Food Hardship in the US During the Pandemic: What Can We Learn from Real-Time Data? (2022)
Summary
Article in the Review of Income and Wealth, 2022, by Sara Ayllón and Samuel Lado of the University of Girona. The authors study the potential effect of the declaration of the state of emergency, the beginning and end of the stay-at-home orders, and the one-off Economic Impact Payments on food hardship in the US during the first wave of the coronavirus pandemic. They use daily data from Google Trends for the search term "foodbank". They report that demand for charitable food begins to decrease once families start receiving the stimulus payments, with the biggest fall after the lifting of the lockdown orders. Their estimates indicate the increased need for emergency help lasted for at least 10 weeks.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
- Issuer
- Source documents
- Document type
- Food Hardship In The Us During The Pandemic What Can We Learn From Real Time Data
- Case
- Food Hardship In The Us During The Pandemic What Can We Learn From Real Time Data
Cite this document
“Food Hardship in the US During the Pandemic: What Can We Learn from Real-Time Data? (2022).”
Original at doi.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/food-hardship-in-the-us-during-the-pandemic-what-can-we-learn-from-real-time-data/
, SHA-256 24ebd55f4398be3684f241206a5a7b4c7ec0a125327363e7082e2aa78da6ff10.Full text
bs_bs_banner
Review of Income and Wealth
Series 0, Number 0, Month 2022
DOI: 10.1111/roiw.12564
FOOD HARDSHIP IN THE US DURING THE PANDEMIC: WHAT CAN
WE LEARN FROM REAL-TIME DATA?
by Sara Ayllón*
University of Girona, EQUALITAS & IZA
AND
Samuel Lado
University of Girona
We study the potential effect of the declaration of the state of emergency, the beginning and end of
the stay-at-home orders, and the one-off Economic Impact Payments on food hardship in the US
during the first wave of the coronavirus pandemic. We use daily data from Google Trends for the
search term “foodbank” and document the development of a hunger crisis, as indicated by the num-
ber of individuals who need to locate a food pantry through the internet. The demand for charitable
food handouts begins to decrease once families start receiving the stimulus payments, but the biggest
fall comes when economic activity resumes after the lifting of the lockdown orders. Our estimates
indicate that the increased need for emergency help among vulnerable families lasted for at least 10
weeks during the first wave of the pandemic, and we argue that real-time data can be useful in pre-
dicting such urgency.
JEL Codes: I32, I38, H53
Keywords: food hardship, pandemic, Google Trends, CARES Act, stay-at-home orders
1. Introduction
The impact of the COVID-19 pandemic on poverty will be devastating around
the world (Sumner et al., 2020; Palomino et al., 2020), and the US will be no excep-
tion. Yet it may take a long time before researchers can actually measure the full
consequences of the pandemic for vulnerable households, largely because of the
unavailability of data. Surveys that were operating in the field during the COVID-19
Note: Support from the projects PID2019-104619RB-C43 and 2017-SGR-1571 is acknowledged.
Participants at the A&E seminar series at Universidad Carlos III (November 2020), the XXVIII
Meeting on Public Economics (May 2021), the XXIII Applied Economics Meeting (June 2021), the
34th Meeting of the European Society for Population Economics (ESPE) (June 2021), the International
Association for Applied Econometrics (IAAE) Annual Conference (June 2021), the XIV Labour
Economics Meeting (July 2021) and the 9th Meeting of the Society for the Study of Economic Inequality
(ECINEQ) (July 2021) are thanked for their useful comments. Sara Ayllón also thanks the Department
of Social Sciences at the University of Eastern Finland for its warm hospitality while writing the revised
version of this paper. Any errors or misinterpretations are our own.
*Correspondence to: Sara Ayllón, Department of Economics, C/Universtat de Girona 10, 17003,
Girona, Spain (sara.ayllon@udg.edu).
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
This is an open access article under the terms of the Creative Commons Attribution-NonCommerc
ial-NoDerivs License, which permits use and distribution in any medium, provided the original work
is properly cited, the use is non-commercial and no modifications or adaptations are made.
1
Review of Income and Wealth, Series 0, Number 0, Month 2022
outbreak had to abruptly terminate all face-to-face interviews.1 While efforts are
being made to keep collecting face-to-face content remotely, the questions that
could not be adapted have been dropped (Sastry et al., 2020). In addition, the qual-
ity of data collected from surveys during the pandemic is of some concern, on
account of the negative effects on response rates, the unrepresentative random sub-
samples (as the virus has affected different socioeconomic groups differently),
comparability of data at different points during the pandemic, respondents’ self-
selection, etc. (Schaurer and Weiß, 2020). Data from multiple surveys are also
expected to be made available to researchers with a substantial delay. However, if
we are to design effective policies to help those most in need, then an accurate diag-
nosis of the economic situation of families is a matter of urgency.
The main objective of this study is to analyze the potential for real-time data
to provide a prompt diagnosis of the effects of the pandemic—and the sudden halt
of economic activity—on one of the most extreme consequences of poverty: food
hardship. With that objective in mind, we use daily data from Google Trends for
the search terms “foodbank” and “food pantry.” The assumption is that families in
need may potentially use the internet to locate the closest foodbank available, once
they run out of goods in their pantries. They might also want to know about open-
ing days and hours and generally how the system works. We want to analyze the
extent to which web search intensities for those terms can be a leading indicator of
subsequent demand for charitable food and ascertain whether real-time data can
serve as an alternative to survey data for poverty analyses. The ultimate goal is to
document the effects of the pandemic—and of the measures imposed to stop its
spread—on the ability of families to provide food for their members. We focus on
data relative to the US, one of the richest countries in the world, with high levels of
food insecurity.2
According to data from the US Department of Agriculture, in 2019, 10.5
percent of households in the US (35.2 million individuals) suffered food insecu-
rity.3 That is, at times during the year, these households were uncertain of having
(or were unable to acquire) sufficient food to meet the needs of all their members,
because they did not have sufficient money or other resources. Some 6.4 percent
of US households qualify as low food secure and 4.1 percent as very low food
secure. In the former case, households obtain sufficient food only by using a num-
ber of strategies, such as eating less varied diets, participating in food assistance
programs or resorting to food pantries. In very low food secure households, the
eating patterns of one or more members have been disrupted and food intake has
1For example, in the US, two major supplements to the Panel Study of Income Dynamics (PSID)—
the Child Development Supplement and the Transition into Adulthood Supplement—were in the field
during the pandemic outbreak and had to cancel all interviews. The same was true of Europe. The most
important data source for the study of poverty in Europe, the European Union–Statistics on Income
and Living Conditions, suspended data collection during the first wave of the pandemic, given the im-
possibility for its interviewers to go into families’ homes. Eurostat has suggested that countries either
postpone/prolong the fieldwork or move from face-to-face to telephone interviewing (Eurostat, 2020).
2We use the term “food hardship” throughout the paper when discussing our foodbank search in-
tensity measure from Google Trends. We refer to “food insecurity” only in those instances in the paper
where the standard definition of food insecurity in the US is used. We would like to thank a referee for
suggesting such use of both terms.
3See www.ers.usda.gov for further details.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
2
Review of Income and Wealth, Series 0, Number 0, Month 2022
been reduced at times during the year because of insufficient money. The preva-
lence of the problem is even higher in households with children (13.6 percent), in
single-mother households (28.7 percent), in black, non-Hispanic households (19.1
percent), and in Hispanic households (15.6 percent). Moreover, there is consider-
able heterogeneity across states, with the lowest prevalence in New Hampshire (6.6
percent) and the highest in Mississippi (15.7 percent)—see Figure A.1 in the
Appendix.
