Abstract
The COVID-19 pandemic dramatically reshaped the labor market, especially for service sector workers. Frontline service sector workers, already coping with precarious working conditions, faced proximate risks of COVID-19 transmission on the job and navigated new workplace safety measures, including masking, social distancing, and staying home while sick, all in a polarized political environment. We examine polarization in the effects of COVID-19 workplace safety measures on workers’ feelings of safety and well-being. Specifically, we examine how support for former President Trump moderates the relationship between COVID-19 safety practices (masking, social distancing, staying home while sick) and workers’ feelings of safety and well-being. To do so, we draw on novel data collected by The Shift Project from 2,039 service sector workers at 89 large firms during the COVID-19 pandemic. We find that workplace safety measures are positively associated with workers’ self-assessments of feeling safe and with mental health, but only for Biden voters.
In an era of intense political polarization (Pew Research Center, 2019), political attitudes increasingly shape how people view the world, including issues that on the surface appear to be non-political. Notably, politics strongly influences individuals’ views of science and their acceptance of scientific knowledge and values (Evans and Hargittai, 2020; Gauchat, 2012, 2015; Kozlowski, 2022; Lee, 2021). In issues ranging from climate change (Dunlap et al., 2016; Hornsey et al., 2016) to routine vaccination (Baum, 2011; Jones-Jang Mo and Noland, 2022), scholars have documented that political ideology and partisan affiliation are associated with polarized attitudes about science. Science is not just a set of beliefs. Acceptance of scientific knowledge can shape individuals’ behavior in the world and can moderate how events in the world shape individuals’ well-being, including at work, a key social context where politics, world events, behavior, and well-being intersect. However, little existing research has moved beyond the role of politics in shaping beliefs about scientific issues to understand how the politicization and polarization of scientific knowledge influences individual well-being. Existing studies most often operationalize politics as political ideology or partisan affiliation, but rarely isolate the specific effect of supporting an individual political leader. Yet the contemporary period is not just ideologically polarized, but is driven by loyalty to particular personalities. Furthermore, the personal consequences of abstract scientific issues like climate change or of diseases that are now uncommonly encountered (thanks to vaccination) can be quite distal for many Americans. How, then, does this polarization influence well-being in contexts where the risks and consequences are much more proximate?
The COVID-19 pandemic is one such example of a real risk that is, to varying degrees, quite proximate to the entire population. Since early 2020, the novel coronavirus has posed serious health risks to nearly all individuals in the U.S. and around the world. Despite the shared experience of living through a “once-in-a-lifetime” pandemic, the American public holds sharply diverging views on the severity of the virus and the need for pandemic mitigation strategies, such as wearing masks and social distancing (Deane et al., 2021). This is due, in part, to the politicization of the issue by former President Donald Trump, who downplayed the severity of the virus and questioned the need for a coordinated government response. As a result, politics has driven non-adherence to pandemic mitigation strategies, making the COVID-19 pandemic a substantively relevant site to investigate the dynamics of polarized views of science (Allcott et al., 2020; Shepherd et al., 2020). The pandemic is also a strategic site for understanding polarization in the context of a proximate and immediate health risk, particularly at work and particularly for frontline workers.
Over the past year, a small but robust literature has identified a clear relationship between politics (including political ideology, party affiliation, and voting behavior) and COVID-19 risk perceptions and preventative behaviors. Partisan differences exist in the propensity for individuals to voluntarily engage in measures to prevent COVID-19 (Allcott et al., 2020; Barrios and Hochberg, 2020) and in individual trust in scientists to understand COVID (Evans and Hargittai, 2020). However, despite the outsized role former President Trump played in politicizing COVID-19, most studies operationalize politics as either political ideology or partisanship rather than support of President Trump specifically. Limited evidence suggests that Trump approval is associated with lower perceived risk of COVID-19 (Shepherd et al., 2020) and greater intentions to defy social distancing measures (Graham et al., 2020). However, this work focuses on either risk perception or adherence to pandemic mitigation strategies, rather than on how these politically polarized beliefs about COVID-19 shape individual well-being. Finally, risks in the population are heterogenous, which makes it difficult to separate the role of politics from actual differences in risk. While everyone is at some risk of contracting and spreading COVID-19, some groups are systematically at higher risk. Much of the extant research speaks to the general U.S. adult population, overlooking the heightened risk of groups like frontline workers. Finally, many of these studies were conducted at the start of the pandemic, where scientific understanding of the virus was fluid, policies to combat it were constantly evolving, and partisan lines perhaps less hardened. Our study centers frontline service sector workers, a social group more proximate to the risk of COVID-19 compared to the general population, and builds upon existing scholarship by examining well-being as our outcome of interest and by focusing on the role of political leaders.
Focusing on frontline service sector workers, we examine polarization in the effects of COVID-19 workplace safety measures on individual workers’ feelings of safety and well-being. Specifically, we examine how the relationship between co-worker masking, customer masking, ability to socially distance, and ability to stay home while sick and workers’ feelings of safety and well-being is moderated by support for former President Donald Trump.
To do so, we draw on novel individual-level data from 2,039 frontline service sector workers employed at 89 of the largest service sector firms in the U.S. collected between September and November of 2020 during the COVID-19 pandemic. This population was unable to engage in remote work and therefore faced heightened risks of COVID-19 infection, interacting indoors with customers and in close proximity to co-workers (Ravenelle et al., 2021; Wolfe et al., 2021). In addition, they were not granted additional COVID-19 paid leave under the Families First Coronavirus Response Act because all worked at large exempted employers, and they generally earned low wages, making it extremely difficult if not impossible to voluntarily leave their jobs. These data give us unique leverage to test open questions about polarization and science in a high-stakes scenario. We are able to examine the dynamics of polarization around political leaders for a group of workers at high objective risk and gauge the consequences not for their beliefs, but for their well-being.
