Social Protection Discussion Paper Series

Social Protection Discussion Paper Series
No. 0315
Social Protection Discussion Paper Series
Benchmarking Government Provision of Social Safety Nets
Timothy Besley, Robin Burgess and Imran Rasul
August 2003
Social Protection Unit
Human Development Network
The World Bank
Social Protection Discussion Papers are not formal publications of the World Bank. They present preliminary and
unpolished results of analysis that are circulated to encourage discussion and comment; citation and the use of such a
paper should take account of its provisional character. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated
organizations or to members of its Board of Executive Directors or the countries they represent.
For free copies of this paper, please contact the Social Protection Advisory Service, The World Bank, 1818 H Street,
N.W., MSN G8-802, Washington, D.C. 20433 USA. Telephone: (202) 458-5267, Fax: (202) 614-0471, E-mail:
[email protected] Or visit the Social Protection website at http://www.worldbank.org/sp.
Social Safety Net Discussion Paper Series
Benchmarking Government Provision of
Social Safety Nets
Timothy Besley, Robin Burgess, and Imran Rasul
August 2003
The findings, interpretations, and conclusions expressed in this paper are entirely those of
the author(s) and should not be attributed in any manner to the World Bank, to its
affiliated organizations or to members of its Board of Executive Directors or the
countries they represent.
Abstract
The question of how much governments should spend on social programs generally, or safety
nets in particular, is of great obvious interest to policymakers but is extremely difficult to
address empirically. The approach in this paper differs from others by assuming that what
governments can potentially do in terms of spending on social programs is given by what
governments across the world are actually observed to be doing on average.
After first briefly reviewing the existing methodologies, their limitations, and what can
be learned, an analysis of 63 countries spending patterns from 1972-1997 is presented using a
comparative benchmarking methodology. Unconditional rankings of spending on safety nets
and other health and education social programs are refined by controlling for various factors
which affect the ability to fund programs. Two sets of factors are examined: (i) structural
features captured by regional dummy variables and characteristics of the underlying
populations; and (ii) quality of government as reflected in measures of corruption, rule of
law, political pressure, and others. Separate analyses are conducted across countries for
selected welfare indicators such as the infant mortality rate and life expectancy at birth and
for states in India, for which additional information is available on macroeconomic factors
and institutional features influencing safety nets spending. The approach generates a picture
as to how states are performing relative to international expenditure norms and may be useful
to policymakers in determining the appropriate level of overall spending.
v
Table of Contents
I. Introduction ................................................................................................................... 1
II. Safety Nets .................................................................................................................... 3
What are Safety Nets and Who are They For? ..............................................................3
The Benefits of Safety Nets ...........................................................................................4
Types of Safety Net .......................................................................................................6
III. Safety Nets and Welfare ............................................................................................ 7
Policy Objectives ...........................................................................................................7
Defining and Measuring Poverty ...................................................................................8
Existing Methodologies .................................................................................................9
IV. Our Contribution...................................................................................................... 15
Summary of Previous Methodologies..........................................................................15
Our Methodology.........................................................................................................15
What Determines the Need for these Programs? .........................................................18
Scope for Institutional Reform.....................................................................................19
Informal Explanation of the Rankings.........................................................................19
Comparisons Across Rankings ....................................................................................22
V. Making Benchmark Comparisons............................................................................ 23
Cross-Country Analysis ...............................................................................................23
Indian State Level Analysis .........................................................................................26
Welfare Outcome Regressions.....................................................................................29
VI. Results........................................................................................................................ 31
Cross-Country Analysis ...............................................................................................31
Indian State-Level Analysis.........................................................................................70
VII. Toward a Constructive Policy Dialogue ............................................................... 81
The Case for Safety Nets .............................................................................................81
Benchmarking ..............................................................................................................82
Constructing the Rankings...........................................................................................83
An Example of Benchmarking.....................................................................................83
Data Appendix................................................................................................................. 85
Cross-Country Data .....................................................................................................85
Indian State-Level Analysis.........................................................................................86
Construction of the Shock Variables ...........................................................................87
References........................................................................................................................ 88
Figures
1: Cross-Country Movements Across Rankings............................................................... 58
vii
Tables
1: Safety Net Expenditures as percentage of GDP ........................................................... 32
2a: Spending On Social Safety Nets (SS/GDP)................................................................ 36
2b: Spending On Social Safety Nets (SS/Total Expenditure)........................................... 38
2c: Spending On Social Safety Nets (SS/Tax Revenues) ................................................. 42
2d: Spending On Social Safety Nets (SS/GDP), by continent.......................................... 46
2e: Spending On Social Safety Nets (SS/Total Expenditure), by continent ..................... 49
2f: Spending On Social Safety Nets (SS/Tax Revenues), by continent............................ 52
3: Cross-Country Safety Net Regressions (standard errors in parentheses) ..................... 59
4a: Cross-Country Welfare Outcome Regressions: Life Expectancy at Birth.................. 63
4b: Cross-Country Welfare Outcome Regressions: Infant Mortality Rate....................... 66
5: Cross-Country Outcome Regressions........................................................................... 69
6: Indian State Safety Net Expenditures (as percentage of state GDP) ............................ 70
7a: Indian State Social Safety Net Expenditures (SS/State production)........................... 73
7b: Indian State Social Safety Net Expenditures (SS/Total state expenditures)............... 75
7c: Indian State Social Safety Net Expenditures (SS/Total state tax revenues) ............... 76
7d: Indian State Social Safety Net Expenditures: Public Food Distribution .................... 79
8: Indian State-Level Outcome Measures......................................................................... 80
9: A Policy Dialogue......................................................................................................... 81
viii
Benchmarking Government Provision of
Social Safety Nets
Timothy Besley, Robin Burgess, and Imran Rasul, Department of Economics,
London School of Economics 1
I. Introduction
The purpose of this paper is to provide some practical guidelines on how much governments
should spend on social safety nets. There are a number of methods put forward in the current
literature for addressing this question. The traditional public economics approach is based
upon calculation of the costs and benefits of each safety net-related government activity.
However, this type of analysis typically requires larger amounts of information than are
typically available, even in countries with reliable household data. More complex approaches
still would emphasize general equilibrium effects and behavioral responses.
Thus while these are the core textbook approaches, in practice it is not clear that they
provide a practical approach in many instances. They may also be difficult for policymakers
to understand, and this can create suspicion about their value.
We could imagine a more tractable approach that measures the objectives of
government from an alternative, normative criterion which may be based upon various
measures of needs—closing the poverty gap, or reducing variability of income to a given
level, or some such thing. The drawback here is we usually don’t have nearly that amount of
money available—we might know how much it costs to provide plausibly delivered targeted
transfers to the whole needy population, just as we might know how much it would cost to
provide a basic package of health services to the whole population. This will inevitably lead
to trade-offs between alternative uses of resources, and we still therefore require some means
of choosing between them.
All of these methods suffer from two additional problems that we shall address in our
methodology. First, they typically do not take account of how effective the government
might be in meeting one need or another. For example, we can expect the quality of
government to affect how efficient safety net resources are in targeting vulnerable groups
rather than being used for rent seeking or other non-productive uses. Secondly, the ability to
extrapolate from studies based on existing methodologies is limited. Thus, we are often
unable to facilitate the comparison of countries to each other, and in particular of economies
1
The authors would like to thank Harold Alderman, Yisgedu Amde, John Blomquist, Kene Ezemenari,
Margaret Grosh, William Jack, Manny Jimenez, Jeni Klugman, Jim Smith and Kalhindi Subbarao for helpful
comments.
.
1
with similar income levels (and levels of institutional development). Such an analysis would
not only be of use in and of itself, but in the context of a constructive policy dialogue with
government officials. Telling policymakers how they perform relative to their economic
neighbors may prove to be an effective means of providing the incentives to increase safety
net expenditures by introducing a form of yardstick competition across economies.
In our approach, we do not assume that governments are optimizing in terms of their
levels of spending on social safety nets. Instead, we relate spending on different types of
safety nets to some benchmark level of performance. This is determined by what countries
are able to do on average for a given set of structural and institutional features of the
economy. This gives us an indication of whether particular countries are spending more or
less than this international norm, and provides a concrete basis for discussing whether
governments ought to be spending more or less on various types of safety net.
In fact, we are able to form three different benchmarks taking different factors into
account. We first examine the share of gross domestic product (GDP) devoted to each type of
safety net, and then rank countries according to this. We have enough variation in this data to
suggest that policymakers cannot all be optimizing. We then try to account for this variation
through certain structural characteristics of the economy, such as the level and distribution of
income, the level of urbanization, etc. We then form a cross-country ranking where the
benchmark is what countries are able to do on average, controlling for their structural
characteristics. Finally, the third benchmark also takes into account the institutional features
of the country, such as the levels of corruption, bureaucracy, and the rule of law. We can thus
decompose the policy advice we can offer to governments into two forms—first, the effects
of changes in the structural characteristics of the economy, and second, the effects of
changing institutional features. We set out the benefits of altering both types of policy in
terms of safety net spending, and what the resulting change would be in any given country’s
performance.
This approach is useful as it will generate a clear picture as to whether particular
countries or states are over- or underperforming relative to international, regional, or national
norms. In fact, our methodology allows us to present simple information to policymakers in
the form of country rankings. This, in turn, enables us to make statements of the form
“country i is spending less on safety nets than we would expect given its ability to finance
such expenditures and the need for safety nets in the country, relative to the international
norm.” We can make similar statements that also take into account the institutional quality of
the country. Furthermore, we can benchmark country i’s performance relative to its
neighbors, which may be perceived by policymakers to have similar structural features. This
information, coupled with the arguments for investment in safety nets that are also outlined
in the paper, can be used during dialogue with policymakers on the appropriate level of
spending on safety nets.
Benchmarks of this type are useful as they can generate incentives to improve
performance to conform with international and regional norms. Pressure to do this can come
both from the international community and from the domestic dialogue which is generated by
the publication of such information. This has often been the case in such areas as labor and
environmental standards, and it seem sensible that this process of “yardstick competition” be
extended to cover the case of social protection.
2
The value of this benchmarking exercise is contingent on data being reflective of
government activities in the safety net area and being comparable across countries. On both
these counts, available data sets are somewhat lacking. This does not detract from the validity
of our methodology but rather points out that gathering more accurate and comparable data
on what is actually being spent on safety nets should be a priority in terms of thinking about
what the appropriate levels of spending should be.
The remainder of the paper is organized as follows. Section two begins by defining
safety nets, and here we emphasize a functional or objectives-based definition, rather than
working with program-based definitions of safety nets. We think of safety nets as playing
both redistributive and risk-reducing roles in the economy. We then detail who ought to be
the recipients of safety net expenditures, and the benefits to the poor and the non-poor that
such expenditures confer, before giving examples of which types of program we may think
of as forming part of a safety net. Section three discusses how the existing methodologies
address the question of how much governments ought to spend on safety nets, the limitations
of each approach, and what we may learn from them.
Section four then argues from first principles what determines the need for such
programs, before providing the intuition behind our methodology and presentation of results.
Section five formally presents our methodology of benchmarking country performance, both
in terms of safety net expenditures, but also for welfare outcomes such as life expectancy and
infant mortality rates. We do this both across countries and at the level of Indian states where
one can more effectively control for common macroeconomic factors and institutional
features. Section six then goes through each of these sets of results in turn.
Section seven frames our discussion in the context of a constructive policy dialogue. First,
it presents a clear way to argue the case for safety net expenditures and to get policymakers to
not only think of such programs in terms of the costs of provision, but also in terms of their
benefits. Secondly, we summarize how the results may be used in such a dialogue, and what
policy implications follow. We show how our rankings can be used to justify increased safety
net expenditures or policy reforms, through yardstick competition.
II. Safety Nets
What are Safety Nets and Who are They For?
There is some debate over what constitutes a safety net. This is partly due to the fact that
such programs have only existed for a half century, beginning in Western Europe and now
gradually being implemented in some form in most developing economies.
Authors such as Atkinson (1995) and Subbarao (1997) argue that the purpose of safety
nets is to alleviate chronic and transient poverty. They identify the mechanisms that help to
mitigate these adverse outcomes as being either private safety nets, which are informally
organized or community-based, and public interventions covering health, education, social
insurance, and publicly-funded transfers such as food subsidies. Alternative views of safety
nets are, for example, those of Barr (1994), where their role is seen in a broader social
context, to not only increase consumption per capita, but also to have a redistributive
function and create political stability. Holzmann and Jorgensen (1999) argue that public
3
interventions assist in better managing income risks, as well as contributing to social
cohesion.
We define safety nets as those public interventions which are designed to serve two key
functions: (i) to play a redistributive role transferring resources toward the poorer members
of society to bring them out of poverty, and (ii) to provide greater opportunities for
individuals to mitigate risks from unforeseen contingencies. Such risks can operate at the
level of the household, say through an unexpected death or unemployment of the household
head, but also at the community or national level due to natural disasters, financial crisis, and
terms-of-trade deteriorations. The correct balance between the redistributive and riskreduction roles of safety nets will ultimately depend upon country-specific factors.
Safety nets therefore do not only protect individuals from transient periods of poverty,
say due to loss of employment, sudden illness, or natural disasters, but also serve to protect
individuals from lifetime poverty that can arise from, say, lack of education and poor health,
particularly in childhood. Hence, when considering how much governments should be
spending on such programs, the long-run intertemporal, intergenerational, and wider social
benefits should all be taken into account as well as considering the short-run alleviation of
poverty.
In all countries, we observe three types of vulnerable individual. Different safety nets
will be able to assist these different groups. First, there are the chronically poor whose
income levels remains below an acceptable minimum, typically set through a poverty line.
These individuals remain in such a state even during periods of economic growth and in the
absence of microeconomic and macroeconomic shocks. The second vulnerable group
consists of the temporarily poor whose income levels fluctuate above and below the poverty
line in times of shocks. Thirdly, we also recognize the existence of those individuals who fall
into poverty in phases of macroeconomic adjustment, for example, people who lost their jobs
during the transition of Eastern European countries toward a market economy, e.g., due to
privatization or the bankruptcy of government-owned enterprises.
However, although safety nets should be targeted toward each of these groups, it should
not be thought that the non-poor do not benefit from such social expenditures. We will
discuss how the non-poor also benefit from such programs in the next section.
The Benefits of Safety Nets
Following on our definition of what types of social assistance constitute safety nets, we can
now be clear about the exact benefits of such expenditures:
•
Redistribution: safety nets aid in transferring resources to the poor, and thus in
protecting them from poverty in both the short- and the long-run. It can be argued
that this raises the welfare of both the poor and the non-poor if the society is averse
to inequality. The programs can be targeted at the individual, household, or
community level to raise the well-being of the poor to levels above the minimum
standards that are accepted nationally and internationally. (Albania, for example,
has a program of social assistance that operates through community-level
institutions). Redistribution need not be at the expense of growth, and a growing
4
•
•
•
body of literature identifies cases where redistributive policies have led to growth
enhancements (see Benabou 1996).
