key: cord-0804069-96dlri53
authors: Zeballos, Eliana; Dong, Xiao
title: The effect of COVID‐19 on food sales
date: 2021-10-13
journal: Appl Econ Perspect Policy
DOI: 10.1002/aepp.13201
sha: bcd285150d0ce77b24568a7e72568a390767ebf3
doc_id: 804069
cord_uid: 96dlri53

Both COVID‐19 pandemic‐related restrictions and recessionary employment loss severely impacted US food sales during 2020. This article estimates the historical relationship between food expenditures and employment at the county level. Using these estimates, we simulate the impact of the loss of employment on food sales and find that, on average, employment loss increased food‐at‐home (FAH) sales by 1.3% and decreased food‐away‐from‐home (FAFH) sales by 0.5% in 2020. We argue differences to the actual 4.8% increase in FAH sales and 19.5% decrease in FAFH sales in 2020 likely stem from the more drastic COVID‐19 pandemic‐related restrictions and behavioral changes.

FAH sales and 19.5% decrease in FAFH sales, respectively, that the United States Department of Agriculture (USDA) ERS's FES reports.

The results of this study make several contributions to the literature and highlight potential economic challenges at the county level. First, we extend the literature on shifts in the overall household food expenditures and FAH and FAFH spending due to recession-driven economic shocks. Second, we focus on the change in food expenditures driven by changes in local economic conditions as we rely on county-level data. Third, we estimate the likely shift in food spending caused by the economic shocks measured by employment rates at the county level in 2020, enabling identification of the counties that potentially experienced the most salient economic effects of COVID-19. Finally, we present evidence that on average and measured by employment levels, the economic effects of COVID-19 on FAH sales, and especially FAFH sales are small compared to the actual changes in food spending at the aggregate level in 2020. From a policy perspective, by isolating the economic effects and pandemic effects on food sales, we can shed light on how food sales will change in response to the potential separate effects from lifting restrictions and economic recovery.

In this article and building on Dong and Zeballos (2021) , we combine and use three different datasets: (1) the National Establishment Time Series (NETS) to examine county-level food spending from 1990 to 2019; (2) employment data at the county level as an indicator of local economic conditions and the severity of the economic shocks in each county from 1990 to 2020; and (3) the FES to measure aggregate changes in FAH and FAFH expenditures from 1990 to 2020.

To examine county-level food spending across time and among different food establishments, we use sales at FAH and FAFH establishments in each county from the NETS database. NETS is a longitudinal database that records the sales, employment figures, growth, and performance against industry peers for specific business locations across time. Dun and Bradstreet and Walls and Associates created NETS by using Dun and Bradstreet's archival data from surveys of establishments (Walls & Associates, 2013) . Each unit of observation in NETS reports the annual sales revenue and the number of employees for each establishment from 1990 to 2019. Due to the detailed and granular level of the data, NETS has been intensively used in recent studies that rely on yearly variations for identification as the database was created to focus on the establishment and firm-level growth and performance across time. As NETS tracks the sales, growth, and number of employees for individual-level establishments, it provides a rich and unique panel dataset that allows us to examine the change in food sales across different food establishment types in varying geographic locations. NETS contains business establishments from a comprehensive list of industries, and the database categorizes establishments using the NAICS and the Standard Industrial Classification numeric codes. To identify food spending at the county level, we use NAICS codes to select only the establishments that sell food, and we base our selection on the methodology used by the USDA, ERS's Food Expenditures Series.

Recent literature has shown NETS captures the food environment relatively well compared to the official Bureau of the Census' Economic Census, and NETS is ideal for researchers attempting to look at the food environment over time (Cho et al., 2019) . Although the number of food establishment employees has similar aggregate trends compared to the County Business Pattern, which is an annual series that provides subnational economic data by industry, aggregate food sales trends do not align with trends observed by the FES and Economic Census. This is likely due to a significant portion of sales data in NETS being imputed based on firm-level employment numbers and in particular with estimated sales based on employment data (Barnatchez et al., 2017; Crane & Decker, 2019) . 5 The sales imputation potentially raises concerns about using NETS data for certain research questions surrounding firm and business dynamics (Barnatchez et al., 2017; Crane & Decker, 2019) as a significant portion of sales data in NETS is imputed based on firm-level employment numbers and in particular with estimated sales based on employment data. Therefore, to accurately estimate food sales, we perform a correction by applying a sales per employee ratio, which is calculated using the Economic Census for each North American Industry Classification System (NAICS) code for the most relevant food industries by state and firm size. Our corrected food sales compare well against the aggregate trends observed by the FES.

