Content uploaded by Fernando Flores Tavares
Author content
All content in this area was uploaded by Fernando Flores Tavares on Dec 14, 2020
Content may be subject to copyright.
Content uploaded by Fernando Flores Tavares
Author content
All content in this area was uploaded by Fernando Flores Tavares on Nov 16, 2020
Content may be subject to copyright.
Content uploaded by Gianni Betti
Author content
All content in this area was uploaded by Gianni Betti on Nov 16, 2020
Content may be subject to copyright.
Content uploaded by Gianni Betti
Author content
All content in this area was uploaded by Gianni Betti on Nov 16, 2020
Content may be subject to copyright.
Content uploaded by Gianni Betti
Author content
All content in this area was uploaded by Gianni Betti on Nov 16, 2020
Content may be subject to copyright.
* Department of Economics and Statistics, University of Siena, Siena, Italy; email: fern.florestavares@student.unisi.it.
** Department of Economics and Statistics, University of Siena, Siena, Italy; email: gianni.betti@unisi.it.
The Pandemic of Poverty, Vulnerability, and COVID-19: Evidence from
a Fuzzy Multidimensional Analysis of Deprivations in Brazil
Fernando Flores Tavares*, Gianni Betti**
This is an accepted manuscript, pre-copyedit version of the article published in World Development.
The final version is available online at: https://doi.org/10.1016/j.worlddev.2020.105307.
Abstract
This paper aims to show how much and in which way people in Brazil are deprived in terms of
indicators directly related to the capacity to prevent and recover from infection with COVID-19. We use
the Alkire-Foster (AF) method and a fuzzy-set approach as complements to measure multidimensional
poverty within the context of the coronavirus pandemic. We propose two pandemic-specific indexes to
account for the vulnerability related to the capacity to prevent infection with and to recover from the
disease. The outcomes reveal structural deprivations in the country and considerable inequality among
regions and ethnic groups. Rank correlation analyses suggest that the proposed indexes can trace the
trends in increasing infection and a higher mortality rate in vulnerable regions. Compared to headcount
ratio results, the fuzzy measures have more precise outcomes and are better able to capture the evolution
in mortality patterns. Our empirical evidence offers an additional warning that the pandemic responses
need to prioritize the most vulnerable groups and reinforces the need for coordinated national action.
Keywords: COVID-19 ∙ Multidimensional Poverty ∙ Fuzzy-set approach ∙ Alkire-Foster (AF) method ∙
Latin America ∙ Brazil
1. Introduction
The COVID-19 outbreak has exposed the inequality and interlinked socioeconomic deprivation
affecting developing countries to a greater extent than before. The fact that some of the population has
these problems not only is related to the pandemic but mainly reveals historical gaps that are exacerbated
by the virus. In Brazil, minority groups are at a disadvantage in terms of economic, social, and health
deprivations (Hoffman, 2018; Fernandes, 2017; Raupp et al., 2017). Planning an efficient response to the
pandemic requires an understanding of the increased risk of exposure, especially among those living in
unsafe conditions. In this sense, interest in analyzing the vulnerability to infection with COVID-19
among subgroups has grown (Pareek, 2020; Khalatbari-Soltani et al., 2020). By examining how much
and in what ways people in Brazil are deprived in terms of indicators directly related to the ability to
prevent infection with and recover from COVID-19, this paper joins others on this topic.
The first confirmed case in Brazil was diagnosed on February 25, 2020. By May 10, the country had
162,699 confirmed cases in a pattern of rapid infection (DATASUS, 2020). Even though Brazil climbed
to second in the worldwide number of confirmed COVID-19 cases on May 22 and in the number of
confirmed deaths from COVID-10 on June 12, its national government is still struggling to recognize the
problem and promote coordinated action (see Lancet, 2020). The pandemic is worsening the quality of
life in entire communities, and the lack of effective policies poses an additional threat to the population.
Families experiencing multidimensional poverty face at least two sets of additional risk factors.
First, people living in poverty might not be able to follow the recommendations for prevention (see
WHO, 2020a, 2020b). Sheltering at home might be infeasible if their housing is inadequate for keeping
them safe and comfortable during a quarantine. It is not always possible to wash hands, clean and
disinfect the home properly if one has inadequate access to clean water and sanitation conditions are
poor. Keeping a safe distance from others is not practicable in an overcrowded residence. Furthermore,
transmission of the virus might be enhanced in high-density communities (Lusignan et al., 2020; Rubin
et al., 2020) and in places with insufficient social distancing (Rubin et al., 2020, Chu et al., 2020); and
the spread of COVID-19 can be mitigated where the mobility control measures are stricter (Kraemer et
al., 2010).
Second, poor living standards and insufficient health services reduce the ability to recover from
COVID-19. Drinking unsafe water and being exposed to improper sewage disposal is highly correlated
with the contraction of preventable diseases (WHO, 2019a, 2019b), which can compromise the immune
system. Families who use highly polluting fuels for cooking might be a risk group as indoor air pollution
is associated with respiratory diseases (WHO, 2018a). Because the schools are closed, food security is
now under threat for families with schoolchildren who depend on schools for daily free meals. The lack
of physicians and intensive-care beds in hospitals is critical for people in need of treatment. The distance
from hospitals is an additional factor in vulnerability, particularly for several Indigenous communities
that live far from urban areas.
1
The literature on infectious disease outcomes for subgroups suggests that risks are higher among
minority groups and in more deprived regions. For instance, Zhao et al. (2016) show that, during the
2009/2010 influenza A(H1N1) pandemic, the risk of mortality in England was higher for non-White
populations than White populations and for people living in the most deprived areas compared with those
in less deprived areas. Lusignan et al. (2020) estimate that, within the Oxford Royal College of General
Practitioners Research and Surveillance Centre primary care network, Black people have higher risk
factors for testing positive for COVID-19 than White people and so do individuals in more deprived
areas.
Studies about racial and ethnic disparities in the United States in terms of infection with and mortality
from COVID-19 also show that minorities are the hardest hit. Laurencin and McClinton (2020)
demonstrate that in Connecticut, the Black population had a proportion of infection and death that
exceeded its share of the population even at the beginning of April. Yancy (2020) shows that this
disproportion is also present in Chicago, Louisiana, Michigan, and New York City. In Chicago, for
example, Black people make up 30% of the population but more than 50% of the confirmed cases and
almost 70% of the deaths. Millett et al. (2020) and Holtgrave et al. (2020) confirm these discrepancies
among racial and ethnic minorities, and they conclude that social characteristics, structural racism, less
access to health care, and other factors might be driving these results.
Research on the impact of the COVID-19 pandemic on minority groups and in different regions in
Latin America is still thin, but it confirms the same outcomes there. A pioneering study by Baqui et al.
(2020) uses the SIVEP-Gripe (Sistema de Informação de Vigilância Epidemiológica da Gripe) dataset
to analyze COVID-19 hospital mortality in Brazil. The analysis selects only observations that account
for ethnicity to assess the relation between health risk, ethnicity, and regional differences. The authors
find that Black and Brown people are at the highest risk of a hospital death. They also show that people
at hospitals in the northern region had comorbidities more often and a higher risk of mortality than people
in most of the central-south region.
To contribute to the pandemic literature on Brazil, we use the Alkire-Foster (AF) method and the
fuzzy-set approach as complementary measures of multidimensional poverty in the context of COVID-
19 (Alkire and Foster, 2011; Betti and Verma, 2008). Because families have multiple difficulties at the
same time, unidimensional poverty measures—which usually focus only on monetary poverty—are
insufficient to account for the reality for these people. Therefore, the methods proposed in this work are
appropriate for collecting clear evidence of overlapping kinds of deprivation. The latter, also seen as the
intersection of multidimensional aspects of poverty, are considered high-risk factors in any
multidimensional approach (Lemmi and Betti, 2006), and this is particularly evident when poverty and
deprivation are analyzed at the regional or subnational level (Betti et al., 2012).