Despite such high levels of prevalence, and according to the same data source,
2019 was the eighth year in a row that food insecurity had declined, from a peak in
2011 of 14.9 percent of US households. That year, 2019, was also the first year that
food insecurity prevalence was statistically significantly below the pre-recession
levels of 2007—see Figure A.2 in the Appendix. However, the outbreak of the
pandemic has abruptly changed this trend. Schanzenbach and Pitts (2020a), using
data from the Census Bureau’s Household Pulse Survey, document the fact that
food insecurity at least doubled among all households and tripled among house-
holds with children during the first weeks of the pandemic. Estimates for the
increase in food insecurity since the beginning of the pandemic vary depending on
the definition of food hardship, the data source used, and the period of analysis,
yet the escalation in the need for charitable food is beyond doubt in the great
majority of studies (Bauer et al., 2020; Feeding America, 2020; Niles et al., 2020;
Raifman et al., 2020; Gundersen et al., 2021).4
Our paper contributes to three strands of literature. First, we enrich the grow-
ing research into the short-term effects of the pandemic on economic well-being
and poverty (Chetty et al., 2020; Dang et al., 2020; Han et al., 2020; Cicala, 2021;
Crossley et al., 2021). Second, our results extend previous findings on the diverse
impact of the different interventions undertaken to stop the spread of the virus
(Kong and Prinz, 2020; Baek et al., 2020). We study the causal effect on the number
of searches for “foodbank” of three events: the announcement of the state of
emergency, the beginning of the stay-at-home orders, and the end of the lock-
downs. These were all declared on different dates across the nation’s states.
Importantly, we also contribute to analysis of the effects of policies implemented
to mitigate the negative effects of the pandemic. In particular, we analyze the effect
of the one-off Economic Impact Payments (EI Payments) under the CARES Act,
by which adults in low-income families received a one-time payment of $1200, plus
$500 for each dependent child under the age of 16. Interestingly enough, the great
majority of payments were received by low-income families on April 15 (Chetty et
al., 2020).5 Finally, we contribute to the literature that discusses the extent to which
(ready available) real-time data serve as a tool to predict economic phenomena,
and whether such a tool can be used for sound policy design (Nuti et al., 2014; Jun
et al., 2018).
4Ahn and Norwood (2021) are the only ones claiming that food insecurity only increased among
households with children, and not in all households. In the next section, we provide a literature review
of the most recent studies on food insecurity.
5Annual income needed to be less than $150,000 for married couples, with the household head’s
income less than $112,500; annual income for single filers had to be below $75,000. However, families
with any immigrant adult without a Social Security number were ineligible (Bitler et al., 2020).
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
3
Review of Income and Wealth, Series 0, Number 0, Month 2022
We document the development of a hunger crisis in the US during the first
wave of the coronavirus pandemic, as indicated by the number of individuals need-
ing to locate a foodbank through the internet. The demand for charitable food
becomes less apparent once families start receiving the stimulus payments, but the
main decline comes once economic activity resumes, with the lifting of the lock-
downs. Our estimates indicate that the increased need for emergency help among
vulnerable families lasted at least 10 weeks during the first wave of the pandemic.
Our findings suggest that real-time data can be useful in forecasting the increased
need for charitable food—partly among individuals who have to find information
on foodbanks for the first time. The policy implications of our findings are very
important, because food hardship has short-and long-term consequences in sev-
eral domains—it is associated with worst health at all ages, including fatigue and
reduced immune response (Gundersen and Ziliak, 2015); chronic diseases (Seligman
et al., 2009); psychological problems and mental disorders (Hamelin et al., 2002;
Davison et al., 2015; Burke et al., 2016); and increased future healthcare costs.6
This paper is organized as follows. After this introduction, Section 2 provides
a literature review of recent papers that have analyzed the impact of the corona-
virus outbreak on food insecurity in the US. Section 3 presents the data used, its
advantages, and limitations. Section 4 details our identification strategy. Section
5 presents our main findings and some robustness checks. Section 6 discusses the
usefulness of Google Trends data for the purposes of our analysis, by comparing
it with information on food insecurity trends from contemporaneous surveys and
with trends on one of its leading causes, unemployment. Finally, there are some
conclusions.
2. Literature Review
There are a handful of studies that analyze the short-term impact of the pan-
demic on food insecurity in the US.7 Bitler et al. (2020) use the COVID Impact
Survey and the National Health Interview Survey (NHIS) to compare food insecu-
rity rates over time. They estimate that the prevalence increased by 12 percentage
points between 2018 and April 2020 (from 11 percent to 23 percent). Low-income
families with children have been hardest hit: their food insecurity rates rose from 13
percent in 2018 to 34 percent in April 2020—an increase of 21 percentage points.
The authors argue that the rise in food insecurity is explained by the rise in unem-
ployment and food prices. In addition, and taking data from the COVID Impact
Survey, the authors estimate that 6.8 percent of the respondents had received food
from a food pantry in the 7 days before the interview (8.3 percent among house-
holds with children), which is far above the previous peak of 2.8 percent in 2014
6See also Alaimo et al. (2001), Jyoti et al. (2005), and Howard (2011) for the effect of food insecu-
rity on children’s development and academic outcomes, and Case et al. (2005) and Currie (2009) for the
long-term consequences of poor health during childhood.
7Part of the literature has also been devoted to analysis of the impact of the coronavirus pandemic
on poverty and income loss in the US. In this respect, some studies document an important deteriora-
tion in the situation of vulnerable families (Dang et al., 2020; Belot et al., 2021; Parolin and Wimer,
2020), while others actually conclude that, thanks to the support packages, poverty was alleviated
during the first wave of the pandemic (Han et al., 2020; Ganong et al., 2020).
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
4
Review of Income and Wealth, Series 0, Number 0, Month 2022
(estimated using the Current Population Survey–Food Security Supplement). The
authors conclude that despite the measures taken by the US government—
extending and increasing unemployment insurance (UI) and Supplemental
Nutrition Assistance Program (SNAP) payments, introducing a new program to
replace missing school meals (Pandemic-Electronic Benefit Transfer—P-EBT) and
sending cash relief payments (EI Payments)—food insecurity has increased.
In a similar vein, Schanzenbach et al. (2020b) analyze data from the COVID
Impact Survey and estimate that food insecurity had doubled overall and tripled
among households with children by the end of April 2020, relative to predicted
rates for March. They also document that 7 percent of respondents reported hav-
ing received assistance from a food pantry in the week before the interview. The
figure jumps to 10 percent when individuals who said they had applied or tried to
apply for food pantry assistance are included.8 Additional estimates using other
sources of data can be found in Schanzenbach and Pitts (2020a) and (2020c). Ziliak
(2021) compares the December Supplements of the Current Population Survey up
to 2019 and the Census Bureau’s Household Pulse Survey for 2020, and shows that
food insecurity tripled between 2019 and July 2020 (from 3.4 percent to 10.8 per-
cent among the adult population).9 In comparison, during the Great Recession the
rate only rose from 3.6 percent in 2007 to 4.9 percent in 2009 (and peaked at 5.1
percent in 2014). The author also documents an increase in the percentage of low-
income adult individuals receiving charitable food: up from 9.1 percent in 2019 to
14.5 percent in June 2020—a total increase of 59 percent. These figures are well
above those of the Great Recession, when there was an increase of only 26 percent
between 2007 and 2009.
Other studies have also analyzed the impact of the changes in social assistance
programs and unemployment benefits on food insecurity in the US since the out-
break of the pandemic. Bauer et al. (2020) study the effect of the P-EBT program,
which provides families with a voucher to purchase groceries for an amount equal
to the value of school meals missed from the start of the school closures at the end
of the 2019/20 school year. The authors estimate a reduction in household food
insecurity levels of 6 percentage points 2 weeks after the P-EBT is paid out to vul-
nerable families with children. Using data from the Understanding Coronavirus in
America (UCA) survey, Raifman et al. (2020) analyze the impact of the expansion
of UI benefits in terms of both size and scope. Their results indicate that receipt of
UI is associated with a 30 percent decline in food insecurity among low-income
households that suffered job loss during the first wave of the COVID- 19
pandemic.10
Despite the timely relevance of the aforementioned studies, research on trends
in food insecurity since the pandemic outbreak is not without its limitations—
mostly because of the need to rely on survey data. It is often the case that trends
8As explained by the authors, these figures are likely to underestimate the share of the population
that interacts with food pantries, because surveys exclude the homeless population and likely underrep-
resent those with unstable housing.