We find that politics is significantly associated with the individual well-being of frontline workers during the COVID-19 pandemic. When frontline workers report high compliance with workplace safety policies (co-worker masking, customer masking, social distancing, and ability to stay home when sick), both Trump and Biden voters experience similarly high levels of safety and happiness and low levels of psychological distress. However, when there is more limited adherence to COVID-19 mitigation policies in the workplace, Biden voters report feeling much less safe and happy, while Trump voters’ feelings of safety and happiness are largely insensitive to these precautions. Seemingly scientifically objective safety standards have entirely different consequences for frontline workers who voted for Trump as opposed to those who voted for Biden. We conclude that polarization shapes the meaning and the consequences of non-wage job amenities for workers during COVID-19.
Background
The Politicization of Science
Over the last several decades, science, scientists, and scientific knowledge and values have become increasingly polarized (Evans and Hargittai, 2020; Gauchat, 2012, 2015; Lee, 2021). Issues that once seemed non-political are now highly partisan. In general, conservatives have become less trusting of science over time, while liberals’ trust in science has remained relatively constant (Gauchat, 2012) or even grown (Lee, 2021), While most of this research focuses on the politicization of science among ideological conservatives specifically, Lee (2021) extends this work by analyzing how changes among both Democrats and Republicans contribute to the growing partisan gap in trust in science. He finds that “party polarization in trust increased between the 1970s and 2010s because Democrats became more likely to trust science while Republicans became less likely to trust science” (Lee, 2021:2). Importantly, Kozlowski (2022) shows how conservatives’ declining confidence in the scientific community was mostly due to growing distrust in scientists among moral rather than economic conservatives. This work highlights how conservatives and liberals are not homogenous, but instead are made up of political actors with heterogenous views and interests.
Conservatives’ distrust in science holds true across multiple contexts, from more general views on “science” to more specific (yet still distal) issues such as science policy and climate change (Ballew et al., 2019). A robust literature has shown that U.S. conservatives are more likely to be skeptical of and dismiss climate change (Ballew et al., 2019; Dunlap et al., 2016). Scholarship has also documented how partisan affiliation influences attitudes toward diseases and their prevention measures (e.g., vaccines) (Baum, 2011; Jones-Jang Mo and Noland, 2022; Suryadevara et al., 2019). For example, Suryadevara et al. (2019) examine the specific case of HPV vaccination rates, finding that Republican-leaning states had significantly lower levels of HPV vaccination compared to Democrat-leaning states, although both had similar levels of Tdap and meningococcal vaccination rates.
While much of this literature focuses on ideological differences (i.e., conservative vs. liberal views), some scholars focus instead on partisan affiliation (i.e., Republican vs. Democrat). In a meta-analysis of 171 academic studies across 56 nations, Hornsey et al. (2016) conclude that the most important predictor of climate change beliefs is partisan affiliation, above and beyond political ideology. In other words, whether one accepts or rejects climate change is more closely related to an individual's identification with a specific political party than to how conservative or liberal they are. Some studies go further, examining the dual role of ideology and partisanship in shaping beliefs about climate change. According to a recent study by Ballew et al. (2019), around 97% of liberal Democrats believe that global warming is happening, compared to 40% of conservative Republicans, while 90% of liberal Democrats are at least somewhat worried about it, compared to only 23% of conservative Republicans.
This outpouring of research on polarization and science has significantly expanded our understanding of how politics have come to increasingly shape trust and belief in science, especially in the domains of vaccines and climate change. We describe three key open questions in this literature below and argue that to fully understand the nature and ramifications of the polarization of science, research must look beyond ideology and even party to consider the role of political leaders in driving polarization and must look beyond beliefs to the consequences of polarization for well-being, especially for individuals at substantial risk of the underlying events. We then investigate the case of frontline service-sector workers during COVID-19 as a strategic site for advancing this literature.
The Influence of Political Leaders
The existing literature on the politicization of science described above generally operationalizes politics primarily as an individual's political ideology and/or partisan affiliation. Doing so, however, overlooks the potentially important role of individual political leaders in shaping public opinion. Indeed, limited research suggests that one mechanism by which politics shapes attitudes about science is the transmission and receipt of cues from political leaders (Jones-Jang Mo and Noland, 2022; Kousser and Tranter, 2018). Voters’ beliefs about policy issues do not form in isolation, but rather are influenced by many external factors, including political leaders. Jones-Jang Mo and Noland (2022) note that “[o]nce citizens perceive controversial science issues as politically controversial matters, they are likely to display partisan-based biases and information behavior.” Kousser and Tranter (2018) study the influence of political leaders on their constituents’ views of climate change in Australia, where (as in the U.S.) conservative politicians are less likely than progressives to believe in climate change caused by humans. The authors evaluate whether voters follow political leaders’ views of climate change, using the Australian Survey of Social Attitudes. They find that voters follow their party leaders’ views on climate change, concluding that national political leaders influence partisan attitudes.
Similarly, political leaders play a key role in influencing the public's beliefs about vaccines. Examining the role of political partisanship in shaping attitudes toward H1N1 (swine flu) and its vaccination program, Baum (2011) argues that the polarization of the mass media starting in the 1990s allows individuals to limit their exposure to attitudes that reinforce rather than challenge preexisting beliefs. The author suggests that the anti-vaccination attitudes among Republican partisans were in part a response to trusted (conservative) opinion leaders who were voicing skepticism. Jones-Jang Mo and Noland (2022) investigate the role of political cues in shaping beliefs about the link between vaccines and autism. Using two experiments, the authors find that political cues (rather than scientific expertise) shape individuals’ beliefs in the vaccine and autism debate. Specifically, based on an experimental manipulation that presented participants with political and/or scientific cues, Republicans tended to follow former President Trump rather than scientists and subject-matter experts, while Democrats follow scientists and not Trump. President Trump has long been considered a distinctly polarizing figure, but the authors’ findings suggest that support for Trump ultimately affects partisan's views about science, too. While these studies are an important first step in establishing the role of political leaders in shaping the politicization of science, the analyses typically lack individual-level measures of voting behavior in the context of the highly salient, ongoing COVID-19 pandemic.