This paper is inspired by the policy briefing on multidimensional poverty and COVID-19 risk factors
written by Alkire et al. (2020). They show that the Global Multidimensional Poverty Index (GMPI)
(OPHI and UNDP, 2019) provides information that is useful for identifying risks and vulnerabilities
related to COVID-19. They estimated that 472 million people in the world face simultaneous deprivation
in terms of water, nutrition, and indoor air pollution.
This paper innovates in at least three lines of research, both theoretical and applied:
1. it proposes two COVID-19-specific multidimensional indices: the COVID-19 prevention index
and the COVID-19 recovery index;
2. it proposes a rank correlation analysis to determine how the vulnerability indexes can capture
the mortality patterns in vulnerable regions;
3. it introduces a fuzzy counterpart to these indices.
To achieve these original contributions, we have moved step by step; the first step was to adapt the
GMPI in the context of COVID-19 in Brazil, creating a multidimensional vulnerability index (MVI). We
selected eight interlinked vulnerability indicators in the dimensions of sanitation, home shelter, physical
distance, and recovery from illness. Five of those indicators are also among the ten GMPI
2
indicators. To
better account for groups and regional disparities, we took a further step in building an appropriate
multidimensional index. The fact that the variables previously selected for the MVI are all interlinked
makes it difficult to observe the immediate relation to COVID-19. Therefore, we propose two
multidimensional poverty indexes related to the COVID-19 pandemic in terms of prevention and the
ability to recover as the first contribution of this paper. In this way, we can obtain a more comprehensive
and detailed picture of deprivation in these two aspects. The indexes reveal considerable inequality
between regions and ethnic groups, confirming the existing evidence that minority groups and vulnerable
regions have more exposure to the virus.
The second contribution of the paper is our estimation of rank correlations to clarify whether the
states with the highest vulnerability are also those with highest death rates. In Brazil, the first cases
emerged in the wealthiest states in the southeast and gradually spread to some of the poorest states in the
north and northeast. By calculating the evolution of the correlations, our indexes identify this path,
showing that the virus is progressively hitting harder the most vulnerable regions. This trend is observed
in the two COVID-19 multidimensional poverty indexes and the unidimensional monetary poverty index.
Interestingly, the index of monetary poverty shows the highest correlations in almost all of the
epidemiological weeks, which suggests that a lack of money is an immediate factor of vulnerability when
people face unexpected shocks and reinforces the importance of using both monetary and nonmonetary
indexes as complementary tools in a multidimensional poverty analysis.
Our third contribution is in using the fuzzy approach to overcome the limitation of standard poverty
measures, which treat poverty as a binary phenomenon (poor/non-poor). Using this approach enriches
the other two contributions. Fuzzy measures are more suitable for analyses at the subnational level and
for subgroups because they have smaller standard errors in the estimation of poverty and are better at
capturing mortality trends by showing higher rank correlations in most of the results.
The paper is organized as follows. Section 2 presents the empirical strategy, as well as the description
and sources of the data, and the scope of the indexes. Section 3 presents and discusses the results, and
Section 4 concludes.
2. Methodology and Data
2.1 Empirical Strategy
2.1.1 The Alkire-Foster Method
The most traditional measure of poverty is the headcount ratio (HCR), also known as the incidence
of poverty or poverty rate, which shows the percentage of people identified as poor. In this approach, by
defining a poverty line, the result is a dichotomic measure that splits the population into the poor and the
non-poor.
The Alkire-Foster methodology (AF), developed by Alkire and Foster (2009, 2011), goes beyond the
traditional approach by measuring multidimensional poverty based on its incidence (HCR) and intensity
(A). The latter is the average share of deprivation across individuals who are identified as poor. The
adjusted headcount ratio (M0), or multidimensional poverty index (MPI), is defined as the product of
incidence and intensity, M0 = HCR*A.
The identification of multidimensional poverty is calculated using the two-cutoff approach. The first
is the deprivation cutoff set for each variable. In this way, individuals can be identified as being deprived
in terms of a specific indicator, which means that we must define a deprivation threshold for each of the
variables. We apply the second cutoff by calculating the weighted sum of deprivation and classifying an
individual as poor if the resulting score is above the chosen poverty cutoff. Because the estimation of M0
is particularly well-suited to ordinal/binary data (Alkire and Foster, 2009), when applying the AF method,
we use our variables as ordinal indicators and transform the continuous variables into binary indicators.
2.1.2 The Fuzzy-Set Approach
Both the traditional monetary approach (HCR) and the MPI approach are based on deprivation cutoffs
(poverty lines), which treat poverty indicators as binary (poor/non-poor); instead, the fuzzy-set approach
treats poverty and multidimensional deprivation as matters of degree, determined in terms of the
individual’s position in the distribution of the monetary variable concerned (either income or
consumption expenditure) and other aspects of living conditions (Betti and Verma, 2008). The state of
deprivation is thus seen in the form of fuzzy sets, to which all members of the population belong in
varying degrees. In particular, within a determined poverty range, the approach uses membership
functions to identify the degree of certainty of individual poverty in a specific dimension (Alkire et al.,
2015).
The fuzzy-set approach was first proposed by Cerioli and Zani (1990) and developed by Cheli and
Lemmi (1995) in the so-called totally fuzzy and relative approach. Later, Betti et al. (2006) proposed the
integrated, fuzzy, and relative (IFR) approach, in which the membership function used for the fuzzy
monetary (FM) measure is defined as:
where is the cumulative distribution function for consumption expenditure, L is the corresponding
Lorenz curve, is the ranked individual sample weight, is individual consumption expenditure, and
is a parameter. The definition of the membership function is based on the monetary variable, in which
the alpha parameter is chosen such that the mean is “anchored” to the headcount ratio. The FM measure,
as defined previously, can also be applied in terms of the generalized Gini measures when we define α =
1.
In a multidimensional context, is an individual composite index, in which the weights of the single
indicators are not predetermined but, rather, follow the prevalence-correlations principles proposed by
Betti and Verma (2008). If the prevalence of an indicator is high, then its weight is low, and if correlations
with other vulnerable variables are high, then its weight is low. In this way, we determine appropriate
weights without the necessity of recurrence in potential arbitrary weight choices.
Another important advantage of fuzzy measures is that they are more informative and have smaller
standard errors (Betti et al., 2018). Therefore, fuzzy measures are more useful for subnational poverty
measures (Betti et al., 2012), which means that we can obtain poverty estimations for areas with relatively
small samples that are more statistically significant than those yielded by other measures.
The fuzzy approach and the AF method are complementary measures. The latter has the advantage
of providing intuitive measures that can be decomposed by population groups. In contrast, the former
has the advantage of overcoming the poor/non-poor dichotomy and enables more precise measures for
subnational regions.
2.2 Data
To construct the multidimensional indexes, we combine different publicly available sources. In this
section, we describe the data sources and the indicators.
2.2.1 Household Expenditure Survey
The primary source of data is the Brazilian Consumer Expenditure Survey (POF) for 2017-18, the
most recent round, released by the Brazilian Institute of Geography and Statistics (IBGE) on May 3,
2020. The POF is a high-quality household survey conducted to investigate the profile of consumption
expenditure, income, and living standards of Brazilian households. The data are widely used in poverty
and inequality research and have particular national importance because they are used to construct
consumption baskets in order to calculate official consumer price indexes.
