key: cord-1045401-dadwlgpq
authors: Ziliak, James P.
title: Food Hardship during the Covid‐19 Pandemic and Great Recession
date: 2020-10-02
journal: Appl Econ Perspect Policy
DOI: 10.1002/aepp.13099
sha: 7605588ecfa15048d67d2dfb5b7618bf0a1e74ec
doc_id: 1045401
cord_uid: dadwlgpq

I compare the extent of food hardships in the United States among adults and seniors before and during the Covid‐19 Pandemic. Food insufficiency increased three‐fold compared to 2019, and more than doubled relative to the Great Recession. Food insufficiency among seniors increased 75 percent during the Covid period, but more than doubled when including reduced intake of food varieties. Receipt of charitable foods among disadvantaged adults spiked 50 percent in the Covid period, but the initial response among seniors was a sharp reduction, before rising. These patterns are consistent with strong social distancing measures enacted in response to the Pandemic.

SNAP are much lower for seniors than younger adults (Vigil 2019; Bitler, Hoynes, and Schanzenbach 2020; Moffitt and Ziliak Forthcoming) .

In this paper, I assess how food hardship in the population overall and among seniors compares in the current crisis to the two decades preceding the Covid-19 pandemic. The data for the pre-Covid period come from the 2001 to 2019 December Current Population Survey (CPS), which is the source of official estimates of food insecurity (Coleman-Jensen et al. 2020) .

Because the CPS supplement on food insecurity spanning the onset of Covid-19 will not be made available until the second half of 2021, for the post-Covid period I rely on data from the Census Bureau's Household Pulse Survey (Pulse). 1 The Pulse is a new survey fielded to provide real time information on a variety of socioeconomic and health outcomes induced by the pandemic (U.S. Census Bureau 2020). The Pulse only asks a couple of the questions found on the CPS, and thus it is not possible to directly compute food insecurity as measured officially by the USDA.

Thus, I consider two measures of food hardship, the share of adults who are food insufficient and the share receiving charitable food. Food insufficiency is a broad measure of food distress, and in fact is used as a screener for eligibility for the full food security module in the CPS (Tiehen, Vaughn, and Ziliak 2019). Charitable food is defined as receipt of free groceries or meals from nongovernmental organizations, family, and friends. I begin the analysis by documenting trends in food insufficiency for all adults and those ages 60 and older. I provide two measures of food insufficiency, one more restrictive whereby the household is food insufficient if they report that they sometimes or often do not have enough food to eat, and the other more inclusive where households reporting that they have enough food to eat but with reduced variety are also classified as being food insufficient. The share of the adult population reporting the more severe form of food insufficiency tripled from 3.4 percent in 2019 to 10.8 percent in July 2020. Seniors also reported higher food insufficiency, but a more attenuated increase of 75 percent, from 2.8 to 4.9 percent. When including those reporting enough food but reduced variety, the share more than doubles for all adults from 18.6 to 44.2 percent, as well as seniors 60 and older (14.5 to 32.8 percent). These sharp increases in food insufficiency during the Covid period are found among the low-income population, as well as across racial groups. Black adults are 2-3 times more likely than whites to report food insufficiency in a typical year, and with the onset of the Pandemic, 1 in 5 Black adults were food insufficient. In contradistinction, overall food insufficiency increased by a more muted 36 percent over the Great Recession among all adults, and by under 10 percent among seniors. I next document trends in the share of adults receiving charitable food. Because of the structure of the CPS questionnaire, I restrict the analysis of free food to those households whose income is less than 185 percent of the federal poverty line or who report being food insufficient (including those with reduced intake of food variety). The share of disadvantaged adults receiving free food rose steadily from just under 6 percent at the start of the sample period to just over 9 percent in December 2019, with no discrete change during the Great Recession. This share leapt to 14.5 percent by June 2020. The share of low-income or food insufficient seniors receiving charitable food exceeded that of adults overall in every year, and also increased steadily (but more slowly) until 2019. However, with the onset of Covid-19 there was an initial sharp drop in this population of seniors receiving free food, followed by a subsequent rebound.

This pattern is consistent with strong shelter-in-place and other social distancing restrictions that were gradually relaxed over time.

