Policy Research Working Paper 9673
Death and Destitution
The Global Distribution of Welfare Losses
from the COVID-19 Pandemic
Francisco H. G. Ferreira
Olivier Sterck
Daniel Mahler
Benoît Decerf
Development Economics
Development Research Group
May 2021
Policy Research Working Paper 9673
Abstract
The COVID-19 pandemic has brought about massive when a more relative approach is taken to poverty lines. In
declines in well-being around the world. This paper seeks both cases, the poverty burden of the pandemic, relative
to quantify and compare two important components of to the mortality burden, is much higher for poor countries.
those losses—increased mortality and higher poverty— The distribution of aggregate welfare losses—combining
using years of human life as a common metric. The paper mortality and poverty and expressed in terms of life-years
estimates that almost 20 million life-years were lost to —depends on the choice of poverty line(s) and the relative
COVID-19 by December 2020. Over the same period and weights placed on mortality and poverty. With a constant
by the most conservative definition, more than 120 million absolute poverty line and a relatively low welfare weight
additional years were spent in poverty because of the pan- on mortality, poorer countries are found to bear a greater
demic. The mortality burden, whether estimated in lives welfare loss from the pandemic. When poverty lines are
or years of life lost, increases sharply with gross domestic set differently for poor, middle-income, and high-income
product per capita. By contrast, the poverty burden declines countries and/or a greater welfare weight is placed on
with per capita national income when a constant absolute mortality, upper-middle-income and rich countries suffer
poverty line is used, or is uncorrelated with national income the most.
This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the
World Bank to provide open access to its research and make a contribution to development policy discussions around the
world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may
be contacted at bdecerf@worldbank.org.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
Death and Destitution:
The Global Distribution of Welfare Losses from the COVID-19 Pandemic
Francisco H. G. Ferreira, Olivier Sterck, Daniel Mahler and Benoît Decerf 1
Keywords: COVID-19; pandemic; welfare; poverty; mortality; global distribution
JEL Codes: D60, I15, I31, I32
1Ferreira is at the London School of Economics and IZA; Sterck is at the University of Oxford; Decerf and Mahler
are at the World Bank. We are grateful to Lykke Andersen, Angus Deaton, Martin Ravallion, Andres Velasco and
participants at the Lemann Center seminar at Stanford University for comments on an earlier version. All
remaining errors are our own, as are all the views expressed.
1. Introduction
Since its onset in December 2019, the COVID-19 pandemic has spread death and disease across the whole
world. Around the time of its “first anniversary” on December 15, 2020, 1.64 million people were counted
as having lost their lives to the virus globally 2 and, because of the likelihood of under-reporting, that was
almost certainly an undercount. Although it is primarily a health crisis – with substantial additional pain
and suffering caused to the tens of millions who have survived severe cases of the disease and, in many
cases, continue to suffer from long-term sequels – the pandemic has also had major economic effects.
The current estimate is that global GDP per capita declined by 5.3% in 2020. Economic contraction was
widespread, with 172 of the 182 countries for which data is available experiencing negative growth in real
GDP per capita in 2020 (World Bank, 2021).
This severe global economic shock has caused the first reversal in the declining trend in global extreme
poverty (measured as the share of the world’s population living under $1.90 per day) since the Asian
Financial Crisis of 1997 – and only the second real increase in world poverty since measurement began in
the early 1980s. 3 This increase in extreme deprivation comes with its own suffering and anguish: jobs and
homes were lost and people struggled to feed their children and themselves. Many asked whether they
“would die of Coronavirus or hunger?” 4
This paper seeks to address two questions. First, what were the relative contributions of increased
mortality and poverty to the welfare losses caused by the pandemic, and did these contributions vary
systematically across countries? Second, how were the aggregate welfare losses distributed across
countries? To do so, we revisit and update some of our earlier findings in Decerf et al. (2020, henceforth
DFMS): that paper did not address the second question above and, conversely, we do not explore
counterfactual herd immunity scenarios here, as we did there. The first question above was also
addressed in the earlier article using mortality and poverty estimates from June 2020, which are updated
here to December 2020.
Our motivation is to look at the impact on the cross-country distribution of well-being with the recognition
that both the health crisis and the economic debacle have caused huge welfare losses. We focus on
extreme outcomes in both domains: mortality in the case of health and falling into poverty as the
economic outcome. This implies that our estimates of welfare losses are clearly a lower-bound: we ignore
the burden of the disease on those who survive it and, furthermore, data limitations mean that we look
only at deaths officially classified as due to COVID-19, rather than the possibly preferable metric of excess
deaths. 5 Similarly, we ignore welfare losses from income declines that do not entail entry into poverty
2
Our World in Data, https://ourworldindata.org/grapher/cumulative-covid-deaths-region.
3
There was also an apparent increase in 1989, which is fully accounted for by China switching from an income to a
consumption indicator. See https://data.worldbank.org/indicator/SI.POV.DDAY.
4 On February 11, 2021, a Google search for “dying of coronavirus or hunger” yielded 2,330,000 results. Some of
the titles on the first page included “The pandemic pushes hundreds of millions of people toward starvation and
poverty” (Washington Post, September 25, 2020) and “More people may die from hunger than from the
Coronavirus this year…” (Forbes, July 9,2020).
5 Even if data on excess mortality were more widely available, it would also include additional deaths caused by
poverty, potentially confounding our comparisons.
2
and, furthermore, we look only at the short-term poverty consequences arising from the
contemporaneous income losses in 2020. We ignore, therefore, the longer-term consequences of any
insults to child development arising from additional undernutrition, or the likely substantial future
consequences of the schooling crisis that resulted from the pandemic (see, e.g. Lustig et al, 2020). These
choices are not intended to minimize the importance of those negative consequences of the pandemic.
On the contrary, the evidence to date suggests that they will be extremely important. Rather, they follow
from a desire to focus on the most severe short-term consequences of the crisis along the two principal
dimensions of health and incomes, using the best available data while avoiding an accumulation of
assumptions and simulations. We are forced to make some assumptions to fill data gaps that inevitably
arise when analyzing an ongoing phenomenon, but they are few and therefore are hopefully clearer and
more transparent than if we had tried to incorporate expected future losses, and so on.
A welfare-based approach requires comparing health and income losses or, in our case, mortality and
poverty costs. As in DFMS, we eschew more traditional methods such as the valuation of statistical life
(VSL, see e.g. Viscusi, 1993). Our approach is theoretically closer to the modeling of social welfare as
aggregate expected lifetime utility, as in Becker et al. (2005) or Adler et al. (2020). But, unlike those
authors, we model the effect of the pandemic on social welfare in a way that allows us to use years of
human life – either lost to premature mortality or spent in poverty – as our unit of comparison. This has
two advantages over the alternative of using a money metric to value human lives: first, we hope it
overcomes the instinctive aversion of many participants in the public debate to the idea of placing a
“price” on human life. Second, the model yields a single, easily understandable normative parameter for
the trade-off between mortality and poverty which has a direct, observable empirical counterpart. We
can then simply present the empirical object for all countries in our sample, and let the reader compare
her own valuation of the normative parameter to the data.
The simple model is presented in DFMS, and we do not repeat it here. The basic ingredients are (i) a
utilitarian welfare function that simply adds up lifetime individual utility across people and time periods;6
(ii) an individual utility function that depends solely on whether one is dead, poor or non-poor; 7 and (iii)
an assumption that the pandemic may have two effects on people: it may (or may not) cause them to die
earlier than they otherwise would, and it may (or may not) cause them to spend some additional years in
poverty before dying, relative to the counterfactual. This simple framework yields the result that the
overall welfare effect of the pandemic is proportional to a weighted sum of the number of years of life
lost to premature (COVID-19-induced) mortality and the number of additional years spent in poverty.