The sample design of the POF is structured to cover the entire territory of the country; it is
representative in terms of the country, major regions, capitals, metropolitan regions, other parts of the
states, and urban or rural areas. The survey sample in 2017-18 totals 69,660 households, providing
information at the household and individual level.
The variables derived from the POF are Drinking water, an indicator that accounts for the
household’s frequency of supply, whether the household has running water, and the system of
distribution; Sanitation, which represents whether the household has at least one indoor bathroom with
shower and toilet, whether it is shared with other households, and whether it is connected to the public
sanitation system; Electricity, which represents whether the household has access to electricity and the
frequency of this access; Housing, which assesses the materials used in the household’s flooring, walls,
and roof; School meals, which, for households that have children who used to have daily free meals at
school, calculates how many children had access to this service, and how many meals per day; Share of
food consumption expenditure, as a proxy for a household’s food security; Overcrowded housing,
calculated as the number of residents per permanent bedroom in the household; Older adults per resident,
calculated as the number of people age 60 or more per number of members of the household who are
younger; Commuting time, which represents the number of members of the household who spend more
than an hour to get to work; Indoor air pollution by cooking fuel, which refers to the kind of fuel used by
the household for cooking; and Private insurance, which shows whether the individual has private health
insurance. The scores are presented in Table 2.
The remaining variables (described below) come from other sources and were merged with the most
possible disaggregated subnational level in the POF (state, capital, metropolitan region, or other parts of
the state). These variables are uniform across the population at the corresponding merged level.
Studies on COVID-19 stress that demographic and social variables matter when it comes to the
consequences of the pandemic (Oke and Heneghan, 2020; Souza et al., 2020). Table 1 shows
demographic and social characteristics estimated from the POF 2017-2018 dataset for the Brazilian
population
3
. Age and gender are two important factors in COVID-19 risk. According to the estimations
of Oke and Heneghan (2020), the case fatality rate is 66% higher for males older than 30 than for women
and 4.47 times higher for people age 80-89 than for those age 60-69. The median age in Brazil is 33 (the
mean is 35.26), while the median age across Europe Union members in 2018 range from 37.3 years in
Ireland to 46.3 years in Italy (Eurostat, 2019). This difference could reflect lower risk in terms of age in
Brazil, but other factors that affect risk remain to be proved.
For instance, as mentioned in the Introduction, few analyses are available about the impacts on ethnic
minorities. In Brazil, this is particularly important because ethnic minorities are at a relative disadvantage
in terms of the risk of infection. As we discuss in the next section, Indigenous people, who make up 0.4%
of Brazil’s population, predominantly live in regions with higher vulnerability to infection with COVID-
19 (22.7% live in the northern region) and have the highest vulnerability scores.
Table 1. Mean of Demographic and Social Variables
Variable
Mean
Gender
Women
51.61%
Men
48.39%
Color/ethnicity
White
44.00%
Black
10.22%
Asian
0.68%
Brown
44.42%
Indigenous
0.38%
Not identified
0.30%
Area type
Urban
85.26%
Rural
14.74%
Age in Years
35.26
Literacy ratio (>14 years)
92.41%
Years of education (>14 years)
9.37
Number of observations
178,431
2.2.2 Data on Access to Health Care and Risk Ratio by Age and Gender
The survey Area of Influence of Cities (REGIC) conducted by the IBGE in 2018 is used to provide
information on the distance that people need to travel from their city to other cities to access intensive-
care health services. To calculate the distance, we used geographic coordinates to measure the length in
kilometers of the shortest path between two cities. The final indicator is the mean for each POF
subnational level (capitals and other parts of the states) of the distance, weighted by the frequency of the
corresponding destination.
The data on the number of physicians and intensive-care hospital beds in the public health system is
available at the city level on the National Registry of Health Facilities (CNES) website. We used the
CNES data processed by the IBGE at the municipal level for December 2019, calculating the mean for
each POF subnational level. Both indicators are calculated per 1,000 people.
The risk ratio by age and gender was built based on the estimation of Oke and Heneghan (2020),
which use Italian data from the Italian National Institute of Health (ISS). The indicator sets the risk
reference score to the age between 60-69 and increases/decreases if the age is above/below this range.
2.2.3 Legal Measures of Social Distancing and Mobility Indexes
In Brazil, to date there has been no coordinated social distancing policy implemented at the national
level. The federal states and municipalities started to adopt measures to contain the spread of COVID-19
regardless of the decisions of the national government. However, these policies were implemented at
different times and in different ways. To capture the differences in the level of each state’s strictness, we
used the index of legal measures for social distancing developed by Moraes (2020a, 2020b). Moraes
considered all the decrees by state legislatures adopted April 6-24, 2020, to construct the index. The
measure considers the suspension or restriction of six types of activities: cultural, athletic, and religious;
bars and restaurants; non-essential services and business; non-essential industries; schools and
universities; and transportation. In this paper, the score was adapted to range from 0 (strict restrictions)
to 10 (no restrictions) (see Table 2).
For the mobility index, we used the Google Community Mobility Report from March 11, 2020, which
is the day on which the World Health Organization (WHO) declared COVID-19 a pandemic, to April
30, 2020. The Google indicator provides population-wide information on the relative change in mobility
in each state and in the following categories: retail stores and recreation, grocery stores and pharmacies,
parks, transit stations, workplaces, and residents. The change in mobility is the percentage change from
a baseline day before the pandemic. We use the mean of the changes in mobility in retail and recreation,
parks, transit stations, and workplace categories as a proxy for changes in behavior regarding daily
activities.
2.2.4 COVID-19 Indicators
The data on confirmed cases of COVID-19 and deaths from it are available on a daily basis on the
coronavirus website of the Ministry of Health (https://covid.saude.gov.br). The first confirmed case was
identified on February 25 and the first confirmed death on March 17, 2020. We collected statistics for
the states and capitals using official data from the Brasil.io (2020) website,
https://brasil.io/dataset/covid19/caso_full/. Based on the number of deaths confirmed as being due to
COVID-19 and the population estimated by the POF, we calculated the number of confirmed deaths per
one million people. It is important to stress that the official number of confirmed deaths from COVID-
19 underestimates the actual number, mostly due to limited testing.
2.2.5 Descriptive Statistics
Table 2 shows the score range and descriptive statistics for all the indicators used in the COVID-19
multidimensional poverty indexes. Figure 1 presents the correlations between each pair of indicators
calculated as Pearson coefficients. The heatmap is colored using a range from -1 (blue) to +1 (red). The
deprivations that are commonly explored in research on multidimensional poverty—such as having clean
drinking water, sanitation, electricity, housing, housing density, and indoor pollution—are all positively
correlated. The correlations of these variables with the share of expenditure on food, the distance from a
hospital, monetary poverty (measured by the household consumption expenditure per capita, $3.20 a day,
in 2018 purchasing power parity [PPP]), and COVID-19 deaths per million people are also positive.
However, they have a negative correlation with population density and indicators related to health-care
resources, such as private health insurance, physicians per 1,000 people, and intensive-care beds per
1,000 people. The correlations with the remaining variables are negative or near zero.
