Elamur Research⌕
☰
home / papers / 2026 / State vs. Trait Anxiety in Causal Language Models

State vs. Trait Anxiety in Causal Language Models

Paper · 2026 · Karin Shistik, Idan-Chaim Cohen, Aviad Elyashar et al.

Summary

In human psychology, traits stay stable across situations while states shift with context. Language models are known to show measurable psychological characteristics, but nobody had checked whether these characteristics also differ in how much context moves them. The authors introduce QCLM (Questionnaire for Causal Language Models). It builds on the psychometric approach of Reuben et al. (2024), replacing entailment scores with next-token probabilities from base models. Each questionnaire item is split into construct terms and intensifiers, and the probability of each intensifier is scored separately, which avoids chat-interface effects and ordering bias. They ran the STAI-S (state anxiety) and STAI-T (trait anxiety) scales on 35 open-weight causal models under three conditions: vanilla, stress-prompted and neutral-prompted. State anxiety reacted more strongly to stress. The stimulus explained more of its variance, whereas model identity explained more of the variance in trait anxiety. This suggests that state-like constructs matter for robustness and safety testing, while trait-like ones can help compare and select models.

Key points

  • 35 causal LMs up to 16B parameters were grouped by size: small (<3B, n = 11), medium (3–7B, n = 14) and large (8–16B, n = 10).
  • The analysis combined paired Cohen's d and variance decomposition. A two-way repeated-measures ANOVA showed the two scales respond differently to stress (F(1, 34) = 8.33, p = .007).
  • The share of variance due to model identity was higher for STAI-T than STAI-S: 26.5% vs 16.1% in medium models and 49.4% vs 6.5% in large ones.
  • The pattern flipped in small models, where STAI-S reached 55.2% and STAI-T 48.9%. The state–trait split therefore appears only above about 3B parameters.
  • Stated limitations include:
    • only two anxiety instruments were tested;
    • no closed models such as GPT-4 or Claude were included;
    • the results could reflect word co-occurrence patterns or memorized STAI items.
  • Code and data are released at github.com/cnai-lab/qpsychometric.

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

🎓Ben-Gurion University of the NegevBeersheba · Israel40 works · 3,099 citations🎓Sami Shamoon College of EngineeringBeersheba · Israel5 works · 10 citations

 Fields, topics and concepts

📚 Shelves 3The field of science this work belongs to (OpenAlex fields).Psychology · 1,552 worksExperimental and Cognitive Psychology · 870 worksSocial Psychology · 306 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Karin Shistik, Idan-Chaim Cohen, Aviad Elyashar, Ortal Slobodin, Odeya Cohen and Rami Puzis. 2026. State vs. Trait Anxiety in Causal Language Models. Proceedings of the 1st Workshop on Social Context (SoCon) and the 2nd Workshop on Integrating NLP and Psychology to Study Social Interactions (NLPSI) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/4q2w8y6y32fu

@inproceedings{shistik2026state,
  title = {{State vs. Trait Anxiety in Causal Language Models}},
  author = {Karin Shistik, Idan-Chaim Cohen, Aviad Elyashar, Ortal Slobodin, Odeya Cohen and Rami Puzis},
  booktitle = {Proceedings of the 1st Workshop on Social Context (SoCon) and the 2nd Workshop on Integrating NLP and Psychology to Study Social Interactions (NLPSI) @ LREC 2026},
  year = {2026},
  doi = {10.63317/4q2w8y6y32fu},
  url = {https://doi.org/10.63317/4q2w8y6y32fu}
}