Equation (1) in DFMS, which contains this result, is reproduced below:
∆������������
= ������������������������������������ + ������������������������ (1)
∆������������������������
6
With no aversion to inequality and no time discounting.
7 Simple restrictions imply that this is a step-function approximation to utility that is increasing and concave in
incomes, but the coarseness we introduce means that people are insensitive to income gains or losses that do not
entail a crossing of the poverty line. This simplification may be seen as the price we pay for converting to a life-year
metric, but it is also consistent with our emphasis – discussed above – on the extreme outcomes of death and
destitution (defined as falling into poverty).
3
where ∆������������ denotes the expected impact of the pandemic on social welfare; ∆������������ ������������ is the difference in yearly
individual utility between being poor and non-poor; and LY and PY are respectively the total number of
years of life lost and the total number of additional years spent in poverty due to the pandemic. α is a
normative parameter that represents the ratio of the individual utility loss from each year lost to
premature mortality (∆������������ ������������) to the loss from each additional year spent in poverty (∆������������������������). α is therefore
the (social) marginal rate of substitution between life- and poverty-years. It can be understood as the
“shadow price” of a lost life-year, expressed in terms of poverty years. We think of it as a societal
parameter, and therefore as some aggregation across individuals of the answer they might give to the
following hypothetical question: “If you could make this bargain, how many years would you be willing to
spend in poverty during the rest of your life in order to add one additional year at the end of your life?”
Clearly, there is plenty of room for individual disagreement about the value of α. Different people might
answer that question very differently, depending on how far above the poverty line they are (or expect to
be); on their expected residual life-expectancy and, of course, on their preferences. We thus choose to
remain mostly agnostic about α. In what follows, we simply present values for LY and PY across as many
countries as possible. When we discuss the relative contributions of mortality and poverty, we present
ratios of PY to LY, which the reader can compare to her own preferred value for α in order to assess which
source is responsible for the larger welfare loss. Next, when we seek to summarize the inter-country
distribution of welfare losses, we suggest a plausible range for α: between five and twenty years.
The rest of this short paper is organized as follows. The next section describes how we compute the
numbers of years of life lost (LY) and additional poverty years (PY) for each country, given the available
data, and presents the estimates for 145 countries. Building on those ingredients, Section 3 summarizes
the evidence on the relative importance of poverty and mortality in lowering welfare around the world.
It also investigates the global distribution of those losses under plausible values for the key normative
parameter, both for a constant absolute poverty line and under a more relativist approach to poverty
identification. Section 4 concludes.
2. Estimating additional mortality and poverty with imperfect data
The two key data sources for our mortality estimates are the UN Population Division database and the
Global Burden of Disease Database (Dicker et al., 2018). Combining data from these two sources and using
smoothing procedures described in the Appendix to DFMS, we obtain estimates for the number of people
of age a living in country j, Naj, for 0 ≤ ������������ ≤ 99. 8 Using these country-specific population pyramids, we can
also construct estimates for age-specific residual life expectancies, ������������ ������������������������. If we observed age-specific COVID-
19-induced mortality rates maj, for all countries in our sample, then our estimate of years of life lost due
to the pandemic in each country would be given by the sum across all ages of the number of deaths from
that age group (������������������������������������ ������������������������������������), times the average residual life-expectancy at that age: 9
8
Those aged 100 or more are counted as being 99.
9 If those who die from COVID-19 are systematically less healthy than those who do not at any given age, then using
the pre-pandemic residual life expectancy to estimate LY may introduce some upward bias.
4
99
������������������������������������ = � ������������������������������������ ������������������������������������ ������������������������������������ (2)
������������=0
The above procedure is exactly what DFMS used for six countries for which we had age-specific COVID-
19-related mortality data. As in that earlier paper, however, we do not observe COVID-19 mortality data
disaggregated by age for most countries in our global sample. In addition to the population pyramid (Naj),
we do observe the overall COVID-19 mortality rate, mj. If we assume a constant infection rate across ages
in each country, ������������ ������������, then the following identity for the number of COVID-19 deaths would hold:
99 99
������������������������ � ������������������������������������ = � ������������������������������������ ������������ ������������������������������������������������ (3)
������������=0 ������������=0
where ������������������������������������ is the age-specific infection fatality rate (IFR) in country j. Since IFRs are also not observed for
each age group for many countries, we follow DFMS in imputing the infection fatality rate estimated for
France by Salje et al. (2020) to all high-income countries in our sample, and the IFR estimated for China
by Verity et al. (2020) to all other countries. 10 Under those assumptions, we can solve Eq. (3) for each
country’s infection rate ������������������������. The estimated age-specific COVID-19 mortality rates for each country, ������������ � ������������������������,
are then given by:
������������
� ������������������������ = ������������������������������������ ������������������������ (4)
Plugging ������������
� ������������������������ into Equation (2) yields our estimates for the number of years of life lost to COVID-19 in
each country, ������������������������������������. We estimate that COVID-19-induced mortality in the year to December 15, 2020 caused
the loss of 19.3 million years of life across the 145 countries in our sample (which account for 96% of the
world’s population). Absolute numbers range from 14 in Burundi to 3,148,000 in the United States. 11
Those aggregates obviously depend a great deal on the country’s population. Figure 1 plots ������������������������������������ adjusted
by population (LY per 100,000 people) against each country’s GDP per capita, with both axes in logarithmic
scale. Two features of the scatter plot are worth highlighting: First there is enormous variation in the
population-adjusted loss of life-years across countries. Even if one discounts Burundi and Tanzania as
outliers where reporting is unlikely to have been reliable, LYs range from roughly one year lost per 100,000
people (in countries as diverse as Papua New Guinea, Thailand and Vietnam) to one or more years lost
per 100 people, in a large set of countries including Brazil, Peru, Mexico, Belgium, the Czech Republic and
the United States. Some of that variation is systematic: Figure 1 reveals a strongly positive (and concave)
10
Alternative approaches to addressing this missing data problem – that we do not observe COVID-19 mortality
disaggregated by age for most countries – are possible. Instead of assuming an age-invariant infection rate for each
country and imputing China’s and France’s IFR to other countries, Heuveline and Tzen (2021) assume that the
shape of the distribution of COVID-19 deaths by age (and gender) in other countries is similar to that of the United
States. Their method and ours generate remarkably similar distributions of years of life lost across countries, with a
correlation in logs of 0.99. Details of the comparison are available from the authors on request.
11 Measurement error is likely to plague a number of these country estimates, with under-reporting being of
particular concern. Deaton (2021) singles out Burundi and Tanzania as likely candidates for under-reporting.
5
relationship between the mortality costs of COVID-19 and level of economic development. 12 Second,
however, there is also considerable variation around the regression line at each income level – particularly
at and above per capita GDP levels of $5,000 or thereabouts: Brazil and Thailand have comparable per
capita income levels, but Brazil lost roughly one thousand life-years for each life-year lost in Thailand,
controlling for population. The disparity is even greater between Bolivia and Vietnam, and still striking
between France and the Republic of Korea.
