key: cord-0723814-zceoppzq
authors: Charlton, Diane
title: Seasonal farm labor and COVID‐19 spread
date: 2021-09-01
journal: Appl Econ Perspect Policy
DOI: 10.1002/aepp.13190
sha: 5a4cbb2dc42dc8412a5a2bf35becce6375c59dcc
doc_id: 723814
cord_uid: zceoppzq

The COVID‐19 pandemic in 2020 caused unprecedented shocks to agricultural food systems, including increased risk to worker health, labor‐related input costs, and production uncertainty. Despite employer precautions, there were numerous worksite outbreaks of COVID‐19. This paper examines the relationship between month‐to‐month variation in historical agricultural employment and changes in the incidence of confirmed COVID‐19 cases and deaths within U.S. counties from April to August 2020. The results show that employment of 100 additional workers in fruit, vegetable, and horticultural production was associated with 4.5% more COVID‐19 cases within counties or an additional 18.65 COVID‐19 cases and 0.34 additional COVID‐19 deaths per 100,000 individuals in the county workforce.

Farm labor demand is often seasonal and uncertain, particularly in high value crops such as fruits and vegetables that are still harvested by hand. In January 2021, the American Farm Bureau Federation president Zippy Duvall stated that "the biggest limiting factor of American agriculture is our labor force" (Harker, 2021) , thus underscoring the importance to industry leaders and policymakers of understanding the relation between major shocks like COVID-19 and farm worker safety and agricultural production risk. In this paper, I measure the association between historical monthly agricultural employment and the incidence of new COVID-19 cases and deaths within U.S. counties from April to August 2020. The findings from this paper The findings show that employment of 100 additional workers in the FVH sector in 2019 was associated with 4.5% more COVID-19 cases within counties or an additional 18.65 COVID-19 cases and 0.34 additional COVID-19 deaths per 100,000 individuals in the county workforce. 4 Results are robust to the inclusion of controls for 2019 monthly employment in post-harvest crop activities, construction, retail, and accommodations and food industries. Further analysis of the relation between specific crop employment and incidence of COVID-19 within counties shows a significant positive increase in COVID-19 cases or deaths associated with employment by grape, other non-citrus fruit producers, greenhouses, and Farm Labor Contractors (FLCs). I find no such statistically significant association between COVID-19 cases or deaths and employment in more mechanized crops, including grains, oilseeds, and field crops, or in livestock. The findings suggest that FVH crops, which frequently depend on a large seasonal workforce, are particularly vulnerable to COVID-related labor supply disruptions. Thus, FVH producers may incur higher costs related to worker health provisions or risk management strategies during a severe event like the coronavirus pandemic.

This paper contributes to the literature examining the effects of farm labor supply shocks along with the emerging literature investigating how social activity and migration influence the spread of COVID-19. Richards (2018) shows evidence of persistent farm labor shortages in subsectors of the California farm labor market. Inward farm labor supply shocks due to changes in immigration enforcement policies have led to increased farm wages and reduced agricultural producer profitability (Ifft & Jodlowski, 2016; Kostandini et al., 2014) , substitution of capital for labor (Charlton & Kostandini, 2020; Clemens et al., 2018) , and increased supply of family farm labor (Luo et al., 2018) . The coronavirus pandemic in 2020 differs from previous shocks to labor supply in that the pandemic is likely temporary and the pandemic shocked multiple labor sectors throughout the global economy at once. However, in an era of tightening farm labor supply as the U.S. is currently experiencing (see for example Charlton and Taylor (2016) and Zahniser et al. (2018) ), even temporary increases in labor costs and uncertainty of labor supply could push producers to invest in more mechanized methods of production or new labor management practices.

This paper also contributes to new literature examining factors that increase the spread of COVID-19 (Dave et al., 2020a,b; Friedson et al., 2020; Mangrum & Niekamp, 2020) . However, unlike much of the COVID literature, this paper does not attempt to identify the mechanisms through which labor-intensive agricultural activities lead to increased incidence of COVID-19 within counties. Rather, the objective of this paper is to identify whether there is a positive association between seasonal, labor-intensive agricultural activities and new COVID-19 cases and deaths. The findings from this paper can help producers and policymakers anticipate and respond to threats to worker health and safety when labor-intensive tasks cannot easily be delayed or performed remotely even during severe events like a pandemic.

Farm labor shortages have been a primary concern to U.S. agricultural producers for many years. Since the late 1990s, the share of farm workers who migrate in the United States has declined by roughly 60% (Fan et al., 2015) . And since 1980, the share of rural Mexicans, the primary source of labor to U.S. farms, working in agriculture has declined by roughly 1% per year (Charlton & Taylor, 2016) . Farm labor shortages became more frequent in the 2000s (Hertz & Zahniser, 2013) . Though labor shortages are typically local and temporary, they can nevertheless be devastating. Farm producers have responded to the tightening farm labor supply using a variety of strategies, including technology adoption (Charlton et al., 2019; Clemens et al., 2018; Hamilton et al., 2020) , contracting H-2A agricultural guest workers (Luckstead & Devadoss, 2019; Zahniser et al., 2018) , and hiring workers through FLCs (Taylor & Thilmany, 1993) .

