![Sample our Economics, Finance,Business & Industry journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5931/TOC-Subject-Sample-2021-Economics-Finance-Business-Industry.jpg"})
ABSTRACT
This paper explores how working conditions in meatpacking plants contributed to the spread of the COVID-19 virus. Data from the Occupational Information Network was used to construct a set of industry-level working condition variables and compare meatpacking to the sample of other manufacturing industries in our comparison group. This novel approach showed that proximity to others in the meatpacking industry is likely the main factor influencing the spread of COVID-19, more than three standard deviations higher in meatpacking than our comparison sample of other manufacturing industries. Subsequently, we performed a county-level analysis on COVID-19 spread, comparing rural counties with a large share of meatpacking workers to nonmetropolitan counties that were similarly dependent on other single manufacturing industries, using the time frame of mid-March to the end of 2020. In mid-April 2020, COVID-19 cases in meatpacking-dependent rural counties rose to more than 12 times compared to rural counties dependent on other single manufacturing industries. This difference disappeared completely by mid-July and held steady throughout the year. We demonstrate that our results are robust to a battery of robustness checks ruling out the set of plausible alternative hypotheses, including examining data on COVID-19 spread among meatpacking workers directly.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 For the list of counties, see Appendix A.
2 Krumel (Citation2017) demonstrates that immigrant workers might “have different compensating wage differentials from natives in the meatpacking industry”, i.e. they are more willing to take these dangerous jobs at lower salaries than a comparable U.S. worker.
3 It is important to note that the authors used two-digit North American Industry Classification System (NAICS) codes in their analysis (likely because of data availability for monthly employment numbers). For context, meatpacking is a four-digit industry classification (3116) and would have been aggregated up to the manufacturing sector (31–33) in their analysis.
4 The values that O*NET provides are cross-industry occupational scores. While occupations will undoubtedly vary from industry to industry, we believe that the core components of the occupations will be the same.
5 Since the formal definition for the work context involves contact with others (i.e. face-to-face, by telephone, or otherwise), this variable will be challenging to interpret in light of COVID-19. Separating occupations with high in-person contact and occupations that can easily switch to remote is not feasible.
6 It is important for us to acknowledge that this weighted average is representative of the average worker in the industry but will obviously not be representative of every single worker (for example, production workers have very different work characteristics than executives).
7 We acknowledge limitations with this specific outcome variable. In Appendix E, we engage with several alternative constructions of our time series that enhance our confidence in the validity of the analysis (most directly by examining curated COVID-19 case data for food manufacturing workers that would not suffer from potential differences in recording rules which might be unobservable in our JHU data).
8 The full table with the associated scores included is provided in Appendix B.
9 Despite the statistically meaningful difference calculated, we want to emphasize that all manufacturing workers still have a high level of contact with others.
10 These primary results are robust. Several alternative constructions of the time series are included in Appendix E. 1) They are robust to aligning counties by point in the epidemiological disease progression, eliminating concerns about statewide spread driving the measured results. 2) They are robust to using the neighbouring rural counties to address some unobserved factors likely to be shared across neighbouring counties. 3) It is further demonstrated that meatpacking plants are likely the vector of observed county-level spread by relaxing the employment threshold, indicating a monotonically decreasing pattern as meatpacking employment becomes less prominent in a county. 4) The results are robust to the utilization of deaths as the variable of interest, indicating that testing is not driving the analysis. 5) The results are also validated externally by using a novel dataset with COVID-19 cases by meatpacking workers and workers in other food processing, exhibiting an identical temporal pattern.
11 Appendix C tracks the data as separate categories rather than presenting the ratio.
12 On June 3, 2020, the start of the convergence, meatpacking-dependent counties had a 2-week-moving average of daily cases of 38, compared to 6 for our comparison group. On July 21, 2020, our data appeared to stabilize at around a 2-week moving average of daily cases of 19 for both groups. For the next two months, the 2-week moving average of daily cases is bounded between a low of 15 and a high of 22. A temporary steady-state appears to be achieved, suggesting the convergence was not driven by our comparison group increasing.
13 Appendix D presents a visualization of the meatpacking production floor adjustments made that was published in the New York Times.
14 We do not have a serious concern about either of these confounding with our results. If a sick individual showed up to work because of an attendance bonus, there still needed to be a mechanism for spreading the virus to their co-workers (i.e. it is still the physical proximity that is likely enhancing the spread). Shared transportation is one of the variables we condition over in our multivariable regression. Unsurprisingly, the variable is positive and statistically significant, though it does not impact the point estimate on our variable of interest.
15 We run an alternative specification where instead of the meatpacking indicator, we include the weighted average for the physical proximity of the county’s employment as our variable of interest.
16 The indicator variables we develop for the different meatpacking thresholds are mutually exclusive to allow for estimation to occur. For example, if an observation has 20% or more employment in meatpacking it will receive an indicator value of one for just that threshold, and zero otherwise. We always select the highest threshold achieved as the indicator variable that receives the value of one.
17 Hancock, Georgia is the only county that has a manufacturing-dependence other than meatpacking in this list, as more than 20% of county employment is in NAICS 3272 Glass and Glass Product Manufacturing.
18 All subsequent results are robust to changing this categorization to the epidemiological disease progression.
19 Artz, Jackson, and Orazem (Citation2010) used 20% as their highest indicator variable cut-off in their analysis.
20 To provide context against our preferred counterfactual, in the category containing 10–20% employment, one county is also considered dependent on a single manufacturing industry. In the category containing 5–10% employment, there are three counties also considered dependent on a single manufacturing industry; and for the final category containing at least one employee in meatpacking but no more than 5% of total county employment, there are 17 counties that are classified as dependent on a single manufacturing industry. There is negligible overlap between these two different measures, meaning the analysis can be viewed as distinct.
21 There is a legitimate concern that because meatpacking plants were testing employees at the door, those tested individuals would skew the number of COVID-19 cases up for the host county, compared to other counties that did not have the same level of systematic testing conducted (this would be especially problematic over the asymptomatic individuals, who would have been found to be infected when similar asymptomatic people in the general population would never have been tested in the comparison group).
22 This external validation is critically important, as we will further leverage this data to conduct a later robustness check.
23 The CDC’s report is not a comprehensive data source on COVID-19 cases among meatpacking workers due to limited response rates by states. Their report, however, is likely representative of meatpacking as a whole since their data set is over 28 different states and covers at least a quarter of the meatpacking industry.
24 Performing a back-of-the-envelope calculation, if meatpacking workers were three times less likely to die from COVID-19 than the general population and individuals in meatpacking-dependent counties were still dying at four times the rate of other manufacturing-dependent counties, this revised estimate is still in line with our primary result. While it does not entirely mitigate the concern that testing drove our results over cases, it should lessen the worry.
25 Late June also has a similar level; however, the blip in mid-July makes it difficult to verify that this was the true start of this pattern.
26 This difference appears to be a data dump that is being attributed to a single day (that we believe likely contained cases from multiple days), which skews the pattern in our figure for the subsequent week. However, since we cannot determine what is truly driving this outlier, we do not correct this datapoint for transparency and replicability.