Figure 1: Life-years lost to COVID-19, and GDP per capita
PER
MKD BEL
COL
BIH BRA MEX
ARM ECU MNEARG
IRN PAN SVN ITA
ESP USA
BOL
MDA BGR CHL CZE GBR
FRA
BLZ SWE CHE
00
ROUHUN LUX
JOR ZAF IRQ HRV POL NLD
CRI PRT AUT IRL
10
HND GEO
TUNALB
PSE UKR
GTM
PRY
RUS ISRMLT CAN
KGZ CPV DOM TUR LTU
GRC
MAR
GUY
SLV LBN SUR SRB
AZE SVK DEU
MDV KAZ
LVA DNK
IND BLR
JAM EST ARE
PHL IDN DZA TTO
NAM
EGY CYP FIN ISL
LY per 100,000 people
GMB NPL MRT NOR
0
BGD MMR
10
KEN PAK
SDN
NIC GAB AUS
YEM
GNB ZWE SENCMR
ZMB
COG LCA URY
LBR HTI
ETH LSO UZB BWA
JPN
MLI GHA AGO MYS
CAF MWI SLE TJK KOR
MDG
TGO COM
NGA LKA MUS
TCD
UGA GIN CIV
10
MOZ RWA
CODNER BFA BEN
CHN
FJI
PNG THA
1
TZA
VNM
BDI
1
0.0.
01
500 1000 2000 5000 10,000 20,000 50,000 100,000
GDP per capita (PPP, constant 2011)
LIC LMIC UMIC HIC
What explains this massive variation in population-adjusted life-years lost to COVID-19 – both across and
within country-income categories? Mechanically, Equation (2) tells us that it must reflect cross-country
differences in three variables: the age structure of the population (������������������������������������ ), residual life-expectancies at each
age (������������ ������������������������), and age-specific mortality rates (������������������������������������). The first two are slow-moving variables that reflect each
country’s historical development; the stage of the demographic transition they are in; access to and the
quality of their health care systems, etc.
The last variable – the country’s age-specific mortality rates, which are themselves the product of infection
rates and infection fatality rates (Eq. 4) – reflects each country’s exposure and response to the pandemic.
12 Deaton (2021) presents a similar figure that plots the log of lives – rather than life-years – lost against log GDP
per capita. He notes that “there is no relationship […] within the OECD.” Goldberg and Reed (2020) also document
that, as of July 2020, the number of lives lost to COVID-19 per million inhabitants was larger in advanced
economies than in developing countries. They suggest that older populations and a greater prevalence of obesity
in developed countries can partially explain this positive association between mortality and development.
6
Infection rates (which vary substantially internationally) initially reflected how quickly the virus arrived in
each country, and then the extent to which health systems were able to prevent spread within the
country. Infection rates are also likely to depend on urbanization and climate. Infection-fatality rates
reflect the quality of health care and the extent to which, for example, hospitals were overwhelmed by
the pandemic at any stage.
Our data suggests that all three of these variables contribute to the positive association between the
population-adjusted mortality burden of the pandemic and national per capita income seen in Figure 1. It
is well-know that COVID-19 mortality varies substantially with age, and that it is much higher for the
elderly. Figure A1 in the Appendix plots the ratio of COVID-19 mortality among those aged 65 and over to
the mortality among 20-39 year-olds in our data: the ratio ranges from around 100 in low and middle-
income countries to between 200 and 300 among high-income countries. 13
Figures A2 and A3 plot indicators for the other two variables, namely the age structure of the population
and age-specific residual life expectancies, both against per capita GDP. Specifically, Figure A2 plots the
share of the population aged 65 or over (which ranges from 3%-4% among the poorest countries to 20%-
25% among some of the richest); and Figure A3 looks at the residual life expectancy at age 65 across
countries – which ranges from 13-15 years among most low-income countries (LICs) to 20-23 at the high-
end, among high-income countries (HICs). These upward sloping curves in Figures A1 – A3 suggest that
all three variables play some role in contributing to the positive slope in Figure 1. COVID-19 mortality is
highly selective on age; richer countries have many more people in the vulnerable, elderly age ranges; and
they tend to have higher life-expectancies at those ages, implying a larger number of years lost per death.
Turning to the estimates of poverty years added by the COVID-19 pandemic, it is important to note, first
of all, that the household surveys from which we generally obtain reasonably reliable estimates of poverty
are not yet available for 2020 in any country. This means that actual data on household incomes or
consumption levels are not available at this time, and one must rely on ex-ante estimates and
approximations.
In that context, following Lakner et al. (2020), our basic approach is to compare “expected” poverty rates
in 2020 under two scenarios: one with COVID-19 and one without. To do this, we use three basic
ingredients. The first is the remarkable collection of household survey microdata from 166 countries
contained in the World Bank’s PovcalNet database. 14 The dates of the latest household surveys in
PovcalNet vary across countries, but all are “aligned” to 2019, using historically documented growth rates
in GDP per capita and a pass-through coefficient to adjust for the fact that growth in mean incomes in
household surveys is typically less than GDP per capita growth measured in the National Accounts. 15 This
procedure, which is carried out internally at the World Bank for nowcasting poverty, assumes no change
in inequality between the last available household survey and 2019.
13
Some of actual variation in this ratio is missing because age-specific mortality data is, as discussed earlier, not
widely available, so we use France’s IFR for all HICs, and China’s for all other countries. This accounts for the sharp
jump between MICs and HICs in Figure A1.
14 We use data from 145 of these 166 countries, for which we can find the required mortality statistics.
15 It is assumed that 85 percent of growth in GDP per capita is passed through to growth in welfare observed in
household surveys in line with historical evidence (Lakner et al., 2020).
7
Our starting point are these distributions of household per capita income or consumption, expressed in
US dollars at PPP exchange rates, aligned for the pre-COVID-19 year of 2019. For any given poverty line z
expressed in per capita terms, the share of the population with incomes below that line in country j is the
headcount measure of poverty incidence, ������������������������2019 (������������). We then obtain our estimate of poverty in the
counterfactual “no-COVID-19 2020” scenario by inflating the 2019 income vector in country j by one plus
the (adjusted) growth rate forecast for 2020 in the January 2020 issue of the Global Economics Prospect
(GEP): ������������������������2020������������ (������������). The poverty estimate for the 2020 COVID-19 scenario, ������������������������2020������������ (������������), is similarly obtained
by inflating the same 2019 income vector in country j by one plus the (adjusted) growth rate forecast for
2020 in the October 2020 Macro and Poverty Outlook report of the World Bank. 16
The idea, of course, is that the January forecasts were produced at a time when COVID-19 was a little
heard-of virus confined to Wuhan province in China, and no macroeconomist had remotely imagined a
pandemic on the scale we have since seen. The October forecasts, on the other hand, were the latest
available for 2020 at the time of writing and reflect the World Bank’s expectations of the pandemic’s
impact on growth around the world. Finally, we assume – conservatively – that the short-run poverty
effect of the pandemic lasts for a single year, so that each additional person in poverty corresponds to
2020������������ (������������) − ������������ 2020������������ (������������).
precisely one additional poverty year: ������������������������
������������ = ������������������������ ������������
As the formula indicates, the number of poverty-years added by the pandemic depends on the poverty
line that is used. Figure 2 presents two of many possible poverty line options: Panel A uses a constant
absolute line for all countries, namely the World Bank’s international (extreme) poverty line of $1.90 per
person per day (Ferreira et al. 2016). While there are good reasons for using the same poverty line in an
international comparison of this kind, there are equally valid arguments for attempting to account for the
fact that basic needs themselves may vary with national income, and that a different (and costlier) bundle
of commodities may be needed to achieve the same welfare (or capabilities) threshold in Austria, say,
than in Afghanistan. This latter view implies that “poverty” means different things (at least in income
terms) in countries where average incomes are vastly different, and often leads to the adoption of relative
or weakly-relative poverty lines (see, e.g. Atkinson and Bourguignon, 2001, and Ravallion and Chen,
2011). 17
Panel B of Figure 2 adopts a (coarse) step-function approximation to a relative poverty line. It uses the
World Bank’s income-class poverty lines proposed by Jolliffe and Prydz (2016). Using a database of
contemporaneous national poverty lines in 126 economies, these authors selected the median values of
per capita poverty lines among low-income countries ($1.90); lower middle-income countries ($3.20);
16
This method assumes that the adjusted growth rates in real GDP per capita accurately reflect the growth (or
shrinkage) in household consumption. With ongoing globalization, the importance of tax havens, and so on, one
might imagine that GDP has further decoupled from household consumption and that other variables from
national accounts--or other economic indicators altogether--could be more informative. Yet Castaneda et al.
