The coronavirus pandemic has created massive social upheaval. Beyond the health and economic consequences, the day-to-day behaviors of billions of people have been altered. And, the end is currently not in sight.
Now as much as ever, behavioral scientists have the ability to address important social problems with rigorous research. Whether its research focused on how to create messages that promote the wearing of masks, research focused on combating the loneliness and isolation people feel due to social distancing, or more basic research that seeks to understand how different theories of social behavior hold up in the current moment, the changes this virus has forced on people have created unprecedented opportunities for social scientists who study human behavior.
Given the nature of the pandemic, an important variable within many studies is location. Participants’ location may affect their behavior or responses within a study because not all people are equally exposed to the virus at the same time. However, accounting for people’s location requires a fair amount of flexibility in data collection. Here, we report how researchers can sample people within specific counties of the U.S., including COVID-19 hotspots.
In a recent study, the CloudResearch team in collaboration with researchers at MITRE sampled people within coronavirus hotspots across the U.S. and compared participant responses to areas where the rate of infection was relatively low. To do so, we matched county-level data on infection rates to specific U.S. zip codes. Then, we divided all U.S. counties into three zones. Counties in the top 10% of infection rates were considered to have a high density of infection (i.e. hotspots), counties between the 90th and 20th percentiles were considered to have medium density of infection, and counties in the lowest 10% of infection rates were considered low-density infection (coldspots).
After defining these three zones, we launched a study using our Prime Panels platform. In the study, we recruited an equal number of participants from each zone (using zip codes to target participant responses). We asked people questions about when, where, and how frequently they engaged in health behaviors to deal with the pandemic.
People from the high-density COVID-19 infection rate zone (hotspots) were 14 percentage points more likely to wear masks compared to people from the low-density zone. Other factors that predicted face mask wearing were age, education, income, and political party affiliation (older, more educated, higher income, and Democratic respondents were more likely to wear a mask).
However, all the above factors that predicted face mask wearing were also associated with living in a high-density region. This is because high-density regions tend to be more densely populated, and people living in densely populated regions, such as large cities and surrounding suburbs, are also more likely to be Democrats, to have a liberal orientation, and to have higher education and income levels. This leads to a key question: Are Republicans less likely to wear masks because they live in cold zones? Or are Republicans less likely to wear masks independent of where they live? These questions can be answered using regression, which measures differences between Republicans and Democrats while statistically controlling for COVID-19 infection rate density.
With county-level statistical controls in place, the only predictors of face mask wearing that remained significant were age and COVID-19 infection rate density. All other variables, including population density, political party affiliation, education, and income no longer significantly predicted face mask wearing. This suggests that person-level demographic, political, and achievement variables may be acting largely as proxies for broader cultural-geographical differences.
Importantly, we are not suggesting that these results provide a definitive demonstration that there are no “real” differences in face mask wearing nationwide between political and demographic groups. What we are suggesting is that polls and statistical models that do not directly control for regional infection rate density may suffer from what is referred to as omitted variable bias. The results of such studies are likely to be skewed as a result and will likely overestimate true differences between political groups. This highlights the need for future studies that examine political differences in COVID-19-specific health practices to include county-level infection rate data in their models.
Finally, our results may help to understand why states with previously low infection rates, like Texas and Florida, have experienced a surge in COVID-19 infections. It is likely that the perception of a proximal health threat and a sense of vulnerability, partially causes people in high infection dense zones to adhere more strictly to recommended health practices. To the extent that these perceptions play a causal role in influencing health practices, people living in regions that have had historically low infection rates may not experience COVID as a threat to which they are vulnerable leading to a false sense of security. This may lead people to be less vigilant with regard to health practices and may paradoxically put people in the safest zones at greater risk of future outbreak. While the national conversation tends to focus on current hotspots, monitoring individuals’ perceptions, attitudes and resultant health behaviors in the cold zones can prevent a spike in infection rates and subsequent transition to a hotspot.
Sampling people within specific zip codes may have important implications for behavioral science researchers in public health, consumer behavior, psychology, sociology, and other fields. Currently, most of the microtask platforms common among academic researchers such as Mechanical Turk do not have the capability to sample people within specific zip codes or counties. In addition, because the federal government and other agencies often collect and report health and economic data at the county level, researchers may be able to gather survey data from people within a specific county and then pair the individual-level data with county-level data. Adding county-level data to survey responses may allow researchers to consider many more predictors of behavior and social phenomena than otherwise could be gathered within a single survey session. And, when these methods are combined to understand behavioral change in response to the pandemic, researchers may be able to arrive at more complete and informative conclusions.
If you’d like to learn more about sampling people within specific geographic locations, get in touch with us today! We can discuss the feasibility of your project and explain how our research team can help make your study a success.
Also, check out coverage of this research in the Wall Street Journal (see MASK WEARING). This research was conducted by professors at Lander College, a division of Touro College, the CloudResearch team, and the MITRE Corp.