Beyond Beliefs: Politics and Well-Being
Beyond the way politics is operationalized as a predictor of beliefs about science, the specific outcomes employed in existing studies likewise have limitations. Most existing research on the topic focuses on attitudes and beliefs about science. Scholarship has examined how politics affects beliefs about science (Gauchat, 2012; Lee, 2021), attitudes about scientific funding and policy (Gauchat, 2015), belief in the existence of climate change (Hornsey et al., 2016), assessments of the risk climate change poses (van der Linden 2015), as well as awareness of vaccines and belief in vaccine misinformation (Jones-Jang Mo and Noland, 2022).
However, notably absent from this body of research is an investigation into the effects of holding politicized views about science on personal well-being. The polarization of science is not merely a set of beliefs that exist in the abstract. Rather, holding such sharply diverging views on scientific reality necessarily leads to different decisions and different actions, which might have significant consequences for individual well-being. For example, individuals with higher levels of concern about climate change are more likely to engage in behavior change, such as recycling and reducing energy use (Semenza et al., 2008). Our understanding of how (or whether) the polarization of scientific belief affects individual decisions and well-being remains limited.
Polarization and Distal vs. Proximate Risks
Finally, the extant literature on the polarization of scientific beliefs primarily involves abstract issues. “Science” is a broad and multidimensional concept. For example, while one may have strong beliefs about climate change that are informed by political worldviews, the actual risks and consequences of climate change are distant for most people in North America. Indeed, although almost half of Americans view climate change as any sort of personal risk (Ballew et al., 2019), the timing of the threat is less imminent compared to the risk of a virulent infectious disease under conditions of high community spread.
Scholars of environmental risk have documented that despite fairly widespread concern about climate change, most people consider it to be of secondary importance in their daily lives, especially compared to more pressing and immediate issues (Fagan and Huang, 2019). However, direct experience with the risks of climate change is typically associated with higher levels of risk perception, as Adelekan and Asiyanbi (2016) find with flood victims in Nigeria. If proximity to risk affects individuals’ risk perceptions and beliefs about disasters and climate change, does polarization lessen in the context of proximate risk, such as that faced by frontline workers during the pandemic?
To summarize, despite this extensive body of work on political differences in beliefs about science, notable gaps remain. First, it remains unclear what element of politics, specifically, influences views on science. Some studies emphasize political ideology and/or partisan affiliation, but few isolate the effect of supporting individual political leaders (but, see Jones-Jang Mo and Noland (2022)). Second, the extant literature mostly examines attitudes, beliefs, and behaviors (e.g., vaccinations) related to science and scientific issues. However, very little focuses on the effect that holding those beliefs has on individual well-being. Finally, the existing research on the politicization of science mostly involves abstract, distal risks whose consequences individuals are removed from in their daily lives. Thus, it remains unclear whether the political polarization documented to date functions similarly when the population faces immediate personal risk.
The Case of COVID-19
The growing polarization of science in the U.S. is particularly pertinent to the ongoing COVID-19 pandemic. Although the coronavirus and its associated prevention measures are fundamentally scientific issues, based on decades of research in medicine and public health, the debates surrounding the pandemic are highly political. For example, intense polarization is evident in the sharp political divides on vaccinations and mask mandates throughout the country (Hamel et al., 2021). Recent scholarship suggests that there is a clear relationship between politics and COVID-19 risk perceptions and preventative behaviors. In general, Democrats are more likely than Republicans to be worried about the risks COVID poses and to engage in preventative behaviors (Allcott et al., 2020; Barrios and Hochberg, 2020; Gadarian et al., 2021). Much of this research takes place at the neighborhood or county level, using aggregate measures of political support and COVID-related beliefs and behaviors. Specifically, areas with more Republicans are less likely to engage in social distancing behaviors, measured at the aggregate level (Allcott et al., 2020), while counties with a higher share of Trump voters are less likely to engage in COVID-related Google queries (Barrios and Hochberg, 2020), suggesting lower risk perception. Measuring political polarization at the aggregate level precludes conclusions about individual-level dynamics in the polarization of COVID-19, as doing so risks an ecological fallacy. The limited scholarship to date that uses individual-level data has found that Democrats are more likely than Republicans to engage in social distancing efforts (e.g., limit socializing with members outside of their household) (Clinton et al., 2021) and to be more worried about the pandemic (Gadarian et al., 2021).
Because COVID-19 and the societal responses to it are novel, there were no pre-existing political fault lines for citizens to base their views on, so when the virus first emerged it was not immediately clear what partisans “should” think based on their party's political platform. However, despite the lack of pre-existing politicization, the pandemic has become highly politicized. Notably, former President Trump and other Republican officials consistently downplayed the severity of the pandemic and lambasted efforts to curtail the virus as assaults on individual freedom. As such, the COVID-19 pandemic is a strategic site to investigate the role of political leaders in shaping the politicization of science.
A limited number of studies specifically identify the role of Trump support in shaping COVID-related beliefs and behaviors. Using individual-level data from early in the pandemic, Shepherd et al. (2020) find that Trump approval specifically is significantly associated with low COVID-19 risk perceptions, above and beyond political party affiliation, while Graham et al. (2020) find that greater faith in Trump is a strong predictor of defying social distancing measures. However, this research measures approval of (Shepherd et al., 2020) or faith in (Graham et al., 2020) Trump, rather than voting behavior. Gadarian et al. (2021) employ intended 2020 presidential vote as one measure of partisanship (along with party affiliation and ideology), documenting political differences in health behaviors, health attitudes, and pandemic-related policy preferences. However, their data were collected in March 2020, in the early stages of the pandemic in the U.S. before partisan divides were crystalized. Taken together, these studies suggest that political leaders, and Trump specifically, may play an important role in shaping beliefs about COVID-19, but further research with more precise measures of voting behavior and data collected later in the pandemic is necessary.