This article is protected by copyright. All rights reserved. Accepted Article I extend the analysis by exploring the determinants of food insufficiency and receipt of charitable food, focusing on both socioeconomic characteristics and the role of state business cycles. Because of the differential trends in food insufficiency and charitable food receipt between younger and older adults, I conduct the analyses separately for adults ages 18-59 years old and those ages 60 and older. The patterns across groups are similar, with risk of food insufficiency higher for those younger in age, members of minority racial or ethnic groups, unmarried, lower educated, lower income, not working, and renters. The more severe form of food insufficiency is unresponsive to state business cycles, though there is evidence of a countercyclical response when including reduced variety in the measure of food insufficiency.

These patterns of effects are similar in the models of charitable food receipt, though with the important difference that charitable food receipt rises with age. An auxiliary analysis of the aggregate time effects from the regression models suggests that food insufficiency and charitable food receipt respond countercyclically to the national business cycle, but during the Covid-19 period most of this macroeconomic shock is unexplained by the unemployment rate, pointing instead to other factors such as public health policies.

The CPS is a nationally representative monthly labor-force survey of about 60,000 households conducted by the Census Bureau for the Bureau of Labor Statistics. 2 The survey is a rotating design where households are in sample for 4 months, out of sample for 8 months, and then in sample for another 4 months. The interviews are conducted primarily in-person during interview months 1 and 5, and by phone in months 2-4 and 6-8. and Week 11 spanning July 9 -July 14, 2020. Estimates are weighted using the person supplement weight for each week, and there are 340,705 persons used in the analysis.

The CPS is the source of official food insecurity estimates, the measurement of which 1. "In the last 12 months, since December of last year, did you ever run short of money and try to make your food or your food money go further?" (variable HES9) 1 Yes 2 No -2 Don't Know -3 Refused -9 No Response and 2. "Which of these statements best describes the food eaten in your household--enough of the kinds of food (I/we) want to eat, enough but not always the kinds of food (I/we) want to eat, sometimes not enough to eat, or often not enough to eat?" (variable HESS1, referred to as the "food sufficiency" question) 1 Enough of the kinds of food we want to eat 2 Enough but not always the kinds of food we want to eat 3 Sometimes not enough to eat 4 Often not enough to eat -2 Don't Know -3 Refuse -9 No Response

If the household's income is below 185% FPL, or they choose options 1 or -2 to question HES9, or they choose options 2, 3, or 4 to the food sufficiency question (HESS1), then they proceed into the food security module (Tiehen et al 2019).

This screener detail is important because the Pulse does not field the 18-item food security module; however, it does ask the same food sufficiency question as in the CPS (HESS1 above), but the reference period is the prior 7 days. The food sufficiency question in the CPS has no explicit time horizon, but presumably refers to the time of the interview. This suggests that the CPS and Pulse should line up well on the metric of food sufficiency since there are no screeners on this question for either survey. I consider two variants of food insufficiency, one more restrictive where the household sometimes or often does not have enough food to eat and a second broader measure that also defines the household as food insufficient if they report that they have enough but not always the kinds of food they want to eat. I refer to the former as food insufficiency and the latter as food insufficiency with reduced variety. 5

The December CPS also asks households a series of questions on whether they received any food assistance from governmental and non-governmental sources. These include aid from the main federal food program Supplemental Nutrition Assistance Program (SNAP), school breakfast and lunch programs, as well as from charitable food banks and pantries, religious organizations, soup kitchens, senior centers, and home-delivered meals like Meals On Wheels.

The Pulse asks a similar set of questions about receipt and sources of free food from non-governmental organizations, but they do not ask about SNAP. 6 For charitable food, I use the summary question 26 in the Pulse on whether anyone in the household received free groceries or meals in the prior 7 days. In the CPS, I aggregate responses on whether the household received meals delivered (HESC1), ate at a community center or soup kitchen (HESC2 and HESC4M), or went to a food pantry or bank (HESCM3). The reference period for each of these questions in the CPS is the prior 30 days. 7 The CPS questions are only asked of those with incomes below 185% FPL or who report food distress, and thus I restrict the analysis of charitable foods in the CPS and Pulse to the similar subpopulation of low-income and food insufficient with reduced variety.

Because the pre-Covid and Covid-period data come from different sources, a brief comparison of the datasets is helpful to understand whether a priori we might expect any differences in food hardships due to survey sample composition.

