Abstract / Summary
Background: The COVID-19 pandemic highlighted the importance of incorporating human behavior into infectious disease models. Doing so requires predicting how individuals will respond to outbreaks and policies. Key questions that limit the integration of dynamic behavior into disease models are: Is behavior change driven by policy, objective disease information, or subjective risk perception? Can disease data approximate risk perception? Is mitigation behavior influenced by information at broad or fine spatial scales? Methods: In this ecological study, we leverage US survey data (4.3 million responses from September 2020 through January 2021) on social distancing and concern about infection, alongside data on COVID-19 cases and mitigation policies. Using a county-level spatiotemporal regression model, we evaluate whether social distancing is associated with policy, social information, or epidemiological information at broad or fine spatial scales. We use transmission models to demonstrate the epidemic consequences of different behavioral drivers. Results: We find that epidemiological and social information are predictive of mitigation behavior: a one standard deviation increase in epidemiological or social information led to a decrease in non-household contacts of 0.10 or 0.47 people per day, respectively. State-level variables are more predictive of social distancing than county-level variables, especially for policies. However, conditions in socially adjacent counties are better predictors of behavior than those in spatially adjacent counties. In our transmission model, epidemiologically and socially informed contact patterns yield similar epidemic dynamics, though broader spatial scales of information explain more behavioral variation. Conclusions: The US population modified its behavior during the COVID-19 pandemic in response to case incidence and policy at broader spatial scales, potentially reflecting the lack of localized surveillance or mandates. Modeling assumptions of behavioral responses to disease incidence at larger spatial scales are reasonable.