World Bank, G2Px, Scaling up Social Assistance Payments as Part of the Covid-19 Pandemic Response (2020)
Summary
World Bank G2Px paper, Scaling up Social Assistance Payments as Part of the Covid-19 Pandemic Response, published in 2020 under the G2Px Initiative launched in early 2020 in partnership with the Bill and Melinda Gates Foundation to improve government-to-person (G2P) payments through digitization. The abstract describes the expansion of social assistance as among the most common public policy responses to the COVID-19 pandemic, at a scale unprecedented in government spending. Sections cover the challenges of scaling up social assistance, the early international experience, the second wave of cash transfer responses, conclusions and references. Annex 1 tabulates country data on cash transfer recipient households and people covered, with totals of 389,775,468 households and 1,737,852,062 people across 67 countries, corrected on August 28, 2020.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Cite this document
“World Bank, G2Px, Scaling up Social Assistance Payments as Part of the Covid-19 Pandemic Response (2020).”
Original at pubdocs.worldbank.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/world-bank-g2px-scaling-up-social-assistance-payments-as-part-of-the-c-5fcc8f3753/
, SHA-256 82577f95fa3e3ae1aa09b1c91a4632022e0f7b7ebe39851ca4881c46f3c72947.Full text
i
© 2020 International Bank for Reconstruction and Development / The World Bank
1818 H Street NW
Washington DC 20433
Telephone: 202-473-1000
Internet: www.worldbank.org
This work is a product of the staff of The World Bank with external contributions. The findings,
interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World
Bank, its Board of Executive Directors, or the governments they represent.
The World Bank does not guarantee the accuracy of the data included in this work. The boundaries,
colors, denominations, and other information shown on any map in this work do not imply any
judgment on the part of The World Bank concerning the legal status of any territory or the endorsement
or acceptance of such boundaries.
Rights and Permissions
The material in this work is subject to copyright. Because the World Bank encourages dissemination of
its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as
full attribution to this work is given.
Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank
Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-
2625; e-mail: pubrights@worldbank.org.
Cover photo: Achmad/World Bank
ii
TABLE OF
CONTENTS
ABOUT G2PX ......................................................................................................... iv
ABSTRACT .............................................................................................................. v
CHALLENGES IN SCALING UP SOCIAL ASSISTANCE IN THE CONTEXT OF A
RESPONSE TO COVID-19 ........................................................................................ 1
THE EARLY INTERNATIONAL EXPERIENCE.............................................................. 3
THE SECOND WAVE OF COVID-19 CASH TRANSFER RESPONSES ......................... 13
CONCLUSIONS ..................................................................................................... 17
REFERENCES ......................................................................................................... 19
ANNEX 1. COUNTRY DATA ................................................................................. 216
iii
About G2Px
The G2Px Initiative was launched in early 2020 in partnership with the Bill and Melinda Gates
Foundation. This initiative contributes to the broader agenda of improving government-to-person (G2P)
payments through digitization by shifting the paradigm beyond program-specific efficiency gains to one
that simultaneously accelerates critical development outcomes such as financial inclusion, women’s
economic empowerment, and government fiscal savings. Through this initiative, the World Bank Group
seeks to establish a global framework, develop best practices and provide upstream technical assistance
to radically improve G2P payments globally, ensuring that all G2P programs aim and design for broader
inclusion and empowerment outcomes through a focus on digitization.
The G2Px Initiative seeks to refine the understanding of both the benefits and challenges in order to
create a roadmap to ensure that governments and the technical advisors that support them are
empowered to design G2P programs not only for short-term efficiency gains for single initiatives but also
for government-wide efficiency gains, and, for recipients, a seamless user experience in the short-term
and maximum empowerment benefits in the long-term. This roadmap, complemented by in-country
case studies and technical assistance, will help to build a global movement for ensuring that all G2P
specifically state financial inclusion and women’s economic empowerment as objectives and design
programs accordingly.
iv
Abstract
Globally, the expansion of social assistance is among the most common public policy responses to the
COVID-19 pandemic. The scale of these programs is unprecedented both in terms of government spending
and the share of the population receiving social assistance. The challenge of quickly determining who
would be eligible for these transfers highlighted gaps in information about workers in the informal sector
and their families and other vulnerable groups. A review of some of these experiences reveals the
importance of certain ‘assets’ such as a robust digital identification system and other key registries in
areas such as tax and social security. Countries with superior and more inclusive assets in this regard were
able to move more quickly than those without them. This may lead governments to expedite the
investments required in this infrastructure and better prepare them for future crises. In the meantime,
the second wave of responses may have to rely on innovative methods of registration and targeting, using
new data sources and leveraging the ubiquity of mobile phones in low and middle-income countries. The
challenge in the years that follow will be to find a way to permanently integrate real time data on the
majority of the population that operates in the shadows of the informal sector.
v
Challenges in scaling up social
assistance in the context of a response
to COVID-191
The first round of reports around the world confirmed that temporary social assistance payments2 will be
an important part of the response to the COVID-19 pandemic. As of June 30, more than 100 countries had
announced plans to scale up social assistance payments by increasing benefit levels, coverage or both. 3
Dozens of developing countries will spend unprecedented amounts to provide direct assistance to
individuals and households suffering from the effects of the pandemic.
This raises two separate, but related challenges. The first is to make social assistance payments safely
and securely in the new context of the pandemic.4 The subject of this note is the second challenge; to
expand the list of eligible social assistance beneficiaries. Determining who is eligible for the new transfers
quickly and with minimal physical contact is complicated by the fact that data on a large segment of the
population is often not present in existing government databases. As illustrated in Figure 1, the social
protection system in developing countries tends to cover workers in the formal sector through social
insurance and poor households through social assistance.
Figure 1. Social Protection Responses to the COVID-19 Pandemic
Source: Own elaboration.
1Drafted by Robert Palacios, Lead Social Protection Specialist, World Bank with contributions from Doug Johnson, World Bank
consultant. Useful comments and suggestions were provided by Vyjayanti Desai, Ana Georgina Marin Espinosa, Luz Stella
Rodriguez, Kenichi Nishikawa Chavez, Alan Gelb, Valentina Barca and Jonathan Marskell. The views expressed here are the
author’s and do not necessarily reflect the views of the World Bank.