To further identify FAH and FAFH spending, we use product and services codes (PSCode), which the Bureau of the Census also uses to prepare product line reports that details the percentage of sales by product and contributing industry. We select all the PS codes related to food and nonalcoholic beverages and classified them as FAH if they were sold for off-premise consumption or as FAFH if they were sold for on-premise consumption used by the USDA, ERS's Food Expenditures Series. Using these percentages, we then calculate the portion of FAH and FAFH sales for each establishment in NETS using its NAICS code.

We use overall total employment at the county level as an indicator of local economic conditions and the severity of the economic shocks in each county from 1990 to 2020. The Burau of Labor Statistics (BLS) constructs county-level employment and unemployment estimates by combining state-level data from its current population survey with county-level counts of unemployment insurance claimants. Data on annual county-level unemployment and employment start in 1990, similar to the NETS database. 6 To estimate the effect of historical economic conditions and shocks on food expenditure, we use annual county-level employment data from 1990 to 2019, which corresponds to the timespan of the NETS data in our regression analysis. 7 To simulate the change in food sales at the county level due to the COVID-19 recession, we use the most recent 2020 county-level employment figures from BLS (2021). The BLS publishes monthly county-level employment numbers for the latest 18 months, and we use the average employment figures from January 2020 to December 2020.

To measure aggregate changes in FAH and FAFH expenditures from 1990 to 2020, we use the FES from the USDA-Economic Research Service. 8 The ERS FES is a comprehensive data set that measures the US food system by quantifying the value of food acquired in the United States by type of product, outlet, and purchaser. The data series measures the value of food acquired, including food and beverage sales (as well as taxes and tips), and the value of food produced at home, donated and furnished to employees and institutionalized persons. Figure 1 shows the historical trends in FAH and FAFH spending across time adjusted for inflation. Spending on both FAH and FAFH has been gradually increasing over the last eight decades, except for decreases in food spending during major economic shocks such as the Great Recession. However, FAFH spending is trending upwards more rapidly over time and surpassed FAH expenditures in 2004. FAFH and FAH both comprise about half of the shares of total food expenditures. While both FAFH and FAH decreased during the Great Recession (2008-09), FAFH decreased more, showing the more salient effect that the recession had on FAFH versus FAH.

In this article, we first estimate the historical relationship between FAH and FAFH expenditures and changes in employment levels at the county level for the years 1990-2019. Second, as we lack food sales at the local level for 2020, we use the estimates from the historical relationship to simulate the impact of employment loss on food sales during 2020 at the county level.

Finally, we aggregate the simulated values to compare with available national-level food expenditure data.

To econometrically estimate the historical relationship between FAH and FAFH expenditures and changes in employment levels at the county level between 1990 and 2019, we use the following specification. 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 We regress the growth rate of food sales on the growth rate of employment at the county level where c indexes county and t indexes year. We approximate the growth rate of food sales by differencing the log of food sales and the growth rate of employment by differencing the log of employment. By differencing the log of food sales, we control for potential seasonality and nonstationarity in the data. We estimate the effects of a recession by using the dummy variable recession t , where recession t equals 1 for the years 1990, 1991, 2001, 2008, and 2009 (periods of recession according to the National Bureau of Economic Research [NBER]). 9 The NBER definition for recession is generally presented quarterly. We assume if any of the recession falls into a year for even a quarter (as in 1991), the recession dummy equals 1 for that year. 10 We rely on the local employment rate from 1990 to 2019 to capture the salience of the economic effect of COVID-19, which is likely to be heterogeneous across space. The discrete binary national variable "recession" serves as a "catch-all" for nationwide effects of entering into a recession such as increased federal stimulus, expansion of benefits, and monetary policy changes. Finally, we control for local time-invariant characteristics with county fixed effects. 11 We deflate the dollar value for the sales using the consumer price index for FAH and FAFH sales from BLS.

