Addressing Bias in Predictive Modeling in Education

This technical report discusses the implications of bias in predictive modeling, particularly in the context of higher education. It outlines how Civitas Learning employs predictive modeling to aid institutions in enhancing student success initiatives while addressing concerns of bias and equity. The report defines predictive modeling and explains its reliance on algorithms that analyze data to predict future behaviors. It identifies potential pitfalls in predictive modeling, including data nonstationarity, poor model training, and malicious intent. Civitas Learning's approach to mitigating bias is detailed, emphasizing the use of influenceable derived variables that can be impacted by institutional interventions. The report also describes the importance of transparency in variable ranking and the necessity of using a broad range of input variables to strengthen model accuracy. Additionally, it provides recommendations for institutions on protecting their predictive models from biased misuse, focusing on appropriate use cases, controlled access to data, and professional development for staff.