Appendix Table 1 responding household is weighted up to account for more of the population, and thus the gap between weighted and unweighted missing income in the CPS has grown over time.

Appendix Table 1 shows that across most demographic characteristics the CPS and Pulse align well. The CPS has slightly more very young and very old persons than the Pulse, and thus The table also shows bins of income-to-needs, defined as midpoint of the household's income bin divided by 9 The temporary drop in weighted income nonresponse in 2008 is suggestive that who selects to be a respondent is affected by the business cycle. 10 Bollinger et al. (2019) report a similar rise in supplement nonresponse in the March CPS, which is used for official estimates of poverty, inequality, and health insurance coverage. The December supplement nonresponse rate is about 10 percentage points higher than the March rate in a typical year. An important difference is that the Census imputes an entire supplement record to the missing household in the March supplement, but not in December. 11 Family income in the CPS refers to the income of household members ages 15 and older who are related by birth, adoption, or marriage, and excludes cohabiting partners and other unrelated individuals. It is generally asked in interview months 1 and 5, and then carried forward to intervening months. Household income in the Pulse includes all members of the household, including cohabiting partners and other unrelated individuals.

household-size specific weighted poverty thresholds for the corresponding year. 12 Here we see many more people in the Pulse as having incomes below the poverty line than the CPS, and when including persons below twice the poverty line, the CPS has a 28 percent share compared to 34 percent in the Pulse. The CPS is likely more accurate because those income bins are in smaller increments, and thus the midpoint of the bin is a better proxy for actual income.

The Census conducts a limited amount of editing to the December CPS to ensure proper skip patterns are followed for each record, and appropriate missing data codes are assigned.

There is no replacement of missing data with imputed values from another record. Likewise, there is a limited amount of editing in the Pulse, but also a limited amount of imputation in the Pulse survey on the demographic characteristics, using a pared-down procedure akin to that employed in the March CPS income supplement. Otherwise there is no imputation in the Pulse.

Appendix Figure 3 shows trends in missing information on the focal food hardship measures of food insufficiency and charitable food. In the CPS, item nonresponse of food insufficiency is effectively zero, though nonresponse on the free food questions is trending upward, albeit at low levels of under 2 percent for all adults and about 4 percent in 2019 for seniors. 13 Nonresponse to the food insufficiency question in the Pulse is about 2-3 percent, and is about 1-2 percent for the free food question. There is no difference among seniors and all adults in the Pulse. Because these rates of nonresponse are low, I drop item nonresponders from the analysis of food hardships and rescale the percentages to sum to 100% among respondents.

12 Poverty thresholds are obtained from https://www.census.gov/data/tables/time-series/demo/incomepoverty/historical-poverty-thresholds.html 13 Income nonresponders in the CPS are placed into the category of incomes above 185% FPL and thus are screened out of the food security module that includes the charitable food questions unless they express some food hardship in either of the initial two screeners HES9 and HESS1 (food sufficiency). Bollinger et al. (2019) show that earnings nonresponse in the CPS is U-shaped across the earnings distribution, and is highest among low earners. This suggests that the practice of placing income nonresponders in the December CPS in the above 185% FPL category likely results in undercounts of receipt of SNAP and charitable foods, and possibly food insecurity.

The takeaway is that the December CPS and Pulse align well across most major socioeconomic characteristics. Rates of missing income are comparable in unweighted data, but because of rising nonresponse to the December CPS supplement there is a wider gap between the CPS and Pulse in weighted estimates of missing income. There is also a difference in income and charitable food nonresponse between all adults and seniors ages 60 and older in the CPS, but this gap is not present in the Pulse. In the ensuing descriptive analyses, I present trends in food hardship including those with missing incomes, and also conditional on low incomes among those with nonmissing income. Low incomes include those with income-to-needs below 185%

FPL as that is an important screener in the CPS. Because the heaping of persons into wider bins in the Pulse make income-to-needs less accurate than in the CPS, I also show results for persons with household incomes less than $50,000. The regression models retain the full sample, and include controls for missing income. This permits an estimate of risk of food insufficiency or receipt of charitable food among those with missing incomes, and how they each compare to that risk among those who reported their income.

In this section I focus on food hardships in the CPS and Pulse, and then turn to the determinants in the subsequent section. I begin with current food insufficiency in Figure This article is protected by copyright. All rights reserved.