(2019) show that, out of more than a thousand variables, the change in real GDP per capita is the second most
predictive variable of changes in household consumption. The only variable that does better is changes in
employment, for which we do not have pre- and post-COVID-19 forecasts for 2020 that are widely comparable
across a large number of countries.
17 Note that this argument is unrelated to differences in prices, which are supposed to be addressed by the use of
PPP exchange rates.
8
upper middle-income countries ($5.50); and high-income countries ($21.70). In Panel B, poverty years are
computed using these income-class specific poverty lines for countries in each income category.
Using the $1.90 line for all countries, we estimate that a total of 121 million additional poverty years were
induced by COVID-19 in the year to December 15, 2020 across the 145 countries in our sample. Absolute
numbers range from -35,800 in Papua New Guinea 18 to 74.2 million in India. Panel A in Figure 2 reveals –
perhaps unsurprisingly - a strongly downward-sloping relationship with GDP per capita, with an average
of 2,568 poverty years per 100,000 people added in low-income countries, as compared to 28.5 years per
100,000 people added in high-income countries.
Figure 2: Poverty-years added by COVID-19, and GDP per capita
Panel A: z = $1.90 Panel B: Income classification poverty lines
00
,0
ZWE
SLE BFA
10
IND
LBR FJI
SDN
MLI YEM BWA
SEN COG BLZ IND
NER RWA NGA
GNB
TCD
COD IRQ
00
TZA CMR
ZMB AGO SVK
EST
ETH LSO NAM ZAF SDN BLZ ARM HRV
KEN CIV HND GRCPRTCZE
,0
MDG BEN
PY per 100,000 people
CAF PHL PHL LKA
PY per 100,000 people
LCA SLE BFA
TGO GMB GEO ALB
LCA
10
MWI HTI ZWE MEX PAN HUN
MUS LTU
UGA LBR JAM DZA
PER ESP
SUR MRT CHL POL ITA
MOZ MRT SEN
KGZ JOR ECU BWA LVA
MLI YEM COL MNE GBR
RWA PAK MDV THA ARG SVN
NPL
COM NIC GEO PER IRQ NER GNB
TCD COG UZB NAM URY TTO FRA
BGD JAM
UZB COD BGD HND NGA CPV IDN ISRMLT IRL
IDN CMR SUR GAB CYP
CPV GTM TZA JPNCAN
00
GHA ETH CIV GTM ZAF
GIN KEN
PAK SLV ECU COL CAF MDG BEN ZMB AGO SLV
MMR
NIC PRY
IRN DEU
BOL DOM ARE
10
BDI MWI TGO
HTI GMB PSE BRA FIN
ALB MNE LSO BOLMAR
LKA MEX UGA COM LBN
FJI ARG MOZ MKD CRI TUR
BEL NLD
MKD GHA GUY EGY
TUN CHN KAZ SWE
AUS USA
BRA PAN NPL AUT
PSE ARM SRB BGR KOR
KGZ ISL
DNK
GUY
00
TJK EGY RUS LUX
GABBGR ROU GIN ROU
CHE
10
PRY BDI
MUS
MMR MAR CRI NOR
VNM TUN MYS
IRN MDA
VNM
BIH
ESP TJK
LVA LTU
0
10
JOR DZA TUR
GRC
BLR
CHN
DOM TTO ITA DNK LUX
0
10
500 1000 2000 5000 10,000 20,000 50,000 100,000
THA CHL PRT
HRV HUN
POL
SVK
EST GBR NLD
AUT
GDP per capita (PPP, constant 2011)
500 1000 2000 5000 10,000 20,000 50,000 100,000 LIC (1.9$ poverty line)
GDP per capita (PPP, constant 2011)
LMIC (3.2$ poverty line)
LIC LMIC UMIC HIC UMIC (5.5$ poverty line)
HIC (21.7$ poverty line)
The negative slope disappears completely in Panel B, where median poverty lines from each country
income category (LICs, LMICs, UMICs and HICs) are used for countries in the respective groups. The linear
regression line across the entire scatterplot has a slope of -0.0286 (p-value = 0.612). The mean number of
PYs/100,000 people is 2,568 for LICs, 2,778 for LMICs, 3,418 for UMICs and 3,330 for HICs. It is perhaps
worth noting that there is nothing mechanical about this particular result: there is no reason why one
would necessarily expect that adopting median poverty lines among groups of progressively richer
countries would completely eliminate the negative association between the poverty burden of the
pandemic and GDP per capita. Using these more generous poverty lines, our estimate for the total number
of additional poverty years induced by the pandemic rises to 300 million.
The next section seeks to combine the country-level PY and LY estimates obtained above in order to assess
their relative and absolute importance in determining overall welfare losses from the pandemic.
18Papua New Guinea is the only country for which the October 2020 growth forecast for 2020 was higher than the
January 2020 forecast.
9
3. Total welfare losses and the relative contributions of death and destitution
Given the estimates of life-years (LYs) lost and of additional poverty-years (PYs) due to the pandemic, we
now ask: first, which of the two sources contributed the most to lowering welfare in each country; and
second, what were those total welfare costs, using our metric of poverty years. The answers to both
questions depend critically on the value of α, the normative parameter that tells us how many PYs cause
as great a welfare loss as a single LY. Given a value of α and the ratio of PYs to LYs observed in any particular
country, we can immediately tell whether poverty or mortality contributed the most to the aggregate
welfare losses from the pandemic in that country. If we denote that empirical ratio by ������������ ������������ , then
������������ ⁄������������������������
������������� = ������������������������
an observer with ������������ > ������������ ������������� will regard mortality as the principal source of welfare loss in country j.
Conversely, if ������������ < ������������
�������������, poverty is considered the greatest source of welfare loss.
Figure 3 plots those observed ratios ������������ ������������� against GDP per capita for all countries in our sample, using the
income-class specific poverty lines used in Panel B of Figure 2. (The line is even steeper if the constant
$1.90 line were used instead.) Two things are immediately apparent: first, the variation in empirical PY/LY
������������� is 15.7 (in United Arab Emirates) and the range is from 0.33 (in Bosnia
ratios is enormous: the median ������������
and Herzegovina) to 5,537 (in Burundi). 19 Second, the poverty to life-years ratio is strongly negatively
correlated with GDP per capita: although the regression lines shown in Figure 3 are for each income class,
a simple linear regression over the entire sample has a slope of -1.043, (p-value = 0.000).
