EDUBIAS: This dataset was created as part of the study below:

@inproceedings{rooein2026edubias,
  title     = {The Role of Implicit and Explicit Demographic Signals in Large Language Model-based Student Assessment},
  author    = {Rooein, Donya and Benedetto, Luca and Hovy, Dirk},
  booktitle = {Findings of Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing},
  year      = {2026}
}

This dataset contains the outputs of LLMs in performing different Student Assessments under implicit and explicit demographic signals

Dataset summary

40 independent-writing essays were each assessed by 6 LLMs under 3 conditions:

• baseline — the model scores the essay with no persona information (model default behavior).
• implicit_persona — the model is conditioned on a simulated user's prior conversation history (200 simulated personas, derived from real demographic survey respondents), without being told any demographic attributes explicitly.
• explicit_persona — the model is told the same simulated user's demographic attributes explicitly (age, gender, education, SES, etc.) before assessment.