[ Figure 1 here]

The upper left panel of Figure 1 shows that across the population of adults 18 and older there has been a dramatic increase in food insufficiency with the onset of the Covid-19

Pandemic. The share of the population reporting food insufficiency in the spring of 2020 is three times higher than 7 months earlier. (less than 185% FPL), and 26 and 47 percent (less than $50,000), respectively. This shift among seniors is consistent with shelter-in-place restrictions that limited mobility, meaning that Covid-19 had a disproportionate effect on seniors ability to acquire foods of wide variety.

[ Figure 2 here] This article is protected by copyright. All rights reserved.

The Covid Pandemic has not been neutral with respect rates of infection and mortality across race, with Blacks facing much greater risks of health complications from Covid-19 (Benitez, Courtemanche, and Yelowitz 2020) . In addition, the extant food insecurity literature has documented persistent racial differences in risk of food hardships (Gundersen and Ziliak 2018) . Figure 2 depicts trends in food insufficiency and food insufficiency with reduced variety separately for whites and Blacks, and within each racial group, among all adults and those ages 60 and older (full sample, including those with missing income). 14 The top panel of the figure makes clear that there is a persistent racial gap in risk of food insufficiency, with Blacks facing rates 2-3 times higher than whites (about 5 percentage points higher in a typical pre-Covid-19

year and 10-15 percentage points higher during Covid). Moreover, the increase in food insufficiency among Black adults was as dramatic as for whites (211 percent increase for Blacks from 2019 to July 2020, 206 percent increase for whites), meaning that 1 out of 5 Black adults reported being sometimes or often without enough food to eat compared to 1 out of 11 whites.

The racial gap holds for seniors too, but the increase in food insufficiency during Covid was slightly greater among whites than Blacks (70 versus 62 percent). Moreover, as seen in the bottom panel, the racial gap also is in evidence with the broader measure of food insufficiency with reduced variety, but that gap is slightly less pronounced than in the top panel. Across both whites and Blacks, all adults and seniors, there was a sharp increase with Covid-19. In July 2020 5.5 out of every 10 Black adults reported food insufficiency with reduced variety compared to 4 out of 10 whites (the estimates are 40 and 30 percent, respectively, among seniors).

[ Figure 3 here] 

I follow the wider food insecurity literature and focus on both socioeconomic characteristics and the role of state business cycles as determinants of food insufficiency and receipt of charitable food (Anderson et al. 2016; Gundersen and Ziliak 2018) . Because of the differential trends between younger and older adults, I conduct the analyses separately for adults ages 18-59 years old and those ages 60 and older. Appendix Table 2 contains weighted summary statistics of the variables used in the regressions, and Appendix Tables 3-4 contain the direct probit coefficients.

[ Table 1 here] Table 1 presents the marginal effects from the probit models of food insufficiency and food insufficiency with reduced variety for the combined 2001-2020 survey years. In columns (1) and (2) we see that relative to adults ages 50-59 the risk of food insufficiency initially declines and then rises with age, which is perhaps correlated with the presence of children in the household. Gender has no effect on food insufficiency, but women are at slightly greater risk of food insufficiency with reduced variety. The Black-white racial gap identified in Figure 2 holds once we control for other confounding factors, with Blacks 1.9 percentage points more likely to be food insufficient compared to whites, and 5.5 percentage points more likely to be food insufficient with reduced variety. 16 Hispanics are also at elevated risk of food insufficiency relative to non-Hispanics, but the effect sizes are less than half the Black-white gap. Marriage is quite protective of food insufficiency, with widowed, divorced, or separated persons at substantially greater risk than those never married. Household size leads to slightly greater risk of food insufficiency, but the number of children appear to be protective of the more severe form of food insufficiency.

Higher levels of human capital are particularly protective against food insufficiency, with high school dropouts 5.5 and 15.5 percentage points more likely to be food insufficient and food insufficient with reduced variety, respectively, than college graduates. Consistent with much of the food insecurity literature, higher levels of income are most protective against food hardships.

Those adults living in poverty are 15 (34) percentage points more likely to be food insufficient 16 Appendix Table 2 reports that the mean rate of food insufficiency among 18-59 year olds is 6 percent and is 28 percent when including reduced variety. The corresponding means for ages 60+ are 3 and 19 percent.