Table 2. Score range and descriptive statistics for the variables used in the COVID-19 multidimensional
poverty indexes
Variable
Score Range
Mean
Standard Error
Min
Max
Drinking water
0-6
0.605
0.003
0
6
Sanitation
0-4
0.494
0.002
0
4
Electricity
0-4
0.056
0.001
0
4
Housing
0-9
1.027
0.003
0
9
School meals
0-16
0.476
0.003
0
16
Share of expenditure on
food
Continuous
0.177
0.0003
0
1
Overcrowded housing
(residents per permanent
bedroom)
Continuous
1.905
0.002
0.333
13
Older adults per
household
Continuous
0.193
0.001
0
4
Commuting time
0-4
0.1501
0.001
0
4
Population density
(inhabitants per km2)
Continuous
1261.859
5.827
0.673
8435.358
Index of legal measures of
social distancing
0-10
3.269
0.004
0.8
6.7
Mobility index (%
reduction from a baseline
day before the pandemic)
Continuous
51.294
0.011
39
61.392
Risk ratio by age and
gender (1 is the risk
reference score set for age
60-69)
Continuous
0.496
0.003
0
8.018
Indoor pollution due to
cooking fuel
0-2
0.010
0.0002
0
2
Private insurance
0/1
0.260
0.001
0
1
Distance from hospital (in
km)
Continuous
30.503
0.101
0
606.544
Physicians per 1,000
people
Continuous
1.174
0.002
0.365
4.695
Intensive-care hospital
beds per 1,000 people
Continuous
0.441
0.001
0
3.01
Note: For ordinal variables, the score ranges are from no deprivation to total deprivation. The variable for private
insurance is the only binary variable, in which 0 means no insurance, and 1 means the person has insurance. The
continuous variables are identified as such.
(1) Drinking water
(2) Sanitation
(3) Electricity
(4) Housing
(5) School meals
(6) Share of food expenditure
(7) Overcrowded housing
(8) Older adults per resident
(9) Commuting time
(10) Population Density
(11) Ind. of legal meas. of social dist.
(12) Mobility Index
(13) Risk ratio by age and gender
(14) Indoor pollution
(15) Private insurance
(16) Distance from hospital
(17) Physicians per 1,000 p.
(18) Int. care hosp. beds per 1,000 p.
(19) Monetary poverty*
(20) COVID-19 deaths per mill.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(16)
(17)
(18)
(19)
(20)
-0.3
0
0.5
1
Figure 1. Pearson correlation matrix of the variables used in the COVID-19 multidimensional poverty
indexes, monetary poverty, and COVID-19 deaths per million people
*Monetary poverty is measured by the household consumption expenditure per capita ($3.20 a day, 2018 purchasing power
parity).
2.3 Multidimensional Poverty Indexes: Defining the Scope
This section proposes in detail the two COVID-19-related multidimensional poverty indexes
(CMPIs), comprising variables directly related to the capacity for preventing infection with COVID-19
and for recovery from it. Table 3 shows the dimensions and variables of each index, the definition of the
deprivation cutoffs and weights used in the AF method, and the resulting prevalence-correlation weight
scores for the fuzzy analysis.
Most of the cutoffs in the AF analysis were adapted from the United Nations sustainable development
goals and consider the Brazilian context and data availability. In the fuzzy application, to avoid
dichotomization of the variables and to obtain more information, we regard the variables as ordinal or
continuous indicators when possible. Only the indicator for private insurance is binary in both
approaches. With respect to the weights in the AF method, for simplicity, we follow the GMPI by
assuming that the dimensions are of equal weight. Also following the GMPI standard, for each
multidimensional index, we consider people vulnerable
4
(VN) to infection with COVID-19 if they are
deprived of at least one-third of the weighted indicators and consider people at severe risk (SR) of
infection with COVID-19 if they are deprived of at least half the weighted indicators.
Table 3. Structure of the COVID-19 multidimensional poverty indexes, variable cutoffs, and weights
Index
Dimension
Variable
AF Deprivation Cutoff
AF
Weight
Prevalence
correlation
weight
COVID-19 prevention
Hygiene
Drinking water
The household does not have daily access to
water, or does not have indoor running water, or
the water does not come from a public water
system.
0.100
0.103
Sanitation
The household does not have indoor bathroom
with shower and toilet, or the bathroom is shared
with other households, or the disposal of human
waste is not connected to a public sewage
system.
0.100
0.097
Staying at home
Electricity
The household has no access to electricity.
0.100
0.103
Housing
The household’s housing materials for at least
one floor, wall, and roof are inadequate.
0.100
0.097
Food security
School meals
One or more children in the household have
breakfast, lunch, snacks, or dinner free at school
every day.
0.100
0.094
Share of food
consumption
expenditure
Food represents 75% or more of the total
consumption expenditure of the household.
0.100
0.106
Household
density
Overcrowded
housing
There are three or more residents per permanent
bedrooms in the household.
0.100
0.098
Older adults per
household
Two or more older adults per members of a
household.
0.100
0.102
Public social
distancing
Commuting time
At least one individual in the household spends
more than an hour to get to work.
0.050
0.063
Population density
The household is in a region* where the pop.
density is higher than the mean of the Brazilian
capitals (> 2,700/km2)
0.050
0.062
Index of legal
measures of social
distancing
The household is in a state where the index is
higher than 2 (out of 10, which is the least
restrictive)
0.050
0.040
Mobility index
The household is in a state where the index is
less than 60% of the relative reduction in
mobility
0.050
0.035
Index
Dimension
Variable
AF Deprivation Cutoff
AF
Weight
Prevalence
correlation
weight
COVID-19 health recovery
Living standards
Electricity
The household has no access to electricity.
0.083
0.087
Housing
The household housing materials for at least one
of floor, wall, and roof are inadequate.
0.083
0.082
Overcrowded
housing
There are three or more residents per permanent
bedrooms in the household.
0.083
0.081
Risk groups
Risk ratio by age
and gender
The indicator is 1 or more;1 is the risk reference
score at age 60-69 (Oke and Heneghan, 2020). It
is an individual-level indicator.
0.125
0.120
Indoor air pollution
due to cooking fuel
The household’s cooking fuel is wood, oil,
kerosene, or another liquid fuel.
0.125
0.130
Healthy immune
system
Drinking water
The household does not have daily access to
water, or does not have indoor running water, or
the water does not come from a public water
system.
0.062
0.057
Sanitation
The household does not have indoor bathroom
with shower and toilet, or the bathroom is shared
with other households, or the disposal of human
waste is not connected to a public sewage
system.
0.062
0.060
School meals
One or more children in the household have
breakfast, lunch, snacks, or dinner free at school
every day.
0.062
0.063
Share of food
consumption
expenditure
Food represents 75% or more of the total
consumption expenditure of the household.
0.062
0.071
Access to health
care
Private insurance
The individual has no private health insurance.
0.062
0.054
Distance from
hospital
The household is in a region* where the
weighted mean distance from a hospital is more
than 100 km.
0.062
0.083
Physicians per
1,000 people
The household is in a region where the mean of
the indicator is less than 1 physician per 1,000
people.
0.062
0.063
Intensive-care
hospital beds per
1,000 people
The household is in a region* where the mean of
the indicator is less than 1 bed per 1,000 people.
0.062
0.050
Notes: AF deprivation cutoff refers to the description of the cutoff in the AF method. A cutoff definition is not necessary in
the fuzzy approach, because it does not treat the variables as binary measures. AF weights are the values of the weights used
in the AF method. Prevalence-correlation weights are the weights calculated in the fuzzy approach analysis.
* Capital, metropolitan region, or other parts of the state.
In Table 4, we present the number and percentage of deprived people in terms of all the variables
used in the two CMPIs. The indicator with the highest percentage of deprived people is the index that
measures mobility reduction, with 95.6% of the population deprived. This means that most of the
Brazilian population lives in states in which the mean reduction in daily mobility was less than 60%
(from March 11 to April 30). The lack of national coordination in social distancing measures, as the result
for the index of legal measures suggest (81.7% of people deprived), is one possible factor in the small
reduction in mobility. Moreover, the participation in protests opposing coronavirus lockdowns and the
continuous calls by the president to end social distancing is another possible factor that demotivated
people to decrease mobility.
