Even with skill-based qualifications, LLM team assignment remains gender-biased: when two teams are equally valid, models favor Core Development for male personas over Interface Design¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q2 · i2 medium
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 i2 When choosing between Core Development and Interface Design for equally qualified personas, all models significantly favor Core Development for male personas compared to female personas (GPT-4.1 OR=2.53, GPT-5.2 OR=2.81, DeepSeek OR=1.43, p<0.05). → Erfan Entezami 2026 q2 i? Skill qualifications guide assignments strongly: 99.2% of all Skill-based assignments matched one of the two ground-truth teams. → Erfan Entezami 2026 (2)

Evidence¶

Erfan Entezami 2026¶

Erfan Entezami, Andrew Lan, and Madeline Endres. (2026). Generative AI May Reinforce Social Biases in Software Engineering Education. arXiv preprint. https://arxiv.org/abs/2609.28483 doi:10.48550/arxiv.2609.28483

q2 · i2 · causal · r2

Pairwise binary logistic regressions over Skill-based persona assignments (Table 5), restricted to on-pair predictions; the models systematically prefer the gender-stereotypical assignment between two equally appropriate teams, with printed odds ratios OR=2.53 (GPT-4.1), OR=2.81 (GPT-5.2), and OR=1.43 (DeepSeek).

"when choosing between theCore DevelopmentandInterface Designteams, all models significantly favor theCore Development team overInterface Designfor male personas compared to female personas (𝑝< 0.05, GPT 4.1: OR=2.53, GPT-5.2: OR=2.81, DeepSeek: OR=1.43)."

Erfan Entezami 2026 (2)¶

Erfan Entezami, Andrew Lan, and Madeline Endres. (2026). Generative AI May Reinforce Social Biases in Software Engineering Education. arXiv preprint. https://arxiv.org/abs/2609.28483

q2 · i? · causal · r2

Distribution of predicted labels across all models for Skill-based personas (Table 4); the article reports this share descriptively, with no effect size or significance test attached to it.

"the models overwhelmingly assign students to one of these intended teams, with 99.2% of all assignments matching one of the two ground-truth labels."

Discussion¶

Related Claims¶

• In LLM-assisted team formation, class standing strongly drives assignments: seniors are far less likely than freshmen to be assigned to Interface Design relative to Core Development — related
• In LLM-assisted team formation with name-only personas, male students are significantly less likely than female students to be assigned to Interface Design and Quality Assurance teams relative to Core Development — related
• In LLM-assisted team formation, persona nationality systematically shifts assignments, with Nigerian names favored for Quality Assurance and South Korean names for Interface Design relative to Chinese names — related
• Gender disparities in CT education show a very small but statistically significant overall effect favoring males (g = 0.106) — related
• Using demographic variables as predictors risks reinforcing biases embedded in training labels, including self-fulfilling prophecies — related
• Learners perceived male PALs more positively and recalled more with them than with female PALs, reflecting gender stereotypes — related