Previous studies on the COVID-19 pandemic has highlighted the importance of behavioral data sources, including human mobility data, in informing infectious disease modeling and forecasting efforts. Human mobility data, which tracks population movement patterns, serves as a pivotal tool in understanding virus transmission dynamics and guiding public health responses. Yet, its integration into predictive and modeling tasks brings forth challenges of potential bias and inequity, especially towards marginalized communities. This dissertation delves into an in-depth examination of how different types of data, particularly mobility data, can influence the accuracy and fairness of COVID-19 forecasting and modeling tasks.

The first study systematically analyzes county-level COVID-19 case predictions in the United States, assessing the value added by mobility data. Findings indicate a modest improvement in prediction accuracy with median correlation improvements of approximately 0.13, albeit with diminished gains for historically marginalized populations. This suggests an underrepresentation in mobility datasets, highlighting the need for equitable data integration strategies.

In the second study, an audit of the CDC’s Forecast Hub reveals significant disparities in error rates affecting minority and rural communities, pointing to systemic biases in COVID-19 forecasting models. The investigation also underscores the lack of impact from mobility data in mitigating these discrepancies, stressing the imperative for algorithmic accountability and fairness in public health modeling.

The third study, which constitutes future work, aims to explore the relationship between mobility data and the economic resilience of small urban businesses during the pandemic. Preliminary analyses indicate a clear association between mobility restrictions and business visitation patterns, and also a disparate recovery pattern particularly in areas predominantly inhabited by Asian and American Indian communities. The proposed research seeks to extend these findings by examining if and how mobility data correlates with or predicts business activity, offering a more comprehensive measure of resilience beyond mere visitation volumes.

Collectively, these studies advocate for a critical evaluation of the use of human mobility and contextual data in COVID-19 forecasting and modeling. By highlighting the potential for both enhanced predictive accuracy and the perpetuation of inequities, this dissertation calls for the development of models that are not only accurate but also equitable. Emphasizing the need for inclusive data representation and rigorous algorithmic audits, the research presented aims to pave the way for more ethical applications of data science in public health and beyond.