FLCs can mitigate the risk of a farm labor shortage by matching workers to farm employers and thus reducing frictions in farm labor markets. However, FLCs have sometimes been known to use their ease of mobility to evade legal detection, such as when suspected of hiring unauthorized workers (Taylor & Thilmany, 1993) . FLCs may encounter particularly large challenges in reducing the spread of COVID-19, since they often house workers together and transport them to farms on buses. Given the high mobility of FLC crews and the long incubation period of COVID-19, the virus could potentially spread quickly among crews of farm workers before being detected. Thus, a major farm labor management tool, the use of FLCs, may be inadequate to mitigate risks associated with worker safety during the pandemic.

Farm workers may be particularly susceptible to COVID-19, not only because their work is essential and cannot be performed remotely, but also because farm workers often share a number of characteristics that are correlated with greater risk for COVID-19. Farm workers and their families often live below the poverty line, reside in dense living quarters, and lack access to healthcare and health insurance. According to the National Agricultural Workers Survey (NAWS), 48.92% of farm workers in 2016 reported that they did not have health insurance, 37.04% had not used any type of healthcare services in the United States in the past 2 years, even fewer had used health services in another country, and 19.15% reported that they did not seek medical care because it was too expensive. 5 Policy interventions can only partially offset some of these disadvantages. For example, the Families First Coronavirus Response Act, effective through December 31, 2020, required employers with fewer than 500 employees to provide paid sick leave for workers affected by COVID-19. However, employers with fewer than 50 employees could request an exemption, and there was no guarantee that workers would select to take time off from work if they or their family members displayed coronavirus symptoms even if they were offered paid sick leave.

UC Berkeley School of Public Health (2020) found that 13% of 1,091 farm workers in Monterey County, California tested positive for COVID-19 between July and November 2020. Antibody tests showed that 20.2% of workers had prevalence of antibodies in September and 19.4% of workers in October. Only 58% of those who tested positive for COVID-19 during the study displayed symptoms, demonstrating how difficult it can sometimes be to detect the virus. Within this sample of workers, 37% reported living in crowded housing defined as more than two people per bedroom, and 19% lived with roommates who were of no relation. Approximately 9% of farm workers lived with someone who had been diagnosed with COVID-19 within 2 weeks prior to their interview, and 43% of workers lived in housing with shared bathroom or bedroom, thus making it impossible to quarantine if infected. A little more than a third of workers said that they traveled to work with people from outside of their household, and 11% said that they had at least one co-worker quarantined or isolated within the past 2 weeks.

Among workers who said they had experienced COVID-19 symptoms, 57% said that they went to work while they had symptoms. In most cases, workers said that they continued to work because they felt well enough to do so. An additional one-quarter said that they came to work because they were concerned about losing pay, and 13% said that they were concerned about losing their job if absent. When asked whether they would get vaccinated when a vaccine became available, only 52% of workers said that they were extremely likely to, and 11% said that they were either unlikely or very unlikely to do so. Thus, even as vaccines become available, the risk of COVID-19 exposure in the farm industry may not disappear, at least not right away, and it may be difficult to determine if someone has been positively exposed to COVID-19, since many either do not show symptoms or feel well enough to continue working while symptomatic.

Lack of legal work authorization can further contribute to farm workers' vulnerability to COVID-19. An estimated 48% of farm workers, excluding H-2A guest workers, are unauthorized immigrants. 6 Many farm workers may be fearful of seeking public services, including healthcare or legal services if they believe that their immigration status might be exposed or if they believe that someone might try to deport them or their friends or family. H-2A workers may be less vulnerable to some of these challenges in obtaining services, since they have legal documentation to work in the United States. 7 However, their working conditions and access to legal and health services likely vary. Since H-2A workers can only legally work for one employer during their residence in the United States, some H-2A workers may not be aware of the services that are available to them or may fear that certain actions would jeopardize future employment.

Farm worker ethnicity also plays a role in susceptibility to COVID-19. An estimated 81% of farm workers in 2016 were of Hispanic descent, not including H-2A workers. 8 Hispanics appear more likely to get COVID-19 and more likely to have severe symptoms. Hispanics experience an estimated 5-7-fold risk of COVID-19 mortality compared with Whites (UC Berkeley School of Public Health, 2020). Even though Hispanics represented only 24% of workers in industries with the highest rates of COVID-19 outbreaks in Utah, they made up 73% of the COVID-19 cases in those industries (Bui et al., 2020) .