Economic efficiency: this can then be improved through each of the following
mechanisms:
(i) We know that if the allocation of resources in a sector characterized by
imperfect information, missing markets (especially in insurance), public
goods, or externalities, is left to market forces, then the equilibrium
allocation will be non-optimal due to these sources of market failure. Safety
net expenditures often serve to correct these types of market failure (see
Benabou 1996).
(ii) The poor will not have to resort to using short-run coping strategies, such as
selling their assets in times of crisis, and so will be left better off in the longrun.
(iii) By becoming less vulnerable to income shocks, the poor will also be able to
invest in their human and physical capital, namely they will be more willing
to spend time and resources on education and machinery, for example.
(iv) There will be reduced incentives for individuals to enter into marginalized
economic activities such as working in the black economy or informal
sectors, or engaging in criminal activities. This should not only save
government resources in preventing such activities, but may generate higher
tax revenues which can then be ploughed back into social assistance.
Political economy: social assistance to the poor may empower them to engage in
the policymaking process at both a local and national level. This may well reduce
the probability of socially inefficient political decisions being made solely for the
benefit of elites or certain special interest groups, or other types of political failure.
The engagement of the poor in the policymaking process can have self-enforcing
effects in that, if a government demonstrates a commitment to reducing poverty,
incentives increase for political organizing by the poor—who, as we have argued,
are a heterogeneous group. The poor are therefore in a better position to place
policies designed specifically for themselves onto the political agenda. This may
have the effect of raising support for the government in the long run.
Social cohesion: safety nets can play an important role in providing for social
cohesion in a number of ways:
(i) At times of macroeconomic crisis or adjustment processes, hard economic
decisions often have to be made. By raising social cohesion, safety nets may
raise the political acceptability of market-based reforms that often need to be
made in the aftermath of economic crisis to enable a country to reach a path
of sustainable growth. They also demonstrate a government’s commitment
to social welfare issues. Venezuela, for example, introduced a package of 14
programs to accompany its policy of structural adjustment.
(ii) If society is averse to having unequal opportunities available to individuals
or income inequality per se, then safety nets can improve social cohesion.
Empirical support is growing for this view of safety nets as a form of social cohesion.
For instance, Sala-i-Martin (1997), using cross-country data, shows that public transfers have
a positive correlation with growth and may therefore also be a productive input into national
output. The mechanism by which public transfers affect output is argued to be that of social
cohesion—increased transfers reduce social unrest, which enhances the conditions for growth
and more than offsets any negative effects of the distortionary taxation required to fund the
5
transfers in the first place. Rodrik (1998) provides cross-country evidence on the efficiency
gains to be had from government provision of social services, and in particular their
provision through effective government-level institutions. His empirical evidence appears to
suggest that those economies with good institutional frameworks are most able to deal with
external shocks in the long run. Economies with weak institutions tend to delay price and
fiscal adjustments in response to shocks, thus diverting resources away from productive and
entrepreneurial activities, and increasing uncertainty in the economic environment. In
essence, high-quality institutions can be seen as one mechanism to deal with internal
conflicts over resources when economies are hit by external shocks.
Types of Safety Net
We can now be precise about the types of programs that make up safety nets. These can take
the form of cash or income transfers, such as pensions, child allowances, unemployment
benefits, or microfinance, or they can be transfers in kind of commodities such as food
subsidies, housing subsidies, or energy subsidies. They may provide income indirectly by
offering vulnerable groups employment in public works programs or more broadly, by
providing services such as health and education. Given our earlier definition of safety net
programs as interventions that are designed to play a redistributive role and to provide greater
opportunities for individuals to mitigate risks from unforeseen contingencies, we focus
narrowly on the following types of policies.
Cash Transfers
The two most common types of cash transfers are forms of social assistance, targeted to
vulnerable groups in society such as the unemployed, children, the disabled, or pensioners,
and forms of financial assistance to families. Often these are based upon the number of
children living in the household.
The World Bank and others have carried out a number of studies of such programs
including family allowances in Hungary, Russia and the Kyrgyz Republic; pensions in India;
unemployment benefits in Jordan (the National Assistance Fund); the Janasaviya Program in
Sri Lanka, which pays for two years of basic training for targeted household heads; and a
Namibian scheme of transfers to children of AIDS-infected parents.
In-Kind Transfers
These can involve transfers of commodities, such as rice or kerosene. The advantage of such
transfers is that they are less susceptible than cash transfers to periods of high inflation, when
the value of the latter can be quickly eroded. By their nature, in-kind transfers are less
fungible than cash transfers, so they are often argued to be a more cost-effective means by
which to raise welfare if they are correctly targeted.
Several types of in-kind transfers have been implemented—general price subsidies
(predominantly in African and Middle-Eastern countries), quantity rationing (South Asia),
food stamps (Latin America), and nutritional interventions such as direct food transfers,
which are prevalent everywhere. In addition, there is also extensive use of housing subsidies
throughout Eastern Europe.
Public Works
These programs are often implemented only during a downturn in the economic cycle. As
well as providing employment for the poor, the programs also serve to build a nation’s
6
infrastructure, an essential component of any development policy. It is important to keep the
costs of participating in such programs low, e.g., by minimizing traveling distances, in order
for them to be effective in reaching the poor. Hence there are non-poor gainers from such
programs. Examples of such schemes include the public employment schemes in Argentina
(the Trabajar Program), Bolivia (Emergency Social Fund), Chile, China, and India.
Informal Safety Nets
In many societies we also tend to observe the existence of informal networks of support,
based perhaps on kinship or community ties, that also seek to mitigate against income
shocks. For example, in many Islamic countries such as Pakistan, the informal institution of
zakat acts as a tax on wealth, collected by mosques and redistributed to the poor. In many
Sub-Saharan countries and in India, there is a system of labor transfers within communities.
Finally, in China, there are structural features of the rural economy, such as universal and
egalitarian access to land, which help to insure individuals against adverse outcomes.
A key issue for policymakers is whether such private transfers are crowded out by
publicly provided social expenditures and, if so, to what extent this crowding out occurs.
There is a large body of literature that deals with exactly this issue and this will be discussed
in more detail later in this paper.
One of the main factors determining the effectiveness of safety nets is their ability to
correctly target the poor. Targeting can be based either on self-reports from individuals
(where incentives must be provided for individuals to truthfully report their well-being) or on
measured household characteristics or regional characteristics.
The other crucial issue regarding the ability of these programs to effectively reach and
be able to help the poor, is the manner in which they are implemented. For effective
implementation we require a supportive institutional framework, i.e., one that is not subject
to corruption or rent-seeking, or that is not plagued by bureaucracy, and where the rule of law
is respected. It is one of the key objectives of the empirical analysis in this paper to examine
how well governments perform in the provision of safety nets, relative to international
benchmarks when institutional quality is explicitly taken into account.
III. Safety Nets and Welfare
Policy Objectives
To be able to address the question of how much countries should spend on safety nets, we
require some objective function to be evaluated. There are two principle ways in which this
can be done. First, we can take our basis from economic theory, which suggests that, when
the marginal benefits of different types of social expenditure are equal to the marginal costs
of raising public funds, an efficient outcome is reached. However, the data requirements for
such an analysis are unreasonable, especially on the benefits side.
An alternative way in which an objective may be defined is in terms of outcomes
directly, such as reducing the poverty gap by x%, reducing income variability, reducing the
percentage of the population affected by natural disasters or communicable diseases,
increasing participation rates in the labor force, or reducing the incidence of child labor.
However, it is still the case that in order to assess whether such a policy is feasible, cost
7
considerations still come into play. Such policy targets may be unattainable given resource
constraints that are not taken into account explicitly.
A third method to address the question of how much governments ought to spend is to
adopt some measure of where a safety net should be set (effectively setting the poverty line),
such as a dollar per day or half the median income, and then see whether social expenditures
do indeed reach those that they should, or identify which subgroups of the poor are most
effectively targeted. Unlike the previous two approaches, there is no need to consider the
costs of provision here as it is taken as given that what governments are observed to be doing
is actually feasible.
Defining and Measuring Poverty
All of these three methods presuppose that poverty can be measured, but this is by no means
a straightforward issue either. The first thing to decide upon is the metric we believe best
approximates welfare, in order to be able to measure poverty. The alternatives available
include income, needs, and capabilities. To take income as a measure of well-being is to
focus on the commodity basis of well-being, but often our intuition suggests that claims over
commodities are not the only factors that contribute to well-being. Uncertainty regarding
one’s economic environment, the set of available opportunities, non-market sources of
welfare, and one’s biological and physical status are not easily captured in the framework of
commodity possession. Furthermore, as has been pointed out by Dreze and Sen (1990) it is
typically at times of economic crisis such as drought or floods that the mapping between an
individual’s income and entitlements to market-produced goods becomes most unclear.
An alternative route is to measure well-being in terms of an individual’s needs. The
issue then becomes what set of needs constitute an individual’s basic needs, which if not met
would imply the individual was poor. Even if a consensus can be reached on this bundle of
basic needs, in practice this approach often tends to revert to converting such a bundle into
the equivalent amount of income required to purchase it.
Using the capabilities of an individual, i.e., what an individual is able to do and be, was
first introduced by Sen (1985). The aim is to be able to incorporate non-market sources of
welfare such as available opportunities, political empowerment, and so forth. The notion of
capabilities underpins the Human Development Index, and has thus been used to provide the
basis of international comparisons in well-being. Measures such as the prevalence of diseases
(such as AIDS or malaria), or hours worked by children, may be particular measures of
capabilities that can be employed.
Ultimately, however, until micro-data is collected on capabilities, it appears as if some
income-related measures, such as consumption or expenditures, will be used in studies that
address the question of how much governments ought to spend on safety nets. The case in
favor of such income-related measures is not only made on the grounds of data availability, but
also the strong evidence that exists that many other welfare outcomes, such as mortality,
nutritional status, and life expectancy at birth, are all highly correlated with income levels.
Hence, income, while far from being a perfect measure of well-being, captures at least to some
degree some of the wider notions of well-being that these alternative metrics capture. Work is
ongoing by the World Bank and others to construct indicators of poverty along the dimensions
of risk and vulnerability, social exclusion, and access to social capital, which may all prove in
8
the long run to be preferable measures of welfare when addressing the particular question of
safety net spending.
Given that we have established some measurement of poverty, we still have to
determine where the poverty line must be set. The literature on this is huge (see Lipton and
Ravallion (1995) or Ravallion (1998) for good reviews), but the main issues are whether the
poverty line should be defined in absolute or relative terms and if, or how, to take account of
the degree of inequality and heterogeneity amongst the poor. Armed with a poverty line, we
are now in a position both to motivate policy and focus the attention of policymakers on the
plight of the poor, and to address the effectiveness and distribution of benefits arising from
safety net expenditures for a given objective.
Existing Methodologies
There is a large body of literature, discussed below, which attempts to assess whether
governments spend the desired amount by assessing the welfare impacts of social
expenditures and also whether such policies are efficiently targeted. The principal methods
by which to assess the welfare impact of social expenditures are benefit incidence studies,
behavioral approaches, and computable general equilibrium (CGE) models. See Selden and
Wasylenko (1992), van de Walle and Neads (1995) and van de Walle (1998) for surveys of
the literature, and Hammer (1997) for an alternative discussion of the issues. Data limitations
have meant that these studies have been largely focused upon developed Central and South
American economies.
Benefit Incidence Studies
Benefit incidence studies are the benchmark public economics cost-benefit approach to
evaluating government interventions. These studies tend to focus on a particular type of
expenditure, rather than public expenditures in general. They proceed by first grouping
households or individuals according to some indicator of living standards. In order to make
valid international and intertemporal distributional comparisons, this welfare indicator may
be adjusted to take account of variations in the cost of living, say between rural and urban
regions, and household demographics through some equivalence scale such as those
discussed by Deaton and Muellbauer (1986) and Browning (1992). Earlier studies such as
Meerman (1979) and Selowsky (1979) used household size as an equivalence scale. Later
studies used equivalence scales but Jarvis and Micklewright (1995) argue that the results of
benefit incidence studies are often found not to be robust to the equivalence scale used.
The method then assumes that the benefit which accrues to households can be proxied
by the value of government expenditures. Having obtained a poverty profile, we can use
observations on the number of actual beneficiaries to form a distribution of social
expenditure across the welfare groups, which is then taken to be an indicator of the benefit
incidence. The method then allows us to classify a program as either being progressive or
regressive and to examine the actual effectiveness of targeting policy.
Benefit incidence studies have been employed for the past two decades, beginning with
the papers by Reynolds and Smolensky (1977), Meerman (1979), Selowsky (1979), LeGrand
(1982), and Gruske (1985). More recently, benefit incidence studies have been conducted by:
Bahl, Kim, and Park (1986) in Korea: Riboud (1990) in Costa Rica; Ravallion, van de Walle,
and Gautman (1995), who look at the distributional impacts of cash benefits introduced to
9
compensate for policy reforms in Hungary; the collection of papers in van de Walle and
Nead (1995) such as Alderman and others on education in Pakistan, Deolalikar on the
impacts of health expenditures on children in different income groups in Indonesia, and
Selden and Wasylenko on educational expenditures in Peru; Prescott (1997), who examines
the efficiency of targeting of education, health, and social transfers in Vietnam; and Hanmer
and others (1998), who examine health expenditures in Zimbabwe.
Such studies are methodologically straightforward to implement and can provide
information on how the benefits of public interventions are distributed across the poor, but
give no clear indication of whether the efficiency criteria on levels of expenditure are
satisfied. In practical terms, the data requirements can be severe, especially to construct
welfare rankings. However, this method is also subject to a number of deeper criticisms.
Most of these apply more broadly to most policy studies, not just those evaluating the
benefits of safety nets.
The fact that such studies take the benefits that accrue to an individual to be well
proxied by the average cost of provision to that individual can also be called into question.
For example, the unit cost of immunization can be considered to be small relative to the
lifetime benefits. The issue is complicated both because well-being is multi-dimensional and
because the estimation of outcomes in the counterfactual world without public spending is
not straightforward. More specifically, this would require us to calculate the extent of the
crowding out of private and informal transfers by public transfers. Such data are typically not
available.
Benefit incidence studies implicitly assume that there is a uniform cost of service
provision over all households, or that the public good is homogeneous. To the extent that this
is not the case, they may lead to incorrect inferences regarding the distribution of benefits of
social expenditures.
The method, being a partial equilibrium analysis, implicitly assumes that relative prices
and real incomes do not change, and that marginal benefits are equal to average benefits.
However, there may be a divergence between average and marginal benefits. For example, if
there are increasing returns to scale from public expenditures, which may be the case for
infrastructural investments such as roads or electrification, then the marginal benefit is likely
to be greater than the average, as such public goods are provided to more households. A
characteristic that this methodology has in common with behavioral approaches is that it
assumes the geographical distribution of the population to be static. In the case of developing
countries, regions well-endowed with public services can induce population inflows or cause
wage differentials to arise which lead to worker migration (see Todaro 1969, Williamson
1988).
