Causal Analysis of Author Demographics in Academic Peer Review
Research finds papers by minority authors ranked 0.42 points lower, highlighting systemic bias as AI enters review.
A new study from researchers Uttamasha Anjally Oyshi, Gibson Nkhata, and Susan Gauch applies causal inference methodology to a dataset of 530 academic papers to move beyond correlation and quantify the independent impact of author demographics on peer review outcomes. Published on arXiv (ID: 2603.06641), the research simulates the selection process by using publication venue prestige as a proxy for review rank. The findings reveal statistically significant causal disadvantages, with authors from minority racial groups seeing an average treatment effect (ATE) of -0.42 points in ranking, female authors at -0.25, and those affiliated with institutions in the Global South at -0.57.
This work arrives at a critical juncture, as the authors note the growing influence of artificial intelligence in academic assessment. The paper argues that these quantified biases present a clear and present danger: if AI review systems are trained on historically biased data or lack explicit fairness guardrails, they risk automating and scaling these existing inequities. The authors emphasize that their findings underscore a 'pressing necessity for fairness interventions' in both traditional and emerging AI-based review processes to protect scientific meritocracy and credibility.
- Causal analysis of 530 papers shows minority authors' work ranked 0.42 points lower on average.
- Female authors and those from Global South institutions faced penalties of -0.25 and -0.57 points respectively.
- Study warns AI review systems could exacerbate these biases without deliberate fairness interventions.
Why It Matters
As AI begins to automate academic review, this research provides crucial data to prevent scaling historical biases.