Many employers implemented policies to reduce the risk of spreading COVID-19 while working in the fields, but measures also need to be taken to increase social distancing in worker housing and transportation. Some FLCs rented additional buses to transport workers to worksites or made additional trips with the same bus to increase social distancing during commute (Beatty et al., 2020) . Some of the employers who provide housing for farm workers housed fewer workers per living unit and designated quarantine housing for workers who showed symptoms of COVID-19. These measures could be vital in preventing a COVID-19 outbreak, but they also increased the marginal cost of employing each worker. Employers in Washington sued the Departments of Health and Labor & Industries, stating that it was impossible to comply with the emergency COVID-19 safety regulations, including restrictions on distance from a hospital emergency room, and employers in Oregon requested the repeal of emergency rules stipulating increased spacing between beds in farm worker housing. Some employers state fears that if emergency housing rules are not repealed, the incidence of farm workers sleeping in informal housing, including vehicles, will likely rise (Rural Migration News, 2021). The increased cost of emergency worker safety provisions could lead some producers to take negligent actions while the risk of COVID-19 spread is still high. Other producers might consider investing in labor-saving technologies to reduce labor costs and mitigate risks to farm worker health or the risk of encountering a farm labor shortage.

Prior to the coronavirus pandemic, many fruit and vegetable producers were already experimenting with new technological innovations to reduce labor inputs in production and harvest. For example, Taylor Farms of Salinas, California began installing robotic arms that package salads in its processing facility. In the fields, they harvest with automated lettuce harvesters that use patented jet knife technology to cut the Romaine heads near the ground. The harvester then delivers the Romaine heads to workers who inspect and sort them on the harvester's platform. These harvesters arguably improve workers' safety by removing the need for workers to bend over rows of lettuce, wielding machetes, for several hours every day. Taylor Farms says these innovations reduce their risk of labor shortages and provide opportunities for workers in higher-skilled, better-paying jobs. 9 Strategies to slow the spread of COVID-19 among farm workers, including extra bus transportation, providing quarantine housing, spacing workers, providing additional sick leave, and providing personal protective equipment is costly to producers. Furthermore, increasing COVID-19 cases during peak harvest seasons poses enhanced risk of a labor shortage and loss of harvested crop. A positive relationship between farm employment and new COVID-19 cases could lead to new innovations to reduce risks to workers' health or to increase mechanization in agricultural production. Historically, some farms and workers have benefited from the development of labor-saving technologies while others did not (Schmitz & Seckler, 1970) . While the long-term effects of the coronavirus pandemic on agricultural production are hitherto unknown, this paper provides an initial step toward understanding the potential sources of risk to worker health and agricultural production during a pandemic.

Sector-specific employment data come from the U.S. Bureau of Labor Statistics QCEW. The QCEW records employment and wages for establishments that report to U.S. Unemployment Insurance (UI) programs. As a measure of FVH employment each month, I sum employment in orange groves, citrus groves, greenhouse, nursery, and floriculture production, vegetable and melon farming, apple orchards, grape vineyards, strawberry farming, other berry farming, fruit and tree nut combination farming, and other non-citrus fruit production (NAICS 11131, 11132, 11141, 11142, 1112, 111331, 111332, 111333, 111334, 111336, and 111339) . I also include a proxy for employees of Farm Labor Contractors (NAICS 115115). Importantly, I drop all counties that do not report any agricultural employment in FVH sectors, grain and oilseed sectors (NAICS 1111), other crop sectors (NAICS 1119), or livestock sectors (NAICS 112).

I make a minor adjustment to the NAICS 115115 employment by FLCs, since employees of FLCs may not work in the county of their employer's address. I proxy for the FLC employment for each county-month observation based on the county share of contract labor expenditures in each state in the 2017 Agricultural Census. I multiply the Agricultural Census county share of contract labor expenditures in the state by the QCEW number of FLC employees in the state each month. I use this product as a proxy for county FLC employment, and if data in the Agricultural Census are missing, I replace the proxy with the county FLC count in the QCEW. The July FLC counts by county in the QCEW and the FLC proxy based on county shares of contract labor expenditures in the 2017 Agricultural Census are plotted in the maps in Figure 1 . As expected, the locations of FLC workers in the proxy are more geographically spread than the counts taken directly from the QCEW. As a robustness check, I repeat the analysis using the county-by-month FLC counts directly from the QCEW instead of the proxy. Results are in the appendix. Figure 2 plots the 2019 national monthly employment in FVH (including FLC) sectors, grain and oilseeds, other crops, livestock, and post-harvest agricultural activities. FVH and FLC employment range from about 375-540 thousand workers, peaking in July. Grain and oilseed sectors employ only a maximum of 40,000 workers during the year, other crops about 55,000, and livestock a little less than 234,000. Post-harvest crop activities employ only a maximum of 85,000 workers. F I G U R E 2 National monthly employment in fruit, vegetable, and horticultural (FVH) and other agricultural sectors (2019). Notes: FVH employment includes employment on orange groves, citrus groves, greenhouse, nursery, and floriculture production, vegetable and melon farming, apple orchards, grape vineyards, strawberry farming, other berry farming, fruit and tree nut combination farming, and other non-citrus fruit production (NAICS 11131, 11132, 11141, 11142, 1112, 111331, 111332, 111333, 111334, 111336, and 111339) F I G U R E 4 Geographic variation in peak fruit, vegetable, and horticultural (FVH) employment (July 2019). Notes: FVH employment includes employment on orange groves, citrus groves, greenhouse, nursery, and floriculture production, vegetable and melon farming, apple orchards, grape vineyards, strawberry farming, other berry farming, fruit and tree nut combination farming, and other non-citrus fruit production (NAICS 11131, 11132, 11141, 11142, 1112, 111331, 111332, 111333, 111334, 111336, and 111339) , and Farm Labor Contractors [Color figure can be viewed at wileyonlinelibrary.com] One concern with the QCEW is that industry employment is sometimes suppressed when there are few firms located in the county and identification of individual firms might be possible. This causes an undercount of employees in some counties. Measurement error from suppression should theoretically be quite small, since only counties with few firms (and thus likely relatively few employees) will be suppressed. However, there may be exceptions and as a robustness check, I repeat the analysis dropping all counties that have suppressed employment in any of the FVH sectors listed earlier, including FLCs. Results are reported in the appendix.