9When considering adults who report having sufficient food but not enough variety, the percentage
escalates from 18.6 percent to 44.2 percent (Ziliak, 2021).
10For a previous analysis of the effect of public assistance on food insecurity, see Borjas (2005),
Ratcliffe et al. (2011), and Shaefer and Gutierrez (2013).
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
5
Review of Income and Wealth, Series 0, Number 0, Month 2022
need to be estimated from different data sources that use different approximations
and definitions of food insecurity and also different reference periods (Ziliak,
2021; Schanzenbach and Pitts, 2020c). Sample sizes are at times relatively small
(Raifman et al., 2020) or are very selective (e.g., families that were below a certain
percentage of the official poverty line before the pandemic), thus leaving aside indi-
viduals who became poor during the first wave of the pandemic. Moreover, most
surveys started to collect data weeks after the first COVID-19 diagnosed case. For
example, the first wave of UCA lasted from April 1 to April 28, the US Census
Bureau’s Household Pulse Survey started collection in the days from April 23 to
May 5, and the COVID Impact Survey took place from May 4 to May 10—thus,
preventing analysis of the very first impact of the pandemic on food insecurity. In
addition, data do not always cover the whole nation, but only certain regions or
cities (e.g., the COVID Impact Survey), it necessarily relies on self-reporting, which
is prone to bias, and it may reach researchers only with a substantial delay.11 In the
next sections, we explore the use of (readily available) data from Google Trends to
document changes in food insecurity.12
3. Data
We use daily data from Google Trends for the search terms “foodbank,” “food
bank,” or “food pantry” for all the states in the US for the period January 1 to June
30, 2020. Data are provided by Google as a search intensity measure that goes from
0 to 100—being 0 the smallest proportion among the queried terms within a given
region and a time frame and 100 being the highest proportion.13 Given that we
extract three queries, we compute a unique dependent variable “search” that con-
tains the highest daily value among the three terms.14 Figure 1 shows the search
intensity index trend. As can be seen, while Google Trends searches for “foodbank”
were relatively low and stable during the first weeks of 2020, the outbreak of the
pandemic meant a sudden upsurge in the number of searches—being particularly
intense 8 weeks after the Centers for Disease Control and Prevention (CDCP) first
public alert about the coronavirus and 1 week before the declaration of the state of
emergency at the national level (issued on March 13, 2020). The search intensity
index escalates rapidly with the stay-at-home orders and until mid-April—with
11One of the most important databases for the analysis of food insecurity in the US, the December
Supplement of the Current Population Survey, will only provide data relative to 2020 in the second half
of 2021.
12Google search intensity data have been used to forecast multiple economic indicators: private
consumption (Vosen et al., 2011); motor vehicles and travel purchases, unemployment benefit claims,
and consumer confidence (Choi and Varian, 2012); job searches (Baker and Fradkin, 2017), etc. Since
the coronavirus outbreak, the use of Google search intensity data has grown exponentially to study
mental well-being (Brodeur et al., 2021); economic anxiety (Fetzer et al., 2021); unemployment claims
(Kong and Prinz, 2020), etc. None of the reviewed literature though contains any analysis of food inse-
curity using data from Google Trends. At a very descriptive level, Mayasari et al. (2020) analyze the
frequency of use in Google Trends of terms related to lifestyle behaviors, nutrients, and herbs and,
briefly, also food insecurity. They determine in which countries of the world searches for terms such as
“food bank,” “free meal,” or “turmeric” are more common.
13More details regarding the computation of the search intensity measure can be obtained from
Google Support at https://support.google.com/trends/answer/4365533?hl=en.
14Throughout the paper, we will simply refer to “foodbank.”
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
6
Review of Income and Wealth, Series 0, Number 0, Month 2022
Google search intensity index
40 60 80
20
1/14
1/07 01/2 5−0
2−0 1/21
01/2 9−0 1/28
2/04
02/0 5−0 2/11
1−0 8−0
02/1
02/1 2−0 2/18
02/2 9−0
6−0 2/25
State 03/0 4−0 3/03
3/10
of em
01/0 01/0 01/1
03/1 erge
8−0 ncy
03/2 5−0 3/24
3/31
04/0
04/0 1−0 4/07
04/1 5−0 4/21 8−0
EI Pa 4/14
04/2 ymen
2−0 ts
04/2 9−0 4/28
5/05
05/0
05/1 6−0 5/12
05/2 3−0
0−0 5/19
05/2 7−0 5/26
6/02
06/0
06/1 3−0 6/09
06/1 0−0
7−0 6/16
06/2 4−0 6/23
6/30
Figure 1. Weekly Average of the Search Intensity Index for “Foodbank” in Google Trends, US,
January 1 to June 30, 2020
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term
within a region and time frame. Data downloaded on October 1, 2020.
Source: Authors’ computation using data from Google Trends.
only a slight decrease at the end of March, coinciding with end-of-the-month pay-
ments. The maximum is reached in the second week of April, when a clear down-
ward trend commences—probably because vulnerable families start to receive the
one-off EI Payments (see below for more detail).
If we take a longer time frame, the data indicate that the search intensity peak
in April 2020 was quite unique in the time that Google Trends data have existed.
Figure 2 shows the monthly data for the search term “foodbank” since January
2004 and up until June 2020. While there are peaks throughout the period, partic-
ularly in the winter months and during the Great Recession, none gets even close
to the peak of April 2020, potentially indicating the severity of the need for chari-
table food brought about by the measures imposed to stop the pandemic.15
Importantly for our analysis and our identification strategy, Google Trends
provides the index by geographical area (in our case, we extract data at the state
level). Figure 3 shows the search intensity trend for all the states, along with the
declaration of the state of emergency in each state (solid line), the beginning of the
15Food insecurity is seasonal in many contexts, as in the winter months many vulnerable families
have to choose between paying for heating costs or food (Bhattacharya et al., 2003; Nord and Kantor,
2006). Also see Lombe et al. (2018) for an analysis of the impact of the Great Recession on increased
food insecurity in the US.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
7
Review of Income and Wealth, Series 0, Number 0, Month 2022
100
Google search intensity index
40 60 80
20
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020
Figure 2. Monthly Search Intensity Index for “Foodbank” in Google Trends, US, January 2004 to
June 2020
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term
within a region and time frame. Data downloaded on November 19, 2020.
Source: Authors’ computation using data from Google Trends.
stay-at-home orders (if any) (dash), April 15, when a large number of low-income
families received a payment via the CARES Act (short dash) and the end of the
lockdown (long dash).16 While there is considerable heterogeneity across the states,
an increase in the search intensity measure for “foodbank" occurs in all of them
after the start of the pandemic, reversing in most states around mid-April.
It is relevant to highlight at this point that the Google Trends search index is
constructed from randomly selected subsets of total search data. Although all que-
ries are stored, Google Trends randomly samples each search and uses only a frac-
tion of the total searches for a term to build the index.17 Sampling is done daily,
which complicates the replication of previously downloaded data (Nuti et al.,
2014). If one wants to get data from different samples, one needs to wait only a day
for the sampling to change. To ensure that our results do not depend on this sam-
pling process, we extracted data on 30 different days. Figure A.3 in the Appendix
shows that the trends that we can derive in each extraction are very close to one
another. The same is true for the results.
The advantages of using Google Search data for our analysis are that: (1) it
allows an analysis almost in real time (data are available with a delay of 36 h); (2)
16See also Table A.1 in the Appendix for the exact dates.