By looking at the data, it is possible to observe that, independent of the pandemic context, a large
proportion of Brazilians do not have access to basic public services. The lack of public health-care
infrastructure is widespread. More than 83% of the population is deprived in terms of intensive-care
hospital beds, and the mean is 0.44 beds per 1,000 people (see Table 2 and Table 4). Moreover, 51% of
the population has access to less than one physician per 1,000 population. In some states, the private
health sector offers proportionately more physicians and hospital beds. In any case, although only 26%
of the population has private insurance, it does not mean that they will have access to all the private
health infrastructure, only to the hospitals and services specified by their contract
5
. It is also evident that
for several people the access to sanitation and drinking water is inadequate: 30.2% and 39.1%,
respectively. Clean water is crucial for preventing infection with COVID-19 because it is needed for
frequent and sufficient hand washing (WHO, 2020b) and is essential for human health and well-being
(WHO, 2019a). Moreover, improper sanitation is a major cause of infectious disease (WHO, 2018b,
2020b) and can compromise the immune system, with a possible impact on recovery from COVID-19.
Table 4. Number and percentage of deprived people per indicator
Variable
AF Pop. Deprived
AF % Deprived
Drinking water
62,511,394
30.18%
Sanitation
80,970,486
39.09%
Electricity
457,742
0.22%
Housing
22,155,805
10.69%
School meals
47,170,894
22.78%
Share of food consumption expenditure
283,702
0.14%
Overcrowded housing (residents per
permanent bedroom)
26,313,116
12.71%
Older adults per household
6,666,591
3.22%
Commuting time
26,390,282
12.74 %
Population density (people per km2)
34,501,439
16.66%
Index of legal measures of social distancing
169,164,425
81.68%
Mobility index (% reduction from a baseline
day before the pandemic)
198,056,358
95.63%
Risk ratio by age and gender (1 is the risk
reference score set for age 60-69)
31,702,592
15.31%
Indoor pollution due to cooking fuel
2,094,513
1.01%
Private insurance
153,306,719
74.02%
Distance from hospital (in km)
8,187,121
3.95%
Physicians per 1,000 people
105,612,592
51.00%
Intensive-care hospital beds per 1,000 people
172,871,751
83.47%
Total population
207,103,790
Note: AF Pop. Deprived and AF % Deprived refers, respectively, to the number and incidence of deprived people using the
cutoff defined in the Alkire-Foster (AF) model.
3. Results
3.1 Multidimensional Poverty Analysis
This section presents the results for both the AF and fuzzy approaches. For the AF method, Table 5
shows the outcomes for the multidimensional headcount ratio (HCR), the poverty intensity (A), and the
adjusted headcount ratio (M0). In addition, the results for each CMPI are shown for VN and SR. For
group decompositions, we use only the HCR indicator, as it is more intuitive and the comparison with
the fuzzy measure is more appropriate.
According to the AF and fuzzy results, between 16.2% and 15.6% of the population is vulnerable to
infection with COVID-19 as measured by the prevention index. This implies that between 32.4 million
and 33.5 million people cannot implement proper prevention measures related to at least one-third of the
weighted indicators. Severe risk, which represents deprivation in half the weighted indicators, is 3.4% in
the AF and 4.1% in the fuzzy results, respectively. In the health recovery index, the two approaches
diverge more. The AF estimates 19.8% of people are vulnerable (41 million people), while the fuzzy
estimate is 13.7% (28.4 million people). In terms of severe risk, the results are 2.2% for AF and 2.6% for
fuzzy.
Table 5. COVID-19 multidimensional poverty indexes per approach and indicators
AF
Fuzzy
Index
M0
VN
HCR
VN
A
VN
M0
SR
HCR
SR
A
SR
VN
SR
Prevention
0.068
16.17%
0.419
0.018
3.43%
0.528
15.64%
4.11%
Health recovery
0.081
19.81%
0.409
0.012
2.15%
0.555
13.72 %
2.60 %
Table 6 presents the results for each state, showing the number of confirmed COVID-19 deaths per
million people and the share of the monetarily poor people measured by household consumption
expenditure per capita ($3.20 a day, 2018 PPP). Alternatively, Figure 2 illustrates in maps the distribution
of confirmed deaths per million, FM poverty, and fuzzy vulnerability for the health recovery index by
state. The outcomes demonstrate the vast regional inequality in Brazil. The northern and northeastern
regions have the highest proportion of vulnerability and severe risk of infection with COVID-19. For
instance, Amazonas state (AM) has the most deaths per million people and among the highest risk: the
incidence of vulnerable people according to the health recovery index is 50.5%. By comparison, in São
Paulo (SP), the state with the most infections in absolute terms, it is 3.9%.
Table 6. COVID-19 death indicator and estimation results for the COVID-19 multidimensional poverty
indexes and unidimensional monetary poverty by state
State
COVID-
19
Prevention Index
Health Recovery Index
Monetary Poverty
Death
per
million
HCR
VN
HCR
SR
Fuzzy
VN
Fuzzy
SR
HCR
VN
HCR
SR
Fuzzy
VN
Fuzzy
SR
HCR
Fuzzy
RO
39.70
36.74%
7.28%
45.72%
2.47%
35.10%
1.43%
16.04%
1.55%
14.28%
14.01%
AC
69.61
52.05%
24.79%
48.19%
22.99%
58.45%
24.64%
44.02%
14.90%
16.42%
16.06%
AM
353.17
45.14%
17.48%
27.34%
11.43%
50.45%
16.18%
27.52%
7.52%
25.48%
23.60%
RR
96.74
26.81%
8.43%
16.17%
2.78%
48.12%
7.72%
19.75%
3.26%
21.00%
20.15%
PA
144.91
41.55%
14.10%
48.98%
17.19%
44.20%
6.22%
32.34%
7.78%
22.13%
20.16%
AP
132.70
41.71%
12.94%
24.69%
6.31%
51.63%
9.02%
25.81%
4.51%
11.77%
12.13%
TO
17.59
20.35%
5.70%
34.35%
4.67%
34.65%
2.47%
13.64%
2.65%
21.57%
19.74%
MA
78.61
44.37%
14.78%
36.19%
18.26%
50.13%
9.14%
35.86%
13.28%
22.34%
21.45%
PI
22.08
29.03%
8.58%
18.84%
9.61%
51.83%
13.79%
23.05%
7.92%
14.68%
15.46%
CE
178.39
8.13%
0.92%
11.89%
4.45%
29.17%
3.56%
21.28%
5.43%
22.81%
20.36%
RN
39.23
29.09%
5.54%
21.27%
5.69%
31.51%
2.74%
22.68%
3.97%
10.81%
10.86%
PB
46.16
23.93%
3.76%
28.85%
9.59%
27.57%
1.92%
25.80%
5.58%
23.89%
21.39%
PE
155.20
20.73%
1.19%
17.51%
5.46%
25.59%
2.10%
22.91%
5.45%
17.93%
16.68%
AL
60.12
22.93%
1.78%
21.34%
5.31%
31.26%
2.44%
25.14%
3.48%
31.07%
27.45%
SE
23.36
15.39%
1.71%
9.12%
3.82%
22.80%
2.19%
13.59%
2.59%
10.02%
10.24%
BA
19.37
19.56%
4.32%
28.92%
8.05%
26.71%
2.34%
20.78%
5.21%
15.71%
14.88%
MG
7.15
8.32%
1.38%
8.64%
1.17%
13.97%
0.88%
9.82%
0.85%
5.70%
6.23%
ES
68.65
12.22%
1.74%
10.15%
0.98%
15.23%
0.31%
8.31%
0.43%
7.68%
8.38%
RJ
152.77
15.79%
2.45%
8.57%
1.36%
9.50%
0.04%
7.77%
0.39%
8.25%
8.86%
SP
103.37
8.17%
1.47%
6.67%
1.20%
3.89%
0.01%
6.19%
0.28%
2.56%
3.65%
PR
11.15
9.26%
1.00%
16.74%
2.22%
12.78%
0.37%
7.61%
0.97%
7.12%
7.40%
SC
11.54
8.04%
0.08%
2.71%
0.38%
16.50%
0.55%
7.92%
0.70%
2.81%
3.42%
RS
12.23
10.22%
1.50%
5.69%
0.51%
16.29%
0.79%
7.38%
0.63%
2.37%
3.22%
MS
5.63
14.47%
2.26%
37.39%
4.00%
21.29%
0.59%
7.69%
0.96%
5.23%
5.64%
MT
8.01
23.32%
5.63%
36.31%
4.54%
48.58%
6.58%
13.79%
0.93%
7.13%
7.51%
GO
10.03
10.18%
0.59%
6.39%
0.40%
19.27%
0.54%
7.79%
0.32%
4.97%
5.76%
DF
18.96
13.66%
1.47%
5.50%
0.83%
7.62%
0.03%
3.85%
0.28%
3.74%
3.66%
Total
75.79
16.17%
3.43%
15.64%
4.11%
19.81%
2.15%
13.72%
2.60%
9.96%
9.96%
Notes: Covid-19 deaths per million as of May 10, 2020.