(with reduced variety) compared to adults with household incomes above four times the poverty line, and those living in near poverty are 10 (29) percentage points more likely to be food insufficient (with reduced variety). Beyond income, homeownership is also protective against food insufficiency, reducing the risk by 2 percentage points.

Lastly, the more severe form of food insufficiency is unresponsive to state differences in the business cycle, though the more expansive measure suggests a countercyclical relationship.

Holding other factors at their mean values, an unemployment rate of 3 percent as commonly found in 2019 is associated with a 23 percent chance of food insufficiency with reduced variety, and a 12 percent unemployment rate as observed in April 2020 is associated with a 27 percent risk. 17 These are modest effect sizes in response to large swings in unemployment.

The patterns of effects among those ages 60 and older in columns (3) and (4) of Table 1 are similar to those found on younger adults with a few notable differences. The age gradient in the risk of food insufficiency is much sharper among seniors than younger adults, with a 60-69

year old 1.7 (9.6) percentage points more likely to be food insufficient (with reduced variety) than a senior ages 80 and older. This effect size falls by half for seniors ages 70-79. A similar age gradient is found in the food insecurity literature, with the young old at substantially elevated risk of food insecurity compared to the oldest old (Gundersen and Ziliak 2018) . Table 1 also suggests that seniors who are members of minority groups (Black, other races, Hispanic) are also at greater risk of food insufficiency than found among younger adults in columns (1) and (2). On the other hand, because of their reduced attachment to the labor force, senior food insufficiency is less responsive to state business cycles compared to 18-59 year olds, with a 3 percent 17 This is found by using the probit coefficients to predict the fitted CDF holding the variables at their mean values but allowing the state unemployment rate to change.

This article is protected by copyright. All rights reserved.

Accepted Article unemployment rate associated with a 15 percent chance of food insufficiency with reduced variety, and a 12 percent unemployment rate associated with a 18 percent risk.

[ Table 2 here] Table 2 contains a parallel set of regressions, but now the dependent variable is receipt of charitable food. Recall that because of the structure of the CPS questionnaire, this analysis is restricted to those adults with household incomes under 185% FPL or who report food insufficiency with reduced variety. 18 In general the marginal effects in Table 2 accord with those found in Table 1 both within age groups and across age groups. 19 For example, across age groups the racial gap in receipt of charitable food is higher in the senior population than among younger adults. A Black senior is 5.8 percentage points more likely than a white senior to receive charitable food, compared to a 1.5 percentage point gap among 18-59 year olds. The Black senior effect is 32 percent of the sample mean of charitable food receipt reported in Appendix Table 2 , while the effect size for 18-59 year olds is 12 percent of that group's mean, suggesting that the larger effect size of race for seniors is not driven solely by the larger baseline risk level for receiving charitable food. The one variable whose effect is the opposite in the charitable food models compared to the food insufficiency models is age. Receipt of charitable foods declines monotonically in age, with a 18-29 year old at a 3.2 point lower odds of receipt relative to a 50-59 year old, compared to only a 0.3 percentage point lower relative odds for a 40-49 year old.

[ Figures 4 and 5 here] 18 In addition, data from the 2001 December CPS is omitted because of inconsistency in recall periods of charitable food compared to all other years. 19 Because the sample includes those with incomes below 185% FPL or who report food insufficiency with reduced variety, the model identifies the income effects for higher income groups. That is, the food insufficient population draws from a much larger part of the income distribution.

Given the dramatic increase in food insufficiency with the onset of Covid-19, coupled with the increase in charitable food receipt among disadvantaged adults and decrease among those ages 60 and older, it is perhaps surprising that the probit models identify such a modest effect of the state business cycle. Notably, the state unemployment rate captures local deviations from national unemployment rates, and thus may understate the total response to the business Figure 5 shows the year effects from the charitable food regressions, and here we find more of an upward trend in the unexplained time effect after the Great Recession among 18-59 year olds compared to seniors, and then diverging patterns with Covid-19 as seen previously in the raw data of Figure 3 .