17When computing the index, Google Trends also excludes minority searches, those that have a low
search volume, duplicate or repeated searches from the same user performed in a short period of time
and searches with special characters, such as apostrophes. It is insensitive to capital letters. This way, the
level of interest is not artificially affected by, for instance, typing errors.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
8
Alabama Alaska Arizona Arkansas California
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Colorado Connecticut Delaware Florida Georgia
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Hawaii Idaho Illinois Indiana Iowa
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
9
Kansas Kentucky Louisiana Maine Maryland
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Massachusetts Michigan Minnesota Mississippi Missouri
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Figure 3. Weekly Average of the Search Intensity Index for “Foodbank” in Google Trends by State, US, January–June 2020
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term within a region and time frame. Emergency state (solid line),
beginning of lockdown (dash), EI Payments (short dash), and end of lockdown (long dash). Data downloaded on October 1, 2020.
International Association for Research in Income and Wealth
Review of Income and Wealth, Series 0, Number 0, Month 2022
Source: Authors’ computation using data from Google Trends.
Montana Nebraska Nevada New Hampshire New Jersey
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
New Mexico New York North Carolina North Dakota Ohio
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Oklahoma Oregon Pennsylvania Rhode Island South Carolina
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
10
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
South Dakota Tennessee Texas Utah Vermont
Review of Income and Wealth, Series 0, Number 0, Month 2022
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Virginia Washington West Virginia Wisconsin Wyoming
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25 0 5 10 15 20 25
Figure 3. Continued
International Association for Research in Income and Wealth
Review of Income and Wealth, Series 0, Number 0, Month 2022
Google search intensity index
0 20 40 60 80
01/2
1/07
1/14 01/2 2−0
9−0 1/28
1/21 02/0 5−0 2/04
2/11
02/1 2−0 2/18
1−0 8−0 5−0
02/1
02/2 9−0 2/25
03/0 6−0
4−0 3/03
State of em 3/10
erge
03/1 ncy
01/0 01/0 01/1 03/2 8−0
5−0 3/24
04/0 1−0 3/31
4/07
04/1 5−0 4/2104/0 8−0 4/14
EI Pa
04/2 ymen ts
04/2 2−0
9−0 4/28
05/0 6−0 5/05
5/12
05/1
05/2 3−0 5/19
05/2 0−0
7−0 5/26
06/0 3−0 6/02
6/09
06/1
06/1 0−0 6/16
06/2 7−0
4−0 6/23
6/30
Foodbank near me Foodbank donation
Figure 4. Weekly Average of the Search Intensity Index for “Foodbank Near Me” and “Foodbank
Donation” in Google Trends (Comparative Search), US, January 1 to June 30, 2020
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term within
a region and time frame. Data downloaded on October 20, 2020.
Source: Authors’ computation using data from Google Trends
in our case, it permits the analysis of a phenomenon that affects individuals often
not well represented in surveys; (3) data are not affected by the stigma or shame
that may arise in interviews; and (4) it possibly captures individuals who need to
resort to charitable food for the first time given the sudden economic shock brought
about by the pandemic.18 As for the limitations of Google Search, we need to bear
in mind that: (1) older individuals are less likely to use internet to find information
than are younger individuals—therefore, our results may not be representative of
the elderly population; (2) heterogeneous effects cannot be analyzed; and (3) we do
not know exactly why a term is being sought.19 However, given the relevance of this
last point, we have investigated it by taking advantage of the Google Trends
18Niles et al. (2020) estimate that one food insecure household in three can be classified as newly
food insecure since the coronavirus pandemic outbreak.
19That said, the Pew Research Center (2021) indicates that internet users include 99 percent of
adults aged between 19 and 29 years; 98 percent of those aged 30–49; 96 percent of those aged 50–64;
and 75 percent of people aged 65 and over. By race and gender, the percentages are indistinguishable.
Together with age, income and education are the main sources of unequal internet use. Only 86 percent
of adults who earn less than $30,000 use the internet, whereas the percentages are higher for people
earning $30,000–49,999 (91 percent), $50,000–74,999 (98 percent), and $75,000 or more (99 percent). As
for educational attainment, 71 percent of people with less than a completed high-school education use
the internet, compared to 86 percent of those with only a completed high-school education, and 98
percent of college graduates. Moreover, Google had 88.1 percent of the search engine market share in
the US in 2020. We also actually considered searches in languages other than English, to see if we could
predict increased need by population subgroup, but the volume of searches was not sufficiently large.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
11
Review of Income and Wealth, Series 0, Number 0, Month 2022
comparative search tool, which allows several search terms to be introduced at the
same time and information to be gleaned about their relative importance within the
same period and context. Figure 4 compares the search intensity measures for
“foodbank near me” and “foodbank donation.” While the search intensity mea-
sure for “foodbank near me” is very similar to the trend presented in Figure 1 and
used throughout our analysis, the search intensity measure for “foodbank dona-
tion” is stable and very low, indicating that the vast majority of searches were not
done by individuals willing to donate food, but rather by those who needed to
obtain charitable food.20
4. Identification Strategy
We use three different strategies to identify the causal effect of the four
interventions we are interested in (the declaration of the state of emergency,
the beginning and end of the lockdowns, and the one-off EI Payments): (1) a
difference-in-differences (DID) approach; (2) regression discontinuity design with
difference-in-differences (RDD-DID); and (3) an event study.
Regarding the DID estimation, we exploit the fact that each state declared
each “intervention” (event) on a different date. At the same time, we compare the
search intensity measure pre-and post-lockdown in 2020 to the search intensity
measure on the same dates in 2019 (and 2018 for robustness) within the same state.
Thus, our identification strategy relies on the differences in timing of the lockdown
between states and the within-state comparison of Google searches before and
after the same dates in different years. For example, in the case of the lockdowns,
our specification would be as follows:
(1) Sir = 𝛼Lir ⋅ Yeari + 𝛽Lir + 𝛾Xir + 𝜇 r + 𝜌i + 𝜖 ir ,
where i refers to a given day and r refers to a state. Sir is the daily search intensity
measure for the term “foodbank.” Lir is an indicator variable that takes the value 1
on the days after the lockdowns were declared, and 0 otherwise (both in 2019 and
2020). Yeari equals 2020. Our parameter of interest is 𝛼 which captures the causal
effect of the event on the intensity of searches. The variable Xir includes controls
for the accumulated number of deaths by COVID-19 (per million) and the accu-
mulated number of diagnoses (also per million). Finally, the model includes fixed
effects for state (𝜇r), for year, for week, and for weekday (summarized in 𝜌i). Results
are weighted by the population in each state, and standard errors are robust and
clustered at the day level.21
20In addition, Google Trends offers the possibility to know the most common searches that include
a given term. We tried with “food bank” (two words) for the same period of analysis throughout the US
and found that “food bank near me,” “food near me,” and “the food bank” were the three most common
searches that included “food bank”—which is reassuring. Figure A.4 in the Appendix shows the list of
the 25 most common terms and the associated search intensity index. Moreover, search data for “food
stamps” and “SNAP,” which can be regarded as an alternative proxy for the need for charitable food,
provided a similar trend to that shown in Figure 1—see Figure A.5 in the Appendix.
21We have also checked our results when clustering our standard errors at the state level and when
we use a double cluster at the state and day level. Our main conclusions do not depend on such a choice.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
12
Review of Income and Wealth, Series 0, Number 0, Month 2022
We complete our DID results with an estimation of the immediate structural
break of each intervention, by adopting an RDD combined with the DID strat-
egy, as presented above. We compare these estimated breaks to those for 2019,
thus creating an RDD-DID. The goal is to obtain the immediate effect in the
few days around each announcement, rather than the average effect of all pre-
announcement observations on all post- announcement observations, which is
what the DID results provide (Brodeur et al., 2021). Formally,
Sir = 𝛼Lir ⋅ Yeari + 𝛽Lir + 𝜆Dir ⋅ (Lir ⋅ Yeari )𝛿Dir ⋅ ((1 − Lir )
(2)
⋅ Yeari ) + 𝜃Dir ⋅ Lir + 𝜂Dir ⋅ (1 − Lir ) + 𝛾Xir + 𝜇 r + 𝜌i + 𝜖 ir ,
where i refers to a given day and r refers to a state (as above). Dir is the distance in
days from, in this example, the lockdown—being negative for the days before the
lockdown and positive for the days after. This running variable is interacted with
the lockdown to allow for different effects on both sides of the cutoff. In the main
results, we use a polynomial of order 1 and the same fixed effects and controls as in
equation (1), whereas in the robustness checks we use polynomials of higher order.