To simulate the economic effects of COVID-19 on total food sales, FAH sales, and FAFH sales, we apply the resulting estimates of Equation 1 into Equation 2. We multiply β 1 , the marginal effect of the growth of county-level employment on the growth of food sales, by the county-level employment growth in 2020. We add β 3 , the effect of a recessionary year on the growth of food sales. Table 1 shows the results of the regression utilizing Equation 1. In the first column, we use the growth of total food sales as the dependent variable and show that a one percentage point decrease in the growth of a county's employed population is associated with a 0.08 percentage point decrease in the growth of total food sales. A recessionary year is associated with a 0.86 percentage point decrease. We further run separate regressions in the second column and third columns with the growth of FAH and FAFH sales as the dependent variable, respectively. Results show that the growth of FAH and FAFH sales are associated with 0.08 and 0.13 percentage point decreases in response to a one percentage point decrease in the county growth of the employed population. However, a recessionary year significantly increases the growth of FAH sales by 0.34 percentage points and decreases the growth of FAFH sales by 3.7 percentage points. Table 2 presents the results of separate regressions with the growth of sales of supermarkets, supercenters, other FAH, limited-service restaurants, full-service restaurants, and other FAFH as the dependent variable. Supermarkets and other grocery (except convenience) stores are establishments primarily engaged in retailing a general line of food. Warehouse, clubs, and supercenters are establishments primarily engaged in retailing a general line of groceries, including a significant amount and variety of fresh fruits, vegetables, dairy products, meats, and other perishable groceries, in combination with a general line of new merchandise, such as apparel, furniture, and appliances. Other FAH includes convenience stores, specialty food stores, all other general merchandise stores, and home delivery and mail orders. Limited-service restaurants are establishments primarily engaged in providing food services (except snack and nonalcoholic beverage bars) where patrons generally order or select items and pay before eating. Full-service restaurants are establishments primarily engaged in providing food services to patrons who order and are served while seated (i.e., waiter/waitress service) and pay after eating. Other FAFH includes drinking places, food service contractors, and other eating retailers that sell food for on-premise consumption.

Results show that a one percentage point decrease in the growth of a county's employed population is associated with decreases in the growth of sales in every food establishment, with other FAFH places being affected the most (0.21 percentage point decrease) and supermarkets and other grocery stores being affected the least (0.08 percentage point decrease). While a recessionary year has no significant effect on the growth of supermarkets and other grocery stores sales, it leads to a 0.87 and 0.90 percentage point increase in warehouse clubs and supercenters, and other FAH sales, respectively. On the contrary, results show that both growth sales of limited and full-service restaurants decrease by 3.4 percentage points and 4.0 percentage points, respectively. Table 3 presents the simulated impacts of the loss of employment on sales of total food, FAH, supermarkets, supercenters, other FAH, FAFH, limited-service restaurants, full-service restaurants, and other FAFH in 2020. On average, total food spending was estimated to increase by 0.8% in 2020 due to the economic effects of COVID-19. FAH sales were estimated to increase by 1.3%, and FAFH sales were estimated to decrease by 0.5%. Food sales at supermarkets and other grocery stores were estimated to decrease by 0.3% while warehouse clubs and supercenters and other FAH store food sales were estimated to increase by 5.1% and 3.2%, respectively. Full-service restaurants sales and other FAFH sales were estimated to decrease by 1.9% and 0.8%, respectively, while limited-service restaurant sales were estimated to increase by 1.7% in 2020. Figure 2 shows the simulated growth in total food sales for each county in the United States in 2020 due to the economic effects of COVID-19. As the economy recovers Each column is regressed separately. Other FAH includes convenience stores, specialty food stores, all other general merchandise stores, and home delivery and mail orders. Other FAFH includes drinking places, food service contractors, and other eating retailers that sell food for on-premise consumption. Recession: Dummy variable is equal to 1 if the year is equal to 1990, 1991, 2001, 2008, and 2009 , and zero otherwise. The National Bureau of Economic Research definition for recession is generally presented quarterly. We assume if any of the recession falls into a year for even a quarter (as in 1991), the recession dummy equals 1 for that year. Additional controls include county fixed effects. Robust standard errors in parentheses. Abbreviations: FAFH, food-away-from-home; FAH, food-at-home. F I G U R E 2 Simulated growth of total food sales in 2020, per county. The first quartile are counties that experienced a decrease in total food sales and an increase up to 0.6%; the second quartile are those counties with an increase in total food sales between 0.6% and 0.8%; the third quartile are those counties with an increase in total food sales between 0.8% and 0.9%; and the fourth quartile are those counties with the largest increase in total food sales (>0.9%). Source: USDA, Economic Research Service (ERS) using data from the National Establishment Time Series and the Bureau of Labor Statistics [Color figure can be viewed at wileyonlinelibrary.com] and employment rates increase, this map helps provide some insight on the changes in food sales at the county level due solely to these changes. The magnitude of the change varies by location as seen on the broad spectrum of color changes. One notable regional pattern is that most of the Midwest and the northern part of the West are in the fourth quartile, the largest increase, shown in blue. When comparing the simulated impacts of the loss of employment on the growth of food sales to the actual growth in food sales captured by ERS's FES (Table 3) in 2020, we find that while we estimate an increase of 0.8% in total food sales, on average, FES shows that food sales decreased by 7.8%. Furthermore, results show that, on average, losses of employment are estimated to have increased FAH sales by 1.3% and decreased FAFH sales by 0.5% in 2020. These numbers are dwarfed by the 4.8% increase in FAH sales and 19.5% decrease in FAFH sales, respectively, that the FES reports ( Figure 3 ).

