Livestock plants and COVID-19 transmission

The disease burden of COVID-19 is not uniformly distributed across the global population. Certain conditions appear to influence the degree to which people spread the virus. Some contexts and social behaviors are believed to lead to superspreading events that disproportionately affect local populations (6, 7). Previous studies have explored links between the incidence of COVID-19 cases and a range of demographic and environmental factors, such as age, occupation, income, race, intergenerational mixing, temperature, and humidity (8–13). Social, commercial, and industrial activities are also believed to affect transmission, for which reason countries worldwide have implemented a range of economic and social-distancing measures (8, 14–20). In the United States, some industries are exempted from shelter-in-place orders and have remained operational due to their necessity to satisfy basic societal needs (21). We investigate the relationship between transmission and one such activity, livestock processing.

The livestock- and poultry-processing industry is an essential component of the global food supply chain. In the United States, it is a large industry, employing 500,000 people. It is also highly concentrated: The largest four companies in beef, pork, and poultry processing capture 55 to 85% of their respective markets (22–27). This degree of concentration stands in contrast to the European Union (EU), for example, where the top 15 meat companies represent 28% of EU meat production (28).

Over the decades, the livestock- and poultry-processing industry in the United States has consolidated its operations into fewer, larger plants, in which meat production per plant has increased threefold since 1976 (29, 30). Today, 12 plants produce over 50% of the country’s beef, and 12 others, similarly, produce over 50% of the country’s pork (30, 31). Early in the COVID-19 pandemic, livestock-processing plants worldwide experienced spikes in infections, facing shutdowns that disrupted meat and dairy supplies (32–35). In the United States, reports of COVID-19 spreading within the livestock-processing industry led to increased attention and updated safety guidance by the Centers for Disease Control and Prevention (CDC) (22). Several plants were forced to shut down until, among other factors, a federal executive order invoked the status of livestock processing as “critical infrastructure” for national security and mandated that these plants remain open (36, 37).

Work routines in livestock processing have several characteristics that make plants susceptible to local outbreaks of respiratory viruses. The CDC includes the following among potential risk factors: long work shifts in close proximity to coworkers, difficulty in maintaining proper face covering due to physical demands, and shared transportation among workers (22). Previous research has proposed occupational exposure to livestock animals as a driver of viral spread, although an experimental study did not find pigs or chickens to be susceptible to the SARS-CoV-2 virus associated with COVID-19 (38–41). Increases in production-line speeds due to technological enhancements as well as policy changes have also been hypothesized to exacerbate COVID-19 transmission (5, 42). Among those we investigate are USDA waivers on poultry-production line-speed limits for plants with strong commercial production practices and microbial control (43).*

The indoor climate of livestock facilities may increase transmission risk. To preserve meat after slaughter, processing areas are maintained at 0 to 12 °C (44), and such low temperatures have been linked to increased COVID-19 risk (45, 46). Though these rooms are kept at 90 to 95% relative humidity to prevent meat from drying and losing weight, the low absolute humidity at near-freezing temperatures may encourage the transmission of airborne viruses such as influenza (47–49). Moreover, studies have suggested that industrial climate control systems used to cool and ventilate meat processing facilities may further the spread of pathogenic bioaerosols, a proposed COVID-19 transmission route (46, 50–53).

Workers’ socioeconomic status and labor practices may also contribute to infection and transmission. Among front-line meat-processing workers in the United States, 45% are categorized as low income, 80% are people of color, and 52% are immigrants, many of whom are undocumented and lack ready access to healthcare and other worker protections that could facilitate COVID-19 prevention and treatment (54–56). In addition, employees at these facilities may face incentives to continue working even while sick through company policies on medical leave and attendance bonuses (5, 22, 57). In addition, through consolidation over the decades, the meatpacking industry has potentially increased its monopsonistic power over labor markets, which has been linked to greater work hazards (58–60).