A second concern is that the QCEW sums employment by industry using UI records, and employers in some states require employers to report H-2A to UI while others do not. H-2A workers constituted an estimated 10% of the full-time equivalent crop workforce in 2019 (Costa & Martin, 2020) . North Carolina, Georgia, Florida, Washington, and California employed more H-2A workers than any other states from 2015 to 2019, and in 2019 (and most previous years), H-2A employment in the least of these states was more than twice that of the next state. 10 Undercounting H-2A workers, particularly in the top five states, could be of concern for this analysis. Personal phone conversation with North Carolina Division of Employment Security in 2019 revealed that North Carolina does not record H-2A workers in the UI. Georgia was unable to reveal whether farms report H-2A in the UI, and according to Rural Migration News (2020) Florida does not record H-2A employees in the UI, while California and Washington do. As a robustness check, I repeat the analysis, dropping all counties with suppressed FVH employment and all counties in North Carolina, Georgia, and Florida. Results are reported in the appendix. 11 COVID-19 case and death data come from the New York Times, which compiles cumulative counts of coronavirus cases and deaths at the county level over time from state and local governments and health departments. I assume that missing values in the COVID-19 data are zeroes, since most counties did not report COVID-19 cases and deaths prior to detecting COVID-19 in the county. Note that there is uncertainty and measurement error in COVID-19 case and death data. 12 However, these data should serve as a reasonable proxy for monthly COVID-19 incidence.

Data are summarized in Table 1 . Employment is measured in hundreds of workers. The mean number of FVH workers (excluding FLCs) in the sample is 141 per county-month observation. Mean FLC employment is 86 workers per county-month, whether using the proxy or the QCEW count. However, the standard deviation for the proxy for FLC workers is smaller than that of the QCEW count. There is a mean of 35 post-harvest workers per county-month observation. The mean number of COVID-19 cases per county grew from 327.67 in April to 2,195.76 in August. Mean COVID deaths rose from 16.44 to 59.69. There was a mean of 428.76 COVID-19 cases and 11.76 deaths per county-month observation.

I measure the association between month-to-month variation in agricultural employment and COVID-19 growth within counties by estimating the following equation:

where Y i,m is the outcome of interest in county i in month m in 2020, FVH i,m is employment in FVH sectors in county i month in 2019, hr k,i,m is employment in another "higher risk" industry k in 2019, γ s,m is a vector of state-by-month fixed effects, April i Á η m is a vector of indicator variables for county exposure to COVID-19 in April 2020 interacted with month fixed effects, ρ i is a vector of county fixed effects, and ε i,m is the error term.

Outcomes of interest include the inverse hyperbolic sine transformation (denoted arcsinh) of new COVID-19 cases reported in the county each month, and new COVID-19 cases and deaths per capita. Previous literature examines the effects of shelter-in-place orders (Dave, Friedson, Matsuzawa, & Sabia, 2020b; Friedson et al., 2020) and early college spring break (Mangrum & Niekamp, 2020) on the natural log of COVID-19 cases, thus accommodating a nonlinear relationship between the explanatory variable and COVID-19 incidence, which is likely appropriate when modeling a contagious virus. This paper differs from previous literature in that I examine COVID-19 spread primarily in rural counties, many of which had none to a few COVID-19 cases in some months from April to August. To account for a potentially nonlinear relationship between agricultural employment variation and COVID-19 case growth and to reduce the influence of outliers (Bellemare & Wichman, 2020) , I use the inverse hyperbolic sine transformation (denoted arcsinh) of COVID-19 cases. The inverse hyperbolic sine transformation is similar to the natural log with the important distinction that it is defined even for values of zero, 13 which is critical in my analysis of rural counties, since many had zero confirmed COVID-19 cases at the start of the panel. Second, I measure effects on COVID-19 cases and deaths per capita, using as a proxy for county population the 2019 monthly county workforce reported by the Bureau of Labor Statistics. I use the workforce because it is reported at the county-by-month level and county population may vary with seasonal labor migration. It is Note: Fruit, vegetable, and horticultural employment includes employment on orange groves, citrus groves, greenhouse, nursery, and floriculture production, vegetable and melon farming, apple orchards, grape vineyards, strawberry farming, other berry farming, fruit and tree nut combination farming, and other non-citrus fruit production (NAICS 11131, 11132, 11141, 11142, 1112, 111331, 111332, 111333, 111334, 111336, and 111339 noteworthy that workforce measures include individuals who are not currently employed but looking for work. Thus, it is a reasonable alternative to a population count. The controls for "higher risk" industries include employment in industries that typically experience seasonal changes in labor demand and where workers cannot easily work remotely. These industries, defined by subscript k, include post-harvest crop activities (which may be highly correlated with FVH employment as this is a downstream stage in the food supply chain), construction, retail trade, and accommodation and food. 14 Since employment measures come from 2019, prior to the COVID-19 pandemic or any knowledge of it, the pandemic had no effect on hiring decisions in the explanatory variable. State-bymonth fixed effects control for statewide growth in COVID-19, including shocks to COVID-19 growth potentially due to changes in state health mandates. County fixed effects control for timeinvariant characteristics of the county. Standard errors are clustered at the county.