Mirroring some of the limitations of earlier research on climate change and vaccination, the COVID-19 literature typically focuses on attitudes surrounding the COVID-19 pandemic and associated behaviors (e.g., compliance with social distancing guidelines), but not on the effect of holding polarized views on individual well-being. For example, Shepherd et al. (2020) examine the effect of political worldviews on health beliefs and practices during the pandemic, but missing from their analyses is a focus on how political differences in those beliefs is associated with individual well-being. Similarly, most existing individual-level research on the polarization of COVID-19 examines risk perceptions and attitudes about the pandemic (e.g., Gadarian et al., 2021) or compliance with prevention measures (e.g., Clinton et al., 2021), but does not account for the effect of polarized risk perception on well-being.
Finally, most studies on the polarization of COVID-19 sample a general population of U.S. adults, but doing so ignores the potential effect of heightened proximity to risk. While the risks associated with COVID-19 are more proximate to daily life for most people compared to issues like climate change, certain characteristics increase one's risk for contracting COVID or facing serious complications. One social group that is particularly proximate to the risks of COVID is frontline workers in the service and retail sectors. Many of the most controversial and highly politicized policy efforts (e.g., wearing masks and social distancing in stores and restaurants) directly affect these workers’ health and safety, as they bear the consequences of whether these policies are implemented by officials and followed by customers. We might expect that when the risks are this proximate, polarization may play a more muted role.
However, little research has directly examined the well-being of frontline workers during the pandemic. Ravenelle et al. (2021) note that pandemic mitigation measures are typically less feasible for precarious workers, who are more likely to use public transit and less likely to be able to effectively socially distance in the workplace. More generally, higher income is associated with larger changes in self-protective behaviors, which is partially explained by the fact that lower income individuals (many of whom are frontline and essential workers) are more likely to report circumstances where adopting self-protective behaviors is more difficult (e.g., inability to work remotely) (Wolfe et al., 2021). Frontline workers, therefore, are more proximate to the risk of contracting COVID compared to the general population, thus representing a theoretically interesting site to investigate the dynamics of the political polarization of COVID-19.
This paper addresses both empirical and theoretical gaps in the literature on the polarization of COVID-19. Specifically, we use mid-pandemic individual-level data, measure polarization in terms of self-reported votes for Trump or Biden (adjusting for ideology and partisanship), and examine well-being outcomes among workers at high risk in their workplaces. Building on published work that examined the early pandemic period (e.g., March or April 2020) when guidance on prevention behaviors was still emerging and unsettled, we examine the period from September to November of 2020 during which political divides on issues like mask mandates sharpened. We address the question: Among frontline workers, does support for Trump or Biden have consequences for individual well-being when faced with heightened risk of COVID exposure in the workplace?
Data and Methods
The Shift Project Data
We draw on novel survey data from The Shift Project, which has collected data on almost 200,000 frontline workers at 140 large firms in the service sector (retail, pharmacy, grocery, hardware, electronics, general merchandise, fast food, casual dining, delivery and fulfillment, hotel). Since 2017, the project has collected ten rounds of cross-sectional data bi-annually, the most recent of which was collected in June 2021. In this paper, we use data from the ninth cross-section, collected between September and November 2020 (N = 2,039).
The Shift Project uses targeted advertisements on Facebook and Instagram to sample and recruit workers at some of the largest retail and food service employers in the U.S. Facebook and Instagram use is widespread, with over 80% of the working-age population having a Facebook or Instagram profile (Greenwood et al., 2016) and 80% of users reporting activity on one of the platforms at least once a day (authors’ calculation from 2018 Pew Research Center survey). The Facebook advertising platform contains detailed information on each user, including their employer. The Shift Project leverages this information, treating the universe of Facebook users in the US with employer information as the sampling frame, which uniquely allows for the targeting of service-sector workers at particular firms. The advertising platform then also serves as the delivery device for survey recruitment. The Shift Project purchases paid advertisements that appear in users’ newsfeeds and include 1) a stock photograph of a worker in a setting similar to their workplace and 2) the text “Working at [employer name]? Take a Short Survey and Tell us About Your Job!” and “Chance to win a $500 gift card.” Once a user clicks on this advertisement, they are redirected to Qualtrics and asked to complete a 20-min online survey.
Because respondents are recruited through advertisements on Facebook and Instagram, The Shift Project sample is a not a probability sample. Accordingly, we are cognizant of the potential for bias from both non-coverage and selection into our sample. As noted above, most American adults are active on Facebook or Instagram, reducing concerns about non-coverage, though by no means obviating them (if 80% of the target population is on these platforms and 80% of them are active, then we potentially only reach 64% of the target population). Further, selection into taking the survey based on observed and unobserved characteristics is likely. To mitigate the potential selection bias based on observed characteristics, we construct and apply post-stratification weights to the data, weighting our sample to the demographic characteristics (i.e., age, gender, and race/ethnicity) of individuals in the American Community Survey (ACS) who work in the same industries and occupations as the Shift sample. While weighting the data fails to account for selection based on unobserved characteristics, previous work also suggests that the Shift sample is broadly reflective of its target population of hourly workers in the service sector. Schneider and Harknett (2022) conducted extensive data validity checks using The Shift Project data, demonstrating that the Shift sample is aligned with other “gold standard” probability samples, such as the Current Population Survey (CPS) and the National Longitudinal Survey of Youth (NLSY97), on both univariate distributions and bivariate relationships.