In the context of developing countries, benefit incidence studies have to take account of
the possibility of resale of public goods, especially due to the presence of a large informal
sector. Moreover, theory suggests that we can expect institutional structures such as
interlinked factor markets, informal labor markets, to mean that the recipients of public
services pass on any actual welfare benefits to moneylenders or landlords. We may expect
this to be particularly the case in agricultural programs designed to raise farm incomes.
10
Benefit incidence studies typically take no account of behavioral responses by
households to the introduction of public programs. Theory suggests that households will
change their behavior, for example with regard to labor supply (females in particular may be
able to devote more time to home production activities such as child care if their spouses are
able to devote more time to the labor market), investment (as households become better off
they may be more able to invest in education, health services, and fixed productive assets),
consumption (households may be able to transfer budgets toward more nutritious foods), and
private transfers of resources or time either within or across households. Empirical evidence
is found in favor of such crowding out by Barro (1974), Andreoni (1990), Jimenez and Cox
(1992) and Cox and Jimenez (1995), although such estimates do not suggest full crowding
out. For example, Jimenez and Cox (1992) find that social security payments in Peru reduced
private transfers from young to old by 20 percent.
Another limitation of this approach is that the use of cross-sectional household data only
allows for the identification of static effects. This means that we do not capture various other
benefits of safety net expenditure. For example, policies that are designed to alleviate chronic
poverty may well have lifetime benefits for the individual. There may also be spillover social
effects on other individuals in the household or between households in a community. Finally,
there may also be intergenerational benefits of social expenditures. If, for example, parents
are provided with employment, they may be more willing to educate their children, which we
would expect to raise the lifetime earnings of the children. In short, not only are short-term
behavioral responses ignored, but also lifetime, social, and intergenerational effects.
In addition, many of the forms of intervention that safety nets take are responding to the
existence of some form of market failure, such as the presence of externalities or public
goods. These elements are usually not captured in benefit incidence studies.
Furthermore, most of the available data are at the household level, yet ultimately we are
concerned with the effects on individual welfare. This requires us to make assumptions about
intrahousehold allocation mechanisms. The most common assumption made in the literature
is simply to take per capita (or some other equivalization) consumption levels. However,
there is much evidence that we may not have such equitable intrahousehold distributions,
especially based upon gender or age (see Haddad and Kanbur 1990, 1993, Deaton, Parikh,
and Subramanian 1994, and Deaton and Paxson 1996). On the other hand, we can also argue
that, if households are not credit-constrained, then consumption will track permanent income,
and so by using consumption as a welfare measure, we are in fact capturing how households
react to dynamic and stochastic income shocks.
Such analysis does not establish the underlying mechanisms through which individuals
respond to social expenditures. In essence, it is only the demand and supply of social
expenditure, across welfare groups, that is identified, but such partial analysis does not allow
us to recover equilibrating prices. Nor can we calculate the marginal incidence of policies. It
is the average incidence which is identified, and this may hide much of the interesting
information about the size of actual benefits across different welfare groups.
Notwithstanding such criticisms, we can still draw some broad conclusions from this
literature. First, most studies find that expenditures on health, education, social transfers, and
food subsidies are progressive inasmuch as they are higher for the poor as a fraction of their
initial income or expenditure. However, it is generally concluded that the absolute benefits
11
tend to also increase with household income. It is also typically found, for those studies that
make such a distinction, that benefits in urban regions are relatively greater than those in
rural locations. However, most studies also highlight the need to disaggregate expenditures as
much as possible. For instance, the progressivity of primary education expenditures is far
greater than that of secondary education expenditures in most cases. The same issues arise
when looking at different types of health service.
Behavioral Approaches
The second class of studies consists of behavioral approaches. These studies, while still
operating in a partial equilibrium framework, do take explicit account of behavioral
responses, and they also estimate the marginal and not the average incidence. The general
methodology is to devise a means by which to evaluate the recipient’s own valuation of the
benefits received. In early studies, this valuation was proxied by the consumer surplus the
individual obtained. It has long been realized, however, that this ignores the income effects
of relative price changes. If preferences are known or can be inferred, then a compensated
demand curve, along which utility is held constant, can be used to calculate the underlying
utility function of consumers (see McKenzie 1983) on which measures of welfare benefits
can be based, such as the real income per adult equivalent, and equivalent and compensating
variations. This is precisely what later studies have done.
There are really two main issues concerning this approach. The first is how to obtain
consistent estimates of estimated parameters from an econometric model and be sure that
such estimated parameters actually correspond to the underlying structural parameters of the
economy. The problems associated with this are again not unique to the analysis of safety
nets, but apply to the evaluation of government policy in general. The second issue is the
same as it was for benefit incidence studies, namely how to obtain some measures of benefits
to undertake welfare analysis. This second factor has been discussed before, so for the
remainder of this section we will concentrate on the first issue of recovering consistent
parameter estimates.
It is well known that problems can arise in identifying consistent preferences if, for
example, behavior does not accord with the underlying assumptions of utility theory. Also,
the very fact that no markets exist for public goods makes the identification of the utility
derived from their consumption problematic. In response to these issues, a literature on
identification of the willingness to pay has been established which specifically studies the
demand for public goods (see, for example, Gertler and others 1987, 1989, 1990). These
studies allow calculation of willingness to pay across income, or other, subgroups. Hence, it
is possible to examine, for example, whether the poor gain more on the margin than the rich
from a given type of social spending. They deal with the issue of missing markets by
proxying prices by the totality of monetary and non-monetary costs of public provision.
Another strand of the behavioral response literature uses non-monetary welfare metrics
such as nutritional status, mortality, or literacy rates to assess the benefits of public
expenditures. In practical terms, they do this by assessing the impact on such outcome
measures of a set of inputs including socioeconomic background, income, prices, and public
expenditures and complementary services. Examples of this approach are Deolalikar (1995)
on health expenditures in Indonesia and Alderman and others (1995) on public schooling in
Pakistan.
12
The main issue to be dealt with in such approaches is how to recover unbiased estimates
of policy effects. The problem is that, by using policy variables as explanatory variables, we
can typically expect these variables to be correlated with the error term, thus leading to
biased estimates for ordinary least squares regressions. This correlation can arise from
simultaneity, omitted variables, selection, or heterogeneity. We shall briefly discuss each of
these in turn.
If policy is targeted using the same welfare indicator as the dependent variable, or
another indicator highly correlated with this one, then policy is actually simultaneously
determined with the distribution of welfare. For example, Besley and Case (1994) suggest
using political variables that influence policy outcomes but are uncorrelated with welfare
levels as an identification strategy. The policy itself is endogenously determined and its
inclusion as an explanatory variable thus leads to standard ordinary least squares (OLS)
endogeneity bias.
Pitt, Rosenzweig, and Gibbons (1995) look at the impact of government placement
programs in Indonesia while explicitly controlling for the non-random allocation of this
policy. They do this by examining the changes in outcomes over time in a given region
(before and after the introduction of the policy). Such a fixed-effects procedure eliminates
any unobservable program placement effects under the identifying assumption that the
region-specific and time-varying shocks that affect program placement are uncorrelated with
region-specific and time-varying shocks in the policy outcome equations. They find the
simultaneity bias to be large enough to reverse the policy conclusions.
Alternatively, there may be some omitted variable that determines both policy incidence
and welfare levels. An example of this may be policies introduced to locations in close
proximity to urban centers, where welfare indicators are higher per se, e.g., due to a higher
level of community assets, even before the implementation of policy.
Similar biases arise using OLS if there is a selection rule operating for those who
receive the policy treatment so that the policy recipients are not a random sample of the
population, or if there is some unobservable characteristic of individuals that influences
whether they are subject to the policy, e.g., if only more able or well-motivated individuals
seek to receive the policy treatment. Typically, the researcher can employ a Heckman
procedure to introduce an additional selection variable in the equation of interest, which
accounts for the potential correlation between the error term and the other covariates in the
equation.
In short, when using cross-sectional data, it is very difficult to separate the effects of
policy on welfare from the effects of all other observables and unobservables. The issues of
bias can be partly ameliorated using fixed-effects estimation in panel data, using instruments
for policy incidence, or using natural experiments when individuals randomly receive the
policy treatment. The use of panel data in theory allows us also to determine both dynamic
and behavioral effects, but this is easier to say than actually implement. There is now a
growing literature exploiting such data sets, such as van de Walle, Ravallion, and Gautman
(1994) and Ravallion, van de Walle, and Gautman (1995) on safety nets in Hungary.
There is also a growing number of studies that seek to exploit natural experiments. Here,
the control and comparison groups occur naturally, and if we can observe all individuals
13
before and after the policy intervention, then using a difference-in-difference estimating
procedure, the researcher can recover consistent estimates of policy effects. The underlying
assumptions required for this method to work are that the disturbance term is additive in
fixed and time effects, that there are common time effects across all individuals, and that
there are no compositional changes in either group over time. The best known of such studies
is that of Card and Krueger (1994) on the introduction of minimum wage legislation in the
United States, and in the United Kingdom there are papers on the welfare-to-work program
that use the natural experiments framework.
Despite having to deal with such a range of potential econometric difficulties, the
literature on behavioral responses has shed light on a number of issues. A robust result from
the majority of studies appears to indicate that demand for public services is price-inelastic
and that there is much variation in this elasticity over income groups, with the poor being
more price-sensitive, as we would expect. In a sense, these results shed little light directly on
the question of how much governments ought to spend. The main conclusions have been
mostly related to identifying the beneficiaries of safety net expenditures, and examining
whether safety net expenditures have been effectively targeted toward the poor.
Computable General Equilibrium Models
Unlike the previously described methods, a general equilibrium model of social expenditure
does not attempt to estimate parameters of the economy. Rather, computations are based
upon given parameter values, and then the researcher attempts to discover how predicted
outcomes change in response to these imputed parameter values.
The key advantage of such a modeling approach is that the researcher can specify a
complete model of behavioral responses to social transfers, and thus incorporate the
crowding-out effects of safety nets, as well as recovering equilibrium prices and so forth. The
issue remains the reliability of the imputed parameter estimates, and whether robust
conclusions can be drawn.
There are relatively few studies that have attempted such an analysis, one being Piggott
and Whalley (1987). They are forced to make a number of simplifying assumptions, such as
efficient provision of public services, in order to be able to make their model tractable. Their
aim is to compare the two approaches above, hence they explicitly take into account how
consumer surplus and the welfare costs of taxation change with increased social
expenditures, and they also calculate the average and marginal welfare gains of public
spending. Their results vary depending upon the imputed values of certain key parameters,
which probably implies that more studies of the former types should be conducted to better
empirically estimate such parameters. Other CGE papers include those of Hertel (1989) and
Parikh and Srinivasan (1989) on agricultural policy.
When asking how much governments ought to spend, this type of study really requires
precisely estimated parameters to have been estimated in prior studies. It is hard to
extrapolate general implications from each of these studies. Given all of the econometric
problems to be dealt with in order to be able to do this, it seems as if there is a long way to go
before we can confidently use such a class of models to form the basis of policy interventions
across countries.
14
IV. Our Contribution
Summary of Previous Methodologies
Ideally, in order to be able to answer the question of how much governments ought to
optimally spend on safety nets, we would have the data available to estimate whether the
marginal benefits of different types of social expenditure are equal to the marginal cost of
raising public funds, thus ensuring resources are efficiently employed. In addition, we should
be able to define poverty and measure welfare in ways that approximate the true well-being
of individuals. However, as we have argued, it is unlikely to be the case that all of the
conditions are adequately satisfied. Hence, we have a number of second best approaches
which may be able to shed light on the same question. Before we detail our alternative
methodological approach, let us summarize the previous discussion on how we may address
the question of how much governments ought to spend on safety nets.
The first approach of benefit incidence studies was seen to rely on the availability of
large amounts of information, and was a partial equilibrium approach, ignoring the potential
crowding-out effects on private transfers, or market failures in general.
The behavioral approach attempts to determine the underlying demand function for
safety net expenditures. The issues here relate to, first, how well such analysis controls for
potential sources of econometric bias arising from omitted variables, the simultaneous
determination of relevant outcomes and program placement, the non-random selectivity of
individuals into a program, unobservable individual heterogeneity, and so forth. Secondly,
we have concerns regarding how well estimated parameters map back to structural
parameters of the economy in order for us to be able to extrapolate our experiences across
programs and countries.
Both of these approaches have been closer to addressing the questions of: (i) whether
the poor are effectively targeted by safety net programs, and (ii) what the distribution of
benefits of safety net expenditures is. We can relate such questions to the central question
here of how much governments ought to spend by noting that, first, if the poor are not
effectively being targeted then that would suggest that resources are being used nonoptimally in that the marginal benefits accruing to the non-poor would typically be less than
those that would accrue to the poor. The policy lesson to be drawn from this is that
governments should either cut back such expenditures or retarget them toward the poor using
some alternative targeting mechanism. Secondly, if it appears that there is wide dispersion of
marginal benefits across the population of the poor, this again suggests a non-optimal level of
spending. A necessary condition for an efficient outcome to have been reached would be the
equalization of marginal benefits across the poor.
Our Methodology
The methodology we employ to address what level of social spending governments ought to
attain is quite different from those previously discussed. We resolve the difficulties of
defining an objective function by taking the view that what governments can potentially do in
terms of spending on various types of social safety net expenditures, as well as the total level
of government expenditures, is given by what governments across the world are actually
observed to be doing on average. We thus take as given that what countries are observed to
15
be doing is an indication of what it is feasible to do. It is this underlying notion of
benchmarking that both serves as the basis of our analysis and provides a useful framework
to convince governments to put more effort into such policies.
In essence, our aim is to be able to obtain some measure correlated with a government’s
effort into effective spending on safety nets, relative to what other governments are seen to
be doing. Suppose that the outcome that we observe on any policy for country s at time t, xst,
can be written as
xst = β est + γ yst + η t + ε st
where est is some measure of government effort, yst is a vector of characteristics affecting the
ability of the economy to produce these outcomes, ηt is a common “shock,” which all the
economies in question experience and εst is an idiosyncratic shock uncorrelated with
everything else. In general, data constraints mean that we can get only an imperfect set of
controls (yst). Our methodology is to use these data to obtain information on the unobservable
level of government effort (est).
Suppose that government effort put into delivering policy is dependent on the costs and
the payoffs that it faces for doing so. Let the payoff be denoted by r (e; i), where i is some
measure of the information available to those designing the reward structure. This payoff can
be interpreted by politicians as the value of holding office. The interesting case is where
improved information raises the marginal benefit of putting in effort. However, this is by no
means inevitable.
The idea of designing meaningful benchmarking is to lead governments to increase the
effort that they put into delivering outcomes. We propose using rank order information as a
measure of government performance. This is useful provided that information about the rank
conveys information about unobserved effort on the basis of which rewards can be designed.
The conditions under which this will happen are that
•
•
•
η t ≠ 0.
the variance in εst is small enough.