As for the event study, we take the 4 weeks before an intervention and the
4 weeks after (both in 2019 and in 2020), while we set to 0 the exact week of the
intervention. The fourth week before the intervention (k = − 4) is the week of ref-
erence. Formally,
∑k=4 ∑k=4
(3) Sir = 𝛼 nk Lkr ⋅ Yeari + 𝛽 nk Lkr + 𝛾 n Xir + 𝜇 nr + 𝜌ni + 𝜖 nir .
k=−3 k=−3
For example, when k = 2, 𝛼 2 is going to inform us of the effect of a lockdown
2 weeks after its implementation and in comparison to the fourth week before
(k = − 4). The same number of fixed effects and controls as in equations (1) and
(2) are included. Population weights and robust standard errors clustered at day
level are used. We run an additional exercise for k = [ − 4, + 16].
5. Results
Table 1 presents the results of the DID estimations for the four interventions
of interest during the first wave of the coronavirus pandemic: the declaration of
the state of emergency (columns 1 and 2), the beginning of the lockdowns (col-
umns 3 and 4), the one-time EI Payments (columns 5 and 6), and the end of the
lockdowns (columns 7 and 8). For each intervention, we show first results that
consider only fixed effects, while in the adjacent column we present results that also
include population weights and controls.22 Each panel presents results for a rele-
vant period of the intervention (detailed in the first row), to consider the fact that
the different events took place in a sequential manner.
The results indicate an important increase in the search intensity for “food-
bank” as a result of both the declaration of the state of emergency and the
22To save space, we do not show results that consider fixed effects and population weights but no
controls, as they are very similar to the results that include controls. These results are available from the
authors on request.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
13
TABLE 1
Difference-in-Difference Results for the Declaration of the State of Emergency (Columns 1 and 2), the Beginning of the Lockdowns (Columns 3 and 4),
the One-off Economic Impact Payments (Columns 5 and 6), and the End of the Lockdowns (Columns 7 and 8) on the Daily Search Intensity Measure for
“Foodbank,” US, January 1 to June 30, 2020
January 1 to April 15 January 1 to April 15 April 1 to April 30 April 15 to June 30
(1) (2) (3) (4) (5) (6) (7) (8)
Emergency*Year 19.15*** 24.62***
(1.47) (2.01)
Lockdown*Year 18.98*** 27.19***
(1.91) (2.28)
Stimulus*Year −2.59 −4.09
(2.63) (3.76)
End*Year −7.49*** −8.41***
(1.59) (1.87)
State FE yes yes yes yes yes yes yes yes
Year FE yes yes yes yes yes yes yes yes
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
14 Week FE yes yes yes yes yes yes yes yes
Weekday FE yes yes yes yes yes yes yes yes
Weights no yes no yes no yes no yes
Controls no yes no yes no yes no yes
Observations 10,450 10,450 10,450 10,450 3,000 3,000 7,650 7,650
Review of Income and Wealth, Series 0, Number 0, Month 2022
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term within a region and time frame. Controls are the accumulated
number of deaths per million individuals and the accumulated number of diagnoses per million individuals. Weights contain total population in the state in 2019 from
the United States Census Bureau. Robust standard errors are shown in parentheses and have been clustered at day level.
*** p < 0.01; ** p < 0.05; * p < 0.1.
International Association for Research in Income and Wealth
Review of Income and Wealth, Series 0, Number 0, Month 2022
TABLE 2
RDD-DID Results for the State of Emergency, Beginning and End of the Lockdown, and the
Stimulus Payments on the Daily Search Intensity Measure for “Foodbank,” US, January 1 to
June, 2020
State of Beginning of Stimulus
Emergency Lockdown Payment End of Lockdown
(1) (2) (3) (4)
Emergency*Year 32.63***
(2.73)
Lockdown*Year 20.84***
(3.14)
Stimulus*Year −3.43
(2.84)
End*Year −11.55***
(3.50)
State FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Week FE Yes Yes Yes Yes
Weekday FE Yes Yes Yes Yes
Weights Yes Yes Yes Yes
Controls Yes Yes Yes Yes
Observations 18,000 15,120 18,000 14,400
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term
within a region and time frame. Controls are the accumulated number of deaths per million individuals
and the accumulated number of diagnoses per million individuals. Day 0 in each event is dropped from
the specification. Weights contain total population in the state in 2019 from the United States Census
Bureau. Robust standard errors are shown in parentheses and have been clustered at day level.
*** p < 0.01; ** p < 0.05; * p < 0.1.
stay-at-home orders (relative to the pre-pandemic search patterns), with coeffi-
cients that range between 19.15 and 27.19, depending on the specification. Thus,
while both interventions were efficient in terms of stopping the spread of the coro-
navirus, they meant an increase in the number of families that needed to resort to
charitable food. Instead the one-off EI Payments that most low-income families
received on April 15 mark the beginning of a declining trend. The associated coef-
ficient to the policy for the period between 1 April and April 30 is not statistically
meaningful, indicating that the decreased need to look for charitable food after the
payments just compensated for the increase associated with the lockdowns. With
the end of the stay-at-home orders, and as the economy resumed its activity, the
search intensity measure for “foodbank” declined substantially, as indicated by the
negative coefficients in columns 7 and 8.
The previous results are confirmed by Table 2, which provides the results for
the RDD-DID estimation of the immediate break caused by each intervention in
the days surrounding it.23 The first column refers to the specification relative to the
state of emergency; the second column refers to the lockdown; the third column
refers to the one-time EI Payments; and the fourth column refers to the end of the
23An advantage of RDD-DID is that results are not restricted to an arbitrarily chosen time period,
as in the DID estimates presented above.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
15
Review of Income and Wealth, Series 0, Number 0, Month 2022
40
30
20
10
0
−10
−4 −3 −2 −1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Weeks elapsed since the start of the state of emergency
Figure 5. Event Study Results for the Search Intensity Index for “Foodbank” in Google Trends, US,
January 1 to June 30, 2020
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term
within a region and time frame. Beginning of the lockdown [1, 5]; EI Payments [5, 7]; End of lockdown
[6, 13]. Controls are the accumulated number of deaths per million individuals and the accumulated
number of diagnoses per million individuals. Weights contain total population in the state in 2019 from
the United States Census Bureau. Robust standard errors have been clustered at day level. Vertical lines
show confidence intervals at 95 percent.
Source: Authors’ computation using data from Google Trends.
stay-at-home orders. The results indicate once more that the immediate effect of
the declaration of the state of emergency and the lockdowns was to increase the
search intensity measure for “foodbank”—with the effect particularly strong after
the declaration of the state of emergency. By contrast, we find an immediate nega-
tive effect for the end of the lockdowns naturally associated with a certain increase
in economic activity.
Event study estimates inform us not only of the direction of the effect of
each intervention, but also of the duration of each event. Figure A.6 in the
Appendix shows the results when we consider the 4 weeks before and the 4 weeks
after each event, while Figure 5 provides estimates for the 4 weeks before the
state of emergency until the end of June. The results indicate that the increase
in the search intensity measure in Google Trends starts with the declaration of
the state of emergency by each state and continues with the stay-at-home orders.
