Characterizing population-level changes in human behavior during the COVID-19 pandemic in the United States

The transmission of communicable diseases in human populations is known to be modulated by behavioral patterns. However, detailed characterizations of how population-level behaviors change over time during multiple disease outbreaks and spatial resolutions are still not widely available. We used data from 431,211 survey responses collected in the United States, between April 2020 and June 2022, to provide a description of how human behaviors fluctuated during the first 2 y of the COVID-19 pandemic. Our analysis suggests that at the national and state levels, people’s adherence to recommendations to avoid contact with others (a preventive behavior) was highest early in the pandemic but gradually—and linearly—decreased over time. Importantly, during periods of intense COVID-19 mortality, adaption to preventive behaviors increased—despite the overall temporal decrease. These spatial-temporal characterizations help improve our understanding of the bidirectional feedback loop between outbreak severity and human behavior. Our findings should benefit both computational modeling teams developing methodologies to predict the dynamics of future epidemics and policymakers designing strategies to mitigate the effects of future disease outbreaks.

Sign up for PNAS alerts.

Get alerts for new articles, or get an alert when an article is cited.

Changes in human behavior at the population level have been known to impact the spread of communicable diseases. It is well understood that frequent and close human contact, such as shaking hands, hugging, or attending crowded events, can increase the risk of transmission of respiratory diseases such as seasonal influenza or the common cold (1). In contrast, regular hand washing and the use of hand sanitizers are also known to significantly reduce pathogen transmission (2). In fact, Semmelweis’s work in the mid 1800s in Vienna, Austria, linking the absence of hand washing to high mortality in maternity wards began a revolution that eventually led to the enshrinement of hand washing as a preventive health behavior in Western societies (3). Historically, behaviors such as the intermittent migration of people from the countryside to densely packed overpopulated cities contributed to the spread of plague in Europe and Asia from the Roman Empire to the 19th century (4). More recently, the rise in the adoption of preventive behaviors such as adhering to stay-at-home recommendations and/or wearing facial masks substantially slowed the spread of infections during the COVID-19 pandemic before vaccines or other treatments were widely available (5–9). Thus, having a clear characterization of how populations behave and how they change their behaviors can in turn help better understand how communicable disease transmission may unfold during an epidemic outbreak.

Human behavior at the population level can change in response to the emergence of major crises such as wars, famines, or pandemic events. Changes in economic activity and weather patterns, as well as local beliefs and political identity, may trigger and/or modulate behavior changes across geographies. In the context of epidemiological outbreaks, people can change their behavior patterns to reduce their risk of infection and potential death in times of high disease transmission (10). For example, during periods of increased mortality caused by the West African Ebola outbreaks of the mid-2010s, people changed their behavior by abandoning the cultural practice of touching the bodies of deceased relatives when they learned that the deadly disease could be transmitted by contact with bodily fluids from corpses (11, 12).

Similarly, during the COVID-19 pandemic, people reduced their visits to crowded places, such as music concerts, movie theaters, and museums, during times when the number of deaths attributable to COVID-19 was high, as will be shown in this study.

In the same way that human behavior changes will influence the temporal evolution of the spread of infection during an outbreak, awareness of the risks that an infection may bring to individuals (such as their potential death) may lead to changes in population-level human behavior, modulated perhaps by local societal practices and/or ideologies. This bidirectional feedback between human behavior patterns and epidemiological dynamics has not consistently been characterized across diseases and outbreaks, in part due to poor epidemiological surveillance and the limited availability (and reliability) of estimates of human behavior indicators.

Accurate and timely epidemiological surveillance is challenging (13). It would be impractical, in terms of economic resources and people’s consent, to test everyone in a population to fully identify the number of infected individuals at all points in time. Instead, the epidemiological community generally estimates the severity of a disease outbreak using various correlates or proxies such as the number of reported infections, hospitalizations, reported deaths, and the amount of viral RNA in wastewater (14–16). Although these proxies are imperfect for fully characterizing the spread of communicable diseases due to limiting factors such as testing availability, test accuracy, reporting delays, heterogeneous ascertainment rates, and underreporting, they can, in some cases, provide us with meaningful estimates of the time evolution of the severity of an ongoing outbreak, especially when multiple proxies are used in conjunction with each other (13). However, in the context of COVID-19, spatially and temporally heterogeneous access to tests, asymptomatic infections, and reporting delays, among other reasons, made it difficult to track the number of infections in multiple locations in a timely manner (17–19).

