The COVID-19 pandemic has exacerbated political divisions in the United States, with stark contrasts in responses to the crisis. While medical experts advocated for social distancing, protests against restrictions occurred, highlighting the deep social and political cleavages. This study aims to understand the underlying drivers of COVID-19 spread within the general population, beyond individual risk factors. It investigates the statistical drivers of the pandemic using county-level data across the US, considering factors like population density, urban status, travel history, socioeconomic factors, physiological risk factors, and the impact of government shutdowns. The study hypothesizes that differing pandemic experiences, shaped by these factors, have influenced social learning and behavioral responses, contributing to the deepening political polarization.

The study draws upon social learning theory to explain how evolved cognitive mechanisms influence adaptive responses to the pandemic. It reviews existing research on social learning mechanisms like in-group information sharing, imitation, and costly punishment. The literature on the evolution of cooperation, emphasizing mechanisms such as costly punishment, contingent cooperation, external forcing mechanisms, memory, and imitation, informs the understanding of how varied pandemic experiences lead to cooperative or resistant behaviors. The study emphasizes the need to move beyond simply providing information to the public and instead address the different realities and information processing mechanisms inherent in varied pandemic experiences.

The study uses county-level data from various US government sources, including data on COVID-19 deaths, demographics, socioeconomic factors, health indicators (e.g., ICU beds, uninsured rates, comorbidities), and economic factors (occupations affected by shutdowns). The analysis initially included the percent non-Hispanic Black population but was later removed due to its high correlation with other risk factors. Due to the disproportionate impact of the pandemic around New York City, the dataset was partitioned into the NYCSA (31 counties) and the rest of the US (3112 counties). Linear regression models were developed to examine the relationship between per capita COVID-19 death rates and the independent variables. Stepwise regression was employed to address multicollinearity among the independent variables, and coefficients were standardized for comparison. Principal component analysis was also used as an alternative approach to handle multicollinearity. The models were run at various time points throughout the early stages of the pandemic to capture evolving relationships.

The models for the NYCSA counties generally explained around 80% of the variance in per capita deaths. Consistent relationships were observed: a positive association with foreign travel history and negative associations with population density and median male age. For the rest of the U.S., only urbanism and foreign travel history showed consistent relationships with COVID-19 deaths, with weaker explanatory power compared to the NYCSA models. Initially, the primary drivers were logistical factors related to transmission (foreign travel, urbanism, population density). Over time, factors such as median income, median male age, comorbidity, and pollution became more significant in explaining death rates outside the NYCSA, indicating a complex interplay of factors beyond simple transmission dynamics. Principal component analysis yielded four components: Comorbidity (high comorbidity and low income); Density/Travel (population density and foreign travel history); Young/Urban (young male age, urban status, and pollution); and Insured (health insurance). The NYCSA models consistently showed Density/Travel as the strongest covariate, while for the rest of the U.S., Density/Travel and Urban Status were most significant, but with substantially lower explanatory power. As of May 8, 2020, the NYCSA COVID-19 death rate was over 15 times higher than the rest of the country, with a ratio narrowing to approximately 4 by October 8, 2020. A negative correlation was found between per capita deaths and the percentage of the vote for Donald Trump in the 2016 election.

The findings suggest that the COVID-19 pandemic's impact varied significantly across the U.S., with differing levels of uncertainty and signal strength about the threat. The initial focus on logistical factors for transmission highlighted the importance of these factors in the pandemic's spread. However, the varying levels of explained variance between the NYCSA and the rest of the country underscore the complex interaction of several factors affecting mortality rates. The observed differences in experiences are explained through the lens of evolutionary psychology, where the health and subsistence needs of individuals are evaluated according to the environment. Urban dwellers in areas with high transmission rates had strong signals reinforcing the threat and prompting adaptive behaviors like sheltering. Conversely, individuals in rural areas with low transmission rates experienced weaker signals, leading to responses prioritizing economic needs over pandemic restrictions. Social learning mechanisms, particularly imitation and in-group information sharing, were likely critical in shaping these responses. The study highlights the critical role of social learning and cognitive biases in shaping individual responses to the pandemic and the resulting sociopolitical divisions.

This study demonstrates that the COVID-19 pandemic's impact has been shaped by a combination of epidemiological factors, socioeconomic conditions, and evolved cognitive mechanisms. The findings emphasize the need for tailored communication strategies addressing the diverse pandemic experiences across the US. Future research could further investigate the dynamic interplay between social learning, information processing, and behavioral responses to public health crises. The study’s insights could inform future pandemic preparedness and response strategies, emphasizing the importance of tailoring communications to specific communities and mitigating the impact of evolved cognitive biases.

The study's limitations include the reliance on publicly available county-level data, which may not fully capture the complexities of individual experiences. The analysis may not account for all relevant factors influencing COVID-19 spread and mortality, and the cross-sectional nature of the data limits the ability to establish causal relationships. The interpretation of principal components relies on subjective judgment and may not fully capture the nuance of the relationships between variables.