Northern region: RO = Rondônia; AC = Acre; AM = Amazonas; RR = Roraima; PA = Pará; AP = Amapá; TO = Tocantins.
Northeastern region: MA = Maranhão; PI = Piauí; CE = Ceará; RN = Rio Grande do Norte; PB = Paraíba; PE = Pernambuco;
AL = Alagoas; SE = Sergipe; BA = Bahia.
Southeastern region: MG = Minas Gerais; ES = Espírito Santo; RJ = Rio de Janeiro; SP = São Paulo.
Southern region: PR = Paraná; SC = Santa Catarina; RS = Rio Grande do Sul.
Central-western region: MS = Mato Grosso do Sul; MT = Mato Grosso; GO = Goiás; DF = Distrito Federal.
The risk of infection with COVID-19 also differs among ethnic groups. Table 7 shows the indexes
for each group. Overall, the picture is not favorable for the Indigenous population, which has the worst
conditions in all indicators. Brown and Black groups also are at a disadvantage to the White and Asian
groups and the total population.
The fact that the virus is spreading toward the northern region is an additional concern, as it is the
region with the highest proportion of the population that is vulnerable and at severe risk of infection with
Figure 2. Distribution of COVID-19 confirmed deaths per million population, fuzzy monetary poverty,
and fuzzy vulnerability for the COVID-19 health recovery index by state
Note: For abbreviations, see note to Table 6.
COVID-19, and, as mentioned previously, where 22.7% of the Brazilian Indigenous population is
concentrated.
Table 7. Estimation results for the COVID-19 multidimensional poverty indexes and unidimensional
monetary poverty by color/ethnicity
Color/Ethnicity
Prevention Index
Health Recovery Index
Monetary Poverty
HCR
VN
HCR
SR
Fuzzy
VN
Fuzzy
SR
HCR
VN
HCR
SR
Fuzzy VN
Fuzzy
SR
HCR
Fuzzy
White
10.57%
1.69%
9.80%
1.84%
13.49%
0.94%
9.53%
1.24%
5.25%
5.79%
Black
19.54%
4.23%
19.12%
5.01%
21.52%
2.35%
14.95%
3.11%
13.06%
12.31%
Asian
6.71%
0.89%
5.72%
0.73%
7.31%
0.47%
6.98%
0.84%
4.19%
4.61%
Brown
20.96%
4.96%
20.71%
6.20%
25.88%
3.28%
17.67%
3.85%
13.98%
13.62%
Indigenous
34.98%
9.46%
28.64%
7.43%
28.58%
8.88%
23.33%
6.61%
14.97%
13.72%
Not identified
10.60%
0.99%
8.87%
0.39%
8.33%
0.03%
5.68%
0.10%
8.21%
9.40%
Total
16.17%
3.43%
15.64%
4.11%
19.81%
2.15%
13.72%
2.60%
9.96%
9.96%
3.2 The Link between the Multidimensional Poverty Indexes and COVID-19 Deaths
The first confirmed cases were in São Paulo, the richest state and one of the least vulnerable to
infection with COVID-19 according to the CMPI. At the beginning of March, it was estimated that 85.3%
of the transmission came from outside the country, with 54.8% probably coming from travelers infected
in Italy, 9.3% in China, and 8.3% in France (Candido et al., 2020). This suggests that in Brazil the initial
infection was concentrated among the middle and upper classes (who could afford to fly outside the
country). Later data show the spread of the virus to the most vulnerable regions (see Table 6 and Figure
2).
In this section, we propose an innovative analysis, measuring rank correlations at the state level for
the unidimensional and multidimensional poverty indexes and the number of COVID-19 deaths in Brazil
per million people for many consecutive weeks. We used Spearman and Kendall rank correlation
coefficients; for each subgroup, the latter calculates the average of the square difference between the two
ranks, and the former is based on the difference in the number of pairs that do and do not match. The
coefficients are between -1 and +1; an extreme value means that the rankings are perfectly associated
either negatively or positively. Even if the analysis does not imply causality, our indexes can be tested
by determining the evolution in the correlation throughout over the epidemiological weeks.
The results are given for week 12 (April 5) to week 20 (May 10). Figure 3 and Figure 4 illustrate the
results for the HCRs of vulnerability and severe risk (blue and orange lines) and the fuzzy indicators of
vulnerability and severe risk (gray and yellow lines). Figure 5 plots outcomes for the HCR of monetary
poverty (green line) and FM poverty (light blue line).
a. Spearman b. Kendall
Figure 3. The evolution per week in the rank correlations between the prevention index and deaths per
million people at the state level
Overall, the outcomes confirm that our indexes capture the trend in infection from the richest regions,
which are less vulnerable, to the more vulnerable regions. Initially, all the measures show a negative
relation between the vulnerability indicators and deaths per million people. Beginning in week 14, the
correlations increase, meaning that states with the highest vulnerability and severe risk scores have
increasing numbers of deaths per million people.
-0,8
-0,6
-0,4
-0,2
0
0,2
0,4
0,6
week
12 week
13 week
14 week
15 week
16 week
17 week
18 week
19 week
20
Correlation
HCR VN SR VN
Fuzzy VN Fuzzy SR
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
week
12 week
13 week
14 week
15 week
16 week
17 week
18 week
19 week
20
Correlation
HCR VN SR VN
Fuzzy VN Fuzzy SR
The health recovery index has greater correlations than the prevention index. The highest Spearman
coefficients in the prevention index are 0.41 and 0.44 (for HCR vulnerability and fuzzy severe risk),
whereas for the health recovery index they are 0.46 and 0.53 (for fuzzy severe risk and fuzzy
vulnerability). In addition, most of the results suggest that fuzzy measures are more appropriate for
explaining the link with deaths because they show the highest correlations in Figure 4 and Figure 5.
a. Spearman b. Kendall
Figure 4. The evolution per week in rank correlations between the health recovery indexes and deaths
per million people at the state level
Finally, beginning in week 14, the correlations are predominantly steeper and higher for
unidimensional monetary poverty than for the other two indexes. One possible interpretation is that
monetary poverty is an immediate factor of vulnerability and risk to shocks. In times of difficulty, money
seems to be the first thing that plays an essential role in addressing the increasing threat from the
pandemic. Because vulnerability is a multidimensional phenomenon, these results reinforce that
nonmonetary and monetary variables are complementary indicators.