2 Social assistance can also take the form of in-kind benefits such as distribution of food. Social assistance here refers only to
monetary benefits which can take various forms in addition to physical cash.
3 These are sometimes referred to as vertical and horizontal expansions. See Gentilini et al (2020).
4 See World Bank (2020a).
1
Typically, a relatively small share of the labor force is found in the formal sector. Their employment status
and labor income can be monitored through their social insurance contributions or through public sector
payrolls. These databases allow governments to introduce wage subsidies, provide loans or extend
unemployment benefits (where these exist). At the same time, existing lists of social assistance
beneficiaries can be used to channel emergency payments quickly. Dozens of countries have increased
the amounts paid, relaxing requirements (e.g., residency status), suspended conditionalities and made
payments in advance. In contrast, the new beneficiaries of the COVID-19 emergency transfers tend to be
those in the non-poor informal sector, the so-called ‘missing middle’. In low and even middle-income
countries, this is often the largest share of the population. These workers and their families may not
appear in any government database and in the lower income countries, many do not have any form of
identification.5
Some countries have created social registries6 that are used to assess needs and poverty status. The
population coverage of these registries varies. Those that cover most of the population, including the non-
poor, informal sector, can be used to quickly add to the beneficiary lists after applying certain criteria. In
most cases, the expansion based on the social registry cross-checks other administrative databases such
as income tax or social insurance to minimize inclusion errors. The use of existing databases expedites the
process and minimizes the need for direct contact with the population. However, this approach depends
on the accuracy and interoperability of databases being utilized.
Countries without this option must essentially create them from scratch. The nature of the pandemic in
terms of minimizing physical interaction as well as the need to move quickly means that digital application
and registration processes are strongly preferred. As in the case of an existing social registry, the newly
collected data can be cross-checked against existing government databases to improve targeting. A few
countries have used both approaches.
Countries with trusted and inclusive foundational identification systems7 that are digitalized and have
close to universal coverage are at a distinct advantage for several reasons.8 First, in the case of new
applications, the uniqueness of the identifier used in the process ensures that the same individual could
not apply to receive more than one benefit. Double dipping can also be avoided by ensuring that someone
already receiving benefits from another existing program does not become eligible for the new payment.
To the extent that the identification system can be used for authentication, ‘ghost’ or non-existent
5 See, https://id4d.worldbank.org/global-dataset
6 Here we refer to both traditional social registries based on data collected at the household level, typically in ‘census sweeps’,
which includes variables that are used to rank the households according to level of poverty often based on proxy-means tests
(e.g, Pakistan or Rwanda) as well as more advanced systems, sometimes called ‘virtual’ social registries. The latter are typically
based on application processes where data are pulled from multiple administrative databases in order to determine whether a
household is qualified to receive a benefit (e.g., Jordan or Turkey).
7 The ten Principles on Identification for Sustainable Development, which have been endorsed by 25 organizations, offer a
useful guiding framework to elaborate on what ‘trusted and inclusive’ means. For more information see:
https://id4d.worldbank.org/principles.
9 An additional advantage of inclusive digital identification systems that is beyond the scope of this note is their ability to
facilitate customer due diligence (CDD) requirements in order to open accounts to be used to channel payments for individuals
that did not already have bank accounts, potentially enabling remote onboarding of beneficiaries with little to no information
requested from them.
2
beneficiaries can be removed.9 An inclusive foundational ID with close to universal coverage can also be
used to quickly expand coverage, starting with the entire population and filtering out only the most
affluent.
Most importantly however, the ability to cross-check various databases using the unique identifier has
made it possible to increase the precision of targeting. Prior to the crisis, Turkey used 28 government
databases ranging from property and auto registries to income tax and social insurance registries to
determine eligibility. The same system was harnessed to temporarily increase the number of eligible cash
transfer recipients from 4 to 12 million households. This was an example of the expansion of an existing
integrated social registry10 where no additional data had to be collected. This contrasts with static social
registries where periodic door to door data collection of certain indicators is utilized for targeting purposes
and which therefore may be outdated at the time of a crisis.
Some countries used on-line registration and then cross-checked different administrative databases using
the national ID number. Thailand accepted 28 million applications in a few weeks and used various
government registries to filter those that would not be eligible according to the criteria applied. The fact
that the unique ID number was linked to existing bank accounts also expedited the response in countries
like India, Chile and Thailand where this mapping for government payments was already standard
practice. Finally, the identifier was used to convey whether someone was eligible for benefits in several
countries. In Pakistan, the individual had only to send an SMS with the number to verify eligibility.
With a few exceptions, countries with social registries that covered almost the entire population also have
digital identification systems with very high coverage. The power if the identification system was further
magnified by the existence of other administrative databases, especially social insurance which helped
quickly distinguish formal sector workers. Conversely, countries which do not have a strong, national level
ID face difficulties ensuring uniqueness (deduplication) as well as using their administrative data to
improve targeting of benefits.
As discussed below, the first wave of countries to expand social assistance payments to new beneficiaries
tended to be those with better identification systems, social registries and other administrative data. The
next section reviews the early international experience of scaling up social assistance in response to the
pandemic.
The Early International Experience
While more than 80 low and middle-income countries have announced new or expanded cash transfer
programs to offset the impact of the COVID-19 pandemic, less than half have already identified the new
9 In the case of the United States stimulus payments, gaps in this system resulted in checks being sent to more than one million
dead people. https://www.gao.gov/products/GAO-20-625
10 See Barca and Chirchir (2020) for a discussion of integrated social registries.
3
beneficiaries and started making payments. In some of these cases, the estimated number of new
beneficiaries has been announced while in others, no estimates or targets are available. Based on actual
and target beneficiary figures for 66 developing countries,11, the total number of new social assistance
beneficiaries is well over a billion. Excluding India, the figure is close to 900 million. Figure 2 shows that
after excluding India, two thirds of the new beneficiaries live in Latin American and East Asia.12
Figure 2. Distribution of individuals in households receiving new cash transfers
13%
SAR
41% 15% Africa
MENA
LAC
15% EAP
16%
Sources: World Bank Social Protection COVID-19 Response database; UNDP COVID-19 response database; various
World Bank project documents and unpublished notes.