One might be concerned that COVID-19 grew faster in counties that had higher exposure to COVID-19 near the start of the pandemic, and this might be correlated with historical changes in farm employment during the agricultural growing and harvesting seasons. To account for this possibility, I additionally control for a vector of indicator variables for intensity of COVID-19 exposure as of April 2020 interacted with month fixed effects. I create the vector for COVID-19 intensity using indicator variables that divide counties into 41 bins based on roughly equal frequency grouping intervals barring the division of counties with equal number of COVID cases as of April. I say "roughly equal" because there is a large concentration of counties with no cases as of April 2020 (11.62% of the sample) or only 1 case (5.43% of the sample). Most of the other bins contains approximately 2% of counties in the sample. 15 As a robustness check, I also repeat the analysis from April to December, controlling for county-specific trends. Inclusion of county-specific trends absorbs most of the variation of interest in seasonal farm labor from April to August, and thus it is important to lengthen the panel. However, by extending the panel into the fall, one might be concerned that the return of students to schools could affect the spread of COVID-19 (Mangrum & Niekamp, 2020) , potentially biasing results if correlated with changes in agricultural employment. I address this issue by dropping counties with a large postsecondary institution. 16 Results are qualitatively similar to the main results and reported in the appendix.

It is important to note the limitations of causal interpretation in this analysis. Specifically, agricultural labor activity may be correlated with other activities that are relatively high risk for spreading the COVID-19 virus. I expect that many correlated activities are absorbed in state-bymonth fixed effects, but changes in housing density, communal living, or shared transportation, all of which may be directly related to changes in farm labor, will not be absorbed in fixed effects. This investigation does not provide sufficient evidence to suggest that on-farm activities themselves increase the risk of spreading COVID-19, but it does provide insights into which crops and activities likely bear greatest risk of COVID-related disruptions to labor supply.

Main results are presented in Table 2 . The dependent variable in columns 1-2 is the inverse hyperbolic sine transformation of new COVID-19 cases at the county-month. The dependent variable in columns 3-4 is new COVID-19 cases per 100,000 individuals in the workforce, and the dependent variable in columns 5-6 is COVID-19 deaths per 100,000. All specifications include county fixed effects, state-by-month fixed effects, and controls for county level of T A B L E 2 Association between historical monthly fruit, vegetable, and horticultural (FVH) employment and incidence of COVID-19 within counties April-August 2020

(1) Robust standard errors clustered at the county. arcsinh is the inverse hyperbolic sine transformation. Employment by industry is measured in hundreds of workers at the county-month.

FVH employment includes employment on orange groves, citrus groves, greenhouse, nursery, and floriculture production, vegetable and melon farming, apple orchards, grape vineyards, strawberry farming, other berry farming, fruit and tree nut combination farming, and other non-citrus fruit production (NAICS 11131, 11132, 11141, 11142, 1112, 111331, 111332, 111333, 111334, 111336, and 111339) , and Farm Labor Contractors (FLCs). FLCs are employees of Farm Labor Contractors using the proxy calculated from the 2017 Agricultural Census county share of contract labor expenditures in the state interacted with the QCEW number of FLC employees in the state in 2019 (NAICS 115115). All specifications include county fixed effects, state-bymonth fixed effects, and an indicator variable for level of COVID-19 exposure in April 2020 interacted with month fixed effects.