Figure 3: PY/LY ratios and GDP per capita
00
BDI FJI
TZA
,0
THA
10
BFA VNM
00
NER LKA
COD RWA MUS
SLE BEN
10
TCD CIV
MLI NGA CHN
MOZ
LBR LCA
Break-even α
UGA
MDG ZWE BWA
̂
CAF TGO SDN
MWI COM SEN
YEM CMR COG AGO UZB URY JPN
GNBETH GIN
LSO GHA IND
HTI
0
ZMB PAK PHL EST KOR
MRT DZA GAB
10
KEN BGD NIC
JAM
MMR NAMIDN SVK
TTO CYP AUS
GMB LVA
MYS
TJK MDV GRC FIN
EGY LTU ARE
NPL KGZ GEO ALB IRQ PRT
CPVSLV SUR HRV ISL
BLZ
HND GTM ARM
JOR KAZ ISRMLT DEU
POL CZE
HUN
10
MAR LBN DOM CAN NOR
PSE GUY PRY PAN DNK IRL
MEX CHL TUR ESP
ITAGBR
ECU ZAF SRB MNEARG
TUNPERCOL CRI SVN FRA
RUS AUTNLD
BOL IRN SWE
BRA
USA LUX
MKD BGR ROU
BEL CHE
1
BLR
MDA
BIH
1
0.
500 1000 2000 5000 10,000 20,000 50,000 100,000
GDP per capita (PPP, constant 2011)
1.9$ poverty line
3.2$ poverty line
5.5$ poverty line
21.7$ poverty line
19Looking only at positive values, and thus excluding Papua New Guinea (see previous footnote) and Azerbaijan
(zero PYs reported).
10
So far, we have been entirely agnostic about the value of α. Indeed, we argue that one advantage of our
approach is that it can encapsulate the normatively challenging trade-off between lives and livelihoods in
a single, easily interpretable parameter, while simultaneously remaining agnostic about its value. In order
to make further progress in interpreting Figure 3, however, it will prove helpful to suggest a “plausible
range” for α, which we set at 5 ≤ ������������ ≤ 20. In terms of the question we proposed earlier as a means to
elicit the normative judgement, this range means that most people in a country value an additional year
of life-expectancy as being worth spending at least an additional five years, and at most an additional
twenty years, in poverty.
We do not yet have robust empirical evidence from surveys or experiments that try to elicit empirical
values of α, so the reader may of course pick a value completely outside that range. 20 But those who are
comfortable with such a range could, on inspection of Figure 3, classify countries into three broad groups.
Those below the band of grey shadow (������������ ������������� ≤ 5) are countries where the social welfare cost of the
pandemic until December 2020 arose primarily from additional mortality, rather than from increases in
poverty. This group includes a wide variety of nations, such as Bolivia, Brazil, the Russian Federation,
Belgium and the United States. The most extreme cases are Moldova, Bosnia and Herzegovina and Belarus
������������� is below the theoretical unity lower bound for α. Belgium and the Czech Republic are close
where ������������
behind. This first group consists primarily of upper-middle and high-income countries and does not include
a single low-income country.
A second group consists of those above the band of grey shadow in the figure (������������ ������������� ≥ 20). In these
countries, increased destitution contributed more to declining social welfare than deaths and the loss of
life years they caused. There are 70 countries in this group, including most low- and lower middle-income
countries. This group also includes most countries frequently identified in the popular media as successful
in combating the pandemic, through early lockdowns strictly enforced and/or well-functioning testing and
tracing systems, such as Australia, China, Japan, Korea and Vietnam. Uruguay is the only continental Latin
American country in this group, while its neighbors Argentina and Brazil (as well as Chile) are in Group 1
above. 21
The third group consists of those countries in the grey band (5 ≤ ������������
������������� ≤ 20), whose empirical PY/LY ratios
fall within our “plausible range” for α. Given that range, we are unable or unwilling to select either poverty
or mortality as the main culprit in lowering social welfare in these countries. In other words, these are
countries where their relative contributions were broadly similar. The group includes countries from every
income category, from Nepal at the poorer end to Norway and the United Arab Emirates at the richer end.
But Nepal and Tajikistan are the only two low-income countries in the group; all other LICs are in Group
2, where poverty dominated mortality as a source of declining well-being from the pandemic.
Can we go beyond the relative contributions of mortality and poverty in each country and compute the
aggregate welfare losses from the pandemic (arising from deaths and additional destitution) in each
country? Strictly speaking, of course, we suffer from the usual problem of choosing a suitable unit for
20
Preliminary results from surveys we have conducted in the United States, the United Kingdom and South Africa
suggest very low values for α. The mean across the three samples was 2.6. Work on these surveys is ongoing, and
these early results are tentative and should be treated with caution. Nonetheless, they suggest that, if anything,
our plausible range for α is on the high side, which would lead the results that follow to place “too much” weight
on mortality, relative to poverty.
21 In the Caribbean, Jamaica, St. Lucia and Trinidad and Tobago are also in Group 2.
11
measuring well-being. Under our simple model, briefly described in the Introduction, the change in
welfare in country j is given by ������������������������������������������������� + ������������������������
������������ �∆������������������������, where ∆������������ ������������ is expressed in utility terms. However, since
we have assumed that the quantity ∆������������������������ is constant across individuals (and countries), we can do the next
best thing and use the weighted sum of life-years and poverty-years ������������������������������������������������� + ������������������������ ������������ � as our measure of social
welfare loss, which corresponds to using additional years spent in poverty as our social welfare metric.
Once again, this exercise does require choosing both an approach to poverty identification (one or more
poverty lines) as well as one or a range of values for ������������ . For consistency, we use the same two sets of
poverty lines used in Figure 2 (a constant line of $1.90, and the set of four median lines for LICs, LMICs,
UMICs and HICs). For ������������ we pick the bounds of our guidance range, namely ������������ = 5 and ������������ = 20. Figure 4
below plots aggregate social welfare losses, ������������������������������������������������ + ������������������������������������, against GDP per capita for all 145 countries in our
sample: The first row (Panels A and B) uses the constant poverty line of $1.90: Panel A uses ������������ = 5 and
Panel B uses ������������ = 20. The second line (Panels C and D) uses the income-classification poverty lines: once
again with ������������ = 5 in Panel C and ������������ = 20 in Panel D. To control for population sizes, in all cases the welfare
costs are expressed per 100,000 people.
As expected, there is a great deal of cross-country variation in the welfare burden of the pandemic,
regardless of which panel one looks at. The unit of measurement along the y-axis, as noted earlier, is
additional person-years spent in poverty per 100,000 people. In panel A, using the most stringent global
poverty line ($1.90 per person per day) and a relatively low value of ������������ , the burden ranges from 26 in
Thailand and 62 in China, to 7,556 in Belgium and 9,811 in Peru. The regression line across the scatter of
countries in Panel A is downward sloping and the negative slope (-0.14) is statistically significant (p=0.036),
indicating a negative association between national income and the aggregate welfare losses from the
deaths and destitution caused by the pandemic. Given the strong positive association between the
mortality burden and GDP per capita seen in Figure 1, this is an important finding: the pandemic appears
to have induced such large increases in extreme poverty in poor countries (Figure 2A) that the combined
burden of poverty and mortality is, on average, greater for them than for richer countries. 22
The pattern changes (and the overall numbers are mechanically higher) in Panel B, where each life-year
lost is counted as being equivalent to 20, rather than 5, added poverty-years. The association with GDP
per capita now becomes positive and statistically significant, because mortality was much higher in richer
countries.
Panels C and D would be preferred by those who take a more relative view of poverty. They use poverty
lines that are typical of countries in their income ranges: $1.90 for low-income countries; $3.20 for lower
middle-income countries; $5.50 for upper middle-income countries; and $21.70 for high-income
countries. Naturally, this raises the number of poverty years in all but the low-income countries, and the
positive slope of the relationship with GDP per capita strengthens further, first with ������������ = 5, and even more
with ������������ = 20. Using more demanding poverty lines in upper middle-income countries means that ranks
change at the bottom, with Tajikistan and Vietnam now reporting the lowest welfare burdens in the world,
22Yet, there is so much variation around the regression line that, even with this parameter configuration, the
greatest losses in welfare were recorded in upper middle-income countries such as Belize, Macedonia and Peru,
alongside rich countries like Belgium. But poorer countries such as Burkina Faso, Sierra Leone and Zimbabwe are
not far behind.