The average share of the population living in a household receiving or targeted to receive a new benefit
across the 66 countries is about 29 percent but this figure varies significantly across countries and regions.
As shown in Figure 3, it ranges from about 13 percent in Africa to about 40 percent in East Asia and Latin
America. In 10 countries more than half the population lives in a household receiving a new transfer.
These are unprecedented figures and demonstrate the magnitude of the response.
11 These figures exclude high income OECD countries as well as Eastern Europe and Central Asia as well as China.
12 The estimate for India in Annex 1 of close to a billion people is based only on the number of women covered by the Pradhan
Mantri Jan Dhan Yojana (PMJDY) program and their families based on average household size. However, it is possible that these
women live in households already receiving a different transfer, so this figure is an overestimate. It is nevertheless important to
note the unprecedented speed with which this huge program was implemented; the third month of payments was completed in
June 2020 for a total of around US$4 billion dollars. The other programs such as widows and the LPG subsidy are not new
programs but receive top ups. Arguably, the construction workers welfare fund members represent new beneficiaries in which
case the total for India would increase by about 100 million to the extent they were all new beneficiaries (based on the households
of around 23 million construction workers). The bottom line is that the vast majority of households in India received either a top
up, a new transfer or both.
4
Figure 3. Median share of the population receiving new social assistance payments as part of the
COVID-19 response in each region
Average share of population covered
39.9% 39.0%
30.4%
24.2%
20.2%
AFRICA MENA SAR EAP LAC
Sources: World Bank Social Protection COVID-19 Response database; UNDP COVID-19 response database; various
World Bank project documents and unpublished notes.
Only about half of these new programs have been at least partially implemented and in about a third, the
first round of payments is currently being paid or, in a few cases, have even been completed. The most
common ways of reaching new cash transfer recipients to date are either to add households based on
existing social registries or to accept on-line or mobile applications and determine eligibility. Several
countries do both while a few have used paper-based applications. In the case of use of social registries,
existing fields in the database such as age can be used. In most cases, there is an attempt to cross-check
with other databases to ensure that, for example, the person is not receiving a benefit through the social
insurance agency.
In the case of new applications, the approach in countries that have implemented quickly has been largely
digital; individuals fill out online questionnaires or submit their data through mobile apps on their own or
with assistance, such as from local governments or social workers. After the data are submitted (typically
with at least a nominal check for consent by the applicant), these are checked against a set of eligibility
criteria using different administrative databases. Advanced countries also verify identity at this stage,
typically using some form of demographic matching. Thailand, for example, used a code on the back of its
national ID card in a similar way to how online credit card transactions us the code on the signature panel
to authenticate the holder. The applicants are then notified as to whether they were accepted or rejected.
In some cases, an explanation is provided. Some systems allow those rejected to file their grievances with
a process in place to respond and rectify as needed.
5
Table 1 shows how 16 of the countries that moved quickly – representing a little less than half of the
developing countries that did so globally – were able to add to their existing social assistance rolls in a
relatively short time.13 All of these countries have completed the vast majority of their first round of
payments within a few months of announcing the policy. In eleven cases, new applications were received
with only the Philippines using paper forms distributed by local governments. In five cases, most of the
additions used only the existing social registries. Brazil, Egypt and Indonesia and Tunisia used both
methods. Out of the 16 countries, 14 used a unique identifier to cross-check their beneficiary list against
other databases to determine eligibility.
Table 1. Methods and Assets used for adding new beneficiaries to social assistance Programsa
Social registry Cross-checks with
New
Country based- administrative Assets used in expansion
applications
expansion databases
Argentina X Yes National ID, functional registries
Brazil X X Yes Social registry, tax ID, functional registries
Chile X Yes National ID, social registry, functional registries
Colombia X Yes National ID, social registry, functional registries
National ID, social registry, functional registries
Ecuador X Yes
geographic (census data) and mobile usage
Egypt X X Yes National ID, functional registries
Indonesia X X Yes National ID, social registry, functional registries
Jordan X Yes National ID, functional registries
Namibia X yes National ID, functional registries
Pakistan X Yes National ID, functional registries, social registry
Peru X Yes National ID, functional registries, social registry
Philippines X No Local government officials
South Africa X Yes National ID, functional registries
Thailand X Yes National ID, functional registries
Togo X No Functional ID (voter)
Tunisia X X Yes Civil registry, functional registries
Note: a. All countries except Thailand provided additional benefits to those already receiving social assistance.
Not surprisingly, the method chosen reflected available administrative data in each country. Countries
with traditional social registries with high population coverage used these exclusively and eschewed new
applications. Those with partial coverage, such as Brazil and Indonesia, added on-line applications to
supplement the limited scope of the social registry. Social registry-based expansions had the advantage
of avoiding new application or enrollment processes, although additional information related to payments
13 Barca (2020) lays out a more detailed taxonomy to differentiate the mechanisms to add new beneficiaries.
6
channels was often still required.14 People were informed that they were eligible through mass-
communication campaigns, combined with websites where people could check whether they had been
declared eligible. In the case of Pakistan, potential recipients were asked to send a text message to a
telephone number; a text message in reply confirmed whether they were eligible or not. The websites
and messages also provided information on how to receive/collect/cash out payments. In all cases, the
notification process relied on their unique ID numbers. This method allowed for new enrollments with no
physical contact and minimal transaction costs. A major limitation of this approach was the fact that some
of the registries were not up to date. At the same time, even had they been updated, the use of a proxy-
means-test to determine poverty status would still be problematic since the correlation of the variables
used for the calculation with consumption levels would have been structurally changed by the pandemic.15
Peru is a good example. Its social registry ranks about 80 percent of the population from poorest to richest.
The first phase of its expansion of social assistance payments was achieved by increasing the cutoff score
within this population resulting in an increase from around 724,000 to 2.7 million households or more
than one third of the population. The second phase also took advantage of the data in this registry but
then applied cross-checking filters using income and social insurance data to eliminate the remaining low-
income households that had formal sector jobs. This was straightforward because the unique ID number
which has almost universal coverage could be linked across databases. Peruvians could then simply go to
a website with their national ID number and check to see if they had been declared eligible for the transfer.