To test whether the aggregate unemployment rate can account for some of these trends in the aggregate time effects seen in Figures 4 and 5 , I run a series of auxiliary time series regressions of the marginal effects of time dummies on the national unemployment rate and a linear trend. In addition, because of the sharp break in the series in 2020, I also run a specification that admits a trend break in the series after 2020, and another that also permits the unemployment rate to differ pre-and post-2020. These regressions are only meant to be suggestive as the time series is short (22 observations for food insufficiency, 21 observations for charitable food). Table 3 reports the coefficients from the auxiliary regressions, with the estimates for adults ages 18-59 in the top panel, and seniors ages 60 and older in the bottom.

[ Table 3 here]

In the upper left of Table 3 , among adults ages 18-59 we see that the direct effect of the national unemployment rate on the food insufficiency time effects is sizable compared to the state unemployment rate presented in Table 1 . The third column, which admits both a trend break in the series and a nonlinear effect of the unemployment rate shows that with the more flexible specification the time effects are driven more by the unexplained trend break than the unemployment rate. This is further underscored in same specification for the food insufficiency with reduced variety, as well as the charitable food regression. The explanatory power of these simple models is high, with R-squares in excess of 0.9. The bottom panel for seniors shows that the aggregate unemployment rate accounts for the unexplained macro time effect in the pre-Pandemic period, but not in the Covid-19 era. On the other hand, neither the national unemployment rate nor the trend does much to explain the charitable food time effects for seniors, pointing instead to other factors such as public health policies.

I provide a descriptive portrait of how food hardships facing adults in the United States compare in the two decades leading up to and during the global Covid-19 health pandemic using nationally representative data from the Current Population Survey and the Census Household Pulse Survey. The results point to unprecedented growth in food insufficiency and charitable food receipt among adults overall, with food insufficiency tripling since 2019 and charitable food receipt among the disadvantaged increasing over 50 percent. Among older adults ages 60 and above, the increase in food insufficiency was more muted, but the robust 75 percent increase swamped the increase of 10 percent in the two years surrounding the Great Recession. When expanding the measure of food insufficiency to include those households facing reduced intake of food varieties, senior food insufficiency increased every bit as much as adults overall, suggesting the mobility of seniors was strongly restricted during the early months of the Pandemic. This restricted mobility is underscored by the dramatic drop in receipt of charitable foods among seniors, falling below rates of all adults for the first time since data collection started in 2002. Participation among seniors quickly rebounded, but by July 2020 still fell below that of adults overall and below pre-Pandemic levels. These patterns, which hold in richly specified regression models, are consistent with strong shelter-in-place and other social distancing measures enacted at the state and local levels in response to the Pandemic that were gradually relaxed over time.

While the results here are descriptive only, they do point to the value and importance of having access in real-time to key metrics of well-being. The full 18-item food security module in the December CPS, along with the other food-related questions, is a crucial part of our nation's data infrastructure on measuring household well being (Bitler and Mackie 2020) . However, the necessary lags in release of the data mean that it is less able to quickly monitor current developments. The Pulse offers a much less comprehensive measurement of food security than the CPS, but with large sample sizes and real time information on a subset of food hardship indicators and repeated observations, the Pulse is valuable for both monitoring and This article is protected by copyright. All rights reserved. 

Beyond Income: What Else Predicts Very Low Food Security Among Children?

The Covid-19 Crisis Has Already Left Too Many Children Hungry in America. The Hamilton Project

Racial and Ethnic Disparities in Covid-19: Evidence from six large cities

2020. A Consumer Food Data System for 2030 and Beyond. Consensus Panel Report on Improving Consumer Data for Food and Nutrition Policy Research for the Economic Research Service

The Social Safety Net in the Wake of Covid-19

Trouble in the Tails? What We Know About Earnings Nonresponse Thirty Years After Lillard, Smith, and Welch

Household Food Security in the United States in 2019. United States Department of Agriculture

Strong Social Distancing Measures in the United States Reduced the Covid-19 Growth Rate

Food Insecurity and Health Outcomes

Food Insecurity Research in the United States: Where We Have Been and Where We Need to Go

Accepted Article nonexperimental evaluations. The addition of a question on SNAP participation in Phase 2 of the Pulse offers the opportunity to expand these monitoring evaluations to the nation's major food assistance program as we await release of the full spectrum of household food security during the Covid-19 Pandemic in fall 2021.

This article is protected by copyright. All rights reserved. This article is protected by copyright. All rights reserved.Accepted Article