The peak is reached 5 weeks after the declaration of the state of emergency
when all the states that were to declare a lockdown had done so. The one-off EI
Payments were received in all states within 5–7 weeks following declaration of
a state of emergency; at this point, a clear downward trend starts and continues
until week 11, during which searches for “foodbank” are not statistically differ-
ent from those in the reference period. Altogether, though, the search intensity
measure was above the level in the reference period for 10 weeks after declara-
tion of the state of emergency.
Our findings document the development of a hunger crisis in the US
during the first wave of the coronavirus pandemic, as indicated by the number
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
16
Review of Income and Wealth, Series 0, Number 0, Month 2022
of individuals who needed to locate a foodbank through the internet. Charitable
food demand starts to decline once families begin to receive the stimulus pay-
ments, but the main fall comes once economic activity resumes, with the lifting
of the lockdowns. Our estimates indicate that the increased need for emergency
help among vulnerable families lasted at least 10 weeks during the first wave of
the pandemic.
5.1. Robustness Checks
We confirmed our main findings with a number of robustness checks that are
summarized here. First, and following Brodeur et al. (2021), we rescaled the search
intensity index to consider the fact that the absolute number of searches may be
different in 2019 and 2020. As daily searches can only be obtained for a period of
less than 9 months, they propose to use weekly data for the period 2019–2020 to
rescale the dependent variable. Our main conclusions remain unchanged when we
do that—likely because of the use of fixed effects by year. Second, we ensured that
our results do not depend on the comparison between 2019 and 2020 exclusively,
so we have included in our sample daily search data from 2018, too. When we do
that, the coefficients increase slightly, while the statistical significance remains at
the same level. Furthermore, we confirmed that 2019 serves well as a “control” year
in our specifications, by running our models for a sample that includes exclusively
data from 2018 and 2019, and by simulating a scenario whereby the different inter-
ventions occurred on the same dates, but in 2019. Figure A.7 in the Appendix
shows that none of the event study coefficients are statistically significant in this
placebo exercise. Third, we transformed our dependent variable into logarithms to
obtain the same (qualitative) findings. The same is true when we consider a hurdle
model that accounts for the number of zeros in our dependent variable. Fourth,
and following Adda (2016) and Qiu et al. (2020), we added to our controls the daily
average temperature in each state, using data from the National Oceanic and
Atmospheric Administration (NOAA).24 Our results are nearly identical to those
presented in Table 1. Finally, we ran our RDD-DID specifications by including a
polynomial of order 2: we obtain smaller positive coefficients regarding the decla-
ration of the state of emergency and the beginning of the lockdowns, and a stron-
ger negative impact for the end of the stay-at-home orders. The level of significance
is the same as presented in Table 2 for all coefficients.
6. Can We Really Learn to Study Food Insecurity from Google Trends?
The previous section showed the impact of the different interventions on
food hardship during the first wave of the pandemic proxied by search inten-
sity data. In this section, we consider how reliable such data are in predicting
demand for charitable food. First, we do that by considering the extent to which
data on food hardship from Google Trends are comparable to that obtained
from contemporaneous surveys. Then, second, we study the extent to which
24We computed daily temperature by extracting the information from all the weather stations
across the US and averaging it by state and date.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
17
Review of Income and Wealth, Series 0, Number 0, Month 2022
Daily data Weekly data
100 20 100
20
40 60 80 40 60 80
Food insecurity prevalence Food insecurity prevalence
10 15 Google search intensity index 10 15 Google search intensity index
20 20
5 0 5 0
01apr2020 01may2020 01jun2020 01jul2020 01apr2020 01may2020 01jun2020 01jul2020
Food insecurity Google search intensity index
Figure 6. Search Intensity Index for “Foodbank” and Prevalence of Food Insecurity According to
Understanding Coronavirus in America, Daily (Left) and Weekly Data (Right), April–June 2020, US
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term
within a region and time frame.
Source: Authors’ computation using data from Google Trends and Understanding Coronavirus in
America.
search intensity data correlate with changes in unemployment, one of the most
important causes of food insecurity in the present context (Bitler et al., 2020;
Restrepo et al., 2021).
With the first objective in mind, we take data from the longitudinal survey
UCA run by the University of Southern California Center for Economic and
Social Research (CESR). This online survey collects data about attitudes, beliefs,
experiences, and behaviors around the pandemic, and also (since the beginning of
April) information on food insecurity.25 The great advantage of the UCA data set
for the purposes of our analysis is that it contains daily data: panelists are assigned
to answer on a specific day within each 14-day wave. We proxy the daily level of
food insecurity with all the positive answers to three questions: (1) “In the past
seven days, were you worried you would run out of food because of a lack of
money or other resources?”, (2) “In the past seven days, did you eat less than you
thought you should because of a lack of money or other resources?”, and (3) “In
the past seven days, did you go without eating for a whole day because of a lack of
25Note that, for example, the Census Bureau’s Household Pulse Survey, which contains informa-
tion on food insecurity, only started data collection in the fourth week of April, and thus does not allow
analysis of the very first impact of the coronavirus outbreak. In the case of the COVID Impact Survey,
only 1 week in April, one in May, and one in June are covered.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
18
Review of Income and Wealth, Series 0, Number 0, Month 2022
Google search intensity index 0 2000 4000 6000
20 40 60 80 Unemployment Insurance claims
01/0
01/0 1-0 1/07
01/1 8-0
5-0 1/14
01/2 2-0 1/21
1/28
01/2
02/0 9-0 2/04
02/1 5-0
2-0 2/11
02/1 9-0 2/18
2/25
02/2
03/0 6-0 3/03
State of em 4-0
erge3/10
03/1 8-0 ncy
3/24
03/2
04/0 5-0 3/31
04/1 5-0 04/0 1-0
8-0 4/07
4/21 EI Pa 4/14
ymen
04/2
04/2 2-0 ts
4/28
05/0 9-0
6-0 5/05
05/1 3-0 5/12
5/19
05/2
05/2 0-0 5/26
06/0 7-0
3-0 6/02
06/1 0-0 6/09
6/16
06/1
06/2 7-0 6/23
4-0 6/30
Google search intensity index UI claims
Figure 7. Weekly Average of the Search Intensity Index for “Foodbank” in Google Trends and Initial
Unemployment Insurance Claims, January–June 2020, US
Note: Google search intensity index goes from 0 to 100 indicating the popularity of a given term
within a region and time frame. Data on unemployment insurance claims are per thousand.
Source: Authors’ computation using data from Google Trends and the unemployment insurance
claims from the Economic Tracker (Opportunity Insights) and the Department of Labor.
money or other resources?”. The respondents can declare “yes,” “no,” or “unsure.”
In this last case, we assume the answer to be negative. Figure 6 shows the high cor-
relation between the prevalence of (daily and weekly) food insecurity and the cor-
responding search intensity index for “foodbank,” highlighting the usefulness of
web search data to predict demand for charitable food.26
Second, if Google search intensity data are useful in monitoring the develop-
ment of a hunger crisis in a given context, one would expect such data to be cor-
related with one of the leading causes behind the upsurge in food insecurity since
the coronavirus pandemic began, and that is, undoubtedly, unemployment. Thus,
we download data from the Economic Tracker (of the Opportunity Insights orga-
nization) on the number of initial weekly UI claims, which we plot together with
the search intensity data for “foodbank” at the national level.27 Figure 7 shows that
the correlation between the two variables is high, with UI claims peaking 2 weeks
before the search intensity measure for “foodbank” reaches its maximum.28 Given
26The correlation between the two variables is 0.73 for daily data and 0.89 for weekly data.
27See www.tracktherecovery.org.
28We use initial claims, rather than continued claims, because the search intensity data for “food-
bank” is likely to reflect new poor (as argued above), in association with unemployment entries.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
19
Review of Income and Wealth, Series 0, Number 0, Month 2022
the delays in the actual receipt of the benefits, it is not surprising that those newly
unemployed and with no savings need to resort to a foodbank soon after losing
their jobs. Both variables decline at a similar rate from the month of April onwards.