The novel coronavirus (COVID-19) was declared a pandemic in early March 2020 and has led to severe economic effects, including loss of employment and a recession. COVID-19 has also significantly altered the entire food supply chain from the agricultural sector (Johansson et al., 2020) to consumer food retail sales (Ellison et al., 2020) . In this study, we estimate the historical relationship between food expenditures and employment levels at the county level and simulate the impact of the loss of employment on food sales during 2020. We then aggregate these simulated estimates and compare them with actual food expenditures measured by the USDA ERS's FES.

Results show that loss of employment leads to larger decreases in FAFH sales compared to FAH sales. Moreover, during a recessionary year, FAH sales increase while FAFH sales decrease. The increase in FAH sales and decrease in FAFH sales during recessionary years indicates substitution toward FAH sales. This is in line with economic theory as households typically treat FAFH as normal goods. We also see that FAFH sales are more elastic than FAH sales to local economic conditions and national economic conditions. Moreover, the shift in food shares, specifically between FAH and FAFH spending, can have implications for diet quality and present another avenue of the impact of economic shocks on health outcomes. A wide literature explores how economic shocks and job loss influence health outcomes in both the United States and other developed countries (Deb et al., 2011; Dehejia & Lleras-Muney, 2004; Economou et al., 2008; Ruhm, 2000 Ruhm, , 2003 Ruhm, , 2005 . Based on the most recent food intake data from the USDA, Agricultural Research Service, and monthly food spending data from the USDA, ERS, FAFH spending per calorie consumed is slightly over twice the per-calorie FAH spending. Other studies have shown that FAFH tends to be of lower nutritional quality than FAH (Lin & Guthrie, 2012) . Todd et al. (2010) found that even after controlling for individual tastes and preferences, consumption of FAFH increases total caloric intake and reduces diet quality among adults and children. 12 When looking at the relationship between employment and sales at the different types of establishments within FAH and FAFH categories, results show that a one percentage point decrease in the growth of a county's employed population is associated with a decrease in the growth of sales in every food establishment. While a recessionary year has no significant effect on the growth of supermarkets and other grocery stores sales, it leads to significant increases in warehouses clubs and supercenters, and other FAH sales, respectively. On the contrary, results show that both growth sales of limited and full-service restaurants decrease during a recessionary year. This is in line with Kim and Leigh (2011) 's findings that limited-service restaurants are "normal goods" for lower-income houses but "inferior goods" for higher-income households, whereas full-service restaurants were "normal goods" for all income levels. Limitedservice restaurants are probably less sensitive to changes in employment and recessions as limited-service restaurants are often cheaper.

Simulated estimates of the economic effects for 2020 show that, on average, total food spending increased by 0.8% caused by changes in household food spending decisions stemming from the economic impacts of COVID-19. Furthermore, results show that, on average, employment loss is estimated to have increased FAH sales by 1.3% and decreased FAFH sales by 0.5% in 2020. When we looked at traditional supermarkets, supercenters, and other FAH stores, the decrease is higher for supermarkets than for supercenters and other FAH stores, which may be partly due to household substitution toward purchasing groceries at other retailers that primarily sell in bulk and price lower. Similarly, we find that the simulated estimates for 2020 were positive for limited-service restaurants but negative for sit-down restaurants, which may be partly driven by limited-service restaurants being less expensive (the average cost of a meal in a sit-down restaurant is 2.2 times higher than in a limited-service restaurant-FES 2020). One limitation of this project is that we cannot capture seasonal substitution effects as we are estimating from yearly data that only change annually.

Finally, results show that the simulated estimates on FAH sales and FAFH sales are very small compared to the actual 4.8% increase in FAH sales and 19.5% decrease in FAFH sales, respectively. Our results show that the economic effects of COVID-19 were not the likely primary driver of food sales change in 2020, and these differences are added effects of the COVID-19 pandemic-related restrictions and behavioral changes on food sales. This seems especially likely for FAFH sales as the pandemic nature of COVID-19 led to mobility restrictions, shutdowns of restaurants, business operating limits, and consumer fears typically not seen in other recessions.