The core Shift Project survey collects information on a broad range of topics, including respondents’ demographic characteristics, employment conditions, economic circumstances, receipt of benefits, and health and well-being. The Shift Project data have several important strengths that we leverage in this paper, including its focus on frontline workers, its detailed measures of job quality, its linking of employers and employees (allowing us to estimate employer fixed-effects models), and the flexibility of the survey itself. This flexibility was of particular importance during the COVID-19 pandemic. In the ninth round of the Shift survey, which was fielded in Fall 2020 (between September and November 2020), the survey was adapted to include detailed measures of workplace safety practices, prior COVID-19 infection, and political orientation. We focus on these uniquely detailed data, which we have for a final analytic sample of 2,039 workers with complete information on all covariates. In robustness checks, we test the sensitivity of our results to the imputation of missing data for our independent variables.
Key Variables
Health and Well-Being. We focus on three aspects of health and well-being as dependent variables in our analysis. First, we measure workers’ feeling of safety at work. We asked respondents “Currently, do you feel safe at your [EMPLOYER NAME] workplace during the COVID-19 pandemic?” The response options were “Yes” or “No.”
Second, we asked respondents about their psychological health using the Kessler-6 index of non-specific psychological distress: “During the past month, how often did you feel…” 1) “…so sad that nothing could cheer you up?”; 2) “…nervous?”; 3) “…restless?”; 4) “…hopeless?”; 5) “…that everything was an effort?”; 6) “…worthless?”. The response options included “All of the time”, “Most of the time”, “Some of the time”, “A little of the time”, and “None of the time”. We assigned each response option a value ranging from 0 (“None of the time”) to 4 (“All of the time”). Higher values for each variable indicate greater psychological distress. We then created an index of the six questions, with values ranging from 0 to 24. Finally, we use a threshold value on the index to dichotomize the psychological distress measure (Prochaska et al., 2012). For our dichotomous measure, scores of 12 or greater (out of 24) on the distress scale were coded as experiencing distress.
Finally, we asked respondents about their happiness: “Taken all together, how would you say things are these days? Would you say you are…?” with response options of “Very happy”, “Pretty happy”, or “Not too happy”. We collapsed “Very happy” and “Pretty happy” into a single category to indicate higher levels of happiness. In robustness checks, we test the sensitivity of our results to using continuous rather than dichotomized measures of well-being.
Workplace Safety Practices. Our key independent variables capture COVID-19-related safety practices in the workplace. We asked respondents the frequency with which their workplaces engaged in four safety practices to mitigate the spread of COVID-19: 1) how often their co-workers (including managers) wear a mask; 2) how often their customers wear a mask; 3) how often they can practice social distancing in their workplace; and 4) how often their employer encourages employees to stay home when sick. The options for each question were “Always”, “Often”, “Sometimes”, “Rarely”, and “Never”. We collapsed these options into two categories: “High Frequency” (always or often) and “Low Frequency” (sometimes, rarely, or never). As above, we test the use of continuous measures of workplace safety practices rather than collapsing them into binary variables.
Politics. Our analyses include Presidential candidate choice as a moderator. We asked respondents about their candidate of choice in the 2020 U.S. Presidential Election: Joe Biden, Donald Trump, Jo Jorgensen, Howie Hawkins, or Other. We then created a dummy variable with “Joe Biden” and “Donald Trump” as the only response options. The small share of the sample (9.6%) who selected a different Presidential candidate were omitted from the analytic sample. Because the survey was in the field from September to November 2020, Presidential candidate choice includes both intended and actual 2020 presidential votes.
We include two other measures of politics as control variables. We asked respondents, “In general, how would you describe your political viewpoint?” The response options consisted of the standard ideological spectrum (“very liberal”, “liberal”, “moderate”, “conservative”, “very conservative”). We also include partisan affiliation (“Democrat”, “Republican”, “Independent”, “Other”). In robustness checks, we test the use of party and ideology as moderators instead of Presidential vote.
Control Variables. To account for potential confounding variables in the relationships between worker well-being, COVID-19 mitigation measures, and politics, we adjust for several demographic and socioeconomic characteristics. We include controls for gender, age, race/ethnicity, marital status, whether the respondent has children, whether a language other than English is spoken at home, educational attainment, and school enrollment. We also control for a variety of employment characteristics, including job tenure, hourly wage, and usual number of hours worked, as well as several variables related to COVID-19, namely the frequency with which workers interact face-to-face as part of their job and whether they or an immediate family member has contracted COVID-19. Finally, we include fixed effects for employer, occupation, and state-by-survey month (to capture the simultaneous geographic and temporal variation in COVID-19 prevalence).
Analytical Approach
There are three main components of our analysis. We begin by providing descriptive statistics on the frequency of four workplace safety practices reported by frontline workers in the service sector.
Next, we estimate OLS models that identify the association between each of these workplace safety practices
Finally, we include the same interaction term, plus political ideology (PI) and political party (PP) as controls (equation 3).
Results
Descriptive Statistics
In Table 1, we provide weighted descriptive statistics on the Shift analytical sample. Half of respondents are female and two thirds of respondents are white, non-Hispanic; 13% are Black, non-Hispanic; 14% are Hispanic; and 7% are other another race or multiple races, non-Hispanic. While more than 40% of the sample is under age 30, there are large shares of older workers, including 16% who are aged 50–59 and 20% who are age 60 or older, in the groups most at risk for serious COVID-19. While one in five is enrolled in school, in addition to the large age distribution, these descriptives also portray a workforce that is far from the stereotype of early career workers just starting out, with 43% married or cohabiting and 46% reporting having own children. Educationally, only one-in-ten workers have at least a BA, significantly below the 38% of the general population with that credential (Census Bureau, 2021). The demographic characteristics of Trump and Biden votes are roughly similar, though Trump voters are more likely to be white, to be married, and to have children compared to Biden voters.
Shift Sample Descriptives.
Note: Statistics are weighted to the demographics of individuals in the ACS working in similar industries and occupations.