The two exercises highlight the usefulness of real-time data for the purposes
of our analysis—by underscoring its comparability with data from high-frequency
surveys and by pointing out its correlation with leading indicators of family need,
such as unemployment. The great advantage of real-time data from Google Trends
is that it is free and readily available to any researcher, policymaker, practitioner,
or NGO manager.
7. Concluding Remarks
This study documents the development of a hunger crisis in the US during the
first weeks of the coronavirus pandemic. We show large increases in the daily num-
ber of individuals who searched the internet for the term “foodbank,” which, we
argue, proxies the need to resort to charitable food during the period. The results
indicate that the declaration of a state of emergency, the beginning of the lock-
downs in each state, and the abrupt halt in economic activity all imposed a burden
on families in need that could not feed their members from their own resources.
The one-time EI Payments under the CARES Act, and particularly the end of the
stay-at-home orders, led to an improvement in the situation. Our estimates from an
event study indicate that the urgent need for charitable food during the first wave
of the pandemic lasted for 10 weeks.
Hunger is difficult to measure. It is often suffered by hard-to-reach popula-
tions with unstable housing. It is sometimes endured in silence, because of
stigma and shame. As a result, surveys are likely to underreport the extent of the
problem. In this paper, we argue that, despite all its limitations, Google search
intensity data for terms associated with hunger can be used as an additional tool
to study trends in food insecurity. Figure A.8 in the Appendix shows that Google
Search real-time data also predict well the increased demand for charitable food
associated with the second and third waves of the pandemic.29 More broadly,
this paper highlights the value and importance of having access to readily avail-
able real-time data for important indicators of well-being. The early detection
of emergency need can help public officials and NGOs mount a more effective
response.
Future additional research will need to confirm our findings in terms of
food hardship and the usefulness of real-time data for the analysis of other
dimensions of economic distress. Food hardship is just one dimension of severe
deprivation, and resorting to a foodbank is only one of the strategies that fam-
ilies in need undertake to cope with falling income. Forthcoming studies should
analyze the usefulness of our approach when times are less extraordinary than
those brought about by a pandemic. Furthermore, we need a better understand-
ing of how real-time data, most likely capturing the short-run effect of a sudden
29Notably, the search intensity measure for “foodbank” did not reach the levels of the first wave.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
20
Review of Income and Wealth, Series 0, Number 0, Month 2022
economic shock, can serve to complement other measures of food scarcity
derived from surveys.
References
Adda, J., “Economic Activity and the Spread of Viral Diseases: Evidence from High Frequency Data,”
The Quarterly Journal of Economics, 131(2), 891–941, 2016.
Ahn, S., and F. B. Norwood, “Measuring Food Insecurity During the Covid-19 Pandemic of Spring
2020,” Applied Economic Perspectives and Policy, 43(1), 162–68, 2021.
Alaimo, K., C. M. Olson, and E. A. Frongillo, “Food Insufficiency and American School-Aged
Children’s Cognitive, Academic, and Psychosocial Development,” Pediatrics, 108(1), 44–53, 2001.
Baek, C., P. B. McCrory, T. Messer, and P. Mui, “Unemployment Effects of Stay-at-Home Orders:
Evidence from High Frequency Claims Data,” The Review of Economics and Statistics, forthcom-
ing, 2020.
Baker, S. R., and A. Fradkin, “The Impact of Unemployment Insurance on Job Search: Evidence from
Google Search Data,” The Review of Economics and Statistics, 99(5), 756–68, 2017.
Bauer, L., A. Pitts, K. Ruffini, and D. W. Schanzenbach, “The Effect of Pandemic EBT on Measures of
Food Hardship. The Hamilton Project,” Economic Analysis, 2020.
Belot, M., S. Choi, E. Tripodi, E. van den Broek-Altenburg, J. C. Jamison, and N. W. Papageorge,
“Unequal Consequences of Covid 19: Representative Evidence from Six Countries,” Review of
Economics of the Household, 19(3), 769–83, 2021.
Bhattacharya, J., T. DeLeire, S. Haider, and J. Currie, “Heat or Eat? Cold-Weather Shocks and Nutrition
in Poor American Families,” American Journal of Public Health, 93(7), 1149–54, 2003.
Bitler, M., H . W. Hoynes, and D . W. Schanzenbach, “The Social Safety Net in the Wake of Covid-19,”
Brookings Papers on Economic Activity, 2020.
Borjas, G. J., “Food Insecurity and Public Assistance,” Journal of Public Economics, 88(7), 1421–43,
2005.
Brodeur, A., A. E. Clark, S. Fleche, and N. Powdthavee, “Covid-19, Lockdowns and Well-Being:
Evidence from Google Trends,” Journal of Public Economics, 193, 104346, 2021.
Burke, M. P., L. H. Martini, E. Çayír, H. L. Hartline-Grafton, and R. L. Meade, “Severity of Household
Food Insecurity is Positively Associated with Mental Disorders Among Children and Adolescents
in the United States,” The Journal of Nutrition, 146(10), 2019–26, 2016.
Case, A., A. Fertig, and C. Paxson, “The Lasting Impact of Childhood Health and Circumstance,”
Journal of Health Economics, 24(2), 365–89, 2005.
Chetty, R., J. N. Friedman, N. Hendren, M. Stepner, and the Opportunity Insights Team, “The
Economic Impacts of covid-19: Evidence from a New Public Database Built Using Private Sector
Data,” NBER Working Papers 27431, National Bureau of Economic Research, 2020.
Choi, H., and H. Varian, “Predicting the Present with Google Trends,” Economic Record, 88(s1), 2–9,
2012.
Cicala, S., “The Incidence of Extreme Economic Stress: Evidence from Utility Disconnections,” Journal
of Public Economics, 200, 104461, 2021.
Crossley, T. F., P. Fisher, and H. Low, “The Heterogeneous and Regressive Consequences of Covid-19:
Evidence from High Quality Panel Data,” Journal of Public Economics, 193, 104334, 2021.
Currie, J., “Healthy, Wealthy, and Wise: Socioeconomic Status, Poor Health in Childhood, and Human
Capital Development,” Journal of Economic Literature, 47(1), 87–122, 2009.
Dang, H.-A., T. L. D. Huynh, and M.-H. Nguyen, “Does the Covid-19 Pandemic Disproportionately
Affect the Poor? Evidence from a Six-Country Survey,” Technical Report 13352, IZA, 2020.
Davison, K., G. Marshall-Fabien, and A. Tecson, “Association of Moderate and Severe Food Insecurity
with Suicidal Ideation in Adults: National Survey Data from Three Canadian Provinces,” Social
Psychiatry and Psychiatric Epidemiology, 50(6), 963–72, 2015.
Eurostat, Eu-Silc Guidance Note on the 2020 Data Collection in the Context of the Covid-19 Crisis.
Technical Report, Eurostat, 2020.
Feeding America, The Impact of Coronavirus on Food Insecurity. Technical Report, Feeding America,
2020.
Fetzer, T., L. Hensel, J. Hermle, and C. Roth, “Coronavirus Perceptions and Economic Anxiety,” The
Review of Economics and Statistics, forthcoming, 2021.
Ganong, P., P. Noel, and J. Vavra, “Us Unemployment Insurance Replacement Rates During the
Pandemic,” Journal of Public Economics, 191, 104273, 2020.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
21
Review of Income and Wealth, Series 0, Number 0, Month 2022
Gundersen, C., M. Hake, A. Dewey, and E. Engelhard, “Food Insecurity During Covid-19,” Applied
Economic Perspectives and Policy, 43(1), 153–61, 2021.
Gundersen, C., and J. P. Ziliak, “Food Insecurity and Health Outcomes,” Health Affairs, 34(11), 1830–
39, 2015.
Hamelin, A.-M., M. Beaudry, and J.-P. Habicht, “Characterization of Household Food Insecurity in
Québec: Food and Feelings,” Social Science & Medicine, 54(1), 119–32, 2002.