∂ 2r
≥ 0.
∂ e∂ i
Our methodology produces rankings that allow us to benchmark country performance.
Our first ranking is based upon the unconditional policy outcome, xst, namely the level of
safety net spending. We then construct rankings conditional on observable features of the
economy, and these will be based upon the information contained in the conditional level of
safety net spending, xst − γˆy st . We do this controlling for structural features of the economy
that determine the ability of the country to finance such expenditures, the need for such
expenditures, and the quality of the country’s institutions.
Each ranking gives us a series of benchmarks to which countries’ relative performance
can be compared, as well as to their neighbors or economies at a similar stage of
16
development. In a sense, we are not asking how much governments should spend; rather, our
methodological approach is to form an impression of the benchmark performance of
governments. We thus avoid many of the data requirements in tackling the question of how
much it is optimal for governments to spend. We also avoid the need to form an indicator of
welfare to assess the benefits of safety nets, because we look at how much governments
actually are spending relative to some international norm. One way to think about our
approach (moving from the unconditional ranking xst to the conditional ranking where we
control for features of the economy, xst − γˆy st ) is to think of it as an attempt to estimate a
demand function for social expenditures and then examine its properties. This equation
embodies some features of the underlying decision process of policymakers.
The first thing to note is that there is wide variation in the levels of expenditures on
safety nets across countries. It is hard to justify these variations simply on the basis that some
governments are better informed about the optimal level of expenditures than others. Rather,
these differences across countries will clearly be related to the levels of poverty and needs for
these social expenditures, to the ability of governments to meet these needs subject to
resource constraints, but also to underlying differences in preferences and objective functions
of policymakers across societies. Furthermore, these differences may be reflective of the
varying institutional factors across countries that we have argued will influence the ability of
governments to effectively reach the poor for a given level of social expenditures. Our
methodology will attempt to account for the variation in the observed levels of expenditures
by first accounting for features that can be considered to be structural to the economy, in the
sense that these capture both the need for safety nets and a country’s ability to pay for them,
and then additionally controlling for how much of the remaining observed variation in
expenditure levels is due to differences in institutional features of each economy.
To be able to disentangle these two sources of variation in safety net expenditures is
important for two reasons. First, it allows us to see what structural features of an economy
contribute to safety net expenditures, and so we uncover a basic demand function for social
expenditures. Secondly, we can separate out these effects from those related to institutional
quality. We can thus make policy prescriptions based upon both the underlying features of
the economy and recommendations related to the reform of institutions. Being able to offer
policy advice on both of these dimensions is not possible using the existing methodologies
discussed earlier. Furthermore, our analysis uses benchmarking to motivate governments to
act in accordance with policy recommendations.
In order to make the relative performance of countries easily comparable, we prefer to
report our results in terms of the ranking of countries relative to each other, when each is
compared to the international norm. The ranks of each country provide a simple summary
statistic that can be presented to governments to argue the case for more (or less) safety net
expenditure. Furthermore, the currently available cross-country data series do not allow us to
control for all the structural features that we believe would drive safety net spending. This
may lead to concerns about potential sources of bias affecting our estimates. In this case, the
use of rankings, rather than a literal interpretation of the parameter estimates, can make our
results slightly more robust. We will discuss this in more detail in section five.
The way in which this international norm is established, and the various rankings that
we shall form, will also be discussed in further detail in section five. For the remainder of
this section, we wish to focus attention upon the determinants of safety net programs. In
17
attempting to fairly compare levels of safety net expenditures across countries, we will first
need to be able to establish what these factors may be.
Summary of Benchmarking
• Benchmarking allows governments to compare their own performance with that of
their neighbors. This can motivate governments to act on policy advice.
• Benchmarking through the use of use of rankings conveys information on the
underlying effort being put in by governments on these policy outcomes, in a very
clear manner. The data requirements to construct these rankings are minimal
compared to alternative methodologies.
• Our methodology allows for the policy debate to be conducted in terms of the
structural features of an economy, as well as the design of institutions within the
country, with a view toward improving rankings.
• We do this by constructing three ranks—first, a ranking based upon unconditional
expenditures on safety nets; second, a ranking conditioning on structural features of
the economy; and third, a ranking also conditioning on the institutional quality
within the country.
What Determines the Need for these Programs?
The real primitives of any economy are tastes, production technology, endowments, and the
distribution of information. From these primitives it is possible to identify situations when
government intervention is either advisable or necessary. Rather than starting from these
primitives, in our analytical framework we prefer to begin one step ahead by looking at
various institutional and country-specific factors that determine the need for social safety
nets. The four factors that will determine the appropriate level of safety net spending are
discussed below. Though we recognize that these primitives may be endogenous to
government intervention, they nonetheless have the advantage that they may (i) be directly
linked to policies and (ii) are measurable in data.
Factor A: Underlying Distribution of Productive Ability
The distribution of such factors as physical assets (e.g., land), human capital, and labor
power will influence the need for social safety nets. Access to assets, levels of education and
skills, and labor will affect individuals' ability to avoid chronic poverty and to protect
themselves from shocks.
Factor B: Institutions for Private Provision
Families, friends, and informal networks represent the main means of social protection in
developing countries. Individuals can also rely on markets to protect themselves from
specific contingencies such as poor health and downturns in income. How well these
institutions of private provision function will therefore determine the need for social safety
nets. This, in turn, will be a function of the degree of social and market development in a
given economy. Therefore, the need for intervention will tend to be less when there is more
equal distribution of productive ability and where institutions for private provision (both
formal and informal networks) function well.
Factor C: Quality of Government
Factors A and B miss out on the fact that government is also an institution, the quality of
which will determine the appropriate level of spending on social safety nets. Therefore,
18
where bureaucratic integrity is low and preferences of the poor are not represented in the
allocation of fiscal resources, it may be inappropriate to expand spending on social safety
nets.
In other words, if the institutional framework of an economy is misallocating resources,
e.g., to special interest groups or those that are most vocal, as opposed to those in the greatest
need, then the correct policy prescription to prescribe may be a reduction in such
expenditures. For the reasons we made clear in section two, incorrectly targeted social
expenditures that primarily benefit the non-poor will likely harm social cohesion and increase
the resources spent on marginalized economic activities such as the black economy, which in
the long run will only serve to reduce a country’s sustainable level of income growth and
weaken its ability to respond effectively to external shocks and crisis. Furthermore,
improving the functioning of institutions may be a necessary precondition before it is
appropriate to increase expenditures on social safety nets. Also, it may be optimal for a
country with a poor quality of government to spend little on social safety nets, as this
expenditure will have only limited impact on poverty alleviation.
Factor D: The Nature of Shocks Affecting the Region or Country
Some countries are prone to natural calamities such as droughts and floods. Integration into
the global economy via trade and other mechanisms may also make some countries more
prone to financial and terms-of-trade shocks. Characterization of the shocks likely to affect a
given region or country is thus necessary when deciding on the appropriate level of spending
on social safety nets. The nature and frequency of shocks will affect both the aggregate need
for safety net spending and the type of safety net spending that is appropriate.
Scope for Institutional Reform
The methodology we are proposing will allow us to assess whether countries or regions are
spending too much or too little on social safety nets relative to some international
benchmark, rather than through some objective function that we seek to optimize. This can
be done both at the aggregate level and also within specific dimensions of safety net
spending. However, it also raises the prospect of viewing direct institutional reform of areas
(A) to (D) as an alternative policy to raising or lowering spending levels. It is difficult to
directly cost institutional reforms and thus to quantitatively contrast their effectiveness in
alleviating poverty with expenditure adjustments. The methodology we are proposing will
nonetheless allow us to isolate important areas of institutional reform that may improve the
efficiency of social safety net spending, as discussed above. The methodology we are
proposing is powerful as it points both to the necessity of controlling for factors (A) to (D)
when thinking about the appropriate level of social safety net spending in a given country
and because it points to institutional reform as an alternative direction of policy reform to
reduce poverty and vulnerability.
Informal Explanation of the Rankings
We begin with the actual levels of government expenditures on various types of safety net
expenditures expressed as a share of GDP. As already noted, there are wide variations in
these pure unconditional levels of expenditures. This forms the basis of the first rank, rank
one (hereafter referred to as R1). In short, this corresponds to each country's starting rank in
level of safety net expenditures. As we do not take account of any other features of the
economy at this stage, neither structural nor institutional, we shall refer to this rank as the
19
unconditional rank of countries. According to R1, the country with the highest ranking is that
country which spends the most on safety net expenditures as a share of its GDP. This ranking
provides a good starting point from which to then compare how a country’s relative
performance changes as we account for the structural and institutional features of the
economy.
We then move on to the second rank. This is constructed by regressing a country’s level
of safety net expenditures on factors we have argued to be structural to the economy, i.e.,
those factors that capture both the level of need for safety nets and the ability of a country to
meet these needs given its resource constraints. Structural factors that we include in our
analysis include: (i) the (log of) per capita income of the country which proxies for the
government budget constraint; (ii) the fraction of the population of working age (between 15
and 64), which captures the level of dependency in a country—in most countries, children
and the elderly are particularly susceptible to poverty by various alternative measures of
welfare that we may employ; (iii) the fraction of the population residing in urban regions,
which again can capture an element of neediness in the population, given that the poor most
often reside in rural locations; (iv) a measure of shocks to income per capita, i.e., how far a
country is from its long-run sustainable income level in any given year, because such cyclical
components could also be correlated with expenditures.
Regressing the level of safety net expenditures on each of the factors gives us a simple
demand function for social expenditures. We should be careful not to imply too much
causality in the relationship between the covariates and the level of expenditures. Our
approach is merely to identify likely correlates of social expenditure levels. Having
controlled for these structural features, we are still left with an unexplained component of the
level of expenditures. It is this unexplained component that is then ranked to form rank two
(R2). The level of social expenditures that we cannot explain controlling for structural factors
can be positive or negative. In other words, for a given set of structural features, a country
may actually be spending more or less than we would predict, taking what the average
country does as being the international norm. This is precisely what an OLS regression
does—it calculates the average effects of each covariate on the outcome of interest. In our
application, we calculate the predicted levels of social spending, controlling for the
aforementioned structural features, relative to this international benchmark.
It may be argued that the benchmark as set by a regression line is rather arbitrary, and
unrelated to any welfare criteria. However, there are three points to be made on this. First,
our approach is to deliberately move away from the optimizing approach in order to answer
how much governments ought to spend, and to do this in a such a way as to avoid having to
specify any welfare criterion. This is in order to be able to attempt to answer the question
without requiring unfeasible amounts of data. Secondly, what countries are actually observed
to be doing on average is a good indication of feasibility constraints and provides a natural
focal point for what countries ought to be able to do. Thirdly, the method is easily
implementable and provides simple summary statistics, in the form of rankings, to present to
policymakers. This ease of presentation should facilitate a constructive policy dialogue and
debate.
If a country spends more than we would predict given its level of structural features, this
would suggest it spends more on safety nets than needs or cost considerations alone would
explain. The opposite is true if we find a country spends less than it should. According to R2,
20
the country with the highest ranking is that country which spends the most in addition to
what we would have predicted, controlling only for structural factors.
We now take account of the institutional framework within a country, which we have
also argued influences the levels of safety net spending and the effectiveness of a given level
of expenditures. The set of institutional features we take into account are the following (all
definitions are in the data appendix): (i) the level of repudiation of government contracts; (ii)
the threat of expropriation of assets by government; (iii) the level of corruption in civil
society; (iv) the effectiveness of the rule of law; and (v) the amount of government
bureaucracy. These factors are argued to be correlated with the quality of government and so
may indirectly affect social safety net spending.
In order to form a cross-country ranking taking into account these institutional features,
we take the amount of unexplained expenditures from R2, and regress these on these various
quality-of-government indicators. We then examine how much of this disturbance still cannot
be explained. Again, countries may be above or below the predicted levels—the benchmark
is again given by the “average” country.
We rank these unexplained components in R3, which now takes account of structural
and institutional factors. A country with a positive unexplained component is interpreted to
be spending more than we would expect given the quality of its institutions and structural
characteristics of the economy. The opposite applies to countries with negative unexplained
levels of spending. The highest rank for R3 is given to the country with the largest positive
residual component.
Summary
The three cross-country rankings we employ are;2
•
•
•
R1
rank one: unconditional ranking of safety net expenditure
R2
rank two: controlling for structural features of the economy (which proxy
for the need for safety nets and the ability of the government to meet these given its
resource constraints)
R3
rank three: controlling for institutional features that proxy for the quality of
government, as well as the structural features controlled for in rank two
The presentation of our analysis in the form of rankings is useful because:
•
It is a simple summary statistic on which to base arguments to policymakers for
changes in spending.
2
Lindert (1994, 1996), using OECD data, examines the traditional view that the deadweight costs of increased
taxation will limit social spending, and finds that this explanation cannot account for the observed variations in
social spending within this group of countries. He finds that the levels of social spending are more determined
by age distribution, level and distribution of income, and level of political participation. While it is clear that
Lindert’s is not a positive analysis, it does suggest that the levels that governments ought to be spending is more
determined by these underlying structural factors in an economy than the marginal cost of raising public
expenditures through taxation.
21
•
•
•
The use of rankings makes our results more robust to econometric concerns arising
from the potential endogeneity of some covariates, as well as omitted variables due
to lack of data.
The data requirements for this approach are minimal. For instance, we do not need
to define any welfare criterion by which to judge optimal levels of spending.
International or regional norms provide a natural point of comparison for how much
effort governments are seen to be putting into safety net spending. They can also be
used to motivate policy discussions.
Comparisons Across Rankings
It will also be useful to compare how a country’s rank changes according to these three
procedures. We may find, for example, that countries which appear to have high levels of
social spending (high R1 rank) actually are not spending as much as we may expect them to
once structural factors are taken into account. Similarly, there may be cases of countries that
appear at first glance to be spending relatively little on safety nets but, once we account for
their ability to pay for such goods and services, or their weak institutional framework, may
actually be spending a lot relative to the international benchmark. Thus, the movements
across rankings contain much information, as well as the rankings themselves.
Of course it is possible not only to compare countries to the international benchmark as
set by the regression fit, but also within regions, whereby we will be able to examine good
and bad performers, and we will also be able to compare the relative performance of
neighboring countries that may appear to have all of the same measurable economic
characteristics. This may help to induce a form of yardstick competition when the results are
presented to policymakers. Namely, policymakers may be induced to put more effort into
safety net spending if they see that their neighbors, who may face similar resource constraints
and levels of need, are able to perform better in the rankings we construct. In section six we
will present a detailed analysis of all of our results, and in section seven we discuss how this
paper can be used in a constructive policy dialogue.
We have conducted this sort of analysis across Indian states, forming R1, R2, R3 across
15 states. The set of covariates used there in the construction of these rankings is slightly
different from that used in the cross-country analysis, although the methodology and
interpretation of the results is identical. The details of both types of analysis are the subject of
the next section.