We find a strong relationship between proximity of livestock plants and the incidence of COVID-19 over time. Fig. 1 plots average COVID-19 case and death rates over time by whether there is a large livestock facility in a given county relative to rates in counties at varying distances from a plant. In both cases, we see an increasing divergence in outcomes beginning in early April based on livestock-plant proximity.

Fig. 1 does not account for county-level differences in terms of density and demographics. In Table 1, we estimate the relationship between livestock plants and COVID-19 incidence as of July 21, 2020, using regression models that control for potential confounding variables, including county-level measures of income; population density and its square; the timing of the first case; the proportions of elderly people, uninsured people, frontline workers, and people using public transportation; racial and ethnic characteristics; average household size; local freight traffic; and populations of nursing homes and prisons. We find that livestock plants are associated with an increase in COVID-19 cases by approximately four per thousand people, representing a 51% increase over the July 21 baseline rate of eight per thousand. Likewise, death rates increase by 0.07 per thousand, or 37% over the county baseline of 0.2 deaths per thousand. The results are robust both nationally and when only considering variation within states after including state fixed effects. We also use an alternate specification with a binary measure of whether a county has one or more livestock plants. Such counties are associated with six additional cases per thousand, or a 75% increase over the baseline, as well as 0.1 additional deaths per thousand, or 50% over the baseline county death rate.^† In addition, COVID-19 appears to arrive earlier in counties with livestock plants (SI Appendix, Table S2).

Angrist and Krueger (65) noted that “one should always be wary of drawing causal inferences from observational data.” We know of no random-assignment design that could address our research question and thereby yield the most reliable path to causal inference. The best we can do here is provide an unusually broad array of observational evidence. This includes (but is not limited to) ruling out the most obvious confounders, a cross-sectional IV, and the event-study analysis leveraging shutdown timing. A still more compelling natural experiment would leverage explicit and exogenous variation that drives livestock-plant shutdowns, i.e., an IV for the shutdowns or their timing. Unfortunately, we know of no such identifying variation.

Readers may disagree on whether our array of analyses has isolated a causal effect. Given this, and in order to be conservative, we avoided causal language throughout our text so as not to overstate the “hardness” of our method (66). This avoidance and caution stands in contrast to other recent, impactful work on COVID-19.

Still, we believe that our array of analyses constitutes the best feasible approach to shed light on the role of livestock-processing plants in the US COVID-19 pandemic. For a question of this importance, we believe there is no “harder” method available (66). As policymakers and industry leaders seek to preserve vital food-supply chains while mitigating the pandemic’s spread, evidence on the potential scope of the issue is particularly valuable, as well as assessment of the relationship between temporary plant shutdowns and subsequent COVID-19 growth dynamics.

Although our estimate that 6 to 8% of COVID-19 cases are associated with livestock plants may appear high, it is important to recall that high levels of geographic heterogeneity in COVID-19 incidence can be explained by some combination of individual behavior, government policy, social-distancing compliance, and economic activity: The United States, for example, has 4% of the world’s population, but approximately a quarter of all cases and deaths as of July 2020. When narrowing the geographic focus, we can imagine the distribution of COVID-19 incidence to be similarly clustered, if not even lumpier.

Kansas provides a telling example of the outsized role of livestock facilities: As of July 20, a total of 3,200 of 23,300 state cases (14%) were directly linked to meatpacking (67). For context, there are 17,200 employees in the animal-slaughtering industry in Kansas (68), or 0.6% of the state’s population, suggesting that livestock plants had a relationship of a magnitude closer in scale to our own estimates (Kansas’ estimate is 23× their labor footprint). Although the figure we are estimating in our study (6 to 8% of all US cases out of a national livestock workforce of 0.15%, or a multiplier of 40 to 53×) is larger, we believe that this finding is plausible, considering follow-on community spread; Kansas’ official tally, though evidently aided by some degree of contract tracing, was reportedly hampered by lags in hiring staff and legislative actions that have inhibited tracing efforts (69). That is, the figure we have calculated could, in fact, be more complete than the Kansas figure in capturing the spread resulting from livestock plants.