12
at 376 and 422 years per 100,000 people respectively (with ������������ = 5). Fiji (15,855) and Peru (13,631) now
occupy the top ranks in the distribution of population-adjusted welfare losses.
Figure 4: Total welfare losses from the pandemic, and GDP per capita
Panel A: z= $1.90, ������������ = 5 Panel B: z=$1.90, ������������ = 20
PER PER
Aggregate welfare losses for α=20 (PL=1.9$)
Aggregate welfare losses for α=5 (PL=1.9$)
00
BEL
BLZ MKD BEL MKD
COL
,0
ZWE COL MEX BRA MEX
SLE BFA IND ECU BIH BRAMNEARG ECU BIH MNEARG PAN SVN ITA
ESP
10
ZAF SVN ITA BLZ ARM IRN GBR USA
LBR SDN BOL ARM IRN PAN ESP
USA BOL
MDA BGR CHL CZE FRA
BGR CHL GBR
CZE FRA
MLI YEM HND MDA IRQ
BWA ZAF ROUHUN SWE CHE LUX
IRQ
RWA
SEN COG
NGA
ROUHUN SWE CHE LUX JOR HRV POL NLD
NER GNBTCD GEO CRI PRT
JOR HRV POL NLD HND GEO AUT
COD NAM ALB SURCRI ALB IRL
CMRZMB AGO GTM
PHL PRT AUT TUN
00
ETH TZA
KEN
LSO IRL INDGTM PRY CAN
CPV TUN ISRMLT
CAF MDG BEN GMB CIV PSE PRY
LCA ZWE PSE UKR
CAN SUR RUS LTU
GRC
,0
HTI
TGO KGZ SLV ISRMLT SLE BFA KGZ CPV TUR
MWI MRT UKR LTU DOM
RUS
GRC SDN SLV
10
UGA JAM TUR LBR MAR
GUY SRB DEU
NPL GUY LBN AZE SVK
MOZ MAR IDN DOM BWA
BGD NIC SRB DEU MLI YEM SEN COG PHL NAM
LBN AZE SVK GNB RWA
COM UZB NGA KAZ
NER TCD GMB JAM MDV LVA DNK
PAK COD KEN ZMB
MRT
GHA ETH CMR AGO IDN BLR
LVA DNK LSO
NPL
00
GIN MDV KAZ TZA LCA ARE
CAF MDG
HTI BEN
EGY BLR CIV
BGD EST
TTO
MWI
10
BDI TGO EGY
LKA ARE NIC
PAK DZA ISL
EST
TTO UGA CYP FIN
FJI MOZ NOR
MMR DZA GAB UZB
ISL COM MMR
CYP FIN GHA
TJK NOR GAB
GIN
00
MUS LKA AUS
BDI
10
TJK FJI
VNM AUS URY
MUS JPN
URY MYS
KOR
JPN
MYS
0
10
KOR VNM
CHN
CHN
0
10
THA
THA
500 1000 2000 5000 10,000 20,000 50,000 100,000
GDP per capita (PPP, constant 2011) 500 1000 2000 5000 10,000 20,000 50,000 100,000
GDP per capita (PPP, constant 2011)
LIC LMIC UMIC HIC
LIC LMIC UMIC HIC
Panel C: “weakly-relative lines”, ������������ = 5 Panel D: “weakly-relative lines”, ������������ = 20
FJI PER
PER
Aggregate welfare losses for α=20
BEL
Aggregate welfare losses for α=5
IND ARM IRQMEX CZE MKD MEX
BLZ COL
HRV
PAN ARM ECU ITA
ESP
BIH BRAMNEARG PAN CZE
COL ESP
ITA SVN
ECU PRT BLZ
MKD MNEARG
CHL HUN SVK
SVN GBR BEL IRQ IRN CHL GBR
FRA
00
GEO ALB GRC EST HUN USA
SDN JOR BRA POLLTU
FRA BOL
MDA BGR HRV PRT
PHL IRN JORFJI POL SWE CHE
,0
LKA
BIH USA GEO ALB ZAF ROU LUX
SLE BFA BOL JAM LCA IND GRC SVK NLD
KGZ
10
ZWE ZAFDZA BGR
MUS IRL CRI LTU IRL
LBR MRT HND LVA ISRMLT SWE HND AUT
SEN
MDA
CPV GTM MDV BWA ROU CAN NLD CHE LUX KGZ TUNPRY ESTISRMLT CAN
MLI YEM PRY SURCRI PSE GTM
PAK TTO AUT CPV PHL
00
RWA PSE
SLV NAM
TUN
IDN DOMTHA URY SDN SUR
DOM TUR
RUS
NER GNB
TCD BGD COG UZB CYP DEU JAM
UKR MDV LVA DEU
COD CMR NGA MAR TUR SLV
MAR LBN
LKA DZA
LBN MRT LCA SRB
,0
GAB SLE BFA ZWE GUY
ETH TZA
KEN CIV GUY RUS JPN ARE LBR MUS TTO
MMR SRB NAM
10
CAF MDG BENGMB ZMB NIC AGO KAZ FIN SEN IDN AZE
BWA KAZ
MLI YEM PAK CYP DNK ARE
MWI HTI
TGO UKR GNB RWA BGD
LSO EGY DNK COG UZB THA URY
UGA COM NER TCD GMB CMR MMR GAB FIN
MOZ NPL AUSISL COD KEN NGA EGY BLR
GHA ETH NIC JPN ISL
CHN NPL
TZA ZMB
CIV
AZE CAF MDG AGO
KOR HTI BEN
MWI TGO LSO AUS NOR
NOR UGA COM
00
GIN BLR MOZ GHA
CHN KOR
10
BDI
MYS
GIN
00
VNM MYS
TJK BDI
10
TJK
VNM
0
10
500 1000 2000 5000 10,000 20,000 50,000 100,000
0
10
GDP per capita (PPP, constant 2011) 500 1000 2000 5000 10,000 20,000 50,000 100,000
GDP per capita (PPP, constant 2011)
LIC (1.9$ poverty line)
LMIC (3.2$ poverty line) LIC (1.9$ poverty line)
UMIC (5.5$ poverty line) LMIC (3.2$ poverty line)
HIC (21.7$ poverty line) UMIC (5.5$ poverty line)
HIC (21.7$ poverty line)
Figure 4 tells us that whether total welfare losses (from deaths and short-term increases in poverty) are
deemed to rise or fall with national income per capita depends on how one chooses to define and compare
poverty across countries, and on the relative welfare weight between mortality and poverty. When one
takes a more relative view of poverty, allowing for the fact that different (more expensive) bundles of
goods are needed to escape poverty in richer countries, then the impact of the pandemic on poverty is
uncorrelated with per capita income (Figure 2B). Since mortality is strongly correlated with income, this
implies that overall losses are greater in richer than in poorer countries on average. This is still true even
13
if a constant extreme poverty line is used, provided the welfare weight of mortality relative to poverty is
high enough (Figure 4B).
The positive association in Figures 4B, C and D is clearly related to Deaton’s (2021) finding that the effect
of the pandemic on economic growth was negatively associated with GDP per capita – that is: on average,
richer countries experienced larger proportional declines in real national income (as well as more deaths
per capita). 23 But, as the author was careful to point out, his “results say nothing about whether the
degree of suffering has been larger or smaller in poor countries” (Deaton, 2021, p.4). Our results are also
far from capturing all the suffering caused by the pandemic: as discussed earlier, disutility from ill health
among survivors and losses from increased malnutrition or paused schooling are ignored, among other
things. Nonetheless, if one takes falling into extreme poverty (relative to the $1.90 line) as an indication
of absolute economic suffering, our results suggest that the positive association can be reversed –
provided the weight placed on that kind of destitution is high relative to mortality.