In comparison to benefit incidence studies or behavioral approaches, our data
requirements are a lot less strenuous. While there are issues of endogeneity to perhaps be
concerned with in our regression analysis, say because we may believe that the level of
urbanization itself cannot be taken as exogenous but itself depends upon the level of safety
net expenditures, nevertheless, unless these forms of bias vary systematically over countries,
the analysis of the various rankings remains unchanged, although we should not then place
too much literal interpretation on the actual regression estimates. We at least manage to make
some crude attempt at taking account of the level of need, ability to pay, and quality of
government in this analysis.
22
Most importantly, when addressing the question of how much governments ought to be
spending, we move away from an optimizing approach to this question altogether, and focus
instead on country performances relative to international norms, and this gives an alternative
way of thinking about the same problem, as well as being able to present policymakers with a
simple summary statistic on how well they do compared to their neighbors, and at least an
indication of why their rank relative to their neighbors is where it is.
V. Making Benchmark Comparisons
Cross-Country Analysis
In this section we formally detail how our rankings are constructed. The data used in the
analysis were drawn mainly from the World Development Indicators database
(http://www.worldbank.org/data/wdi/home.html) and the IMF Government Finance Statistics
Yearbook. These are the only sources of consistent and comparable statistics which are
available over a reasonable time period (see data appendix for variable definitions). The
drawback is that (i) different types of safety net statistics are not disaggregated and (ii) social
assistance and social insurance measures are reported together, whereas our interest is mainly
in the former. There is thus a clear need to build more detailed data sets on different aspects
of safety net spending to which the analytic framework could be applied. This exercise is,
however, beyond the scope of the current project.
R1: Unconditional Safety Net Expenditures
We constructed a cross-country panel data set over the period 1972–97 with data averaged
over five-year periods. We have data on the levels of safety net spending in country i of type
j in period t, as a share of GDP, denoted sijt. Rank one (R1) is simply the ranking of an
individual country in sijt for each type of safety net spending (j), averaged over all time
periods. We shall denote this time average of social spending of type j in country i as s ij .
Hence, R1 corresponds to the unconditional ranking of the level of safety net expenditures as
a share of GDP, across countries. The higher a country’s ranking, the higher the amount that
nation spent on safety nets of type j on average over the time period 1972–97.
The types of safety net expenditure variables that are available to us are (complete
definitions are given in the data appendix):
•
•
Transfers to Organizations and Households: transfer payments to private
institutions which are not operated as enterprises; current payments in cash to
households adding to their disposable income; and
Social Security and Welfare: transfer payments to compensate for loss in income
or inadequate earning capacity.
In addition, for completeness, we also look at the two most commonly used observed
forms of social spending—education and health. However, we would not typically think of
these as constituting a safety net as defined in section two. For example, health spending
includes expenditures on all medical instruments and medical research. Similarly, for
education expenditures, one would not want to classify tertiary and university expenditures
as safety nets.
23
The remaining safety net categories are close to our earlier definition of safety nets,
although the availability of more disaggregated data, consistently defined across countries
and time, would be ideal, as this would enable a closer matching between our definition and
the expenditure types actually considered.
R2: Controlling for Structural Features
Moving to the construction of rank two (R2), we run a linear OLS regression of the form;
(1)
sijt = β Xit + γ Z i + vijt
where sijt denotes safety net expenditures in country i on safety net j at time t
X it = A vector including log(real per capita GDP per capita), a measure of shocks to
GDP per capita, the fraction of the population aged between 15 and 64, the fraction of the
population residing in urban regions.
Z i = Set of regional dummies for Latin America and the Caribbean, sub-Saharan Africa,
North America, South Asia, Middle East and North Africa, East Europe and Central Asia,
East Asia and Pacific, with Western Europe being the omitted category.
We deliberately exclude country fixed effects precisely because we want to try to
unpack what this fixed effect may be comprised of, and how in particular it may be related to
institutional features. With the inclusion of the set of regional dummies, we are effectively
calculating the regression (1) in terms of deviations from regional means.
After having run (1) we obtained a residual, v̂ijt . This is the variation in safety net
spending that we are unable to explain after controlling for structural features of the
economy—these variables have proxied for the ability of the country to pay for safety nets
and the need for such expenditures.
We then average this residual over the sample period for each country to form v̂ ij . It is
this averaged residual that we use as the basis for our second ranking, R2. This timeaveraged residual gives us a single summary statistic for each country that facilitates
comparisons across countries. To be clear, R2 is based upon the time-averaged unexplained
component of safety net expenditures once the structural factors of the economy, X it , which
control for the needs and budget constraints of the economy, are taken into account. This
averaged residual may be negative, implying that, over time, the country is spending less on
safety nets of type j, given its structural parameters, than we would have expected given what
other countries are spending on average. Given that we control for regional dummies ( Z i ) in
(1), the comparison group is the set of countries in the same region as i. The opposite applies
if this averaged residual is positive. The lower the ranking by R2, the lower (more negative)
the residual.
R3: Controlling for Institutional Quality
Finally, we move to the construction of the third ranking, R3, which summarizes how much
of the unexplained variation in the level of safety net expenditure for country i can be
explained by controlling for the quality of government in that country.
24
More precisely, we regress the time-averaged fitted residual on a series of measures
capturing institutional quality. These measures are averaged over time because there is not
much variation in each of them over time. Thus, it is as if we are running a cross-sectional
regression having averaged over all time periods. The specification of the regression that
forms the basis of R3 is then given by:
(2)
vˆ ij = γ + δQi + ωij
where these measures of quality of government, Qi , are:
•
•
•
•
•
Repudiation of Government Contracts: indicates the risk of a modification in a
contract taking the form of a repudiation, postponement, or scaling down;
Expropriation Threat: risk of outright confiscation or forced nationalization;
Corruption: special payments demanded by high officials, and illegal payments
expected throughout lower levels of government;
Rule of Law: reflects the degree to which the citizens of a country are willing to
accept the institutions established to make and implement laws and adjudicate
disputes; and
Government Bureaucracy: autonomy from political pressure and strength and
expertise to govern without drastic changes in policy.
The data source for these measures is the International Country Risk Guide (ICRG), a
monthly publication of Political Risk Services. Full details are again found in the data
appendix. All of these indices decrease as quality worsens.
Having run equation (2) we obtain a fitted residual, ω̂ ij . This residual is the unexplained
component of safety net spending that cannot be explained after controlling for both
structural and institutional characteristics.
The third ranking system, R3, is based upon this residual, ω̂ ij . Again, it can be the case
that this residual may be positive or negative. A positive residual implies that the country is
spending more on social safety nets once structural features of its economy and institutional
quality have been controlled for, than we would otherwise expect given the expenditures of
other countries in the same region. The opposite is true if this residual turns out to be
negative. The lower the ranking of R3, the lower (more negative) the value of ω̂ ij .
Summary of Rankings
We summarize the basis of our three alternative ranking systems below:
R1 time-averaged unconditional ranking of safety net spending of type j, based on sij.
A lower ranking means that the country unconditionally spends less on safety nets of
type j.
R2 ranking of social spending conditional on structural factors, v̂ ij . A lower ranking
means that safety net expenditures are lower than we would expect compared to the
countries in the same region, controlling for structural features of the economy.
25
R3
ranking taking into account quality of government, ω̂ ij . Again, a lower ranking
means that safety net expenditures are lower than we would expect compared to the
countries in the same region, controlling both for structural and institutional features of
the economy.
•
•
Once we take account of how much a country is actually able to spend on safety
nets, we may find its international ranking increases (moving from R1 to R2), and if
its ranking further increases as we move to R3, it may be possible to infer that the
country has institutions that are effective in translating whatever resources it has
into safety net spending.
Similarly, if we find a country’s ranking to be falling as we move from R1 to R2 to
R3, we could infer that, although at face value the country appears to perform well
in terms of how much it spends, it actually performs worse than we would predict
given the characteristics of its economy and the quality of its institutions, relative to
the international or regional benchmark.
Indian State Level Analysis
We now turn our attention to the analysis at the level of Indian states. Again, we seek to use
our methodology to make some benchmark comparisons across states regarding how much
each state spends on safety nets relative to the others. We use panel data on 15 major Indian
states over the period 1960–92.
The approach closely mirrors that of the cross-country analysis, where we establish
three rankings on states—R1, the unconditional ranking by the level of expenditure in
various categories of safety nets; R2, the ranking conditional on structural features of the
state economy which again are designed to capture both the level of need for safety nets in
the state and the state government’s budget constraints; and R3, which conditions both on
structural factors and on state institutional factors. The interpretation of each of the rankings
remains identical to that for the cross-country analysis.
The main differences between this analysis and that for the cross-country data set are
that: (i) the classifications of safety net expenditures are different; (ii) the set of structural
features we control for are different from those in the cross-country analysis; and (iii) our
measures of the quality of government are also different from the earlier analysis.
Unlike in the cross-country analysis, the use of panel data allows us to control for
unobservable structural factors that are common to all states, in the construction of the
ranking R2, and to control for common unobservable quality-of-government variables in the
construction of R3. Furthermore, in this data set we can be confident of having controlled for
all common macroeconomic shocks that may determine safety net expenditure levels. We do
this through the inclusion of state-level fixed effects. In short, then, we probably have more
reason than in the cross-country analysis to be confident in the actual estimated effects of
structural and institutional features on the level of safety net expenditures, as well as the
rankings themselves.
26
R1: Unconditional Safety Net Expenditures
We have data on the levels of safety net spending in state i of type j in period t, denoted as
sijt. Rank one (R1) is simply the ranking of an individual state in sijt for each j, averaged over
all time periods. We shall denote this time average of social spending of type j in state i as sij.
Hence, R1 corresponds to the unconditional ranking of the level of safety net expenditures
across states. As with the cross-country analysis, the higher a state’s rank, the higher the
amount that state spends on safety nets of type j on average over the time period.
The types of safety net expenditure variables that are available to us at this level of
analysis are (complete definitions are given in the data appendix):
•
•
•
•
•
•
•
Health
Education
Social Expenditures
Food Subsidies
Calamity Expenditures
Development Expenditures
Public Food Distribution3
R2: Controlling for Structural Features
To form the second ranking, R2, we run a linear OLS regression of the form;
(3)
sijt = β Xit + α i + vijt
where α i are state fixed effects, and X it is the set of structural variables proxying for the
ability of the state to finance such expenditures, and the need for them. These include: state
income per capita, which captures the budget constraints facing the state government; rural
and urban headcount measures, which proxy for the level of need for safety nets; and rural
and urban income gini coefficients, which again are included in an attempt to proxy the level
of needs there may be for types of social spending—the greater the degree of inequality, the
greater the number of individuals in poverty who would benefit from such expenditures.
Estimating the state level safety net “demand” function (3), we obtained a residual, v̂ijt .
After averaging this residual over t to form v̂ ij , we obtain the basis of R2. In other words, R2
is based upon the time-averaged unexplained component of safety net expenditures once the
structural factors of the economy are taken into account.
Just as in the cross-country analysis, this averaged residual may be negative, implying
that, over time, the state is spending less on safety nets of type j relative to other Indian
states, given its structural parameters. The opposite applies if this averaged residual is
positive. The lower the ranking by R2, the lower (more negative) the residual.
3
Unlike all of the other forms of safety net that we consider, which are in monetary terms, this transfer is in
kind, and corresponds to public distribution primarily of rice and wheat.
27
R3: Controlling for Institutional Quality
We move to the construction of R3, which tries to see how much of the unexplained variation
in the level of safety net expenditure for state i can be explained by controlling for the quality
of government, once structural factors have also been controlled for. Again, the method is to
regress the time-averaged fitted residual on a series of measures capturing institutional
quality. The available measures for the state-level quality of government are:
•
•
•
•
•
Voter turnout: the percentage of eligible voters in the state that actually voted in
the last elections for the state legislature;
Political competition: the relative number of seats in state legislatures of the
Congress party vis-à-vis its closest rival political party;
Literacy: male and female literacy rates;
Variance of Social Spending: the variance of social spending that cannot be
explained by the variance of state income and natural calamities;
Deviation from State Means of Level of Social Spending: the deviation from
state means of social spending, controlling for mean income levels and the
occurrence of natural calamities.
These measures are poorer proxies for the quality of government than those available at
the cross-country level. The rationale behind their inclusion is the following. The level of
voter turnout measures political participation, which we may expect to be higher if
individuals feel that the political process accurately reflects their preferences. In some sense,
voter turnout can be thought of as measuring the extent to which citizens feel political
institutions are legitimate. The measure of political competition is designed to capture the
responsiveness of state governments to the electorate’s preferences. State literacy rates may
be used as a proxy for how well-informed individuals are about state government policies,
and hence reflects the ability of state governments to make policy responsive to voters’
preferences.
The final two measures reflect the variability of social spending, the argument being that
lower-quality governments are more subject to pressures from special interest groups or rentseeking of government officials and therefore we may expect such governments to have more
variable expenditure levels. Cyclical movements can also lead to variations in levels of
spending, but as these are common to all states they should not affect the ranking we form
across states. Full details of how these measures are constructed are given in the data
appendix. Formally, to construct the third ranking R3, we run a regression of the form:
(4)
vˆ ij = γ + δQi + ωij
where the set of (safety-net-specific) quality-of-government measures are denoted as Qi .
Note that, unlike the cross-country analysis, there is enough variation here in our quality-ofgovernment measures over time to be able to run the equation as a panel regression, rather
than in cross-sectional form.
R3 is based upon the time-averaged residual from (4), ω̂ ij , i.e., that portion of the
unexplained residual from regressing safety net expenditures on structural features that still
28
cannot be explained by the quality of government. Again, this residual may be positive or
negative. A positive residual implies that the state is spending more on social safety nets once
structural features of its economy and institutional quality have been controlled for, relative
to other Indian states, than we would otherwise predict just based on these factors. The
opposite is true if this residual turns out to be negative. The lower the rank of R3, the lower
(more negative) the value of ω̂ ij .
We summarize the basis of our three alternative ranking systems below:
R1
time-averaged unconditional ranking of safety net spending of type j, sijt .
R2
R3
ranking of social spending conditional on structural factors, v̂ ij .
Ranking also taking into account quality of government, ω̂ ij .
As with the cross-country analysis, movements across each of these rankings can be
used to make inferences regarding the structural features or institutional quality of the state in
comparison with other states.
Welfare Outcome Regressions
Up until this point we have been concerned with benchmarking the relative performance of
countries and Indian states with regard to safety net expenditures. We now follow a similar
line of reasoning to benchmark their performance in terms of some key welfare indicators.
We thus extend the analysis to see how structural and institutional features affect various
welfare-related outcomes. We do this both across countries and at the Indian state level.
