Abstract for: Data-informed parameter estimation in behavioral epidemic models

Behavioral epidemic models, also called coupled behavior-disease models, represent dynamics of infectious diseases incorporating the feedback loop between human risk response and disease spread. In such models, parameter estimation through model calibration is essential for model validation and accurate projections. However, few models attempt to estimate behavior-side parameters from data. To address this, we conduct simulation experiments to test the effect on parameter estimation accuracy of a) the stage of a pandemic, b) model structure used for calibration, and c) the availability of public behavior data. Our findings show that, in early pandemic stages, estimation of behavior parameters is biased, with less accurate predictions despite accurate data, relevant model assumptions, and behavior data availability. Parameters are better estimated after the first peak of the pandemic and with additional data on human behavior. A large volume of data is not useful if the model used for calibration fails to formulate human behavior. Conventional SEIR models that neglect behavioral changes may provide a good fit early in the pandemic but with large errors after the first peak of the disease. Overall, the results provide insights into the challenges of calibration and validation of behavior-disease models.