Han, J., B. D. Meyer, and J. X. Sullivan, “Income and Poverty in the Covid-19 Pandemic,” Brookings
Papers on Economic Activity, 2020, 2020.
Howard, L. L., “Does Food Insecurity at Home Affect Non-Cognitive Performance at School? A
Longitudinal Analysis of Elementary Student Classroom Behavior,” Economics of Education
Review, 30(1), 157–76, 2011.
Jun, S.-P., H. S. Yoo, and S. Choi, “Ten Years of Research Change Using Google Trends: From the
Perspective of Big Data Utilizations and Applications,” Technological Forecasting and Social
Change, 130, 69–87, 2018.
Jyoti, D. F., E. A. Frongillo, and S. J. Jones, “Food Insecurity Affects School Children’s Academic
Performance, Weight Gain, and Social Skills,” The Journal of Nutrition, 135(12), 2831–39, 2005.
Kong, E., and D. Prinz, “Disentangling Policy Effects Using Proxy Data: Which Shutdown Policies
Affected Unemployment During the Covid-19 Pandemic?” Journal of Public Economics, 189,
104257, 2020.
Lombe, M., K. Wang, Y. Chu, and V. E. Nebbitt, “The Impact of the Recession on Food Insecurity
Among Households Who were Low Income: Findings from the 2005–2014 National Health and
Nutrition Examination Surveys,” Journal of Poverty, 22(5), 437–53, 2018.
Mayasari, N. R., D. K. N. Ho, D. J. Lundy, A. V. Skalny, A. A. Tinkov, I.-C. Teng, M.-C. Wu, A.
Faradina, A. Z. M. Mohammed, J. M. Park, Y. J. Ngu, S. Aliné, N. M. Shofia, and J.-S. Chang,
“Impacts of the Covid-19 Pandemic on Food Security and Diet-Related Lifestyle Behaviors: An
Analytical Study of Google Trends-Based Query Volumes,” Nutrients, 12(10), 3103, 2020.
Niles, M. T., F. Bertmann, E. H. Belarmino, T. Wentworth, E. Biehl, and R. Neff, “The Early Food
Insecurity Impacts of Covid-19,” Nutrients, 12(7), 2096, 2020.
Nord, M., and L. S. Kantor, “Seasonal Variation in Food Insecurity is Associated with Heating and
Cooling Costs Among Low-Income Elderly Americans,” The Journal of Nutrition, 136, 2939–44,
2006.
Nuti, S., B. Wayda, I. Ranasinghe, S. Wang, R. Dreyer, S. Chen, and K. Murugiah, “The Use of Google
Trends in Health Care Research: A Systematic Review,” PloS ONE, 9(10), e109583 2014.
Palomino, J. C., J. G. RodríÂÂguez, and R. Sebastian, “Wage Inequality and Poverty Effects of
Lockdown and Social Distancing in Europe,” European Economic Review, 129, 103564, 2020.
Parolin, Z., and C. Wimer, “Forecasting Estimates of Poverty During the Covid-19 Crisis,” Poverty and
Social Policy Brief, 4(6), 2020.
Pew Research Center, “Demographics of Internet and Home Broadband Usage in the United States,”
2021. Retrieved July 23, 2021, from https://www.pewre search.org/inter
net/fact-
sheet/
inter
net-
broadband/?menuItem=6b886b10-55ec-44bc-b5a4-740f5366a404.
Qiu, Y., X. Chen, and W. Shi, “Impacts of Social and Economic Factors on the Transmission of
Coronavirus Disease 2019 (Covid-19) in China,” Journal of Population Economics, 33, 1127–72,
2020.
Raifman, J., J. Bor, and A. Venkataramani, “Unemployment Insurance and Food Insecurity Among
People Who Lost Employment in the Wake of Covid-19,” medRxiv, 2020.
Ratcliffe, C., S.-M. McKernan, and S. Zhang, “How Much Does the Supplemental Nutrition Assistance
Program Reduce Food Insecurity?” American Journal of Agricultural Economics, 93, 1082–98,
2011.
Restrepo, B. J., M. P. Rabbitt, and C. A. Gregory, “The Effect of Unemployment on Food Spending and
Adequacy: Evidence from Coronavirus-Induced Firm Closures,” Applied Economic Perspectives
and Policy, 43(1), 185–204, 2021.
Sastry, N., K. McGonagle, and P. Fomby, “Effects of the Covid-19 Crisis on Survey Fieldwork:
Experience and Lessons from Two Major Supplements to the U.S. Panel Study of Income
Dynamics,” Survey Research Methods, 14(2), 241–45, 2020.
Schanzenbach, D., and A. Pitts, How Much Has Food Insecurity Risen? Evidence from the Census
Household Pulse Survey. Institute for Policy Research, Northwestern University, 2020a.
, Estimates of Food Insecurity During the Covid-19 Crisis: Results from the Covid Impact Survey,
Week 1 (April 20–26, 2020). Northwestern University,Institute for Policy Research, 2020b.
, Food Insecurity in the Census Household Pulse Survey Data Tables. Institute for Policy Research,
Northwestern University, 2020c.
Schaurer, I., and B. Weiß, “Investigating Selection Bias of Online Surveys on Coronavirus-Related
Behavioral Outcomes,” Survey Research Methods, 14(2), 103–08, 2020.
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
22
Review of Income and Wealth, Series 0, Number 0, Month 2022
Seligman, H. K., B. A. Laraia, and M. B. Kushel, “Food Insecurity is Associated with Chronic Disease
Among Low-Income Nhanes Participants,” The Journal of Nutrition, 140(2), 304–10, 2009.
Shaefer, H. L., and I. A. Gutierrez, “The Supplemental Nutrition Assistance Program and Material
Hardships Among Low-Income Households with Children,” Social Service Review, 87(4), 753–79,
2013.
Sumner, A., C. Hoy, and E. Ortiz-Juarez, “Estimates of the Impact of Covid-19 on Global Poverty.”
Technical Report 2020/43, WIDER, Helsinki, Finland, 2020.
Vosen, S., and T. Schmidt, “Forecasting Private Consumption: Survey-Based Indicators vs. Google
Trends,” Journal of Forecasting, 30(6), 565–78, 2011.
Ziliak, J. P., “Food Hardship During the Covid-19 Pandemic and Great Recession,” Applied Economic
Perspectives and Policy, 43, 132–52, 2021.
Supporting Information
Additional supporting information may be found in the online version of this
article at the publisher’s web site:
Figure A.1: Percentage of low food secure households in the US, 2019
Figure A.2: Percentage of low food secure households and very low food
secure house-holds in the US, 1995–2019
Figure A.3: Weekly average of the search intensity index for “foodbank” in
Google Trends, data from 30 different extractions, US, 1 January –30 June 2020
Figure A.4: Most common related terms searched in Google Trends that
include “food bank,” US, January –June 2020
Figure A.5: Weekly average of the search intensity index for “food stamps”
and “SNAP” in Google Trends, US, 1 January –30 June 2020
Figure A.6: Event study results for the search intensity index for “foodbank”
in Google Trends, US, 1 January –30 June 2020
Figure A.7: Event study results for the search intensity index for “foodbank”
in Google Trends, placebo exercise, 2018–2019, US
Figure A.8: Weekly average of the search intensity index for “foodbank” in
Google Trends, US, 1 January 2020 –7 February 2021
Table A.1: Dates of the state of emergency declaration and beginning and end
of the stay-at-home orders by state by 30 June 2020, United States
© 2022 The Authors. Review of Income and Wealth published by John Wiley & Sons Ltd on behalf of
International Association for Research in Income and Wealth
23
File and source
- File
- food-hardship-in-the-us-during-the-pandemic-what-can-we-learn-from-real-time-data.pdf
- Size
- 256,719 bytes
- SHA-256
- 24ebd55f4398be3684f241206a5a7b4c7ec0a125327363e7082e2aa78da6ff10
- Original
- doi.org