-0,6
-0,4
-0,2
0
0,2
0,4
0,6
week
12 week
13 week
14 week
15 week
16 week
17 week
18 week
19 week
20
Correlation
HCR VN SR VN
Fuzzy VN Fuzzy SR
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
week
12 week
13 week
14 week
15 week
16 week
17 week
18 week
19 week
20
Correlation
HCR VN SR VN
Fuzzy VN Fuzzy SR
a. Spearman b. Kendall
Figure 5. The evolution per week in rank correlations between the unidimensional monetary poverty
indicators and deaths per million people at the state level
4. Concluding Remarks
This paper contributes to the literature on the potential social impacts of the COVID-19 pandemic.
We use the AF method and fuzzy-set approach as complements to measure multidimensional poverty
within the context of the coronavirus pandemic in Brazil. We propose two multidimensional poverty
indexes to account for the vulnerability related to the ability to prevent the spread of infection and to
recover from infection with the virus.
The data reveal structural deprivations in the country due to the fact that a big part of the population
cannot fully implement the recommended preventive measures and because the social conditions and the
health-care system do not meet the basic requirements for avoiding preventable deaths. Moreover, the
estimations of the indexes illustrate the considerable inequality among regions and ethnic groups. This
is in line with the extensive literature on inequality in Brazil.
Two of the innovations in this paper are presenting pandemic-specific indexes and proposing a rank
correlation analysis that can trace the increasing spread of infection and higher mortality rate in
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
week
12 week
13 week
14 week
15 week
16 week
17 week
18 week
19 week
20
Correlation
HCR Monetary Fuzzy Monetary
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
0,5
week
12 week
13 week
14 week
15 week
16 week
17 week
18 week
19 week
20
Correlation
HCR Monetary Fuzzy Monetary
vulnerable regions. Most of the correlations increase weekly, which means that most states with the
highest vulnerability and severe risk outcomes also have the largest increase in death rates. The monetary
poverty indicator has the highest correlation when compared with the two CPMIs for almost all the
epidemiological weeks. This indicates that money is very important in battling the threat of the pandemic
and that nonmonetary and monetary indexes are complementary variables, rather than competing
variables, in multidimensional poverty analysis.
Another innovation is the application of fuzzy measures, which are more appropriate for the
characteristics of the vulnerability variables because they avoid a binary split between deprivation and
non-deprivation, have more precise measures in subnational analysis, and have higher rank correlation.
Despite the limitations of the data on confirmed deaths from COVID-19, our empirical evidence
offers an additional warning that responses to the pandemic need to prioritize the most vulnerable groups,
and our analysis reinforces the need for coordinated national action. In the short run, rapid measures are
needed to stop the virus from spreading, to ensure that the entire population follows the recommendations
for prevention as well as they can, and to guarantee universal coverage by public health services. In the
medium and long run, this analysis reinforces the urgent necessity of public policies that promote health,
adequate housing, and sanitation.
Acknowledgments
We are grateful to Francesca Bettio, Antonella D'Agostino, Ugo Pagano, Ana Cecília Kreter, Isabel
Sofia Fanciulli, Thiago D. Oliveira, Kamaiaji de S. Castor, Guilherme S. Morlin, Chandni Dwarkasing,
Marwil J. Dávila-Fernández, Henrique C. de Mesquita, Mario A. M. P. de Almeida, the Associate Editor
Christelle Dumas, and two anonymous referees for their helpful suggestions and comments.
References
Alkire, S., & Foster, J. (2011). Counting and multidimensional poverty measurement. Journal of Public
Economics, 95(7-8), 476-487.
Alkire, S., & Foster, J. (2009). Counting and multidimensional poverty measurement (OPHI Working
Paper No. 32). Oxford Poverty and Human Development Initiative, University of Oxford.
Alkire, S., Roche, J. M., Ballon, P., Foster, J., Santos, M. E., & Seth, S. (2015). Multidimensional poverty
measurement and analysis. New York: Oxford University Press.
Alkire, S., Dirksen, J., Nogales, R., & Oldiges, C. (2020). Multidimensional poverty and COVID-19 risk
factors: A rapid overview of interlinked deprivations across 5.7 billion people, OPHI Briefing 53,
Oxford Poverty and Human Development Initiative, University of Oxford.
Baqui, P., Bica, I., Marra, V., Ercole, A., & Schaar, M. V. (2020). Ethnic and regional variations in
hospital mortality from COVID-19 in Brazil: A cross-sectional observational study. Lancet Global
Health, 8(8).
Betti, G., Cheli, B., Lemmi, A., & Verma, V. (2006). Multidimensional and longitudinal poverty: An
integrated fuzzy approach, in Lemmi, A., & Betti, G. (eds.), Fuzzy Set Approach to
Multidimensional Poverty Measurement. New York: Springer, 111-137.
Betti, G., Gagliardi, F., & Verma, V. (2018). Simplified Jackknife variance estimates for fuzzy measures
of multidimensional poverty. International Statistical Review, 86(1), 68-86.
Betti, G., Gagliardi, F., Lemmi, A., & Verma, V. (2012). Sub-national indicators of poverty and
deprivation in Europe: methodology and applications. Cambridge Journal of Regions, Economy
and Society, 5, 149-162.
Betti, G., & Verma, V. (2008). Fuzzy measures of the incidence of relative poverty and deprivation: a
multi-dimensional perspective. Statistical Methods and Applications, 17(2), 225-250.
Brasil.io. (2020). COVID-19: Boletins informativos e casos do coronavírus por município por dia.
Retrieved August 2020, from https://brasil.io/dataset/covid19/caso_full/.
Candido, D. D. S., Watts, A., Abade, L., Kraemer, M. U., Pybus, O. G., Croda, J., ... & Faria, N. R.
(2020). Routes for COVID-19 importation in Brazil. Journal of Travel Medicine, 27(3), taaa042.
Cerioli, A., & Zani, S. (1990). A fuzzy approach to the measurement of poverty, in Dagum, C., & Zenga,
M. (eds.), Income and Wealth Distribution, Inequality and Poverty, Studies in Contemporary
Economics. Berlin: Springer, 272-284.
Cheli, B., & Lemmi, A. (1995). A totally fuzzy and relative approach to the multidimensional analysis
of poverty, Economic Notes, 24, 115-134.
Chu, D. K., Akl, E. A., Duda, S., Solo, K., Yaacoub, S., Schünemann, H. J., . . . Reinap, M. (2020).
Physical distancing, face masks, and eye protection to prevent person-to-person transmission of
SARS-CoV-2 and COVID-19: A systematic review and meta-analysis. The Lancet, 395(10242),
1973-1987.
DATASUS. (2020). Coronavírus Brasil. Retrieved August 2020, from https://covid.saude.gov.br.
Eurostat. (2019). Median age over 43 years in the EU. Retrieved August, 2020, from
https://ec.europa.eu/eurostat/web/products-eurostat-news/-/DDN-20191105-1
Fernandes, G. A. D. A. L. (2017). Is the Brazilian tale of peaceful racial coexistence true? Some evidence
from school segregation and the huge racial gap in the largest Brazilian city. World Development,
98, 179-194.
Hoffman, R. (2018). Changes in income distribution in Brazil. In Amann, E., Azzoni, C., & Baer W.
(Authors), The Oxford Handbook of the Brazilian Economy (pp. 467-488). New York: Oxford
University Press.
Holtgrave, D. R., Barranco, M. A., Tesoriero, J. M., Blog, D. S., & Rosenberg, E. S. (2020). Assessing
racial and ethnic disparities using a COVID-19 outcomes continuum for New York State. Annals
of Epidemiology, 48, 9-14.
