Causal Inference Networks: Unraveling the Complex Relationships Between Curriculum Complexity, Student Characteristics, and Performance in Higher Education

Numerous research studies have explored the influence of curriculum complexity on student performance, primarily focusing on factors like retention and graduation rates. Many of these investigations have employed conventional machine learning and data analysis methods, often yielding results that are challenging to interpret and convey effectively. Furthermore, these studies have generally lacked a comprehensive framework for elucidating how variables such as student gender and prior academic preparation contribute to selecting specific university programs, each characterized by its structural complexity. These studies have yet to present foundational models to elucidate the fundamental mechanisms underlying the causal relationship between the complexity of university programs, student attributes, and success metrics. Our present study introduces an innovative causal inference network model that conceptualizes the university as a dynamic system with interrelated causal relationships among its various components, encompassing students, programs, colleges, graduation rates, and more, each with their respective dependencies. This model allows us to comprehend and visually represent the direction of causality between different variables, enabling us to investigate how changes in one variable, the causal factor, impact another variable. This implementation of causality not only facilitates predictive tasks, like other conventional machine learning models (i.e., hypothetical causation), but also enables us to conduct objective “what-if” analyses (i.e., counterfactual causation) within the research context. In this study, we leverage real-world student data from 30 universities across the United States. The richness and diversity of our dataset empower us to draw robust insights into the causal relationships among various factors that influence student performance, particularly the complexity of the curriculum. A key finding from our causal analysis indicates that an increase in program complexity by 20 points is correlated with a decrease of 3. 74% in the likelihood of graduating within four years. Moreover, our counterfactual scenarios demonstrate that for students with specific demographic profiles, such as males with a certain HSGPA not receiving Pell Grants, an increase in complexity could inversely affect their graduation prospects. These nuanced discoveries underscore the importance of curriculum design in alignment with student demographics and preparation, challenging educators to balance academic rigor with the facilitation of student success. The breadth and scale of our dataset significantly enrich the quality of our conclusions, providing valuable guidance for future educational strategies and policies.