The results of this study also provide some insights into heterogeneous effects of local economic conditions on total food, FAH, and FAFH spending and estimates that the economic effects of COVID-19 on food spending vary across counties. These results may help shed light on the potential impacts of the economic recovery at the local level, which will likely be staggered. 9 Recessionary periods : July 1990 to March 1991 March 2001 to November 2001 December 2007 to June 2009  February 2020 to April 2020. 10 We perform robust checks using half year in a recession as the definition for the recession variable. This new definition only changes 1 year (1991), and the results are robust. 11 We also perform a robustness by estimating if the effects are similar between metropolitan and nonmetropolitan counties, where metro and nonmetro areas are defined by the ERS Rural Urban Continuum Codes: https://www.ers.usda.gov/topics/rural-economy-population/rural-classifications/. We see no significant difference between metro and nonmetro counties. 12 This is less clear-cut for children. Those qualifying for reduced-price school meal programs may eat healthier away from home.

The Impact of COVID-19 on United States Meat and Livestock Markets

An Assessment of the National Establishment Time Series (NETS) Database

The New Normal? US Food Expenditure Patterns and the Changing Structure of Food Retailing

Food Consumption Behavior During the COVID-19 Pandemic

COVID-19 and Food Supply Chains

Capturing the Complete Food Environment With Commercial Data: A Comparison of TDLinx, ReCount, and NETS Databases, TB-1953

Business Dynamics in the National Establishment Time Series (NETS)

How Does the Business Cycle Affect Eating Habits

The Effect of Job Loss on Overweight and Drinking

Booms, Busts, and Babies' Health

COVID-19 Working Paper: The Effects of COVID-19 on Food Sales, AP-088

Are Recessions Harmful to Health After All? Evidence from the European Union

Examining Food Purchase Behavior and Food Values During the COVID-19 Pandemic

Food Insecurity During COVID-19

Shopping Around: How Households Adjusted Food Spending Over the Great Recession

The Consumption Response to Income Changes

Unprecedented Crisis Calls for Unprecedented Policy Responses

Are Meals at Full-Service and Fast-Food Restaurants "Normal" or "Inferior"?

Food Spending Adjustments During Recessionary Times

Nutritional Quality of Food Prepared at Home and Away from Home

Consumer Food Buying During a Recession

COVID-19 Induced Stigma in US Consumers: Evidence and Implications

The Effect of Unemployment on Food Spending and Adequacy: Evidence from Coronavirus-Induced Firm Closures

Are Recessions Good for your Health?

Good Times Make you Sick

Healthy Living in Hard Times

The Impact of Food Away from Home on Adult Diet Quality, ERR-90

Changes in Eating Patterns and Diet Quality Among Working-Age Adults

Food Expenditure Series

Internal document provided to USDA, Economic Research Service. How to cite this article: Zeballos, Eliana, Xiao Dong. 2021

The findings and conclusions in this publication are those of the authors and should not be construed to represent any official USDA or the US Government determination or policy. This research was supported by the US Department of Agriculture, Economic Research Service. The analysis, findings, and conclusion expressed in the paper should not be attributed to Walls and Associates.

The authors declare no conflicts of interest.

Eliana Zeballos and Xiao Dong had equal roles in formulating the research question and writing the article. Eliana Zeballos analyzed the data. ENDNOTES 1 https://www.cdc.gov/media/releases/2020/p0121-novel-coronavirus-travel-case.html. 2 https://www.who.int/director-general/speeches/detail/who-director-general-s-opening-remarks-at-the-mediabriefing-on-covid-19-11-march-2020. 3 https://www.defense.gov/Explore/Spotlight/Coronavirus/Timeline/. 4 Decreased food spending does not necessarily indicate less food volume is consumed. Reduced food spending could stem from substitution to lower quality foods, cheaper alternatives, and fewer prepared foods that require more labor hours. 5 The main concern from Crane and Decker (2019) revolves around examining micro-level business dynamics using NETS. However, as we aggregated to the county level and focused on food sales, our results are not relevant to the concerns raised in Crane and Decker (2019) . Barnatchez et al. (2017) showed that in correlations of county-level aggregates between NETS and Census data, the official standard is high, indicating countylevel aggregates are relatively accurate. In addition, at the spatial location by industry by establishment level, correlations between NETS and official sources are reassuringly strong in appropriately restricted samples. In our study, we focused on both a restricted sample (the food industry) and aggregate to the county level, which meet both criteria suggested in Barnatchez et al. (2017) . 6 County-level labor data from BLS can be found on their website: https://www.bls.gov/lau/. 7 For the concern that sales are correlated to unemployment since firm sales are imputed on firm employment, we used overall total employment at the county level across all industries and not at the firm level nor the food sales level. Overall, employment may not be correlated with employment of specific industries during recessions, and business cycles in the food industry (consumption) are typically different than business cycles in other industries. 8 Data from USDA, Economic Research Service (2020) found online on the ERS website Food Expenditure Series.