InfoAmazon. (2020). Distantes de UTIs e respiradores, indígenas da Amazônia tentam se blindar do
vírus. Retrieved May 20, 2020, from https://infoamazonia.org/pt/2020/05/distantes-de-utis-e-
respiradores-indigenas-da-amazonia-tentam-se-blindar-do-virus/
Khalatbari-Soltani, S., Cumming, R. G., Delpierre, C., & Kelly-Irving, M. (2020). Importance of
collecting data on socioeconomic determinants from the early stage of the COVID-19 outbreak
onwards. Journal of Epidemiology and Community Health, 74, 620-623.
Kraemer, M. U., Yang, C. H., Gutierrez, B., Wu, C. H., Klein, B., Pigott, D. M., ... & Brownstein, J. S.
(2020). The effect of human mobility and control measures on the COVID-19 epidemic in China.
Science, 368(6490), 493-497.
Lancet (2020). COVID-19 in Brazil: So what? (Editorial). The Lancet, 395 (10235), 1461.
Laurencin, C. T., & Mcclinton, A. (2020). The COVID-19 pandemic: A call to action to identify and
address racial and ethnic disparities. Journal of Racial and Ethnic Health Disparities, 7(3), 398-
402.
Lemmi, A., & Betti, G., eds. (2006). Fuzzy set approach to multidimensional poverty measurement. New
York: Springer.
Lusignan, S. D., Dorward, J., Correa, A., Jones, N., Akinyemi, O., Amirthalingam, G., . . . Hobbs, F. D.
(2020). Risk factors for SARS-CoV-2 among patients in the Oxford Royal College of General
Practitioners Research and Surveillance Centre primary care network: A cross-sectional study.
Lancet Infectious Diseases.
Massuda, A., Hone, T., Leles, F. A., Castro, M. C., & Atun, R. (2018). The Brazilian health system at
crossroads: Progress, crisis and resilience. BMJ Global Health, 3(4), e000829.
Millett, G. A., Jones, A. T., Benkeser, D., Baral, S., Mercer, L., Beyrer, C., . . . Sullivan, P. S. (2020).
Assessing differential impacts of COVID-19 on black communities. Annals of Epidemiology, 47,
37-44.
Moraes, R.F.D. (2020a). Medidas legais de incentivo ao distanciamento social: comparação das políticas
de governos estaduais e prefeituras das capitais no Brasil. Nota Técnica n.16, Instituto de Pesquisa
Econômica Aplicada (IPEA).
Moraes, R.F.D. (2020b). COVID-19 e medidas legais de distanciamento social: tipologia de políticas
estaduais e análise do período de 13 a 26 de abril de 2020. Nota Técnica n.18, Instituto de Pesquisa
Econômica Aplicada (IPEA).
Oke, J., & Heneghan, C. (2020). Global COVID-19 case fatality rates. Centre for Evidence-Based
Medicine. Retrieved April 2020, from https://www.cebm.net/covid-19/global-covid-19-case-
fatality-rates/.
OPHI & UNDP (2019). Global multidimensional poverty index 2019: illuminating inequalities. United
Nations Development Programme and Oxford Poverty and Human Development Initiative.
Retrieved April 2020, from https://ophi.org.uk/wp-content/uploads/G-
MPI_Report_2019_PDF.pdf.
Paim, J., Travassos, C., Almeida, C., Bahia, L., & Macinko, J. (2011). The Brazilian health system:
History, advances, and challenges. The Lancet, 377(9779), 1778-1797.
Pareek, M., Bangash, M. N., Pareek, N., Pan, D., Sze, S., Minhas, J. S., . . . Khunti, K. (2020). Ethnicity
and COVID-19: An urgent public health research priority. The Lancet, 395(10234), 1421-1422.
Raupp, L., Fávaro, T. R., Cunha, G. M., & Santos, R. V. (2017). Condições de saneamento e
desigualdades de cor/raça no Brasil urbano: Uma análise com foco na população indígena com
base no Censo Demográfico de 2010. Revista Brasileira De Epidemiologia, 20(1), 1-15.
Rubin, D., Huang, J., Fisher, B. T., Gasparrini, A., Tam, V., Song, L., . . . Tasian, G. (2020). Association
of Social Distancing, Population Density, and Temperature with the Instantaneous Reproduction
Number of SARS-CoV-2 in Counties Across the United States. JAMA Network Open, 3(7),
e2016099.
Souza, W. M., Buss, L. F., Candido, D. D., Carrera, J., Li, S., Zarebski, A. E., . . . Faria, N. R. (2020).
Epidemiological and clinical characteristics of the COVID-19 epidemic in Brazil. Nature Human
Behaviour.
WHO (2018a). Household air pollution and health (World Health Organization Fact sheets). Retrieved
April, 2020, from https://www.who.int/news-room/fact-sheets/detail/household-air-pollution-and-
health/.
WHO (2018b). WHO Housing and health guidelines. Geneva: World Health Organization; 2018.
Licence: CC BY-NC-SA 3.0 IGO.
WHO (2019a). Water, sanitation, hygiene and health: a primer for health professionals. Geneva: World
Health Organization, WHO reference number: (WHO/CED/PHE/WSH/19.149).
WHO (2019b). Drinking-water (World Health Organization Fact sheets). Retrieved April 2020, from
https://www.who.int/news-room/fact-sheets/detail/drinking-water/.
WHO (2020a). Coronavirus disease (COVID-19) advice for the public. Retrieved May 26, 2020, from
https://www.who.int/emergencies/diseases/novel-coronavirus-2019/advice-for-public/.
WHO (2020b). Water, sanitation, hygiene, and waste management for the COVID-19 virus. World
Health Organization, WHO reference number: WHO/2019-nCoV/IPC_WASH/2020.3. Retrieved
May, 2020, from https://www.who.int/publications-detail/water-sanitation-hygiene-and-waste-
management-for-the-covid-19-virus-interim-guidance/.
Yancy, C. W. (2020). COVID-19 and African Americans. JAMA, 323(19), 1891.
Zhao, H., Harris, R. J., Ellis, J., & Pebody, R. G. (2015). Ethnicity, deprivation and mortality due to 2009
pandemic influenza A(H1N1) in England during the 2009/2010 pandemic and the first post-
pandemic season. Epidemiology and Infection, 143(16), 3375-3383.
Notes
1
InfoAmazonia (2020) estimates that, in the Amazon forest region, the Indigenous tribes live on average about 315 km away
from public hospitals equipped with intensive-care departments.
2
The GMPI indicators are nutrition, child mortality, years of schooling, school attendance, cooking fuel, sanitation, drinking
water, electricity, housing, and assets.
3
The color/ethnicity classification follows the POF/IBGE, in which the individuals in the survey declared their race identity
without any influence from the interviewer. The categories are White, Black, Yellow (people that claimed to have Asian
origin), Brown (people that claimed to be parda, mulata, cabocla, cafuza, mameluca, or black mixed-race), indigenous, and
not identified (not declared). For an ethnic background, a discussion about race as a social construction, and segregation in
Brazil, see Fernandes (2017).
4
In GMPI deprivation, the cutoff for being considered poor is deprivation of one-third of the ten indicators. In the GMPI, the
term “vulnerable” is used differently: a person is considered vulnerable to poverty if he/she is deprived of between one-fifth
and one-third of the indicators.
5
Brazil’s health-care is provided by both public and private sectors, and people can use the two sectors depending on
accessibility and ability to afford costs. The public health system, through the Unified Health System (SUS), aims to offer
universal, free of charge, health service provision. The private sector offers services mainly through health plans, insurance
premiums, out-of-pocket payments, and provides services for the SUS. For more details, see Paim et al. (2011) and Massuda
et al. (2018).