Countries with no social registry took applications, typically in a matter of a few weeks, but in some cases
much longer. In Namibia, for example, one million applications were received in two weeks. About 60
percent were declared eligible and paid into bank accounts within two more weeks. Its larger neighbor,
South Africa, opened applications on May 11th and by June 18th had only managed to pay half of the 2.6
million accepted applicants.16 Most of the other countries using this approach have managed to pay out
more quickly than South Africa with the exception of the Philippines where it has taken several months.
In countries that relied on applications, the Philippines stands out as an outlier in several ways. First,
despite having a social registry that covered around 75 percent of the population, the government opted
for an application-based process. This was due to the fact that the social registry was already five years
old and the update that was already in progress was interrupted by the crisis. Second, the application
process was done with paper forms with the consequent need for physical contact with the majority of
the population. The data was encoded locally but with mixed speed and accuracy, which reduced the
ability to verify and deduplicate beneficiaries. Finally, there was no real possibility to cross check with
functional registries or even to deduplicate applications due to the lack of a foundational ID system.
14 In the case of Ecuador, the social registry approach was used to add 800,000 people while census data is being used to find an
estimated additional 100,000 households in poor areas not yet registered. In El Salvador and Guatemala, electricity consumption
levels were used to determine eligibility.
15 There are additional factors that affect how useful social registries can be in a crisis response. For a good discussion of these
criteria see Barca and Beazley (2019) https://www.dfat.gov.au/about-us/publications/Pages/building-government-systems-for-
shock-preparedness-and-response-the-role-of-social-assistance-data-and-information-systems
16 Gelb (2020), https://www.sabcnews.com/sabcnews/sassa-working-on-appeal-system-for-r350-grant-applicants/.
7
In fact, with this one exception, all countries listed in Table 1 were able to rely on a national level, unique
identifier for deduplication, an important advantage in order to move rapidly while minimizing duplication
and fraud. In thirteen of the fifteen countries, the solution was a national ID with high, close to universal
coverage. In the case of Brazil, the tax ID was used effectively (albeit with some problems initially). In
Togo, a biometrically deduplicated voter ID was used. 17 An additional bonus for Togo was the fact that
occupational status had been collected as part of voter registration and was available in the database
allowing for targeting on this basis rather than a functional registry.
India also had a ubiquitous foundational ID that greatly facilitated the COVID-19 cash transfer response.
However, it was unique in that most of the new beneficiaries were women that had been part of a financial
inclusion program dating back several years. These accounts were opened using the biometrically
deduplicated Aadhaar number ensuring that each beneficiary was unique. While the goal of that program
was financial inclusion and there was an emphasis on opening accounts for women, the fact that such a
registry existed and covered more than 200 million women made it possible to channel emergency
transfers to them very quickly with the third monthly payment completed in June 2020.18
Each of the approaches had its own shortcomings.19 Traditional social registries were not up to date
forcing countries to supplement beneficiary rolls with other methods. The case of the Philippines was
already mentioned. In the case of Ecuador, geographic targeting using census data is being utilized to add
close to 100,000 households to the beneficiary list. Pakistan’s social registry was also in the process of
being updated making it difficult to target accurately. Finally, social registries were not designed to
capture information that would be relevant for the kind of economic shock that the pandemic
represented.
The application-based approach faced several constraints. First, the speed with which data had to be
collected led to technical problems as volumes exceeded website capacity in several countries. Digitally
challenged applicants could also face hurdles in accessing the internet or navigating mobile application
processes. Figure 4 is taken from a popular newspaper in Brazil. The cartoonist points out the difficulty
the poor may have to access the app required to apply for the benefit. While this rapid, real time data
collection potentially yielded more timely and relevant information, they were also difficult to verify. In
cases where the governments relied on other administrative databases for cross-checks, the quality of
various administrative databases was tested. As Gelb and Mukherjee20 point out, it is important to have
human back up processes in place to deal with problems. The actual performance of these systems will
only become evident in the coming months when survey data is available and compared to reported
administrative data.
17 The voter registry contains 3.6 million out of an estimated 3.85 million adults aged 18 and over. The database was up to date
due to the fact that the elections were held in February 2020. See Boko et. al. (2020).
18 Studies suggest that a significant proportion of these benefits have not actually been cashed out by beneficiaries for different
reasons including difficulty of movement during lockdown or lack of awareness. See for example,
https://www.indiaspend.com/40-of-jan-dhan-account-holders-could-not-access-govts-covid-19-relief-survey/
19 For a more extensive discussion, see Barca et al. (2020) https://socialprotection.org/discover/publications/space-guidance-
note-rapid-expansion-social-protection-caseloads.
20 Gelb and Mukherjee (2020).
8
Figure 4. “Have You Downloaded the App?”
Source: Nando Motta (2020); see https://www.brasil247.com/charges/baixou-o-app
Thailand, which has various cash transfer programs but does not use a social registry to determine
eligibility, targeted informal sector workers and set up a website to accept applications. More than 28
million people applied corresponding to about half of the working age population. This figure was much
higher than the government expected forcing it to revise its original target from 3 to 9 and eventually to
more than 15 million declared eligible. Brazil combined the social registry and application methods. First,
around 11 million people already in the social registry were added as new beneficiaries (the others
received a top-up). The social registry only covered about one third of the population so the remaining
informal sector workers would have to use the application process. The website and mobile application
were quickly rolled out and more than 100 million applications were received with 36.4 million declared
eligible by the end of the first phase. In both countries, the massive number of applications were not
anticipated and tested their systems. Nevertheless, these cases and several others (e.g., Namibia and
Turkey) showed how quickly benefits could be extended to the majority of the population. The websites
from each country are shown below. The role of mobile phone (including feature phones through USSD)
and internet based self-application combined with the ability to check the information with other
databases was crucial in most of the early implementers.
9
Figure 5. COVID-19 Crisis Cash Transfer Application Websites, Thailand and Brazil
As mentioned earlier, almost all countries used existing functional registries like social insurance to cross-
check for eligibility. This has revealed deficiencies in these databases, especially in terms of their
synchronicity. In many cases in Thailand for example, applications by individuals who had eligible jobs in
urban transport or other services were rejected because they had not been removed from the registry of
farmers. In South Africa, delays appear to have been caused partly by cross-checking with unemployment
10
rolls and unclear eligibility criteria. In many countries, civil registries cannot collect reliable data on deaths
resulting in many ‘ghost’ beneficiaries. The rapid pace of the enrolments taking place during the crisis will
inevitably result in errors of both inclusion and exclusion. Nevertheless, some countries will be in a better
position to correct errors and add or subtract from beneficiary lists.