The motivation for performing this type of analysis is similar to before—what we
observe are large variations in welfare indicators across countries. We would like to be able
to assess the relative performance of countries (or states) by taking account of (i) the
structural features of the economy, which should include the level of a social safety net in the
country as well as the budget constraints facing the economy; and (ii) the institutional
features of the economy, which may by correlates of the effectiveness of safety net
expenditures to be translated into welfare improvements for vulnerable groups. Our aim is
thus to construct three rankings analogous to our earlier analysis: R1, which is the
unconditional rank of the welfare indicator; R2, which is the rank once we have accounted
for the level of safety net expenditures and income levels; and R3, which takes into account
the quality of institutions. By doing this we should then be able to comment on the extent to
which structural or institutional features are correlates of these welfare outcomes.
Hence, it may be the case that we observe a country performing relatively poorly on a
welfare indicator unconditionally. Once we take account of how much it is actually spending
on safety nets, and what it is actually able to spend on safety nets, we may find its
international ranking increases, and if its ranking further increases as we move to R3, it may
be possible to infer that the country has institutions that are effective in translating whatever
resources it is placing in safety nets into welfare enhancements. Similarly, if we find a
country’s ranking to be falling as we move from R1 to R2 to R3, we could infer that,
although at face value the country appears to perform well on welfare indicators, it actually
performs worse than we would predict given the characteristics of its economy and the
29
quality of its institutions, relative to the international benchmark set by what countries
manage to do on average.
We examine the following welfare outcomes—for the cross-country analysis, life
expectancy at birth and the infant mortality rate (IMR) of children aged less than one (per
1000 live births) are both available. At the Indian state level, only the latter of these is
available. We briefly go through the construction of the rankings for these welfare outcomes.
R1: Unconditional Welfare Outcomes
We shall refer to our welfare indicator for country i in period t as Ω it . R1 is based on simply
ranking these unconditional levels. We construct R1 for both welfare indicators, so a country
has a higher R1 ranking if its life expectancy is higher, and if its child mortality rate is lower.
R2: Controlling for Structural Features
In moving to this ranking we want to take account of the factors that we would typically
expect to be correlated with these welfare outcomes. Again, due to econometric issues such
as the potential endogeneity of some of the regressors, we would not place too literal an
interpretation on the regression that sets the OLS benchmark. Again, for purposes of forming
the correct ranking across countries, these potential sources of bias in the actual estimates are
unproblematic as long as these biases do not differ across countries. Hence, in order to
construct to a ranking which accounts for these structural features in the cross-country data,
we run a regression of the form:
(5) Ω it = β Xit + γ Z i + vit
where Ω it refers to either of our welfare outcome measures of life expectancy or the IMR;
Xit includes the real per capita GDP per capita, a measure of shocks to GDP per capita, the
fraction of the population aged between15and 64, the fraction of the population residing in
urban regions, and the levels of per capita expenditures on health and education; and Z i is a
set of regional dummies with Western Europe being the omitted category.
The inclusion of income per capita in the set of regressors proxies for the government
budget constraints, shocks to GDP account for cyclical movements, the measures of working
population and the level of urbanization are both included to proxy for poverty levels (and it
is the potential endogeneity of these regressors that may concern us most). We include health
and education spending as these types of safety net would be expected to alleviate lifetime
poverty. It was decided not to include the other forms of safety net expenditures in order to
maintain as large a sample as possible.
Again, we omit country fixed effects precisely because we wish to be able to shed light
on how much country-specific factors derive from these structural features and how much
they derive from institutional factors.
The residual estimated from (5) thus captures how much of the variability in the welfare
indicator cannot be explained by controlling for structural factors alone. The rank R2 is then
based upon the time-averaged residual from (5), v̂ i . The interpretation of this is as before—
lower rankings (more negative residual) indicating that the country performs worse on the
welfare rank than we would predict given its structural characteristics, where we are
30
comparing to the international benchmark as determined by how well countries are doing on
average with regard to the same welfare indicator.
R3: Controlling for Institutional Quality
We now wish to construct R3 in order to assess how much of this unexplained component
can in fact be accounted for by institutional features of the country. We do this by first
regressing time-averaged residuals from (5) on the set of time-invariant cross-country
quality-of-government measures discussed before:
(6)
vˆ i = βQi + ωi
R3 is then based upon the residual from this regression,. interpreted as the amount of the
welfare outcomes that cannot be accounted for by either structural features of the economy or
institutional features. This can be positive or negative, the interpretation of which is exactly
as before.
When we move to the analysis of welfare outcomes at the Indian state level, where we
only have the welfare outcome of IMR, the set of structural factors we are able to control for
in constructing R2 are: (i) safety net expenditures on health, education, social assistance,
food subsidies, and calamities; (ii) state income per capita; (iii) urban and rural headcounts;
(iv) rural and urban gini income inequality indices. In the construction of R3, we employ
voter turnout, political competition, and literacy rates as our proxies for institutional quality.
The interpretation of the three ranks is the same as that for the cross-country analysis.
VI. Results
Cross-Country Analysis
We now turn to the analysis of our results, where we shall essentially go through the
estimation of equations (1) to (3) both for cross-country data and the Indian state-level data,
and interpret the rankings, R1 to R3, derived from them. We begin with the cross-country
analysis. All the data are from the World Development Indicators and the IMF Government
Finance Statistics Yearbook, available over the period 1972–97, averaged over five-year time
intervals.
Table 1 presents the levels of spending across countries for the two main types of safety
nets that we consider, namely social spending and welfare, and transfers to organization and
households, both as shares of total GDP. The table also gives the figures for health and
education expenditures; discussion of these is in the appendix. There are two things of note in
this table—first, the expenditure shares are broadly consistent with what we might have
expected, with the shares of Western Europe and North America being the highest in most of
the safety net categories, although when looking at the total size of the government sector,
Eastern Europe and the Middle East spend shares comparable to those of developed
countries. Secondly, there is much variation in the expenditure shares across countries,
implying variation in the levels of expenditures across countries. This occurs both within and
across each continent. This confirms our earlier point that, given such variation, there is
clearly something to explain concerning the underlying determinants of such shares. In
31
constructing R2 and R3 we will be trying to account for these levels of variation by the
structural and institutional features of the economy. There is sufficient variation in the data to
shed some light in the course of this analysis on how much governments ought to spending.
Table 1: Safety Net Expenditures as percentage of GDP
SS and
Welfare
Transfers to
Orgs/HH
Health
Education
Total Govt
Exp
Albania
6.729
7.171
1.739
0.712
31.007
Argentina
5.673
6.159
0.274
0.922
13.947
Australia
6.836
7.297
2.638
1.784
23.666
Austria
16.822
16.448
4.695
3.594
37.232
Bahamas
1.167
1.346
3.067
3.771
19.267
Bahrain
0.855
0.642
Barbados
4.696
Belarus
12.216
Belgium
20.028
Belize
1.310
Benin
1.625
Bhutan
0.622
Bolivia
3.534
Brazil
Bulgaria
Country
2.607
3.988
32.171
3.707
6.131
30.902
10.361
1.372
2.771
35.792
23.281
0.864
6.888
48.450
2.197
3.932
26.044
1.097
3.424
17.706
0.729
2.357
3.908
36.060
2.643
0.960
3.562
18.764
7.468
7.282
1.775
1.029
26.538
11.074
11.089
1.271
1.601
45.872
Burkina Faso
0.463
0.776
2.114
12.045
Burundi
1.191
1.390
1.142
4.153
24.268
Cameroon
0.890
1.096
0.787
2.621
18.556
Canada
8.266
9.388
1.419
0.791
22.751
Central African Republic
1.353
1.119
3.877
22.001
Chad
0.199
0.446
1.527
17.173
Chile
8.699
Colombia
2.174
Comoros
Congo, Republic of
2.341
3.929
27.671
0.673
2.941
13.156
0.017
2.756
9.570
43.679
1.763
1.919
4.470
38.017
0.265
Costa Rica
4.086
6.341
4.865
5.204
24.041
Cote d'Ivoire
1.079
0.927
1.228
6.539
27.836
Croatia
14.104
12.032
6.695
2.800
43.661
Cyprus
6.532
7.644
1.980
3.555
32.261
Czech Republic
16.588
17.586
6.203
0.201
33.608
Denmark
15.139
7.030
0.965
4.079
37.179
3.578
4.890
35.094
Djibouti
Dominica
34.685
1.288
Dominican Republic
0.830
0.150
1.603
1.910
15.561
Egypt, Arab Republic of
4.634
6.056
1.124
4.539
42.779
Estonia
9.712
12.283
4.665
2.828
30.437
Ethiopia
1.222
0.881
0.840
2.558
23.390
Finland
10.997
2.484
4.247
32.137
France
17.973
22.119
6.774
3.304
41.312
Gambia, The
0.740
0.967
1.903
3.044
25.034
Greece
6.788
5.525
2.447
2.728
33.508
Guatemala
0.651
0.853
1.663
11.612
Guinea
0.461
21.923
32
SS and
Welfare
Transfers to
Orgs/HH
Health
Education
Total Govt
Exp
Guinea Bissau
1.944
0.684
3.555
4.699
52.123
Guyana
2.805
3.111
6.360
62.661
Haiti
0.657
1.077
1.243
16.935
Honduras
0.985
1.828
3.277
16.233
Hungary
14.343
17.623
2.189
1.110
51.419
Iceland
5.375
6.463
Country
India
Indonesia
0.954
0.883
6.150
3.619
28.997
0.240
0.297
13.504
0.410
1.615
18.590
Iran, Islamic Republic
2.307
1.557
4.145
31.728
Ireland
11.814
6.157
5.475
41.732
Israel
10.518
14.897
3.182
5.688
60.504
Italy
13.301
17.478
4.517
3.921
42.514
Jamaica
1.178
0.270
2.857
6.514
37.372
Japan
8.291
0.696
0.341
1.374
16.987
Korea, Republic of
1.217
2.500
0.211
2.987
16.512
46.505
Kuwait
4.661
9.071
2.507
4.840
Latvia
12.674
14.193
2.309
3.428
30.713
Lebanon
2.545
4.116
0.915
2.562
34.320
Lesotho
0.709
0.585
3.574
8.073
47.083
Liberia
0.294
1.705
3.404
23.532
Lithuania
8.805
8.629
2.091
1.687
25.719
Luxembourg
19.320
20.283
0.890
3.531
39.879
Madagascar
0.611
1.002
2.060
16.700
Malaysia
1.154
1.507
5.397
26.285
Maldives
1.806
3.344
5.793
45.289
1.458
Mali
0.965
1.056
0.697
2.766
21.703
Malta
13.837
14.340
3.796
4.177
41.362
Mauritania
1.470
1.157
4.001
43.003
1.895
3.471
22.977
0.465
2.948
17.317
0.561
0.833
21.270
Mauritius
4.159
México
2.880
Mongolia
5.100
3.802
4.818
Morocco
1.885
1.014
5.202
31.197
Myanmar
0.702
0.828
1.794
13.924
0.719
1.751
Namibia
1.451
37.222
Nepal
0.088
15.098
Netherlands
18.587
22.913
6.223
6.073
50.275
Netherlands Antilles
7.502
6.595
1.679
1.339
25.805
Nicaragua
2.709
4.184
3.024
3.496
33.537
0.713
2.889
16.529
14.100
2.251
2.420
35.837
0.233
0.365
20.220
28.608
Niger
0.306
Norway
12.306
Pakistan
0.419
Panama
4.120
6.147
4.707
4.870
Paraguay
2.112
1.475
0.443
1.444
10.752
2.210
0.950
3.121
16.239
15.718
4.375
3.095
42.740
Peru
Poland
21.358
Portugal
7.828
9.907
2.454
3.156
37.472
Romania
7.865
10.630
1.394
1.923
38.113
Russian Federation
7.386
7.222
0.406
0.722
26.291
Rwanda
0.319
1.051
0.715
2.754
16.170
33
Country
Senegal
SS and
Welfare
Transfers to
Orgs/HH
1.163
Health
Education
Total Govt
Exp
1.125
4.225
21.394
Seychelles
8.494
7.893
4.479
6.554
57.140
Singapore
0.511
1.684
1.315
3.696
20.350
South Africa
1.652
1.702
0.555
1.897
28.546
Spain
13.166
14.154
1.579
1.783
28.302
Sri Lanka
4.537
4.624
1.520
2.741
28.948
St. Kitts
2.805
1.990
3.640
5.174
30.244
St. Lucia
1.461
2.200
3.363
6.809
30.290
Suriname
2.755
2.178
6.341
41.374
Sweden
18.780
21.316
0.676
3.596
39.769
Switzerland
10.152
10.506
3.039
0.667
20.955
Syrian Arab Republic
1.417
0.488
2.780
32.734
Thailand
0.587
0.597
0.972
3.417
16.710
Togo
2.212
0.186
1.764
5.355
35.686
Tonga
0.473
3.578
5.450
43.054
Trinidad and Tobago
2.616
5.264
2.020
3.776
29.218
Tunisia
3.949
6.820
2.148
5.958
32.275
Turkey
0.538
0.523
3.122
20.292
United Kingdom
10.513
12.622
5.020
1.094
37.777
United States
14.122
14.786
1.254
2.182
26.738
Uzbekistan
6.817
9.403
2.879
0.502
Virgin Islands (U.S.)
1.333
1.847
3.576
West Bank and Gaza
0.275
0.704
3.324
6.623
Yemen
2.863
1.644
7.692
Yugoslavia
1.722
1.169
5.776
Latin America and the
Caribbean
2.913
3.273
2.209
3.727
24.937
Sub-Saharan Africa
1.443
1.525
1.582
4.093
28.597
North America
11.194
12.087
1.336
1.486
24.744
Western Europe
13.566
14.792
3.495
3.376
36.551
South Asia
1.494
2.677
1.402
2.476
26.520
Middle East and North
Africa
4.661
6.725
1.846
4.779
38.363
East and Central Europe
10.307
11.662
2.882
2.164
34.186
East Asia and Pacific
2.373
2.293
1.426
3.179
22.902
World average
5.494
6.998
2.139
3.493
30.060
Standard deviation
5.606
6.382
1.570
1.873
11.399
Note: Figures represent averages of five-year period averages for spending data from 1972-1997. Blank cells
indicate data not available.
ss = social security
Orgs/HH = Organizations or households
Source:IMF Government Finance Statistics Yearbook. All years, 1972-1997.
R1: Unconditional Safety Net Expenditures
R1 is constructed from the ranking of the unconditional share of GDP of each type of social
expenditure. This is given in table 2a. What we notice is that, following from table 1, it is the
developed economies that unconditionally spend the highest shares of GDP on each type of
34
safety net, although notable exceptions include Guyana, which actually has the highest share
of government expenditures out of GDP in our data set, followed by Israel. In terms of
neighboring countries, we find that, for instance, Sri Lanka and Pakistan have large
differences in the shares of GDP they devote to social security; Mexico and Brazil have
dramatic differences in terms of social security, as do Jamaica and Trinidad and Tobago in
transfers to organizations and households.
To see how much of these variations can be explained by structural and institutional
factors, we then run regression (1), controlling for structural factors. This leads to the
construction of the second ranking, R2.
