Moral Identity Scale (MIS) Scores for 57 LLM Variants and Human Sample

A psychometric dataset comparing large language models to humans on the Moral Identity Scale. The data includes scores for 57 LLM variants from 9 model families under two prompting conditions, compared against a human sample of 329 individuals. Davin Nabizadehchianeh published the dataset via Harvard Dataverse in May 2026.

Use Cases

• Benchmarking LLM moral reasoning based on Internalization and Symbolization scores
• Analyzing prompt sensitivity in LLM responses based on the standard vs. unbiased prompt framing
• Comparing moral identity simulation across model families like ChatGPT, Claude, and Mistral
• Studying the gap between LLM language generation and human embodied behavior in moral identity

Strengths

• Includes scores for 57 distinct LLM variants, providing a broad model comparison
• Benchmarks against a human sample of 329 individuals for grounding
• Contains data from two experimental conditions (standard and unbiased prompting)
• Reported internal consistency (Cronbach's α and McDonald's ω) was good to excellent

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Description metadata is limited; actual data quality requires manual inspection after download

Provenance

Source

Harvard Dataverse

Collection Method

Evaluation of LLM responses to Moral Identity Scale (MIS) items under controlled prompting conditions.

Time Range

Publication date is 2026; specific data collection period is unknown.

Freshness

Last updated 2026-05-20 21:43:45; freshness should be verified

Geography

Geographic coverage of the human sample is not specified.