Finally, this type of data sharing or ‘data aggregation’ also raises questions as to whether it violates privacy
or personal data protection regulations. While consent may be part of the application process, most
individuals will not be aware of how their data are being used. Moreover, they are not in a strong position
to object as it would result in foregoing the transfer. In Colombia, the government authorized public and
private entities to share data but issued a decree mandating that they could only be used for this purpose
and guaranteeing their confidentiality.21
The early experiences have revealed many challenges, but they have also demonstrated the feasibility of
quickly expanding the role of social assistance to non-traditional beneficiaries to offset the negative
impact of the pandemic. A key lesson has been the importance of infrastructure and data, especially the
underlying identification system and functional registries such as social insurance. Fifteen of the sixteen
countries reviewed here have robust forms of unique identifiers with high coverage while fourteen of the
sixteen were able to leverage existing functional databases.
Figure 6 below expands the sample of countries in order to better illustrate the importance of initial
conditions for rapid implementation of programs for informal sector workers and their families. All of the
66 countries shown in the figure announced a policy that involved adding new social assistance
beneficiaries between the end of March and the first week of May 2020. The countries marked in yellow
had finished a first round of payments before June 30th 22 while the countries marked in blue had not
(although many had made top up payments to beneficiaries of existing programs). About half of the
countries in the sample here were able to make their first round of payments quickly.
The y axis is an indicator based on the average of the coverage rate of the identification system and the
coverage of the social insurance registry. As discussed earlier, a robust identification system that covers
most of the population facilitates registration and determination of eligibility by ensuring uniqueness and
allowing for cross-checking various administrative databases. It also facilitates account opening for
payments for those without bank accounts. Digital, social insurance registries, especially when they can
be mapped to the national ID, allow for rapid cross-checking of applications from workers claiming to be
in the informal sector. Also, the larger the coverage of the formal sector schemes, the smaller the share
of the population that must be added to the social assistance rolls since there are other mechanisms
available to provide them with benefits. The y-axis is the estimated share of the adult population with
bank accounts. The latter indicator serves as a proxy for the ease with which payments can be made.23
21 https://ingresosolidario.dnp.gov.co/documentos/DECRETO_518_DEL_4_DE_ABRIL_DE_2020.pdf
22 Some countries made a single payment but most made between three and six months of payments. India, for example,
announced the scheme in March and completed three monthly payments by the end of June.
23 The first wave countries averaged 90 and 58 percent on the financial inclusion and ID indicators compared to 34 and 68 for the
late starters.
11
While there are many factors24 that affected how quickly countries implemented these new payments,
Figure 6 is suggestive of the importance of the initial conditions that allow for rapid registration and
payment. Notably, most of the countries that were able to create a new list of beneficiaries and start to
pay them during this period are found in the top right quadrant of the figure with high scores on both
indicators.25 There are two obvious exceptions, Pakistan and the Philippines. One explanation is the
availability in Pakistan of a social registry that covered more than three quarters of the population, as well
as the ability to cross check with bank account, passport, property and other databases using the unique
national ID number. The Philippines also has a social registry with similar coverage. However, it was not
used for the coverage expansion as it was outdated and a new census round was under way when the
crisis struck. Moreover, it does not have a national ID and is therefore unable to cross-check other
databases including social insurance.
Figure 6. Relationship between initial conditions and implementation of new social assistance payments
100%
Iran Mauritius
90%
India Thailand
% of adults with a bank account
80% Chile
70%
60% Zimbabwe
50% Bangladesh
40% Philippines Togo
30%
Vietnam
Myanmar
20% Pakistan
10%
0%
0% 10% 20% 30% 40% 50% 60% 70% 80%
Identification index
Source: Author’s calculations using data in Annex 1, Global FINDEX for share with bank accounts, ID4D Global
database for having an ID and World Bank pension database for social insurance coverage.
Note: The circles marked in yellow are countries that made or were close to completing their first round of payments
by June 30, 2020.
24 Additional factors include different levels of urgency depending on the impact of lockdowns, fiscal constraints and the capacity
of the public sector generally. Seeking (2020) provides an example of limited government capacity in the case of South Africa.
25 Note that coverage was also correlated with these initial conditions. The early implementers covered about 39 percent of the
population on average compared to 19 percent for the late starters.
12
All the countries that were not able to complete a first round of payments had relatively low scores on
both indicators. Vietnam, for example, issued the resolution establishing the new payments to informal
sector workers (along with existing social assistance beneficiaries) on April 11th but was unable to register
these individuals by the end of June. Zimbabwe is an interesting case and seemed likely to be among the
first wave of countries responding. It has the advantage of a very high penetration of mobile money
accounts and a national ID with high coverage and announced a package that would make three, monthly
payments for one million vulnerable households on March 31st. A major obstacle is related to a question
of personal data protection related to questions about MNOs sharing national ID data of their subscribers
with the government. Payments had not been made as of early July. In contrast, Namibia announced a
one-time payment on April 9th which was paid to most adults in the country by May 8th.
The second wave of COVID-19 cash
transfer responses
The countries that have not yet started to make payments to new beneficiaries tend to be lower income
and have larger shares of informal sector workers, many of whom do not have forms of identification that
can be easily checked. In the longer run, the crisis may encourage governments to invest in systems that
allow them to react more quickly and some projects were already in motion at the time of the crisis.
Ethiopia, for example, had planned to launch a new digital ID along the lines of India’s Aadhaar system. In
the short run however, there is no digital ID in Ethiopia so that alternative approaches will be required. In
contrast, Rwanda is likely to be in a good position to move quickly due to the high coverage of its digital
ID system and a social registry that covers more than 90 percent of the population.
