Suresh Subramanian's Dissertation Defense

The success of any large-scale policy intervention from public health campaigns to urban planning and economic policy ultimately depends on how individuals and organizations respond to these programs. This presents a fundamental and persistent challenge in predictive modeling of real-world environments: the need to systematically capture the complex, uncertain, and socio-cultural dimensions of human decision-making. Traditional approaches often treat behavior as static, rule-based, homogeneous input and overlook the dynamic interaction of beliefs, social influences, and personal contexts. Furthermore, these approaches are often constrained to specific scenarios and lack the design flexibility to adapt or extend across diverse problem landscapes. This dissertation addresses this core problem by introducing a unified computational framework to model behavior with greater granularity and socio-cultural nuance across diverse problem landscapes. To assess the generalizability and utility of the unified framework we applied it to study diverse real-world problems in the particularly challenging health policy domain characterized by complex socio-cultural dynamics, diverse stakeholders and limited data.

The core of our methodology is a generalizable architecture that addresses two central aspects: complex socio-behavioral drivers of human decision-making and the specific systemic dynamics of the health environment. The proposed framework provides a structured approach to translate both quantitative data and qualitative information including abstract social theories into quantifiable, computable representations of beliefs and attitudes, while systematically handling data uncertainty and incompleteness. The behavioral component of our framework is designed to interact with established health system models (e.g., compartmental epidemic models) through novel feedback mechanisms, allowing for the co-evolution of social dynamics and health outcomes.

We demonstrate the framework's validity and utility through two distinct case studies. First, we use the framework for a micro-meso level analysis of physician adoption of Type 2 Diabetes guidelines. Here, our behavior-infused model not only significantly outperforms baseline models in predicting adoption rates across various physician demographics but also provides richer explanatory insight into the observed adoption patterns. The second case study focuses on a macro-level analysis of public behavior during the 2009 H1N1 and 2020 COVID-19 epidemics. In this case study we demonstrate the framework's ability to create a co-evolutionary feedback loop between a behavioral model and a standard epidemiological (SEIR) model. By making the disease transmission rate a dynamic function of the population's evolving risk perception, the integrated model produced significantly more accurate infection trajectory forecasts than the baseline epidemic models.
This dissertation's primary contribution is a unified, generalizable framework for combining complex social theories and a variety of data sources into quantitative, interpretable, and predictive computational models. By demonstrating the systematic process for developing and integrating behavior-driven models with established health system models, our research offers a structured approach for building more realistic and analytically powerful models of human decision-making in complex real-world contexts.

Question? Contact Suresh Subramanian.