Our analysis of individual meatpacking companies may present an opportunity to explore how differences in corporate structure and operating practices may account for their differential public health outcomes. In particular, the evidence that shutting down plants temporarily may be related to decreases in COVID-19 case growth presents a potentially powerful transmission mitigant. In addition, the positive relationship between COVID-19 transmission and production-line speed waivers issued to poultry plants, particularly those during the 2020 pandemic, is notable, given that these waivers are intended for plants with safe commercial production practices and microbial control.^‡‡ This finding suggests a need for additional examination of this program.

An implication of this study is that some aspects of large meat-processing plants render them especially susceptible to spreading respiratory viruses. One potential explanation is that large plants simply entail more activity and employ more people. Because these plants provide a central location for moving products, it is plausible that a linear increase in the potential infected within the plant would entail a nonlinear response, owing to the complex and exponential nature of disease-transmission dynamics (70). Another driver may be the large physical spaces where processing occurs. Larger rooms tend to be louder and, thus, require more shouting (53), and they may require stronger climate control, which we note in our introduction may aggravate COVID-19 spread. A larger space that employees must navigate in reaching their workstations may also increase the number of workplace interactions.

More broadly, the finding that meatpacking plants may contribute to high levels of community spread underscores the potential negative public health externalities generated by the industry, which may be attributable to industrial concentration, operating practices, and labor conditions. Complicating this matter from an economic standpoint is the supply-chain choke point created by large plants disrupted by COVID-19, causing food shortages, driving up prices, and incurring substantial upstream and downstream economic losses. Cataloging and addressing the underlying factors that produced this systemic risk in the first place could not only strengthen the US food system in the face of COVID-19 and future disruptions, but also help illuminate analogous weak points in other industries and supply chains.

Our analysis used a county-level dataset of COVID-19 cases and deaths from the New York Times, based on reports from state and local health agencies (71). Included in counts are both confirmed and probable deaths, as categorized by states. The five county boroughs of New York City are grouped into one unit. We limited the analysis to the continental United States. Our baseline model specification takes the following form: where $outcom{e}_i$ is the COVID-19 case or death rate in county $i$, $\beta$ is the coefficient of interest, $control{s}_i$ is a vector of county-level covariates, ${\alpha }_s$ is a dummy for fixed effects in state $s$, and ${ϵ}_i$ is the error term.

Covariate data include county-level race, ethnicity, and age structure data from the US Census and mean county-level income data from the US Bureau of Economic Analysis (72, 73). Data on nursing-home populations, incarcerated populations, uninsured populations, average household size, and work-commuting methods come from the 2014–2018 American Community Survey (74–77). Data on manufacturing establishments come from the American Economic Survey (68). Number of frontline workers were derived from Center for Economic Policy Research data (54), transforming from Public Use Microdata Area-level to the county level, assuming even allocation. The freight index is from the Federal Highway Administration’s Freight Analysis Framework (78) using the variable AADTT12, the annual average daily truck traffic in 2012, which we sum across all listed highways in a given county. Data on state-level social-distancing policy come from a dataset synthesizing news articles tracking these policy measures (79–81).

Locations and characteristics of livestock processing facilities come from the USDA FSIS (82). Beef and pork livestock plants were filtered to include plants with volume of all processed products greater than 1 million pounds per month (categories 4 and 5), which account for the vast majority of US production. Poultry livestock were filtered to include plants with volumes greater than 10 million pounds per month (category 5) because that category alone accounts for the majority of US production. County-level mobility data were made accessible to COVID-19 researchers by Google (83). County-level COVID-19 testing data came from a dataset gathered from 31 state health agencies (84). Data on line-speed waivers came from the USDA FSIS (85). Data on plant closures and opening dates came from a dataset assembled from various local news reports, building on a dataset from the Midwest Center for Investigative Reporting (86, 87). Historical livestock-production data are from the 1959 USDA census of agriculture, accessed via the Inter-University Consortium for Political and Social Research (88).