The experience of Timor-Leste, the lowest income country to have managed to complete its first payment
by the end of June, is instructive for the second wave countries. Like Togo, it was fortunate to have
recently had an election and voter registration rates were high. This provided a ubiquitous and relatively
robust form of identification. While direct authentication with the voter database was not possible, checks
for duplicates and erroneous number structures helped reduce potential errors and fraud. Combined with
the demographic data that had been collected at the village level and digitized, this provided a reasonable
starting point for the government teams to go out using software on tablets26 and prepopulated data and
issue cash payments (with social distancing). While not ideal, the digitization of the processes at each step
including grievance redressal, increased transparency and appears to have limited diversion and fraud.
Out of 310,000 households, complaints were filed by about 12,000 or four percent.
Similarly, Cambodia expects commune level officials to collect data using tablets already available from
previous registration efforts for subsidized health insurance. The first-round payments started in July and
used the existing health insurance registry that already contained about 550 thousand households. In the
26 The tablets provided to the 452 teams that moved from village to village making and recording payments were available due
to their use for the 2020 census. See Pinxen (2020).
13
last few months, they have added another 60 thousand households using the existing community-based
process in rural areas. While the government is discussing further expanding to cover urban informal
sector workers affected, this has not yet happened. Luckily, after a recent campaign, most Cambodians
now have a relatively robust national ID and mobile penetration is very high in urban areas.
Figure 7. Data collection for targeting using a tablet in Cambodia
Source: https://www.giz.de/en/mediacenter/75045.html
Cambodia and Timor-Leste are predominantly rural. The task is more complicated in countries where the
target populations are a subset of informal sector workers in urban or semi-urban areas. In Vietnam for
example, commune level officers must collect data on occupational status with paper forms to be digitized
and reviewed to determine eligibility. Aside from the physical contact required, there are several
problems with this approach: First, the answers to these questions are very difficult to verify. Second,
while Vietnam’s national ID covers most of the population, it is not possible to authenticate individuals
electronically to ensure uniqueness or to ensure that applicants are not actually covered by social
insurance or social assistance programs already.
These constraints have reinforced the need to build identification systems that can be harnessed for rapid
responses to crises and for more efficient government administration in normal times. It has also
highlighted the gap in information about the so-called ‘missing middle’, the non-poor, informal sector
which comprises anywhere from 40 to 90 percent of the labor force. Investments in these areas will take
a few years to bear fruit, however. In the meantime, some countries are exploring innovative ways to take
advantage of new techniques for targeting, especially methods consistent with social distancing.
14
One of these techniques relies on the relatively recent phenomenon of high mobile phone penetration in
poor countries27. Where mobile penetration is very high, correlations between ‘ground truth’ data
collected through sample surveys to estimate consumption and poverty at a specific point in time with
mobile phone use patterns has been shown to predict the poverty status of households at least as
accurately as traditional methods.28 However, accuracy is directly related to the amount of information
about individuals available raising new concerns over privacy, including consent.
An alternative is to use anonymized mobile phone data for geographic targeting. In Zimbabwe, for
example, ground level data suggest a strong correlation between density in the capital, Harare, and
poverty rates. Zimbabwe has the additional advantage of high mobile money penetration making it
possible to make payments directly to the e-wallets of all phone numbers mapped to the targeted areas.
These techniques have the additional advantage of tracking behaviors in real time so that the impact of
the transfer is easier to monitor. Several countries, including Bangladesh, Nigeria and the Democratic
Republic of Congo are currently considering this approach in the context COVID-19 pandemic but it could
potentially be used in other situations such as natural disasters.29
Another approach is to combine census data with high precision images from aerial drones or satellites in
order to establish algorithms with machine learning that can predict various indicators relevant for
eligibility. This has been demonstrated to work well at a neighborhood and reasonably at the level of
households in the case of Colombia and is being explored in several other Latin American countries.30
Figure 8 shows the level of geographic granularity in Cartagena and Rwanda using these two alternative
targeting methods. It would be interesting to see how much accuracy might be improved by combining
these methods.
Figure 8. Innovative methods of geographic targeting in real time
Sources: Terraza et al (2020); Blumenstock et al (2015)
27 As shown in the last section, mobile phones can also be effectively harnessed for rapid data collection through a self-registration
process that avoids traditional door-to-door surveys.
28 See for example, Steele and others (2017) and Blumenstock, Cadamuro, and On, (2015).
29 It is important to distinguish between the use of granular data to help determine eligibility at an individual or household basis
versus geographic targeting. While both are possible, the use of identifiable data raises important issues about consent and data
protection.
30 Terraza et. al., (2020).
15
These methods lend themselves especially well to urban areas. This is important in the COVID-19 context
for two reasons. First, most social assistance beneficiaries in low income countries are found in rural areas
where poverty rates tend to be higher. Second, many of the most seriously affected occupational groups
are found in urban and semi-urban areas (for example, street vendors, rickshaw drivers, domestic help
etc.). The first expansion of social assistance to a larger number of poor, rural households was relatively
straightforward in countries like Indonesia and Vietnam. It has been a much slower process to identify
urban, informal sector workers affected by the economic downturn. In Vietnam, vertical expansion of
social assistance benefits took place within a month of their announcement. Three months since the new
benefits for urban informal sector workers was announced, only 35 thousand out of the targeted 5 million
workers have received any payment.
Myanmar is a useful example. In less than a decade mobile penetration has risen from negligible to more
than 85 percent. Figure 9 based on census data from 2017 shows that even higher rates of mobile phone
ownership prevailed for informal sector workers in the kind of occupations affected by the pandemic.
These figures will almost certainly have risen by 2020. The government has recently announced a plan to
provide social assistance to around five million households or about 40 percent of the population. In the
context of few other assets for delivery, harnessing mobile telephony for targeting, registration and even
payment seems worth exploring.
Figure 9. Mobile phone ownership by occupation in urban areas in Myanmar
Refuse Workers and Other Elementary
Workers
Food Processing, Woodworking, Garment
and Other Craft and Related Trades Workers
Labourers in Mining, Construction,
Manufacturing, and Transport
Drivers and Mobile Plant Operators
Sales Workers
Food Preparation Assistants
Street and Related Sales and Services
Workers
Building and Related Trades Workers
(excluding Electricians)
0% 20% 40% 60% 80% 100%
Source: Myanmar 2017 census data.
