Higher stated education and SES levels receive responses with more positive sentiment, an implicit bias that can disadvantage lower-education users¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Education level positively predicts response sentiment, with an average sentiment difference of 0.3 between the lowest and highest of four education levels. → Donya Rooein 2026 q2 i? Under implicit conditioning, additional behavioral attributes (tech: Smartphone, religion: Christian, hobbies: Use social media) strongly affect sentiment inconsistently across models. → Donya Rooein 2026 (2)

Evidence¶

Donya Rooein 2026¶

Donya Rooein, Luca Benedetto, Dirk Hovy. (2026). The Role of Implicit and Explicit Demographic Signals in Large Language Model-based Student Assessment. https://arxiv.org/abs/2609.16993 doi:10.48550/arxiv.2609.16993

q2 · i? · causal · r2

Multivariate regression on sentiment of QA responses across 200 demographic profiles, with sentiment on a {0,1,2,3,4} scale and σsentiment = 0.07. Education carried positive coefficients for the two Llama models (Exp) and, with a smaller coefficient, Qwen-30B (Imp).

"This indicates that higher education level receives responses with more positive sentiment, which can be problematic especially with Exp: indeed, the dataset contains 4 education lev- els (from‘High school or below’to‘Doctorate or above’) and the coefficient indicates that, on aver- age, there is a difference of 0.3 between the senti- ment of the lowest and higher education level"

Donya Rooein 2026 (2)¶

Donya Rooein, Luca Benedetto, Dirk Hovy. (2026). The Role of Implicit and Explicit Demographic Signals in Large Language Model-based Student Assessment. https://arxiv.org/abs/2609.16993

q2 · i? · causal · r2

Same sentiment regression analysis of the implicit condition: several behavioral attributes exceeded the sentiment standard deviation although absent from explicit-condition results. The authors suggest LLMs pick up topics in these users' prompt histories, but effects are not consistent across models.

"for Imp there are several attributes which have a strong effect on the sentiment of the responses (|µb|≥σ sentiment) even though they do not appear from the analysis onExp:Tech: Smartphone,reli- gion: Christian,hobbies: Use social media."

Discussion¶

Related Claims¶

• Implicit demographic cues produce larger and less predictable effects in open-ended tasks, including longer responses for laptop users and education-related length effects — related
• Explicitly mentioned education and SES cues lead most models to produce less readable responses for higher-education users, showing partial readability adaptation — related