4. Conclusions
In this paper we have sought to address two sets of questions: First, what were the relative contributions
of increased mortality and poverty to the welfare losses caused by the pandemic, and did those
contributions vary systematically across countries? Second, how large were the aggregate welfare losses,
and how were they distributed across countries? We focused on welfare losses caused by extreme
outcomes along both the health and income dimensions: death and destitution (defined as falling into
poverty). Following our earlier work in DFMS, we have used years of human life (either lost to premature
death or lived in poverty) as our unit of measurement.
Measuring the mortality burden in terms of years of life lost, we found that this burden increases
systematically and markedly with per capita GDP. There were approximately one hundred times as many
years of life lost per capita among the high- and upper-middle income countries that lost the most, as in
a typical low-income country. This massive disparity was driven by the fact that COVID-19 kills older
people disproportionately, and considerably larger fractions of the population are elderly in richer than in
poorer countries. Higher residual life-expectancies among the elderly in richer countries also contributed.
The association between the poverty burden (measured in terms of additional years spent in poverty) and
GDP per capita depends on how the poor are identified. Using a constant absolute poverty line such as
the international (extreme) poverty line of $1.90, the poverty burden is strongly negatively associated
with GDP per capita. The world’s poorest countries experienced poverty burdens between one hundred
and one thousand times greater than the richest. However, when poverty lines typical of each of the four
income categories (low income, lower middle-income, upper middle-income and high income) are used
instead, that relationship effectively disappears.
Either way, the relative contribution of poverty (vis-à-vis mortality) to the aggregate welfare burden is
much higher in poorer countries. In fact, the ratio declines systematically with GDP per capita across the
whole range. This leads to an important first conclusion: the economic consequences of the pandemic in
23The fact that economic contractions were deeper in countries where the loss of life was greater is also consistent
with the finding by Andersen and Gonzalez (forthcoming) that reductions in economic mobility (using data from
Google’s Community Mobility Reports) and the loss of life years were positively correlated across countries.
14
terms of increased poverty cannot be treated as being of secondary importance. Even at our most
conservative rate for comparing life- and poverty-years (twenty of the latter to one of the former), there
are 70 countries in our sample where poverty was a more important source of declining well-being than
mortality. That number rises to 108 countries (three-quarters of our sample) at the lower rate of five
poverty-years to one life-year. Most (but not all) of those countries tend to be poor. They are not the
countries where the medical and social scientists, journalists and global civil servants that set the terms
of the “global” public debate are located. The importance of the poverty consequences of the pandemic,
relative to those of mortality, has not been given its proper weight in the global discussion.
Our second main conclusion relates to the distribution of the aggregate welfare burden across countries
– and it is nuanced. We show that the association between the welfare burden and initial per capita
incomes is not always, unambiguously either positive or negative. Instead, the shape of the relationship
depends on two key factors: how poverty is defined, and the welfare weight placed on it relative to
mortality. When poverty in a country is assessed in terms of poverty lines typical of countries at similar
levels of development, a positive association arises: richer countries have suffered a greater loss in welfare
than poor ones in this pandemic. That conclusion still holds even if the international extreme poverty line
(IPL) is used instead, provided sufficient weight is placed on mortality relative to poverty. But if the IPL is
combined with a lower welfare weight for death relative to destitution, the association reverses and
poorer countries bear a greater welfare loss from the pandemic on average.
The fact that the association can be positive under plausible parameter configurations at all reflects once
again the magnitude of the income gradient of COVID-19-mortality, one version of which is our Figure 1.
Although this gradient does reflect population age structures, including residual life-expectancies, it is
important to note that demography is not destiny. Japan, by some measures the world’s “oldest” country,
suffered welfare losses orders of magnitude lower than Belgium, Germany and the United States. China,
Korea, Norway and Australia did even better. This is probably one of those cases where the variation
around the regression line matters more than the variation along it. That is the variation that reflects,
among other things, differences in policy responses: the speed with which the virus was contained upon
arrival, either by effective testing and tracing protocols or by early and well-enforced lockdowns, by early
and widespread use of masks and social distancing, or some combination of the above. Our study has
nothing to say about that fundamental source of variation.
Four final caveats are warranted. First, and as noted earlier, our results are sensitive to measurement
error. The possibility that COVID-19-related mortality is substantially under-estimated in many countries
is of particular concern. If the underestimation is concentrated among poor countries, this could alter
both of our main findings. Second – and also noted earlier – we have used “aggregate” or “total” welfare
loss as a shorthand for losses arising from mortality and entry into poverty. The pandemic has
undoubtedly had other effects on well-being, both current and future; both among the sick and among
those only indirectly affected.
Third, the pandemic is still ongoing, and the distribution of both the poverty and mortality burdens in
2021 may turn out to be very different from that of 2020 – particularly as access to vaccination is spreading
unequally around the world. This could cause an eventual “post-pandemic” version of the global
distribution of welfare losses to differ from all of those in our Figure 4, with poorer countries doing even
worse.
15
Fourth, we have nothing to say on the important issue of how losses in well-being are distributed within
countries. If, as seems likely, richer countries have been better able to cushion the losses among their
poorer residents than poor countries have, it is quite possible that the world’s very poorest people have
suffered the most. An investigation of the distributional consequences within countries will be extremely
important, but it will require post-pandemic household survey data and goes beyond the remit of this
paper.
Despite these important caveats, our analysis does suggest that the poverty consequences of the
pandemic should be given as much importance in the global policy conversation as its (horrendous)
mortality consequences. For most poor and middle-income countries, greater economic deprivation has
in fact been a more important source of loss in well-being than premature mortality. Ignoring the large
welfare costs of destitution would lead us to wrong conclusions about the distribution of the burden of
the pandemic across countries, exaggerating the share of suffering visited on richer, older countries, to
the detriment of poorer ones.
16
References
Adler, M. D., R. Bradley, M. Ferranna, M. Fleurbaey, J. Hammitt, and A. Voorhoeve (2020). “Assessing the
wellbeing impacts of the Covid-19 pandemic and three policy types: Suppression, control, and
uncontrolled spread.” Think20, Saudi Arabia.
Andersen, L. E. and A. Gonzales Rocabado (forthcoming). “Life and Death during the First Year of the
COVID-19 Pandemic: An analysis of cross-country differences in changes in quantity and quality of
life.” Latin American Journal of Economic Development, 35. (Forthcoming, May 2021).
Atkinson, A. B. and F. Bourguignon (2001). ‘‘Poverty and Inclusion from a World Perspective,’’ in J. Stiglitz
and P. A. Muet (Eds.), Governance, Equity and Global Markets. New York: Oxford University Press.
Becker, G. S., T. J. Philipson, and R. R. Soares (2005). "The quantity and quality of life and the evolution of
world inequality." American Economic Review 95(1): 277-291.
Castaneda A., R. A., D. G. Mahler, and D. Newhouse (2019). "Nowcasting Global Poverty." Paper presented
at the Special IARIW-World Bank Conference “New Approaches to Defining and Measuring Poverty in
a Growing World” Washington, DC, November 7-8, 2019.
Deaton, A. (2021). “COVID-19 and global income inequality.” National Bureau of Economic Research
Working Paper 28392.