*p < 0.1, **p < 0.05, ***p < 0.01. exposure to COVID-19 in April interacted with month fixed effects. Even columns additionally include controls for historical monthly employment in other seasonal industries where working from home might not be possible. Explanatory variables of monthly employment by sector are measured in hundreds of workers. The results in columns 1-2 show that 100 additional workers in the historical monthly FVH workforce were associated with 2.27%-4.54% more COVID-19 cases within counties. 17 These semi-elasticities are based on mean FVH employment (including employees of FLCs) of T A B L E 3 Association between historical monthly crop employment and incidence of COVID-19 within counties April-August 2020 Note: Robust standard errors clustered at the county. arcsinh is the inverse hyperbolic sine transformation. Employment by industry is measured in hundreds of workers at the county-month. Fruit, vegetable, and horticultural employment includes employment on orange groves, citrus groves, greenhouse, nursery, and floriculture production, vegetable and melon farming, apple orchards, grape vineyards, strawberry farming, other berry farming, fruit and tree nut combination farming, and other non-citrus fruit production (NAICS 11131, 11132, 11141, 11142, 1112, 111331, 111332, 111333, 111334, 111336, and 111339) . Farm Labor Contractors (FLCs) are employees of Farm Labor Contractors using the proxy calculated from the 2017 Agricultural Census county share of contract labor expenditures in the state interacted with the QCEW number of FLC employees in the state in 2019 (NAICS 115115). All specifications include county fixed effects, state-by-month fixed effects, and an indicator variable for level of COVID-19 exposure in April 2020 interacted with month fixed effects. *p < 0.1, **p < 0.05, ***p < 0.01. 227 workers per county-month. Similarly, 100 additional historical FVH workers were associated with 15.78-18.65 additional COVID-19 cases and 0.21-0.34 deaths per 100,000 individuals in the workforce. All coefficients are statistically significant at the 1% level. I find a statistically significant negative coefficient on historical employment in post-harvest crop activities in columns 2 and 6. However, given that post-harvest crop employment is correlated with FVH employment, one should use caution in interpreting these coefficients apart from the coefficients on FVH employment. I find that 100 additional construction workers were associated with 47.02 additional COVID-19 cases and 1.27 additional deaths per 100,000 individuals in the workforce. I do not find statistically significant coefficients on employment in other seasonal non-farm sectors. More importantly, inclusion of the additional controls leads to slightly larger coefficients on FVH employment, but no substantive changes on the coefficients of interest. Table 3 shows the results from measuring the association between historical employment in more specified crop industries and COVID-19 cases and deaths in 2020. The dependent variable in the first column is the inverse hyperbolic sine transformation of new COVID-19 cases at the county-month, in the second column, new COVID-19 cases per 100,000 individuals in the workforce, and in the third column, new COVID-19 deaths per 100,000. The results show a statistically significant positive relationship between COVID-19 incidence and employment on grape vineyards, other non-citrus fruit farms, and in greenhouses.

The coefficients on greenhouse employment in columns 1 and 2 are very large (though statistically significant only at the 10% level). These results would suggest that 100 additional greenhouse employees in 2019 were associated with a 2.28% rise in COVID-19 cases within counties or 305 new cases per 100,000 individuals in the workforce. Risk of COVID-19 spread might be particularly high among greenhouse employees because work is performed indoors. However, one should take caution in interpreting coefficients causally, as employment in various fruits and vegetables may be correlated. Results are nevertheless suggestive of what fruits and vegetables might have been exposed to particularly high risk of COVID-related labor disruptions in 2020.

Coefficients on strawberry employment and other berries might not be statistically significant because there are relatively few counties where berries are grown. Citrus fruit was mostly harvested prior to April, so it would be unlikely to see a statistically significant relationship to COVID-19 incidence. Finally, some vegetables and melons are mechanically harvested and may require less physical interaction between workers compared with grapes and other non-citrus fruit. However, it is notable that FLC employment, which is often used to mitigate the risk of farm labor shortages, was associated with a statistically and economically significant increase in the incidence of COVID-19. Employment of 100 additional FLC workers in 2019 was associated with 2.58% more COVID-19 cases within counties and 31.16 additional COVID-19 cases and 0.51 deaths per 100,000 individuals in the workforce. I find no significant association between employment in more mechanized crops, including grains and oilseeds and other crops (mostly field crops), or animal and livestock and new COVID-19 cases or deaths. This is not surprising given the relatively small number of workers needed to harvest these crops.

The findings in this paper suggest that there is a large, statistically significant positive association between month-to-month changes in agricultural employment within counties and new COVID-19 cases. However, the analysis does not illuminate the precise mechanisms of this relationship. It is notable that increased FLC employment is associated with positive, statistically significant estimated marginal effects. Employees of FLCs are migratory and thus may be both more susceptible to picking up the virus as they travel between farms and to spread viruses with other workers who may share housing, transportation, or social activities.

Numerous factors may correlate with both changes in agricultural employment and new COVID-19 cases. For example, the majority of farm workers in the United States are Hispanics, and Hispanics appear to be particularly susceptible to COVID-19. In California, there were 1750.3 COVID-19 cases for every 100,000 Hispanic residents in August 2020, compared with just 514.3 cases per 100,000 White residents and 438.9 cases per 100,000 Asian residents (Ibarra et al., 2020) . 18 In addition, newspaper reports suggest that misinformation surrounding COVID-19 is more prevalent in Hispanic communities, likely stemming from greater distrust in government, worse access to medical care, and language barriers (Klepper et al., 2020) . If Hispanics are more susceptible to COVID-19 than other races, then the migration of Hispanic farm workers into communities as agricultural labor demand rises, or increased activity and increased interactions between different social circles of Hispanic households during peak farm labor seasons, could increase COVID-19 cases within the county.