16
Conclusions
The options available to governments seeking to provide financial relief to people in the context of the
COVID-19 pandemic quickly are at least partly constrained by their infrastructure and the availability of
data. The quality and coverage of identification systems and the availability of a trusted and inclusive
foundational ID system, social registries, and the administrative databases used by social insurance, social
assistance, tax, and other agencies are especially important when these transfers must be paid quickly.
Out of 16 countries reviewed, 15 had robust foundational ID systems with close to universal coverage. It
is also important to note that these assets allowed governments to implement the new social assistance
relatively quickly while minimizing the risk of spreading the virus since data did not have to be collected
through physical contact.
Nevertheless, the first wave of responses did reveal flaws in existing databases and raised important
privacy concerns related to data aggregation. The crisis has brought the deficiencies of delivery systems
in many countries into sharp relief and may speed up plans to improve identification systems and digitize
government more broadly. It may also encourage governments to go beyond traditional social registries
to harness public and private sector data in new ways that provide real time information, especially on
the informal sector households that comprise the ‘missing middle’ in developing countries.
In the short run, more creative solutions will have to be employed particularly in poor countries where
the assets required for these assets are limited. Geographic targeting using recent advances in data
analysis and machine learning may offer some solutions and may even come to replace traditional, more
invasive methods such as census-sweep based targeting approaches at a much lower cost. The ubiquity
of mobile phones can be leveraged in some countries both to quickly collect data and process applications
as well as for payments, whether through mobile money at a cash payout point.
One of the key lessons that the pandemic holds for social protection practitioners is the supreme
importance of having data about individuals and households. Governments with robust and
comprehensive data were able to respond quickly while those without have struggled.31 This is especially
true for the expansion of emergency social assistance payments. Data in national identification systems
that cover most of the population played a key role in ensuring that payments were not paid to the same
person and combining information from multiple administrative databases for the purpose of determining
eligibility. Some countries, like Turkey, were already doing this using more than two dozen administrative
databases. Others like Brazil and Thailand were able to leverage the databases that they had quickly
because they all contained the unique identifier. Many countries used the social security and income tax
database to filter informal sector workers but there were many other useful databases ranging from
electricity consumption (El Salvador and Guatemala) to auto and land registries (Jordan and Turkey) to
registries for farmers and fishermen (Thailand, Maldives). Each of these countries were able to expand
their social assistance programs dramatically in less than three months.32
31
See Palacios (2020).
32
Notably, their main challenges and the reason for many complaints was the fact that some of these databases
were not up to date.
17
In countries that lack such administrative data, innovative ways to harness other available data sources
are being explored. Several Latin American countries are using satellite or drone imagery that can be
correlated with survey and census data to predict which areas in cities are poor. Several African countries
as well as Bangladesh are partnering with mobile network operators (MNOs) to use mobile phone
behavior data such as call frequency and location to predict geographic distribution of poverty and drops
in economic activity caused by the pandemic. Mobile penetration and the use of mobile money in some
of the poorest countries (e.g., Somalia and Zimbabwe) is very high and matches that of much richer
countries. In Myanmar, over 90 percent of informal sector worker affected by the pandemic have mobile
phones and almost all would have access to someone with a phone.
Good data not only allowed for quick registration of new beneficiaries, it also facilitated digital payments.
Millions of new bank accounts were opened in countries like Indonesia, Brazil and Morocco on the basis
of credible beneficiary lists linked to unique identifiers. In India, digital payments to more than 200 million
women could be made within days of the government announcement because their unique ID number
was linked to their bank account. Regulators in many countries were able to relax customer due diligence
(CDD) rules to open accounts specifically for COVID-19 payments using data provided by the agency
responsible for social assistance.
The benefits of real time data were highlighted by the pandemic but those working towards extending
social protection, including health insurance, to the informal sector have been aware of the information
gaps for the ‘missing middle’ for a long time. Policies such as those outlined in “Protecting All”33 that aim
for universal coverage and to delink social insurance from formality require this information at all times,
not just during crises. Now that so many countries are building these temporary registries for the COVID-
19 response, the challenge is how to leverage this massive effort to incorporate the informal sector and
permanently fill the data gap.
33
Packard et. al. (2019).
18
References
Aiken et al. (2020), “Targeting Development Aid with Machine Learning and Mobile Phone Data”,
Assumpcao, R. (2020), “COVID-19 G2P Cash-Transfer Payments Country Brief: Brazil”, World Bank, June
2020.
Barca, V. (2020), “Options for rapid expansion of social assistance caseloads for COVID-19 responses”,
DFID and GIZ, May 2020.
Barca, V. and Beazley (2019) “Building government systems for shock-preparedness and response: the
role of social assistance data and information systems.”, https://www.dfat.gov.au/about-
us/publications/Pages/building-government-systems-for-shock-preparedness-and-response-the-role-of-social-assistance-
data-and-information-systems
Barca, V. and R. Chirchir (2020), “Building an integrated and digital social protection information
system”, GIZ, January 2020.
Boko, J. et al. (2020). “Technical Note on Togo’s Novissi Program”, World Bank, June 2020.
Blumenstock et al. (2015), “Predicting Poverty and Wealth from Mobile Phone Metadata”, Science,
Volume 350.
GAO (2020), “COVID-19: Opportunities to Improve Federal Response and Recovery Efforts”, June 2020.
Gelb, A. (2020) “South Africa: Summary of Response to Economic Dislocation Caused by Covid-19”,
mimeo.
Gelb, A. and A. Mukherjee (2020). Digital Technology in G2P Transfers for Covid Relief: Lessons from
Selected Cases, Center for Global Development.
Gentilini, U. et al. (2020), “Social Protection and Jobs Responses to COVID-19: A Real-Time Review of
Country Measures”, June 2020.
Government of Myanmar (2017), Census.
Holzblatt, J. and M. Karpman (2020), “Who did not get the Economic Impact Payments by mid to late
May and Why”, Urban Institute, July 2020.
Khan, A. (2020), “COVID-19 G2P Cash-Transfer Payments Country Brief: Pakistan”, World Bank, June
2020.
Martin Moreno, Juan (2020), “Vertical and Horizontal Expansion of Social Protection during Social
Distancing in Argentina”, World Bank, June 2020.
Morgani, Matteo (2020). “The Case of Brazil”, Powerpoint presentation, World Bank, June 2020.