Workers in the Shift sample are also disadvantaged in working conditions, with a median hourly wage of $12.70 and 32 usual work hours per week. Workers also generally have short tenure on the job, with 24% having a less than a year on the job and just 32% more than 6 years. Union membership is uncommon. Almost all workers, 89%, report at least often interacting face-to-face with customers. Three percent of respondents reported that they had contracted COVID-19, the same share as found in Pew's national survey (Kramer, 2020) conducted at the same time as the Shift survey (fall of 2020). A much higher share, 1 in 10 respondents, reported that an immediate family member had contracted COVID-19.
These data portray a population at risk, working in-person for many hours a week at low wages and in frequent contact with customers. In Figure 1, we show respondents’ reports of COVID-19 safety practices at work. By far the most common practice in the Fall of 2020 was co-worker masking, with 96% of respondents reporting that their co-workers masked “always” (86%) or “often” (11%). Customer masking was less prevalent, with 77% reporting that customers did so “always” (35%) or “often” (42%). A slightly higher share, 84% of respondents, reported that their employer “always” (72%) or “often” (11%) encouraged them to stay home when sick. The least common safety practice was social distancing, with 63% of workers reporting that they could “always” (30%) or “often” (33%) do so.
Frequencies of workplace safety practice by 2020 vote (% always or often).
As shown in Table 2, most frontline workers (80%) reported feeling safe at work, but that left a significant minority, one in five, who did not. Many workers were psychologically distressed (33%) or unhappy (29%). Workers reported a range of political ideology, party affiliation, and voting intention. While 10% reported that they were very liberal and 24% reported being liberal, 25% of workers reported being conservative or very conservative. In line with this distribution of ideology, 42% reported being Democrats, 31% Republicans, and 19% independents. Four-in-ten workers reported that they intended to or had voted for Donald Trump for president in 2020, somewhat below the 48% of Americans voting for Trump in the election (as a percent of all votes cast for Biden or Trump) (New York Times, 2020).
Descriptive Statistics of Outcome and Predictor Variables.
Note: Statistics are weighted to the demographics of individuals in the ACS working in similar industries and occupations.
COVID-19 Mitigation and Well-Being
Table 3 contains four panels, one for each of the four measures of COVID-19 safety practices that serve as independent variables, and three sets of models. In Model 1, for each outcome, we show the direct association between COVID-19 safety practices and worker outcomes. We find strong associations between each COVID-19 mitigation practice and workers feeling safe at work. Co-worker masking (β = 0.36, p < 0.001), customer masking (β = 0.27, p < 0.001), ability to socially distance (β = 0.28, p < 0.001), and employer encouragement to stay home when sick (β = 0.44, p < .001), are each positively and significantly associated with worker reports of feeling safe at work. These associations are substantively large, ranging from 0.68 to 1.09 in standard deviation terms. For example, taking predicted probabilities we see that 58% of respondents report feeling safe in low masking workplace environments against 86% of respondents in high masking environments.
Worker Well-Being, Workplace Safety Practices, and 2020 Presidential Candidate Choice.
All models include demographic, work, and COVID-19 controls, as well as state-by-month, occupation, and employer fixed effects.
***p < 0.001, **p < 0.01, *p < 0.05.
These relationships are strongest for feeling safe at work, but we also find clear and consistent evidence that these COVID-19 safety practices are associated with significantly less psychological distress among workers as well. Compared with feelings of safety, these effect sizes for the psychological distress outcome are somewhat smaller, ranging from 0.35 of a standard deviation for customer masking to 0.72 of a standard deviation in the case of employers encouraging employees to stay home when sick.
We find somewhat weaker direct associations between these COVID-19 safety practices at work and workers’ reports of overall happiness, with insignificant coefficients on the two masking variables, but significant associations between social distancing and happiness (β = 0.06, p < .05) and employers encouraging workers to stay home while sick and happiness (β = 0.21, p < .001). For the latter, the effect size is equivalent to 0.45 of a standard deviation.
While these models are associational and do not identify the causal effects of COVID-19 safety practices on worker well-being, we note that the models include a detailed set of controls, including demographic characteristics and other aspects of job quality. Moreover, the introduction of state-by-month effects significantly constrains the scope of comparisons, as do the introduction of employer fixed-effects which allow for within-employer comparisons.
In sum, workers who report that their co-workers and customers mask feel safer at work and experience less psychological distress, while workers who can socially distance and are encouraged to stay home when sick feel safer at work, are less psychologically distressed, and are happier. These results provide important evidence that in addition to preventing the transmission of COVID-19 among workers and consumers, adherence to these mitigation strategies is also significantly associated with worker well-being. However, these average associations may mask significant heterogeneity in the benefits of adherence to mitigation for workers’ well-being.
Polarization in the Association Between COVID-19 Mitigation and Well-Being
The relationship between COVID-19 mitigation practices and worker well-being may be subject to the influence of political polarization. In Model 2 of Table 3 we report coefficients on the interaction between each COVID-19 workplace mitigation strategy and if the respondent intended to vote or voted for Donald Trump in the 2020 presidential election in models predicting each of the three well-being outcomes. In Model 3, we additionally control for both respondent political ideology and for respondent party identification. These controls do little to alter the interactions between Trump vote and the relationship between COVID-19 mitigation practices and well-being outcomes.
For feeling safe at work, we find that the relationship between mitigation practices and worker well-being depends upon support for Biden or Trump. We find significant interactions between co-worker masking, customer masking, and ability to socially distance and voting for Trump, net of controls for party and ideology, the large set of observable worker characteristics, and month, state, occupation, and employer fixed-effects. In Figure 2, we plot the predicted values of feelings of safety from these models by voting and by COVID-19 safety practices, using the Stata margins post-estimation command and holding control variables at their mean values. The left panel is illustrative. We find that when co-worker masking is high, that is respondents report that their co-workers “always” or “often” mask at work, then Trump and Biden voters report similar high-levels of workplace safety – approximately 80%. However, when co-workers do not mask consistently, we see that Biden voters are much more negatively affected, with feelings of safety dropping to 25%, a very large substantive decline. In contrast, Trump voters are much less sensitive and responsive to low masking environments, with their feelings of safety dropping far more modestly to 61%. The coefficient in M3 of Table 3 in essence in provides the difference-in-difference estimate – Biden voters experience a 27 percentage-point larger decline in feelings of safety in the absence of consistent co-worker masking.