R2: Controlling for Structural Features
We now control for structural features of the economy discussed earlier, namely, the log of
per capita income (this enters in logarithmic form because we expect there to be diminishing
marginal benefits of safety nets as income rises), shocks to income (as a proxy for business
cycle effects), the fraction of the population of working age, the level of urbanization, and
regional dummies. Having run this regression, which is reported in table 3, we obtain a
residual which captures the unexplained component of safety net expenditures, controlling
for these structural features. Taking the time average of the residuals for each country, we
obtain the basis of R2, the ranking of countries controlling for structural characteristics.
From table 2 we can see that most of the movement across countries moving from R1 to
R2 is at the lower ranks of R1, i.e., having controlled for structural features, the countries that
had the highest unconditional rank of social expenditures, still tend to have the highest ranks.
However, there are some notable exceptions. Switzerland, Israel, and Cyprus both have large
falls in their ranks moving to R2, implying that, according to the international benchmark of
what the average country spends on safety nets given its structural characteristics, these
countries spend less than we would predict. Conversely, Nicaragua and Indonesia improve
their rankings moving to R2, implying that their performance relative to the international
benchmark improves when taking account of their structural characteristics, i.e., the level of
need and the ability to finance social expenditures.
R3: Controlling for Institutional Quality
We now examine how much of the unexplained residual from (1) can be explained by
country-specific institutional factors. In order to do this, we run regression (2), obtain the
residual, ω̂ ij and then rank this to form R3. Again, we observe less variation for the countries
ranked highest than for those below them. For social security, there are some dramatic falls
for countries such as Denmark, Korea, and Honduras, implying that controlling for both
institutional quality and structural features, they perform worse relative to regional norms
than when only structural characteristics are controlled for. Other countries, such as Chile,
Egypt, and Brazil improve their level of relative performance with respect to expenditures on
social security. In terms of transfers, Paraguay and Mali do worse when we move to R3,
indicating that, relative to the benchmark, these countries would appear to spend less than we
would expect controlling for institutions and structure. Hence, this evidence may suggest that
such countries ought to spend more on safety nets, all else equal.
35
Table 2a: Spending On Social Safety Nets (SS/GDP)
Social Security and Welfare
Name
Transfers to Orgs and HH
Rank1
Rank2
Rank3
Belgium
1
1
1
Luxembourg
2
12
Sweden
3
Netherlands
France
Name
Health
Rank1
Rank2
Rank3
Belgium
1
3
5
19
Netherlands
2
1
7
7
France
3
2
4
2
5
Sweden
4
5
4
3
Luxembourg
5
Austria
6
6
4
Austria
6
Germany
7
10
9
Poland
Denmark
8
13
10
Israel
Uruguay
9
3
2
Uruguay
Name
Education
Rank1
Rank2
Rank3
France
1
2
2
1
Netherlands
2
3
3
Germany
3
4
4
Ireland
32
27
Iceland
8
11
7
6
8
16
9
5
Name
Total Government Expenditure
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Belgium
1
1
2
Guyana
1
2
1
3
Cote d'Ivoire
2
5
3
Israel
2
1
2
4
5
Jamaica
3
4
8
Guinea-Bissau
3
3
3
4
1
1
Guyana
4
3
1
Netherlands
4
4
7
5
5
7
Suriname
5
13
7
Belgium
5
6
8
United
Kingdom
6
8
6
Netherlands
6
2
6
Kuwait
6
33
32
10
Costa Rica
7
9
12
Tunisia
7
7
11
Egypt, Arab
Rep.
7
8
5
12
Panama
8
6
4
Israel
8
10
13
Poland
8
5
4
2
Austria
9
14
13
Ireland
9
8
4
Italy
9
14
23
Malta
10
5
6
Spain
10
11
9
Italy
10
10
8
Malaysia
10
11
16
Ireland
10
15
21
Italy
11
16
26
Norway
11
34
30
Malta
11
11
10
Togo
11
12
23
Suriname
11
16
11
20
Spain
12
8
18
United Kingdom
12
14
19
Guinea-Bissau
12
7
9
Costa Rica
12
14
12
Malta
12
13
Norway
13
51
39
Switzerland
13
43
45
Israel
13
27
19
Morocco
13
15
14
France
13
18
13
Ireland
14
14
15
Portugal
14
10
14
Guyana
14
13
11
Zimbabwe
14
16
19
Luxembourg
14
46
58
Finland
15
60
60
United States
15
31
35
Switzerland
15
46
44
Panama
15
20
18
Sweden
15
39
39
Israel
16
26
47
Canada
16
39
41
Nicaragua
16
12
16
Kuwait
16
9
10
Romania
16
19
16
United
Kingdom
17
23
23
Cyprus
17
24
25
United States
17
31
24
Guinea-Bissau
17
18
20
Congo, Rep.
17
7
6
Switzerland
18
68
64
Australia
18
48
47
Jamaica
18
15
14
Egypt, Arab
Rep.
18
17
9
United Kingdom
18
20
22
Chile
19
19
14
Brazil
19
7
8
Australia
19
32
34
Congo, Rep.
19
22
21
Portugal
19
17
19
Japan
20
35
56
Denmark
20
46
44
Kuwait
20
24
21
Finland
20
19
15
Jamaica
20
12
18
Canada
21
65
63
Tunisia
21
12
16
Finland
21
44
50
Senegal
21
38
28
Austria
21
37
36
Romania
22
17
11
Iceland
22
47
48
Portugal
22
26
37
Malta
22
27
35
Denmark
22
36
33
Portugal
23
21
24
Costa Rica
23
13
21
Greece
23
30
41
Iran, Islamic
Rep.
23
23
17
Norway
23
41
42
12
Brazil
24
9
12
Argentina
24
27
33
Chile
24
16
28
Denmark
24
24
26
Togo
24
11
Australia
25
70
69
Panama
25
15
6
Norway
25
51
42
Chile
25
26
29
Gabon
25
9
9
United
States
26
64
55
Egypt, Arab
Rep.
26
18
13
Suriname
26
28
20
Italy
26
21
32
Nicaragua
26
22
15
Greece
27
43
40
Greece
27
36
18
Tunisia
27
18
23
Trinidad and
Tobago
27
31
41
Greece
27
21
10
Cyprus
28
50
49
Trinidad and
Tobago
28
40
38
Trinidad and
Tobago
28
38
46
Singapore
28
32
27
Germany
28
40
45
Argentina
29
49
48
Sri Lanka
29
9
7
Cyprus
29
55
63
Iceland
29
34
38
Syrian Arab
Republic
29
23
28
Iceland
30
71
72
Nicaragua
30
17
17
Congo, Rep.
30
20
31
Sweden
30
25
25
Tunisia
30
26
24
Kuwait
31
74
74
Korea, Rep.
31
44
43
Gambia, The
31
22
15
Austria
31
33
30
Cyprus
31
34
30
Egypt, Arab
Rep.
32
25
17
Peru
Venezuela RB
32
36
39
Venezuela RB
32
30
24
Finland
32
57
49
32
35
39
36
Social Security and Welfare
Name
Transfers to Orgs and HH
Name
Rank1
Health
Rank2
Rank3
Name
Education
Rank1
Rank2
Rank3
Honduras
33
21
17
34
Brazil
34
33
31
Zimbabwe
35
23
30
22
Togo
36
17
21
20
Dominican
Republic
37
35
Name
Total Government Expenditure
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Sri Lanka
33
15
21
South Africa
33
29
23
Panama
34
33
22
Paraguay
34
33
Costa Rica
35
36
43
Malaysia
35
37
Tunisia
36
38
31
Cameroon
36
Mexico
37
58
65
Mali
37
Guyana
38
44
41
Gambia, The
38
23
15
Spain
38
62
65
Honduras
38
40
Suriname
39
56
58
Cote d'Ivoire
39
19
32
Iran, Islamic
Rep.
39
34
32
Portugal
39
39
Name
Cyprus
33
41
34
40
Luxembourg
34
28
36
Morocco
34
25
26
29
Nicaragua
35
36
40
Trinidad and
Tobago
35
44
44
22
Thailand
36
35
49
Iceland
36
54
55
43
France
37
29
22
Sri Lanka
37
24
35
48
Panama
38
31
17
44
South Africa
39
27
27
Iran, Islamic
Rep.
Rank1
Rank2
Rank3
33
30
25
Nicaragua
40
28
38
Indonesia
40
22
36
Sri Lanka
40
29
25
Turkey
40
37
47
Spain
40
52
61
Trinidad and
Tobago
41
69
71
Japan
41
45
40
Malaysia
41
47
33
Peru
41
42
56
Cote d'Ivoire
41
28
34
Iran, Islamic
Rep.
42
52
45
Guinea-Bissau
42
20
24
Canada
42
66
69
Gambia, The
42
57
60
Zimbabwe
42
35
38
Togo
43
22
29
Thailand
43
26
26
Romania
43
68
70
Korea, Rep.
43
44
43
Chile
43
43
40
Colombia
44
57
57
Guinea
44
25
28
Singapore
44
59
61
Mexico
44
45
53
Uruguay
44
45
37
Paraguay
45
45
53
Gabon
45
41
42
Uruguay
45
56
51
Colombia
45
43
33
Brazil
45
29
29
GuineaBissau
46
18
16
Jamaica
46
38
37
Cote d'Ivoire
46
39
45
Niger
46
53
37
Malaysia
46
32
41
Morocco
47
37
36
Colombia
47
42
46
Senegal
47
40
30
Syrian Arab
Republic
47
46
45
Gambia, The
47
49
53
Congo, Rep.
48
46
37
Togo
48
28
29
Egypt, Arab
Rep.
48
42
38
Mali
48
48
57
Costa Rica
48
47
46
South Africa
49
59
54
Haiti
49
43
35
Sri Lanka
49
49
62
Australia
49
73
68
Syrian Arab
Republic
50
55
50
Morocco
50
49
52
Greece
50
47
31
Ethiopia
50
38
31
Zimbabwe
51
39
52
Madagascar
51
37
27
Cameroon
51
55
46
Canada
51
69
71
Venezuela
RB
52
72
70
Thailand
52
61
55
Ethiopia
52
50
54
United States
52
61
54
Ethiopia
53
11
8
Denmark
53
71
66
Norway
53
54
39
Guinea
53
42
51
Korea, Rep.
54
67
68
Peru
54
57
59
Uruguay
54
62
52
Mali
54
50
60
Jamaica
55
61
67
Luxembourg
55
75
73
Burkina Faso
55
60
58
Senegal
55
56
50
Senegal
56
42
25
Belgium
56
70
75
Madagascar
56
58
55
Switzerland
56
79
78
Malaysia
57
62
66
Guatemala
57
58
57
Romania
57
73
65
Singapore
57
70
65
Cote d'Ivoire
58
32
35
Ethiopia
58
25
47
Dominican
Republic
58
64
61
Turkey
58
55
57
Honduras
59
48
46
Cameroon
59
50
53
South Africa
59
66
59
Pakistan
59
51
52
Mali
60
27
34
Burkina Faso
60
41
56
Australia
60
56
50
Venezuela RB
60
72
69
Indonesia
61
40
27
Niger
61
48
36
Spain
61
61
66
Indonesia
61
48
43
Cameroon
62
47
42
Mali
62
45
26
Guatemala
62
68
72
Cameroon
62
58
48
Dominican
Republic
63
66
62
Sweden
63
74
72
Indonesia
63
52
42
Mexico
63
65
70
Gambia,
The
64
29
32
Colombia
64
60
67
Paraguay
64
69
70
Japan
64
63
66
Haiti
65
41
33
South Africa
65
69
49
Japan
65
59
67
Haiti
65
62
59
37
Social Security and Welfare
Name
Transfers to Orgs and HH
Rank1
Rank2
Rank3
Guatemala
66
54
51
Madagascar
67
30
28
Name
Rank1
Rank2
Health
Rank3
Name
Education
Rank1
Rank2
Rank3
Turkey
66
64
68
Syrian Arab
Republic
67
63
54
Name
Total Government Expenditure
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Congo, Dem.
Rep.
66
63
64
Thailand
66
60
63
Haiti
67
72
69
Madagascar
67
64
62
Thailand
68
53
61
Mexico
68
67
64
United Kingdom
68
67
73
Niger
68
53
47
Turkey
69
63
59
Paraguay
69
65
62
Brazil
69
65
63
Korea, Rep.
69
68
72
Singapore
70
73
73
Indonesia
70
54
60
Argentina
70
74
68
Peru
70
67
77
Burkina
Faso
71
34
44
Japan
71
76
76
Canada
71
70
71
Honduras
71
66
67
Pakistan
72
31
30
Argentina
72
72
74
Switzerland
72
75
74
Dominican
Republic
72
71
74
Niger
73
24
13
India
73
19
18
United States
73
51
51
Argentina
73
75
75
Congo,
Dem. Rep.
74
20
20
Pakistan
74
53
48
Pakistan
74
76
76
India
74
10
14
Congo, Dem.
Rep.
75
52
58
India
75
6
5
Congo, Dem.
Rep.
75
59
56
Korea, Rep.
76
73
71
Germany
76
71
75
Correlation Matrix
rank1
Correlation Matrix
rank2
rank1
1
rank2
.3931
1
rank3
.3388
.9468
Rank test (p-value)
rank3
1
Correlation Matrix
rank1
rank2
rank1
1
rank2
.5488
1
rank3
.5486
.9565
Rank test (p-value)
rank3
1
76
76
73
77
74
64
Guatemala
78
77
76
Paraguay
79
78
79
rank1
rank2
rank3
Correlation Matrix
rank1
rank1
Colombia
Burkina Faso
rank2
rank3
1
rank1
rank1
rank2
.9064
1
rank3
.8792
.9591
Rank test (p-value)
1
Correlation Matrix
rank2
rank3
1
rank1
rank2
.7470
1
rank3
.6957
.9480
Rank test (p-value)
1
1
rank2
.8058
1
rank3
.8071
.9872
Rank test (p-value)
rank1=rank2
.9063
rank1=rank2
rank1=rank2
.0139
rank1=rank2
.2888
rank1=rank2
.2007
rank2=rank3
.7122
rank2=rank3
rank2=rank3
1
rank2=rank3
1
rank2=rank3
.7163
Orgs and HH = Organizations and households.