The results do not indicate that COVID-19 is necessarily spreading on farms, and understanding how or why COVID-19 is spreading in farm labor communities is beyond the scope of this paper. Many labor activities on farms could involve relatively high risk of spreading COVID-19, particularly when it is difficult to distance workers. Nevertheless, many farms and FLCs have taken precautions to reduce social contact on the farm, including assigning workers to the same work crews each day, taking additional trips to transport workers to worksites on buses, taking workers' temperatures each day, spreading workers in the field, and providing additional wash stations in the fields. If workers share seasonal housing with one another, cook and eat meals together, or share childcare responsibilities among several families, among other shared activities, then COVID-19 could spread in seasonal farm labor communities even if the work environment is relatively safe.

Identifying the mechanisms that link labor-intensive agricultural crop production to new COVID-19 cases is beyond the scope of this paper. Yet, the positive relationship between FVH employment and the incidence of new COVID-19 cases found in this paper illustrates how a pandemic like COVID-19 could drastically increase labor uncertainty and health risk during critical stages of agricultural production. Traditionally, FLCs have helped reduce labor supply uncertainty and labor market frictions by matching employers to work crews for seasonal or temporary tasks. However, if FLC workers are similarly vulnerable to catching or spreading the virus, as the results in this paper suggest, hiring workers through FLCs may not be a viable farm labor management strategy.

Despite growing COVID-19 incidence in geographic locations and months when farm employment was increasing, there have been relatively few major disruptions in agricultural production. According to , there were few COVID-19 outbreaks on farms in 2020, and fruit shipments remained steady throughout the year. Nevertheless, heightened farm labor supply uncertainty and increased labor costs could induce more producers and managers up and down the food supply chain to make increased investments to shield workers from healthrelated risks and agricultural production from input supply uncertainty. This could take the form of investments in a less migratory labor supply (such as through the creation of more year-round jobs), increased safety precautions in farm procedures, or labor-saving technologies, among others. 4 The workforce is estimated using monthly 2019 county-level data reported by the Bureau of Labor Statistics (BLS). 5 Statistics based on author's analysis of the NAWS conducted by the Department of Labor, Employment and Training Administration, which surveys farm workers at the workplace and is designed to be nationally representative of the crop workforce excluding H-2A workers. 6 Statistics based on author's analysis of the NAWS conducted by the Department of Labor, Employment and Training Administration, which surveys farm workers at the workplace and is designed to be nationally representative of the crop workforce excluding H-2A workers. 7 H-2A workers constitute an estimated 10% of the full-time equivalent farm workforce (Costa & Martin, 2020) . 8 Statistics based on author's analysis of the NAWS conducted by the Department of Labor, Employment and Training Administration, which surveys farm workers at the workplace and is designed to be nationally representative of the crop workforce excluding H-2A workers. 9 See Taylor Farms. "The Automated Farm" August 22, 2017. https://www.taylorfarms.com/news/theautomated-farm/. Accessed August 28, 2020. 10 Based on author's analysis of the Office of Foreign Labor Certification Disclosure Data. 11 Another source of undercounting are farms with few employees, which are not required to file with UI in certain states. Since only small employers are exempt from filing this should have a relatively small source of measurement error. 12 Note that the Centers for Disease Control and Prevention asserts that COVID-19 counts are provisional and may need to be updated. See, for example, Centers for Disease Control and Prevention (September 1, 2020) "Daily Updates for Totals by Week and State: Provisional Death Counts for Coronavirus Disease 2019 ." https://www.cdc.gov/nchs/nvss/vsrr/covid19/index.htm. Accessed September 1, 2020. 13 The inverse hyperbolic sine of is defined arcsinh x ð Þ ¼ ln x þ ffiffiffiffiffiffiffiffiffiffiffiffi x 2 þ 1 p À Á . 14 I also repeated the analysis additionally controlling for employment in transportation and warehousing, healthcare and social services, other service industry, and the meat processing sector. I do not include all of these controls in the main analysis because these industries do not typically experience large variation in employment within the year. However, the addition of more sector employment controls had little to no effect on the estimated coefficients on FVH employment. 15 I performed the analysis using 12 bins and 41 bins and leaving out April exposure controls entirely. Results are not sensitive to the number of bins used to create the vector of controls or the inclusion of these controls. 16 I define large postsecondary institutions as those with student enrollment of 10,000 or more students. Data on postsecondary institution location and enrollment come from the Integrated Postsecondary Education Data System. 17 Semi-elasticities are calculated by multiplying the coefficient by the mean value of the explanatory variable (Bellemare & Wichman, 2020) . 18 Race/ethnicity was unknown for 231,446 (34.1%) of cases in the cited study. The only ethnicity that surpasses the Hispanic COVID-19 rate is Native Hawaiian and Pacific Islander with 1834.3 cases per 100,000 residents.