Feel safe at work by COVID safety practices and voting.
We see similar, if substantively smaller, interactions when examining customer masking and ability to socially distance. Trump and Biden voters report similar levels of safety when customers consistently mask, but Biden voters experience a much larger reduction in feelings of safety in the absence of consistent customer masking than do Trump voters (β = −0.16, p < .01). This pattern is also observed for ability to socially distance, where Trump voters are relatively unaffected by the absence of this safety practice, but Biden voters report feeling significantly less safe when they are unable to socially distance. The key exception to this pattern is that both Trump and Biden voters experience significant and substantively large reductions in feelings of safety when they report that their employer does not encourage workers to stay home when sick.
Feeling safe at work is an important aspect of work-related well-being. However, polarization in the consequences of lacking access to COVID-19 safety practices extends beyond this proximate outcome to broad and fundamental measures of well-being. In Figure 3, we plot the predicted values of psychological distress by the interaction of voting and COVID-19 safety practice. Here, we see the strongest polarization in response to social distancing and staying home when sick. Biden voters experience significantly more psychological distress when they are unable to socially distance at work (β = 0.13, p < .05) and when their employers do not encourage workers to stay home when sick (β = 0.18, p < .05). In contrast, we do not find polarization in the consequences of inconsistent co-worker and customer masking – Biden and Trump voters exhibit a similar increase in psychological distress in such situations.
Psychological distress by COVID safety practices and voting.
Finally, we find moderate evidence, plotted in Figure 4, of polarization in the association between COVID-19 safety practices and happiness. This is most evident in the wide gaps in how inconsistent co-worker masking affects Biden and Trump voters, with the two groups reporting similar levels of happiness in the presence of consistent masking, but with Biden voters reporting significantly depressed happiness when co-workers do not mask (β = −0.26, p < .05), against no difference for Trump voters. However, we find no evidence of polarization in the happiness response to inconsistent customer masking, ability to socially distance, or employers encouraging workers to stay home while sick.
Happiness by COVID safety practices and voting.
Robustness
Our results are robust to several alternate specifications, including to 1) the removal of weights; 2) continuous measurements of predictor and outcome variables; 3) the use of other measures of politics; and 4) the imputation of missing data. First, throughout our descriptive and multivariate analyses, we apply post-stratification weights which align the demographic characteristics of our sample with those of individuals in the American Community Survey (ACS) who work in the same industries and occupations. To test the sensitivity of our results to the inclusion of these weights, we reran all our analyses without them (results not shown). The key results all remain statistically significant. Second, we operationalize both our outcome and key predictor variables as dichotomous measures, which makes our results easier to interpret. However, several of these variables are also measured continuously. We tested whether our findings are robust to alternate specifications of these variables, obtaining similar results when evaluating both well-being and workplace safety practices as continuous rather than dichotomized variables. Third, one key premise of the paper is that party leaders have a particularly important influence on polarized behaviors and feelings of safety, above and beyond party membership or ideology. In supplementary analyses, we substituted party and ideology for intended 2020 Presidential vote and found that the moderating effects of voting for Biden or Trump are stronger and more consistent compared with both party and ideology. Finally, in the results we present, we omit cases with missing data on any model covariate. As a robustness check, we multiply imputed missing data on our independent variables. We then reran our models on the sample with imputed data and obtained similar results to those presented.
Discussion
The year 2020 witnessed an extraordinary shock to health and daily life in the U.S. and around the world owing to the COVID-19 pandemic. To avoid transmission of this infectious and potentially serious and deadly disease, the public health advice was to limit time spent indoors, to avoid contact with others when sick, and, when indoors, to wear a facial mask and maintain social distance of at least six feet. During the pandemic, many white-collar workers shifted their work time to their homes to avoid indoor exposure in the workplace. For service sector workers, however, working from home was not a viable option, and the nature of their job responsibilities required them to regularly spend long stretches of time indoors in high-traffic workplace settings. These conditions put service sector workers at elevated and continual risk of exposure to COVID-19. Further, their risk of exposure to COVID-19 was shaped not just by their own health and safety practices but very much depended upon the behavior of those around them with respect to masking, social distancing, and avoiding visiting their workplace when sick.
In this paper, we focus on service sector workers, who faced fraught and risky workplace settings during the pandemic, and consider how workplace health and safety practices affected their feelings of safety, distress, and happiness and how these effects were further shaped by a climate of intense political polarization. Drawing on Shift Project survey data from over 2,000 workers collected in Fall of 2020, we find that adherence to masking among co-workers and customers, ability to maintain social distancing, and the ability to stay home from work when sick each have substantial implications for workers’ feelings of safety. For example, 58% of workers reported feeling safe in low masking workplace environments compared with 86% of workers in high masking environments. We also find that these workplace safety practices are associated with reductions in workers’ psychological distress and with increases in workers’ feelings of happiness. This work complements research on the health consequences of the pandemic for workers at elevated occupational risk by now demonstrating that risky workplace settings not only may raise the risk of COVID-19 transmission, but also have implications themselves for workers’ well-being.
However, the value of these health and safety precautions for workers was politically polarized. We show that the relationship between safety practices and worker well-being was highly contingent on which of the major Presidential candidates a worker supported. Workers who had supported Donald Trump in the 2020 Presidential election were less responsive to changes in safety environments. During the pandemic, former President Donald Trump downplayed the severity of the COVID-19 health risk and at times espoused messages that ran counter to public-health recommendations to mask and maintain social distance. Those who supported Donald Trump also witnessed their preferred political leader regularly flout public health recommendations and, ultimately, contract and recover from COVID-19. Although Trump's messages and behavior – if emulated – would expose his supporters to greater COVID-19 health risks, we find a silver lining for these workers, which could be characterized as “ignorance is bliss.” In the face of exposure to objectively riskier workplace settings, Trump supporters experienced some protective benefit in terms of greater peace of mind and mental well-being.
