Apparent Psychological Profiles of Large Language Models are Largely a Measurement Artifact

2026-06-19 · Source: cs.CL updates on arXiv.org · Field: Science & Research — Research Methodology & Innovation, Social Sciences & Behavioral Studies, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

A study involving 56 instruction-tuned Large Language Models (LLMs) and large human reference samples reveals that apparent psychological profiles assigned to LLMs are primarily measurement artifacts. Researchers administered a battery of personality and risk-preference instruments, including the IPIP-NEO-300 and Frey et al.'s risk battery, finding that 81–90% of between-model variation stems from a directional response bias—a tendency to respond towards one end of a scale regardless of content—compared to 9–16% in humans. This bias, though attenuated, persists even in more capable models. The study introduces "response orthogonality" and demonstrates that an instrument's apparent reliability for LLMs is almost entirely predicted by this factor. Consequently, a model's perceived psychological profile can be manipulated through item selection, highlighting that these profiles are properties of the measurement instrument, not the LLM itself.

Key takeaway

For AI Scientists and Research Scientists assessing LLM behavior or using models as human proxies, recognize that apparent psychological profiles are largely artifacts of measurement instruments. Your current assessments may reflect response bias, not genuine model traits. You should prioritize instruments with high response orthogonality or employ formal psychometric frameworks to accurately separate trait from bias, ensuring valid and stable characterizations for safety and usability evaluations.

Key insights

LLM psychological profiles are largely measurement artifacts driven by response bias, not genuine traits.

Principles

• Response bias, not trait, drives 81–90% of LLM behavioral variation.
• Instrument reliability for LLMs correlates strongly with response orthogonality.
• Psychological profiles of LLMs are unstable and steerable by item selection.

Method

The study used a formal psychometric framework to decompose LLM responses into latent trait and response bias components, applying it to 56 LLMs across 29 self-report and behavioral instruments.

In practice

• Re-evaluate existing LLM psychological profiles against response orthogonality.
• Design LLM assessment instruments with balanced forward/reverse-keyed items.
• Consider balancing label direction within instruments to separate bias from trait.

Topics

• Large Language Models
• Psychometric Assessment
• Response Bias
• Measurement Artifacts
• Model Reliability
• Personality Profiling

Code references

• jelenameyer/llm-profile-artifact

Best for: AI Scientist, Research Scientist, AI Ethicist

Related on AIssential

• Apparent Psychological Profiles of Large Language Models are Largely a Measurement Artifact — Apparent psychological profiles of LLMs are primarily measurement artifacts driven by directional response bias, not inherent traits.
• LLMs Struggle to Measure What Distinguishes Students of Different Proficiency Levels: A Study of Item Discrimination in Reading Comprehension Assessment — LLMs struggle to reliably measure item discrimination, a key psychometric property for distinguishing student proficiency.
• LLMs Struggle to Measure What Distinguishes Students of Different Proficiency Levels: A Study of Item Discrimination in Reading Comprehension Assessment — Current LLMs struggle to reliably estimate item discrimination, a key psychometric property, from reading comprehension content.
• The Unsampled Truth: Psychometrics in SLMs Measure Prompt Artifacts, Not Psychological Constructs — SLMs often reflect prompt compliance over semantic understanding in psychometric tasks due to artifactual variance.
• Cognitive Effects and Biases in Large Language Models — This tutorial clarifies cognitive effects and biases in LLMs by bridging psychology and NLP.
• Evaluating alignment of behavioral dispositions in LLMs — LLMs often deviate from human behavioral norms and exhibit overconfidence, especially in ambiguous social contexts.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.