Natarajan, H. et al. (2020) “COVID-19 G2P Cash-Transfer Payments Country Brief: Jordan”, World Bank,
June 2020.
Packard, T. , U. Gentillini, M. Grosh, P. O’Keefe, R. Palacios, D. Robalino and I. Santos (2019), “Protecting
All: Risk Sharing for a Diverse and Diversifying World”, World Bank.
Pinxen, J. (2020), “COVID-19 G2P Cash-Transfer Payments Country Brief: Timor-Leste”, World Bank, July
2020.
Risso, F. and Randall, D. (2020), “COVID-19 G2P Cash-Transfer Payments Country Brief: Peru”, World
Bank, June 2020.
19
Rodriguez, Luz et al. (2020), “COVID-19 G2P Cash-Transfer Payments Country Brief: Colombia”, World
Bank, June 2020.
Seeking, J. (2020), “Bold promises, constrained capacity, stumbling delivery: the expansion of social
protection in response to the COVID-19 lockdown in South Africa’, Center for Social Science
Research, University of Capetown, South Africa, July 2020.
Steele et al. (2017), “Mapping poverty using mobile phone and satellite data”, J. R. Soc.
Interface.1420160690.
Terraza et al. (2020), “A methodology to find, improve and strengthen vulnerable homes”, Powerpoint
presentation, World Bank, June 2020.
World Bank (2020a). “G2P Payments: Key Considerations and Policy Options for Emergency Cash-
Transfer Payments”, G2Px Initiative, World Bank. World Bank (2020). “Bolivia: Social Safety Nets
Project, Project Document”
World Bank (2020b). “Multiphase Programmatic Approach, COVID-19 Strategic Preparedness and
Response Program”
World Bank (2020c). “Thailand Economic Monitor”, World Bank, July 2020, Bangkok, Thailand.
20
Annex 1. Country data
HHs or estimated Individuals in share of 0-not started,
Countries unique recipients recipient HHs Population 1-started
Algeria 322000 1899800 0.05 1
Angola 1600000 8258298 0.27 0
Argentina 5854000 20750438 0.47 1
Bangladesh 5000000 24710691 0.15 1
Bhutan 23000 115000 0.15 1
Bolivia 5279211 0.47 1
Brazil 33679000 102705000 0.49 1
Cabo Verde 42600 178920 0.33 1
Cambodia 610000 2806000 0.17 0
Cameroon 65000 325000 0.01 0
Central African Republic 70000 350000 0.07 0
Chile 2000000 7000000 0.37 1
Cote D'Ivoire 500000 2850000 0.11 0
Colombia 4500000 15750000 0.32 1
Congo,Dem.Rep 250000 1136548 0.02 0
Congo,Rep. 100000 430000 0.08 0
Costa Rica 532820 1864870 0.37 1
Dominican Republic 1951000 6828500 0.64 1
Ecuador 950000 3610000 0.21 1
Egypt 2500000 10250000 0.10 1
El Salvador 1026000 3283200 0.51 1
Ethiopia 855000 3933000 0.04 0
Guatemala 1600000 7680000 0.45 1
Guinea 150000 960000 0.08 0
Haiti 1500000 6600000 0.59 0
Honduras 500000 2000000 0.21 1
India* 200000000 960000000 0.71 1
Indonesia 21480000 85920000 0.31 1
Iran 7000000 24500000 0.30 1
Jamaica 379000 1174900 0.40 1
Jordan 300000 1092000 0.14 1
Lao 270000 1350000 0.19 0
Lebanon 200000 1000000 0.15 0
Liberia 40000 201414 0.04 0
Madagascar 189400 890180 0.03 0
Malawi 184248 811765 0.04 0
Malaysia 8500000 34000000 0.74 1
21
Maldives 20000 113997 0.22 0
Mali 75000 430898 0.02 0
Mauritania 30000 183000 0.04 0
Morocco 4300000 24811000 0.69 1
Mozambique 1000000 4444670 0.15 0
Myanmar 5400000 22680000 0.42 0
Namibia 576000 2851478 0.80 1
Nigeria 2028416 9330714 0.05 0
Pakistan 9000000 62100000 0.29 1
Panama 317845 1271380 0.30 1
Paraguay 330000 1518000 0.22 1
Peru 3580000 13604000 0.43 1
Philippines 13400000 62980000 0.59 1
Rwanda 100000 430000 0.03 0
Sierra Leone 64000 379125 0.05 0
Somalia 1200000 7080000 0.47 0
South Africa 8000000 29600000 0.51 1
South Sudan 430000 2557628 0.23 0
Sri Lanka 795324 3181296 0.15 1
Thailand 11280000 41736000 0.60 1
Timor-Leste 73328 393071 0.31 1
Togo 567000 3039120 0.39 1
Trinidad 97488 389950 0.26 1
Tunisia 883000 3532000 0.31 0
Turkey 8000000 32000000 0.38 1
Uganda 750000 3600000 0.08 0
Venezuela 5687000 22748000 0.80 0
Vietnam 6000000 24000000 0.25 0
WestBank and Gaza 68000 272000 0.06 0
Zimbabwe 1000000 4100000 0.28 1
Total
389,775,468 1,737,852,062
without India
189,775,468 777,852,062
Total number of countries covered 67
Total number of countries started/finished
by end June 36
Simple average of share of population 0.29
median of share of population 0.26
weighted average of share of population 0.44
weighted average excluding India 0.30
Note: Country table data was corrected on August 28, 2020
22
*India includes an unknown number of pre-crisis SA beneficiaries and is therefore an overestimate of new
beneficiaries.
Sources: Gentilini et al. (2020); UNDP (2020); various.
Notes: 1/ Figures in column 3 refer to cash transfer recipient households where the payment that is targeted to an individual is
assumed to benefit the entire household. This is equivalent to the number of payments that would be made. Column 4 represents
the number of people living in the households and is calculated by multiplying column 3 by the estimated number of household
members. Column 5 divides column 4 by the total population.
23
24
File and source
- File
- world-bank-g2px-scaling-up-social-assistance-payments-covid-19.pdf
- Size
- 3,767,017 bytes
- SHA-256
- 82577f95fa3e3ae1aa09b1c91a4632022e0f7b7ebe39851ca4881c46f3c72947
- Original
- pubdocs.worldbank.org