ARE Update Special Issue: Implications of the Cornoavirus Pandemic on California Food

Elasticities and the Inverse Hyperbolic Sine Transformation

Racial and Ethnic Disparities Among COVID-19 Cases in Workplace Outbreaks by Industry Sector-Utah

Can Technology Compensate for a Labor Shortage? Effects of 287 (g) Immigration Policies on the

A Declining Farm Workforce: Analysis of Panel Data from Rural Mexico

Innovations for a Shrinking Agricultural Workforce

Immigration Restrictions as Active Labor Market Policy: Evidence from the Mexican Bracero Exclusion

COVID-19, U.S. Agriculture, and USDA's Coronavirus Food Assistance Program (CFAP)

Coronavirus and Farmworkers: Farm Employment, Safety Issues, and the H-2A Guestworker Program

The Contagion Externality of a Superspreading Event: The Sturgis Motorcycle Rally and COVID-19

When Do Shelter-In-Place Orders Fight COVID-19 Best? Policy Heterogeneity across States and Adoption Time

Every Single Worker Has Covid at One U.S. Farm on Eve of Harvest

Why Do Fewer Agricultural Workers Migrate Now?

Did California's Shelter-In-Place Order Work? Early Coronavirus-Related Public Health Effects

Farm Labor Productivity and the Effect of Mechanization

Duvall and Ford Say Health of Members '21 Priority

Is There A Farm Labor Shortage?

Coronavirus in California by the Numbers

Is ICE Freezing US Agriculture? Impacts of Local Immigration Enforcement on US Farm Profitability and Structure

Distrust of Authority Fuels Virus Misinformation for Latinos

The Impact of Immigration Enforcement on the U.S. Farming Sector

Technology Adoption under Production Uncertainty: Theory and Application to Irrigation Technology

The Importance of H-2A Guest Workers in Agriculture

The Impact of LAWA on the Family Labour Supply among Farm Households

JUE Insight: College Student Travel Contributed to Local COVID-19 Spread

COVID-19 and Farm Labor after 6 Months

Migrant Farmworkers, Many Coronavirus Positive, Move North from Florida to Other States

Immigration Reform and Farm Labor Markets

A Tale of Two States: Farm Labor in CA and FL

Mechanized Agriculture and Social Welfare: The Case of the Tomato Harvester

The Future of the Pandemic in the U.S.: Experts Look Ahead

Worker Turnover, Farm Labor Contractors, and IRCA's Impact on the California Farm Labor Market

Prevalence and Predictors of SARS-Cov-2 Infection Among Farmworkers in Monterey County, CA

Farm Labor Markets in the United States and Mexico Pose Challenges for U.S. Agriculture

This paper measures the relationship between historical month-to-month variation in agricultural employment within counties and new cases of COVID-19 and deaths each month from April to August 2020. The results show that 100 additional workers in the 2019 monthly FVH workforce in were associated with 4.5% more COVID-19 cases within counties. Similarly, 100 additional FVH workers were associated with 25.17 additional COVID-19 cases and 0.48 COVID-19 deaths per 100,000 individuals in the workforce. If farm workers are less likely to get tested for COVID-19 as some reports suggest (UC Berkeley School of Public Health, 2020), then these estimated coefficients could be lower-bounds for the actual association between FVH employment and the incidence of new COVID-19 cases and deaths.All specifications include controls for state-by-month fixed effects and indicator variables for level of exposure to COVID-19 in April 2020 interacted with month fixed effects. However, this does not preclude the possibility that other social activities correlated with agricultural employment could be responsible for the observed relationship between labor-intensive agricultural activities and COVID-19 spread. The findings from this paper have important implications for agricultural workers, producers, and consumers. Potential health risks to workers are a public concern, and new strategies may need to be implemented to help protect farm workers' safety. Outbreaks of COVID-19 in agricultural communities when farm employment is on the rise could also increase the risk of farm labor shortages at critical stages of production and harvest. Labor shortages could be costly to individual producers, and in the extreme event, lead to regional or national food shortages.Understanding sector-specific vulnerabilities associated with unanticipated shocks to worker health risk and labor supply is paramount for policymakers, industry leaders, rural planners, and agricultural producers to make informed risk management strategies. Employer policies to increase social distancing in the workplace likely slow the spread of COVID-19. However, these precautions come at considerable cost to the employer and do not entirely remove the risk of COVID-19 outbreaks.One of the often used strategies to reduce labor market frictions when the farm labor supply is tight is to hire workers through FLCs. However, findings in this paper suggest that FLC crews were similarly vulnerable to the spread of COVID-19 in 2020. Investments in new labor-saving or labor-augmenting technologies that increase labor productivity were already underway for FVH crops prior to the pandemic due, in large part, to the tightening farm labor supply over the past several decades (Charlton et al., 2019; Hamilton et al., 2020) . Increased risk with respect to worker safety and labor supply during the pandemic might induce more producers to invest in mechanized solutions and labor-saving technologies as a risk management strategy (Koundouri et al., 2006) . Long-term effects of the coronavirus pandemic on agricultural production are beyond the scope of this paper. In time, future research may illuminate some of the long-term effects. 

Additional supporting information may be found in the online version of the article at the publisher's website.How to cite this article: Charlton, Diane. 2021. "Seasonal farm labor and COVID-19 spread." Applied Economic Perspectives and Policy 1-19. https://doi.org/10.1002/aepp. 13190