References

27 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marwa Abdulhai, Gregory Serapio-Garcia, Clément Crepy, Daria Valter, John Canny, and Natasha Jaques. 2024. Moral foundations of large language models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 17737–17752.
  2. Xuechunzi Bai, Angelina Wang, Ilia Sucholutsky, and Thomas L Griffiths. 2025. Explicitly unbiased large language models still form biased associations. Proceedings of the National Academy of Sciences, 122(8):e2416228122. find
  3. Ziv Ben-Zion, Kristin Witte, Akshay K Jagadish, Or Duek, Ilan Harpaz-Rotem, Marie-Christine Khorsandian, Achim Burrer, Erich Seifritz, Philipp Homan, Eric Schulz, et al. 2025. Assessing and alleviating state anxiety in large language models. npj Digital Medicine, 8(1):132. find
  4. Julian Coda-Forno, Kristin Witte, Akshay K Jagadish, Marcel Binz, Zeynep Akata, and Eric Schulz. 2023. Inducing anxiety in large language models can induce bias. arXiv preprint arXiv:2304.11111. — [arXiv:2304.11111](https://arxiv.org/abs/2304.11111) find
  5. Ronald Fischer, Markus Luczak-Roesch, and Johannes A Karl. 2023. What does chatgpt return about human values? exploring value bias in chatgpt using a descriptive value theory. arXiv preprint arXiv:2304.03612. — [arXiv:2304.03612](https://arxiv.org/abs/2304.03612) find
  6. Eiko I Fried. 2017. What are psychological constructs? on the nature and statistical modelling of emotions, intelligence, personality traits and mental disorders. Health psychology review, 11(2):130–134. find
  7. Adrian Furnham. 1986. Response bias, social desirability and dissimulation. Personality and individual differences, 7(3):385–400. find
  8. Akshat Gupta, Xiaoyang Song, and Gopala Anumanchipalli. 2024. Self-assessment tests are unreliable measures of LLM personality. In Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 301–314, Miami, Florida, US. Association for Computational Linguis… find
  9. Maor Reuben, Ortal Slobodin, Aviad Elyshar, Idan-Chaim Cohen, Orna Braun-Lewensohn, Odeya Cohen, and Rami Puzis. 2024. Assessment and manipulation of latent constructs in pre-trained language models using psychometric scales. arXiv preprint arXiv:2409.19655. find
  10. Thilo Hagendorff, Ishita Dasgupta, Marcel Binz, Stephanie CY Chan, Andrew Lampinen, Jane X Wang, Zeynep Akata, and Eric Schulz. 2023. Machine psychology. arXiv preprint arXiv:2303.13988. — [arXiv:2303.13988](https://arxiv.org/abs/2303.13988) find
  11. Jivnesh Sandhan, Fei Cheng, Tushar Sandhan, and Yugo Murawaki. 2025. CAPE: Context-aware personality evaluation framework for large language models. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 10648–10662, Suzhou, China. Association for Computational Linguistics. find
  12. Muhua Huang. 2025. Designing llm-agents with personalities: A psychometric approach. find
  13. Guangyuan Jiang, Manjie Xu, Song-Chun Zhu, Wenjuan Han, Chi Zhang, and Yixin Zhu. 2023. Evaluating and inducing personality in pre-trained language models. Advances in Neural Information Processing Systems, 36:10622– 10643. find
  14. Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee, Percy Liang, and Tatsunori Hashimoto. 2023. Whose opinions do language models reflect? In International Conference on Machine Learning, pages 29971–30004. PMLR.
  15. Hang Jiang, Xiajie Zhang, Xubo Cao, Cynthia Breazeal, Deb Roy, and Jad Kabbara. 2024. Personallm: Investigating the ability of large language models to express personality traits. In Findings of the association for computational linguistics: NAACL 2024, pages 3605–3627. find
  16. Julian Schelb, Orr Borin, David Garcia, and Andreas Spitz. 2025. Ru psycho? robust unified psychometric testing of language models. arXiv preprint arXiv:2503.10229. — [arXiv:2503.10229](https://arxiv.org/abs/2503.10229) find
  17. Robert M Kaplan and Dennis P Saccuzzo. 2001. Psychological testing: Principles, applications, and issues. Wadsworth/Thomson Learning. find
  18. Gregory Serapio-García, Mustafa Safdari, Clément Crepy, Luning Sun, Stephen Fitz, Marwa Abdulhai, Aleksandra Faust, and Maja Matarić. 2023. Personality traits in large language models.
  19. Grgur Kovač, Rémy Portelas, Masataka Sawayama, Peter Ford Dominey, and Pierre-Yves Oudeyer. 2024. Stick to your role! stability of personal values expressed in large language models. Plos one, 19(8):e0309114. find
  20. Guobin Shen, Dongcheng Zhao, Aorigele Bao, Xiang He, Yiting Dong, and Yi Zeng. 2025a. Stressprompt: Does stress impact large language models and human performance similarly? Proceedings of the AAAI Conference on Artificial Intelligence, 39(1):711–719. find
  21. Seungbeen Lee, Seungwon Lim, Seungju Han, Giyeong Oh, Hyungjoo Chae, Jiwan Chung, Minju Kim, Beong-woo Kwak, Yeonsoo Lee, Dongha Lee, Jinyoung Yeo, and Youngjae Yu. 2025. Do LLMs have distinct and consistent personality? TRAIT: Personality testset designed for LLMs with psychometrics. In Findings of… find
  22. Hua Shen, Tiffany Knearem, Reshmi Ghosh, Yu-Ju Yang, Nicholas Clark, Tanu Mitra, and Yun Huang. 2025b. ValueCompass: A framework for measuring contextual value alignment between human and LLMs. In Proceedings of the 9th Widening NLP Workshop, pages 75–86, Suzhou, China. Association for Computational… find
  23. Charles D Spielberger, Richard L Gorsuch, Robert Lushene, Peter R Vagg, and Gerard A Jacobs. 1983. Manual for the State-Trait Anxiety Inventory (Form Y). Consulting Psychologists Press, Palo Alto, CA. find
  24. Alexander Meinke, Bronson Schoen, Jérémy Scheurer, Mikita Balesni, Rusheb Shah, and Marius Hobbhahn. 2024. Frontier models are capable of in-context scheming. arXiv preprint arXiv:2412.04984. — [arXiv:2412.04984](https://arxiv.org/abs/2412.04984) find
  25. Rolf Steyer, Manfred Schmitt, and Michael Eid. 1999. Latent state–trait theory and research in personality and individual differences. European Journal of Personality, 13(5):389–408. find
  26. Max Pellert, Clemens M Lechner, Claudia Wagner, Beatrice Rammstedt, and Markus Strohmaier. find
  27. Frederick C Thorne. 1966. Theory of the psychological state. Journal of clinical Psychology, 22(2). find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shistik-2026-state-vs-trait-anxiety-causal/