*3.3. Control Variables*

We also employed a host of control variables in our analysis to control for many plausible alternative explanations for perception of health risks from COVID-19. First, it is possible that people who experience higher levels of anxiety on a daily basis perceive higher levels of risk of becoming infected or dying from COVID-19. We took advantage of a generalized measure of anxiety in the UAS survey to control for anxiety through a measure that indicates how many days the participant had felt anxious in the past two weeks, ranging from 0 (not at all) to nearly every day (3).

Second, given prior literature on discrimination and stigma on the health of minority communities, we created a discrimination index from felt discrimination related to COVID-19. Participants were asked whether: (1) people had acted afraid of them, (2) they had received poorer service, (3) had been threatened or harassed, or (4) treated with less courtesy and respect due to others suspecting they had COVID-19. After each of these questions were recoded to become dichotomous variables (0 = no or unsure; 1 = yes), we compiled an index by adding up the total score across all four questions.

Third, given the importance of minority languages as barriers to public health services and health literacy, we included a measure of language in our models [46,47]. The survey was available for participants to complete in either English (0) or Spanish (1), so we include a dichotomous variable that captures if a participant took the survey in Spanish or not to reflect their level of comfort with completing the survey in English. Based on literature showing that infectious disease outbreaks can create disproportionate adverse effects for linguistic minorities, we accounted for Spanish as a potential limitation to properly accessing health services [46]. Our expectation was that those completing the survey in Spanish were more likely to have greater difficulty accessing public health services in the US due to many of these services not being offered in Spanish.

Fourth, it is likely that having health insurance would shape people's perceptions of health risk during the pandemic, especially in our models relating to the risk of dying from COVID-19 if they were to contract the virus. We expected that participants with health insurance were less likely to be worried about the health risks of COVID-19, so we included a dichotomous measure of having health insurance in our models.

Fifth, race could also play a role in perceptions of health risk during the COVID-19 pandemic. As a result, we also included a host of dichotomous control variables capturing whether a participant identifies as White, African American, Native American, Asian, or Hawaiian or other Pacific Islander.

Finally, we included variables for household income, whether an individual is disabled, their level of education, whether they are currently employed, their gender, and their age. Table 1 presents the operationalization, coding scheme, and descriptive statistics for all variables used in this analysis.


**Table 1.** Descriptive Statistics.