Characterizing human behavior (e.g., risk-averting or risk-exposing actions) at the population level in the context of disease transmission is also a challenge. Multiple novel data sources have been used to monitor changes in human behavior during an epidemic outbreak. For example, Nsoesie et al. (20) used aerial images to monitor the number of cars in hospital parking lots to estimate the population’s need for medical attention, thus providing an indicator of the incidence of respiratory viruses. Commercial airline traffic data have been used to estimate interregional human movement around the world, thus allowing the estimation of potential risks of importation of pathogens (10, 21–23). Human mobility estimates obtained from aggregated and anonymized location-enabled mobile phones have also been used to estimate changes in human contact patterns that have sometimes been linked to spatial and temporal changes in disease transmission (24–29). However, notable limitations about the use of mobile phones records to quantify human mobility trends have been identified that could lead to ambiguous and sometimes uninformative results (24, 30–32). Although existing survey data have been helpful in characterizing patterns of behavior in the population during the COVID-19 pandemic (33–36), survey approaches are frequently limited due to low sample sizes and inconsistent (and frequently short) time periods of deployment. Additional efforts have also been made to infer human behavior through the Oxford Stringency Index (OSI) (37–40), an index that attempts to quantify the stringency of government mandates to mitigate disease transmission (41, 42).

However, various works highlight the variation in receptiveness and attitudes toward preventive public health policies like mask-wearing and physical distancing (43–45) and the limitations of using a policy-based measure like OSI as a proxy for behavioral adoption. It will be shown in this study that government mandates may not reflect how individuals in a community choose to behave.

Having incomplete and poor characterizations of the interplay (feedback) between changes in human behavior and the evolving dynamics of disease outbreaks frequently leads public health officials and decision-makers to design mitigation strategies in the face of imminent disease outbreaks, based on intuition rather than observed evidence. Furthermore, researchers designing and implementing mathematical models of infectious disease transmission to predict upcoming disease events often do not include this important feedback in their formulations, leading to discrepancies between model predictions and eventually observed epidemic trajectories (46, 47).

In this study, we used data from a large-scale national representative survey conducted in the United States during the COVID-19 pandemic to characterize temporal changes in risk-averting (for example, avoiding contact with others) and risk-exposing behaviors (e.g., going to visit a friend). We analyzed the temporal evolution of 15 behaviors at the state and national levels in the United States, aggregated from survey respondents, and evaluated how their temporal trends changed as COVID-19 mortality fluctuated over the first 2 y of the pandemic, at the state and national levels.

We hypothesized 1) that people’s choice to adopt risk-averting behaviors (e.g., avoiding contact with others) would be highest during the earlier months of the pandemic, when uncertainty about the biology and consequences of COVID-19 infections was highest, and that the adoption of protective behaviors would decrease over time, due perhaps to personal fatigue, the availability of successful treatments and vaccines, or the eventual perception of proportions of the population that their infection risk was low; 2) that people’s perception of risk would increase in times when COVID-19 mortality was high, and that such perception would trigger a higher proportion of people to adopt preventive or risk-averting behaviors (conversely, in times of low mortality, we hypothesized that people would be more prone to engage in risk-exposing behaviors); 3) that changes in risk-exposing behaviors (e.g., going to visit a friend) would be temporally associated with subsequent changes in detected COVID-19 cases, hospitalization, and deaths; 4) and that population-level changes of behaviors due to hypotheses 1, 2, and 3 would vary across states based on differences in political leaning. Finally, we hypothesized 5) that state-level adoption of government-directed policies would demonstrate substantial variation even in states with similar levels of nonpharmaceutical intervention stringency as measured by the OSI (48).