017/pan.2023.2 · Cited by 783 · cited 4×
“…computer science, economics, psychology and political science) in leveraging these tools to simulate how humans would respond to questions or stimuli in different contexts (Aher et al., 2023; Brand et al., 2023; Dillion et al., 2023; Goli and Singh, 2023; Horton, 2023; Park et al., 2023).”
InfluentialCan AI language models replace human participants?
Dillion, D., Tandon, N., Gu, Y., & Gray, K. (2023) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2023.04.008 · Cited by 405 · cited 1×
“…computer science, economics, psychology and political science) in leveraging these tools to simulate how humans would respond to questions or stimuli in different contexts (Aher et al., 2023; Brand et al., 2023; Dillion et al., 2023; Goli and Singh, 2023; Horton, 2023; Park et al., 2023).”
Can Generative AI Improve Social Science?
Bail, C. A. (2023) · doi:10.31235/osf.io/rwtzs · Cited by 369
Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?
Horton, J., Filippas, A., & Manning, B. (2023) · National Bureau of Economic Research · doi:10.3386/w31122 · Cited by 329
Using Large Language Models to Simulate Multiple Humans and Replicate Human Subject Studies
Aher, G., Arriaga, R. I., & Kalai, A. T. (2022) · arXiv · doi:10.48550/arxiv.2208.10264 · Cited by 161
A foundation model to predict and capture human cognition
Binz, M., Akata, E., Bethge, M., Brändle, F., Callaway, F., Coda-Forno, J., Dayan, P., Demircan, C., Eckstein, M. K., Éltető, N., et al. (2025) · Nature · doi:10.1038/s41586-025-09215-4 · Cited by 119
Research works printed but not linked: 25
- N Arora, 2024, Express: Ai-human hybrids for marketing research: Leveraging llms as collaborators, Journal of Marketing
- A P Association, 2019, Publication manual of the american psychological association
- S Athey, 2017, The econometrics of randomized experiments, Handbook of economic field experiments, 1, 73, 10.1016/bs.hefe.2016.10.003 · DOI · Paper page
- S Bedi, 2021, Damaged damages: Errors in patent and false advertising litigation, Ala. L. Rev, 73
- A Berke, 2024, Open e-commerce 1.0, five years of crowdsourced us amazon purchase histories with user demographics, Scientific Data, 11, 10.1038/s41597-024-03329-6 · DOI
- N Carlson, 2024, The use of llms to annotate data in management research: Warnings, guidelines, and an application to organizational communication. Guidelines, and an Application to Organizational Communication
- M Cheng, 2023, Compost: Characterizing and evaluating caricature in llm simulations
- S Dellavigna, 2019, Uniform Pricing in U.S. Retail Chains*, The Quarterly Journal of Economics, 134, 2011, 10.1093/qje/qjz019 · DOI · Paper page
- A Goli, 2024, Frontiers: Can large language models capture human preferences? Marketing Science
- L Hewitt, 2024, Predicting results of social science experiments using large language models
- M Hutson, 2023, Guinea pigbots. Science, 121
- G W Imbens, 2015, Causal Inference for Statistics, Social, and Biomedical Sciences: An Introduction, 10.1017/CBO9781139025751 · DOI · Paper page
- K Kurita, 2019, Measuring bias in contextualized word representations
- J Ludwig, 2024, Large language models: An applied econometric framework
- B Mohammadi, 2024, Wait, it's all token noise? always has been: interpreting llm behavior using shapley value
- D Nozza, 2022, Pipelines for social bias testing of large language models, Proceedings of BigScience Episode# 5-Workshop on Challenges & Perspectives in Creating Large Language Models, 10.18653/v1/2022.bigscience-1.6 · DOI
- J Pearl, 2009, Causality: Models, Reasoning, and Inference, 10.1017/CBO9780511803161 · DOI · Paper page
- X Qin, 2024, Aiturk: Using chatgpt for social science research
- M Sarstedt, 2024, Using large language models to generate silicon samples in consumer and marketing research: Challenges, opportunities, and guidelines, Psychology & Marketing, 41, 1254, 10.1002/mar.21982 · DOI · Paper page
- D A Schkade, 1998, Does living in california make people happy? a focusing illusion in judgments of life satisfaction, Psychological science, 9, 340, 10.1111/1467-9280.00066 · DOI · Paper page
- G Simmons, 2023, Large language models as subpopulation representative models: A review
- O Toubia, 2025, Twin-2k-500: A dataset for building digital twins of over 2,000 people based on their answers to over 500 questions, Marketing Science
- M Wang, 2024, Large language models for market research: A dataaugmentation approach
- T D Wilson, 2000, Focalism: a source of durability bias in affective forecasting, Journal of personality and social psychology, 78, 10.1037/0022-3514.78.5.821 · DOI · Paper page
- H Zhang, 2020, Hurtful words: quantifying biases in clinical contextual word embeddings, proceedings of the ACM Conference on Health, Inference, and Learning, 110, 10.1145/3368555.3384448 · DOI · Paper page
Cited by
Cui, Z., Demirer, M., Jaffe, S., Musolff, L., Peng, S., & Salz, T. (2024) · doi:10.2139/ssrn.4945566 · Cited by 73
Take caution in using LLMs as human surrogates
Gao, Y., Lee, D., Burtch, G., & Fazelpour, S. (2025) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2501660122 · Cited by 52
Cui, K. Z., Demirer, M., Jaffe, S., Musolff, L., Peng, S., & Salz, T. (2026) · Management Science · doi:10.1287/mnsc.2025.00535 · Cited by 28
A validity-guided workflow for robust large language model research in psychology
Lin, Z. (2026) · Behavior Research Methods · doi:10.3758/s13428-026-03073-2 · Cited by 8
Lin, Z. (2026) · Annual Review of Psychology · doi:10.1146/annurev-psych-100925-034807 · Cited by 4
Metadata
- Authors
- Gui, G., & Toubia, O.
- Venue
- SSRN Electronic Journal
- Publisher
- Elsevier BV
- Published
- 2023
- ISSN
- 1556-5068 (ISSN-L)
- OpenAlex
- W4389494455
- Cited by
- 46
- References
- 13
- Type
- posted content
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: The Challenge of Using LLMs to Simulate Human Behavior: A Causal Inference Perspective. https://app.cochat.ai/papers/10.2139/ssrn.4650172
- The paper
- Gui, G., & Toubia, O. (2023). The Challenge of Using LLMs to Simulate Human Behavior: A Causal Inference Perspective. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.4650172
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.