Conversely, workers who reported that they supported President Joe Biden in the 2020 election were more sensitive and responsive to safety environments. Biden supporters felt far less safe than Trump supporters in workplace environments with low adherence to safety practices. Prior research has highlighted the health consequences when recommendations to mask, distance, and quarantine while sick are not adhered to in terms of heightened risk of COVID-19 infection (Su et al. 2021). Our research documents additional negative externalities associated with a lack of adherence to health and safety practices in the workplace – namely, diminished feelings of safety, increased distress, and decreased happiness. The mental toll of risky workplace environments was stratified by politics and disproportionately borne by Biden supporters.
Conclusion
Overall, we find that political polarization moderates the relationship between workers’ perceived safety and their well-being during the pandemic, but one exception to this pattern is noteworthy. Workplace environments in which workers were not able to stay home when sick were associated with diminished feelings of safety for Trump and Biden voters alike. One possible explanation for this pattern is the idea that working while sick is a more tangible and less contested threat to health and safety compared with the more politicized discourse around masking as impinging on personal freedoms (Scoville et al., 2021).
In interpreting our results, we note a few limitations that should be kept in mind. First, The Shift Project survey sample is a non-probability sample, and as such may differ in some respects from the broader population of service sector workers. To address this concern, we apply survey weights that align the characteristics of our survey sample with workers in the American Community Survey on demographics. In addition, separate work has shown that The Shift Project data closely resembles two gold standard, probability data sources – the Current Population Survey and National Longitudinal Survey of Youth – in terms of job conditions and on key relationships such as that of job tenure and wages (Schneider and Harknett, 2022). In the current study, we note that our sampling approach attracted large shares of Trump and Biden supporters, which were not far off from the shares of supporters in the overall population. Nevertheless, the possibility of differential non-response on unobserved attributes remains a potential source of bias. There may also be some unobserved selection into our sample, if workers who felt unsafe or distressed in their workplaces were systematically more likely to leave those jobs compared to those who felt safe or were not distressed. However, in the context of the COVID-19 pandemic, it is also important to note that many service sector workers live paycheck to paycheck, and the expanded Unemployment Insurance (UI) in the early phases of the pandemic was unavailable to those who voluntary left their jobs.
Second, our data are cross-sectional and thus safety practices and feelings of safety and well-being outcomes are measured at the same point in time. It is possible that those who feel unsafe, distressed, or unhappy would perceive workplace environments differently than their counterparts who feel safer, less distressed, and happier. However, even if reporting on safety practices is influenced by a worker's own feelings of safety and mental health, the analysis of political polarization is nevertheless revealing. Among workers who report environments with less adherence to safety practices, support for Trump versus Biden is strongly associated with feelings of safety, distress, and happiness.
Third, our analysis omits respondents who did not intend to vote for Donald Trump or Joe Biden in the 2020 Presidential election, including those who voted for third party candidates, those who were undecided, and those who did not intend to vote. The focus of our paper is on the unique role of individual political leaders in patterning the polarization of COVID-19 workplace safety practices, so we chose to focus exclusively on those who voted for one of the two major party candidates.
Finally, our analysis is unable to pinpoint the precise theoretical meaning of the polarized responses by support for Trump to workplace safety policies. For example, the political differences in the effect of low masking environments on workers’ well-being may be a proxy for beliefs about the efficacy of masks; in other words, if you do not believe masks are effective, then your feelings of safety are unlikely to change substantially when there is less masking. If this were the underlying logic, then our results also suggest a potential explanation for partisan differences in compliance with COVID-19 mitigation strategies – Trump partisans may simply believe them to be less effective. However, it could also be a proxy for something more interpersonal, if workers’ feelings of safety or happiness in these environments stem from the implications of what it says about their coworkers.
Our paper harnesses novel data in a distinctive and extraordinary historical period to make several contributions. We focus on service sector workers who face heightened, regular risk of exposure to COVID-19 in their workplaces. These workers are not able to work from home, and their workplaces are indoors and involve frequent face-to-face interactions with customers and co-workers. Extending prior research from the early months of the pandemic, we draw on data from Fall of 2020, several months into the pandemic when scientific debate was largely settled that masking, social distancing, and staying home when sick constituted essential COVID-19 preventive strategies. Our study reveals stark political polarization in the mental health consequences of non-adherence to recommended workplace health and safety practices. In an era of political polarization and strong factional allegiance to Donald Trump, we demonstrate the powerful influence of support for political leaders in shaping responses to safety environments and, ultimately, workers’ peace of mind and mental health. In the context of the massive health threat and stressor of the COVID-19 pandemic, we find that Biden supporters were sensitive to the risks in their workplaces, demonstrated by their diminished feelings of safety and happiness and increased psychological distress in response to non-adherence to COVID-19 safety measures. In contrast, Trump supporters’ feelings of safety and mental well-being were far less influenced by the absence of workplace COVID-19 safety practices. Together, our findings provide evidence of inequality across the political spectrum in the mental health toll of COVID-19 workplace risks.
Footnotes
Acknowledgements
We gratefully acknowledge support from the Bill & Melinda Gates Foundation (Grant Nos. INV-002665 and INV-016942), the Robert Wood Johnson Foundation (Award No. 74528), and the W. T. Grant Foundation (Grant No. 188043). The findings and conclusions contained within are those of the authors and do not necessarily reflect positions or policies of these foundations. The authors received excellent research support from Connor Williams, Elaine Zundl, Annette Gailliot, and Evelyn Bellew.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the William T. Grant Foundation, Robert Wood Johnson Foundation, Bill and Melinda Gates Foundation, (grant number 188043, 74528, INV-002665, INV-016942)