Decerf, B; F.H. G. Ferreira, D. G. Mahler, and O. Sterck (2020). “Lives and Livelihoods: Estimates of the
Global Mortality and Poverty Effects of the COVID-19 Pandemic.” Policy Research Working Paper
9277. World Bank, Washington, DC.
Dicker, D., G. Nguyen, D. Abate, K. H. Abate, S. M. Abay, C. Abbafati, N. Abbasi, H. Abbastabar, F. Abd-
Allah, J. Abdela, et al. (2018). Global, regional, and national age-sex-specifc mortality and life
expectancy, 1950-2017: A systematic analysis for the global burden of disease study 2017. The Lancet
392 (10159), 1684-1735.
Ferreira, F. H., S. Chen, A. Dabalen, Y. Dikhanov, N. Hamadeh, D. Jolliffe, A. Narayan, E. B. Prydz, A.
Revenga, P. Sangraula, U. Serajuddin, and N. Yoshida (2016). “A global count of the extreme poor in
2012: data issues, methodology and initial results.” Journal of Economic Inequality 14 (2): 141-172.
Goldberg, P. K. and T. Reed (2020). “The effects of the coronavirus pandemic in emerging market and
developing economies: An optimistic preliminary account.” Brooking Papers on Economic Activity 6
(1).
Heuveline, P. and M. Tzen (2021). “Beyond deaths per capita: comparative COVID-19 mortality indicators”,
BMJ Open, 11: e042934. doi:10.1136/bmjopen-2020-042934
Jolliffe, D. and E. B. Prydz (2016). “Estimating international poverty lines from comparable national
thresholds.” Policy Research Working Paper 7606. Washington, DC.
Lakner, C., D. G. Mahler, M. Negre, and E. B. Prydz (2020). “How much does reducing inequality matter
for global poverty?” Global Poverty Monitoring Technical Note 13.
Lustig, N., G. Neidhöfer, and M. Tommasi (2020). “Short and Long-Run Distributional Impacts of COVID-
19 in Latin America.” Working Paper No. 2013. Tulane University, Department of Economics.
Ravallion, M., & Chen, S. (2011). “Weakly relative poverty.” Review of Economics and Statistics, 93 (4):
1251-1261.
17
Salje, H., C. T. Kiem, N. Lefrancq, N. Courtejoie, P. Bosetti, J. Paireau, A. Andronico, N. Hoze, J. Richet, C.-
L. Dubost, et al. (2020). “Estimating the burden of sars-cov-2 in France.” Science 369.6500: 208-211.
Verity, R., L. C. Okell, I. Dorigatti, P. Winskill, C. Whittaker, N. Imai, G. C. Dannenburg, H. Thompson, P. G.
Walker, H. Fu, et al. (2020). Estimates of the severity of coronavirus disease 2019: A model-based
analysis. The Lancet Infectious Diseases 20.6: 669-677.
Viscusi, W. K. (1993). "The value of risks to life and health.” Journal of Economic Literature 31.4: 1912-
1946.
World Bank (2021). “Global Economic Prospects, January 2021.” Washington, DC: World Bank.
https://openknowledge.worldbank.org/handle/10986/34710
18
Appendix: Age-specific mortality estimates; share of the population over 65; and residual life-
expectancies at 65 all correlate with GDP per capita.
Figure A1: Relative COVID-19 mortality among the old and the young and national per capita income
Mortality rate over 65 / Mortality rate among 20-39
0
40
URY
LTU JPN
0
FRA
ITA
30
GRC DEU
ESP BEL
LVA GBR SWE CHE
AUT
PRT ISR
PAN DNKUSA NOR
AUS LUX
EST FIN ISL
CANNLD
CHL
HRV MLT
SVN
CYP
POL
HUN KOR IRL
CZE
SVK
0
TTO
20
BOL BLZ LCA
HND CPVSLV
VNM GUY
GTM DOM ARG ARE
ECU
JAM COL
MDV
SUR CRI MEX
BGD NIC ALB
PRY
GEOLBN PER BRATHA ROU
ARMNAM BLR
MNE TUR
HTI
ETH NPL PSE
PAK
SDN UKR
PHLJOR BIH DZASRB MUS RUS
BGR
CAF
COD MWI TCD BEN
MDG GMBZWE
COMLSO KGZ TUN
LKA GAB
IRQ KAZMYS
MOZ
LBR SLE BFA GIN
UGA MLI YEM TJKSENMRT
ZMB IND
AGO UZB
MAR EGY MKD AZE
BDI NER TZA MDA IDN CHN
BWA IRN
GNB RWA
TGO KENCMR
CIV GHA COG MMR ZAF
0
PNG NGA FJI
10
0
500 1000 2000 5000 10,000 20,000 50,000 100,000
GDP per capita (PPP, constant 2011)
LIC LMIC UMIC HIC
Figure A2: Share of the population aged 65+ and national per capita income
.3
JPN
5
.2
ITA
PRT FIN
GRC
Share of pop. over 65
BGR HRV MLT DEU
LVA LTUSVN FRA
EST
HUN CZEESP SWE
DNK
NLD
AUT
BEL
.2
SRB ROU CHE
POL GBR
BIH CAN
NOR
UKR SVK USA
MNE KOR AUS
BLR RUS ISL
GEO URY
ALB IRL
MKD CYP LUX
5
.1
THA
MDA MUS ISR
CHL
ARM CHN
LKA ARG TTO
CRI
LCA
BRA
.1
JAM COL
TUNPER TUR
SLV PAN
VNM MAR KAZ
BOL ECULBN DOM MEX
GUY MYS
IND PRY SUR
DZA AZE IRN
MMR IDN
NPL NIC PHL FJI EGY ZAF
HTI BGD
LSO KGZ GTM
HND CPV UZB
BLZ
PAK BWA
5
SDN JOR MDV IRQ
.0
LBR ETH BEN PNG NAM GAB
COD
CAF TGO
SLE
MOZ MDG
GNB RWAGIN ZWETJK
COM
YEM SEN
CIV
CMR
MRTGHAPSE
COG
NGA
BDI NER MWI GMB TZA
TCD MLI
BFA KEN
UGA ZMB AGO
ARE
0
500 1000 2000 5000 10,000 20,000 50,000 100,000
GDP per capita (PPP, constant 2011)
LIC LMIC UMIC HIC
19
Figure A3: Residual life expectancy at 65 and national per capita income
25 COL
JPN
FRA CHE
PER ESP
ITA CAN
ISR
KOR AUS
IRQ PAN
Residual life expectancy at 65
NIC ISL
MDV THA SWE NOR
CRI PRT CYP FIN AUT
BEL IRL
TUR SVN
GRC GBR NLD LUX
20
CHL MLT DEUDNKUSA
JOR ECU DZA
CPV BLZ BRA EST
MEX POL CZE
SLV
GTM ALB
PRY
LKA
LCA DOM ARG URY
IRN
LBN HRV TTO
PSE JAM TUN MKD SVK
HUN
HND CHN MUS LVA LTU
BGD VNM ZAFSUR ROU
MDA MAR ARM BIH MNEBGR RUS
TJK KGZ NGA BOL BLR MYS
MRT SDN NAM BWA
ETH TZA GUY
UKR SRB KAZ
RWA YEM
UGA NPL MMR IND GEO IDN GAB ARE
COM KEN
15
MWI MLI
BEN SEN PHL
NER LBR TGO GHA PAK AZE
CMR ZMB
CIV AGO EGY
BFA
HTI GMB
COD TCD
SLE MDG FJI
MOZ ZWE UZB
BDI GIN COG
GNB
LSO
CAF PNG
5 10 0
500 1000 2000 5000 10,000 20,000 50,000 100,000
GDP per capita (PPP, constant 2011)
LIC LMIC UMIC HIC
20