38
1
Table 2b: Spending On Social Safety Nets (SS/Total Expenditure)
Social Security and Welfare
Name
Transfers to Orgs and HH
Name
Health
Name
Education
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Uruguay
1
1
1
Uruguay
1
2
1
Iceland
1
1
1
Name
Cote d'Ivoire
1
2
Rank3
2
Germany
2
7
8
France
2
3
10
Costa Rica
2
2
3
Costa Rica
2
1
4
Luxembourg
3
14
18
Luxembourg
3
36
26
Germany
3
4
4
Colombia
3
3
1
Switzerland
4
15
12
Sweden
4
8
8
Panama
4
3
2
Malaysia
4
14
13
Spain
5
2
2
Argentina
5
1
2
France
5
5
6
Honduras
5
10
10
Sweden
6
5
9
Netherlands
6
12
6
Switzerland
6
11
12
Thailand
6
6
15
Austria
7
4
3
Belgium
7
14
15
Ireland
7
9
7
Jamaica
7
16
14
France
8
13
13
Austria
8
4
7
United Kingdom
8
8
10
Tunisia
8
13
21
Belgium
9
17
11
United States
9
16
18
United States
9
12
14
Senegal
9
18
7
Argentina
10
3
4
Spain
10
7
4
Austria
10
20
15
Peru
10
11
12
Denmark
11
16
10
Switzerland
11
24
30
Netherlands
11
13
11
Zimbabwe
11
21
20
Japan
12
20
28
Norway
12
32
37
Italy
12
16
9
Singapore
12
5
5
Netherlands
13
28
39
Canada
13
21
29
Honduras
13
7
5
Mexico
13
22
17
Canada
14
22
22
Poland
14
15
21
Australia
14
15
23
Korea, Rep.
14
15
11
Norway
15
47
43
United Kingdom
15
22
20
Nicaragua
15
6
18
Venezuela RB
15
4
3
Italy
16
36
49
Australia
16
41
42
Dominican
Republic
16
10
17
Burkina Faso
16
17
29
Malta
17
8
5
Israel
17
42
40
Venezuela RB
17
14
22
Niger
17
37
23
Finland
18
51
57
Portugal
18
11
12
Malta
18
19
13
Panama
18
31
37
Chile
19
6
6
Panama
19
5
3
Chile
19
18
32
Morocco
19
24
24
United States
20
55
37
Brazil
20
10
14
Portugal
20
43
50
Guatemala
20
35
33
Australia
21
59
61
Cyprus
21
20
36
Guatemala
21
24
26
Turkey
21
20
18
Brazil
22
10
15
Costa Rica
22
6
9
Jamaica
22
23
16
Togo
22
38
51
United Kingdom
23
45
38
Iceland
23
46
47
Finland
23
30
45
Suriname
23
32
25
Ireland
24
49
55
Tunisia
24
13
11
Greece
24
31
53
Iran, Islamic Rep.
24
27
19
Portugal
25
24
27
Trinidad and
Tobago
25
38
27
Gambia, The
25
17
8
Mali
25
30
38
Greece
26
39
63
Egypt, Arab Rep.
26
18
24
Trinidad and
Tobago
26
34
49
Cameroon
26
45
48
Romania
27
18
17
Denmark
27
47
45
Guinea-Bissau
27
21
30
Belgium
27
12
31
39
Social Security and Welfare
Transfers to Orgs and HH
Health
Name
Education
Name
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Cyprus
28
52
62
Greece
28
37
38
Singapore
28
40
51
Nicaragua
Name
28
39
50
Paraguay
29
12
14
Sri Lanka
29
9
5
Tunisia
29
28
43
Chile
29
25
34
Iceland
30
72
72
Korea, Rep.
30
40
41
Norway
30
52
40
Finland
30
9
6
Mexico
31
38
51
Paraguay
31
17
17
Brazil
31
36
36
Paraguay
31
53
49
Israel
32
69
70
Peru
32
26
22
Burkina Faso
32
22
46
Trinidad and
Tobago
32
28
53
Costa Rica
33
23
29
Nicaragua
33
19
23
Zimbabwe
33
27
29
Iceland
33
26
41
Sri Lanka
34
11
16
Gambia, The
34
23
13
Canada
34
54
56
Dominican
Republic
34
44
44
Colombia
35
43
45
Cameroon
35
29
28
Cyprus
35
62
64
Madagascar
35
47
45
Panama
36
34
32
Malaysia
36
39
33
Haiti
36
29
27
Netherlands
36
23
28
Tunisia
37
44
24
Indonesia
37
35
43
Madagascar
37
25
21
Gambia, The
37
56
55
Nicaragua
38
19
30
South Africa
38
34
32
Kuwait
38
44
37
Ireland
38
42
26
Egypt, Arab
Rep.
39
33
35
Japan
39
48
46
Malaysia
39
53
31
Kuwait
39
8
9
Kuwait
40
73
74
Mali
40
27
16
Thailand
40
51
48
Denmark
40
34
42
Trinidad and
Tobago
41
68
68
Cote d'Ivoire
41
25
34
Suriname
41
37
33
Guyana
41
50
30
Iran, Islamic
Rep.
42
54
56
Thailand
42
31
19
Peru
42
46
34
Ethiopia
42
62
71
Korea, Rep.
43
70
69
Guinea
43
30
31
Israel
43
65
44
Cyprus
43
55
47
Suriname
44
60
65
Colombia
44
45
48
Iran, Islamic Rep.
44
39
35
Egypt, Arab Rep.
44
43
39
Venezuela RB
45
71
71
Gabon
45
44
44
Sri Lanka
45
41
28
Congo, Rep.
45
61
58
Togo
46
25
21
Guinea-Bissau
46
28
35
Senegal
46
26
25
Italy
46
33
32
Guatemala
47
53
42
Jamaica
47
43
39
Togo
47
35
38
Portugal
47
59
62
Togo
48
33
25
Morocco
48
50
47
Guyana
48
38
19
Malta
48
51
56
Honduras
49
35
33
Colombia
49
45
54
Congo, Dem.
Rep.
49
48
52
Indonesia
50
58
52
Spain
50
63
55
Israel
50
54
57
South Africa
51
66
64
Cote d'Ivoire
51
48
41
Sri Lanka
51
65
68
Dominican
Republic
52
61
54
Uruguay
52
55
62
Austria
52
46
36
40
Social Security and Welfare
Name
Transfers to Orgs and HH
Name
Rank1
Rank2
Health
Rank3
Name
Education
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Ethiopia
53
9
7
Congo, Rep.
53
49
57
Name
Sweden
Rank1
Rank2
Rank3
53
29
22
Zimbabwe
54
42
48
Niger
54
47
39
Greece
54
58
54
Senegal
55
40
31
Cameroon
55
50
58
Luxembourg
55
19
16
Cameroon
56
48
53
Paraguay
56
57
42
Guinea-Bissau
56
70
69
Guyana
57
56
50
Romania
57
69
73
Syrian Arab
Republic
57
63
59
Mali
58
27
19
Mali
58
42
20
Indonesia
58
52
46
Malaysia
59
65
66
Ethiopia
59
33
65
Uruguay
59
64
64
27
Burkina Faso
60
26
44
Morocco
60
61
63
France
60
40
Cote d'Ivoire
61
46
41
Mexico
61
64
47
Australia
61
36
35
Congo, Rep.
62
57
58
Denmark
62
70
69
Argentina
62
69
60
Guinea-Bissau
63
21
23
Egypt, Arab Rep.
63
56
60
Norway
63
49
40
Syrian Arab
Republic
64
62
46
Turkey
64
66
70
Haiti
64
73
74
Madagascar
65
32
26
Luxembourg
65
75
67
Spain
65
57
61
70
Haiti
66
41
40
Indonesia
66
59
68
South Africa
66
74
Thailand
67
63
59
Argentina
67
68
74
Japan
67
60
66
Jamaica
68
64
67
South Africa
68
72
59
Romania
68
76
73
Gambia, The
69
29
25
Sweden
69
74
71
Brazil
69
72
72
Singapore
70
74
73
India
70
32
24
Canada
70
66
65
Turkey
71
67
60
Belgium
71
73
76
Switzerland
71
67
63
Pakistan
72
37
36
Congo, Dem. Rep.
72
58
66
United Kingdom
72
71
75
Niger
73
30
20
Syrian Arab
Republic
73
67
61
United States
73
41
43
Congo, Dem.
Rep.
74
31
34
Japan
74
76
75
India
74
7
8
41
Pakistan
75
60
52
Pakistan
75
75
76
Korea, Rep. of
76
71
72
Germany
76
68
67
Social Security and Welfare
Name
Rank1
Rank2
Transfers to Orgs and HH
Rank3
Correlation Matrix
Rank1
Rank1
Rank2
Rank3
Correlation Matrix
Rank2
Rank1
1
Rank2
.5605
1
Rank3
.4924
.9459
Rank test (p-value)
Name
Health
Rank3
1
Name
Rank1
Education
Rank2
Rank3
Correlation Matrix
Rank1
Rank2
Rank1
1
Rank2
.6806
1
Rank3
.6504
.9295
Rank3
1
Rank test (p-value)
Name
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Correlation Matrix
Rank1
Rank2
Rank1
1
Rank2
0.8431
1
Rank3
0.7438
0.9179
Rank test (p-value)
Rank3
1
Rank1
1
Rank2
0.8075
1
Rank3
0.7594
0.9393
Rank test (p-value)
rank1=rank2
.9063
rank1=rank2
.7660
rank1=rank2
0.4096
rank1=rank2
.4160
rank2=rank3
.8043
rank2=rank3
.3368
rank2=rank3
0.5446
rank2=rank3
.4704
Orgs and HH = Organizations and households.
42
1
Table 2c: Spending On Social Safety Nets (SS/Tax Revenues)
Social Security and Welfare
Name
Transfers to Orgs and HH
Name
Health
Name
Education
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Kuwait
1
1
1
Sweden
1
9
6
Kuwait
1
1
1
Kuwait
Name
1
1
Rank3
1
Germany
2
17
16
France
2
7
14
Guinea-Bissau
2
2
2
Guinea-Bissau
2
2
3
Uruguay
3
3
3
Uruguay
3
2
1
Panama
3
3
4
Iran, Islamic
Rep.
3
3
2
Spain
4
4
6
Argentina
4
1
2
Iceland
4
5
5
Cote d'Ivoire
4
6
5
Sweden
5
19
20
Belgium
5
16
16
Costa Rica
5
4
3
Colombia
5
20
11
Romania
6
2
2
Netherlands
6
15
11
Germany
6
8
8
Costa Rica
6
4
4
Switzerland
7
23
17
Austria
7
6
10
Iran, Islamic Rep.
7
6
7
Panama
7
11
23
Austria
8
12
10
Canada
8
14
17
France
8
11
9
Malaysia
8
8
9
Malta
9
8
5
United States
9
19
18
Ireland
9
7
6
Honduras
9
24
31
Luxembourg
10
64
66
Switzerland
10
20
21
Nicaragua
10
15
29
Jamaica
10
13
20
Argentina
11
7
9
Spain
11
11
8
Switzerland
11
16
16
Suriname
11
60
67
10
Belgium
12
35
22
Luxembourg
12
39
38
United Kingdom
12
12
19
Peru
12
12
France
13
26
21
Brazil
13
3
5
United States
13
19
24
Zimbabwe
13
21
18
Denmark
14
25
19
Norway
14
32
37
Italy
14
24
15
Mali
14
38
46
Canada
15
28
29
Poland
15
26
31
Honduras
15
9
10
Morocco
15
17
15
Japan
16
11
14
Panama
16
5
4
Austria
16
21
14
Niger
16
23
22
Italy
17
36
51
Israel
17
34
32
Malta
17
28
22
Tunisia
17
22
26
Brazil
18
5
4
United Kingdom
18
24
22
Netherlands
18
33
39
Thailand
18
7
7
Netherlands
19
50
57
Portugal
19
8
9
Dominican
Republic
19
26
30
Senegal
19
36
29
Chile
20
21
23
Australia
20
41
41
Australia
20
55
64
Mexico
20
19
24
Finland
21
58
64
Cyprus
21
17
36
Gambia, The
21
25
20
Burkina Faso
21
46
57
Norway
22
65
62
Egypt, Arab Rep.
22
10
12
Greece
22
13
26
Turkey
22
10
12
United States
23
54
49
Greece
23
22
27
Jamaica
23
20
17
Singapore
23
28
28
Ireland
24
29
33
Costa Rica
24
12
15
Chile
24
41
60
Syrian Arab
Republic
24
39
35
Portugal
25
9
8
Tunisia
25
13
7
Haiti
25
22
18
Togo
25
40
44
United Kingdom
26
51
48
Sri Lanka
26
4
3
Madagascar
26
29
25
Madagascar
26
33
38
Greece
27
20
34
Iceland
27
45
45
Brazil
27
14
11
Ethiopia
27
64
63
43
Social Security and Welfare
Name
Australia
Transfers to Orgs and HH
Rank1
Rank2
Rank3
28
70
72
Name
Trinidad and
Tobago
Health
Rank1
Rank2
Rank3
28
37
28
Name
Education
Rank1
Rank2
Rank3
Portugal
28
10
12
Rank1
Rank2
Rank3
Nicaragua
Name
28
61
60
Cyprus
29
16
43
Denmark
29
47
43
Guatemala
29
17
27
Guatemala
29
25
32
Iran, Islamic
Rep.
30
15
18
Peru
30
25
23
Venezuela RB
30
57
65
Korea, Rep.
30
18
17
Israel
31
69
68
Nicaragua
31
21
26
Cyprus
31
23
40
Egypt, Arab
Rep.
31
49
50
Guinea-Bissau
32
6
7
Paraguay
32
23
20
Finland
32
54
61
Cameroon
32
41
39
47
Sri Lanka
33
10
13
Korea, Rep.
33
43
42
Zimbabwe
33
27
21
Guyana
33
59
Panama
34
22
26
Guinea-Bissau
34
18
24
Guyana
34
51
35
Gambia, The
34
66
71
Mexico
35
33
53
Cameroon
35
28
29
Suriname
35
68
71
Venezuela RB
35
55
59
Paraguay
36
13
12
Mali
36
27
13
Tunisia
36
34
44
Chile
36
58
65
Iceland
37
71
71
Malaysia
37
38
35
Sri Lanka
37
35
37
Congo, Dem.
Rep.
37
14
13
Costa Rica
38
18
15
South Africa
38
35
30
Burkina Faso
38
42
68
Cyprus
38
16
14
Colombia
39
55
54
Indonesia
39
40
46
Israel
39
65
47
Belgium
39
44
48
Egypt, Arab
Rep.
40
39
39
Gambia, The
40
29
19
Canada
40
62
57
Congo, Rep.
40
51
43
Tunisia
41
46
28
Japan
41
48
47
Singapore
41
56
63
Malta
41
71
69
Nicaragua
42
38
41
Cote d'Ivoire
42
31
39
Peru
42
37
31
Ireland
42
27
19
Syrian Arab
Republic
43
53
31
Thailand
43
33
25
Malaysia
43
30
23
Sri Lanka
43
56
66
Suriname
44
67
69
Guinea
44
30
33
Romania
44
66
59
Finland
44
29
30
Ethiopia
45
14
11
Gabon
45
44
44
Congo, Rep.
45
45
51
Paraguay
45
35
33
Trinidad and
Tobago
46
72
70
Colombia
46
46
48
Togo
46
39
45
Dominican
Republic
46
62
56
Morocco
47
41
40
Jamaica
47
42
40
Norway
47
70
62
Iceland
47
30
25
Togo
48
30
27
Togo
48
36
34
Trinidad and
Tobago
48
61
72
Israel
48
53
58
Mali
49
31
30
Ethiopia
49
48
58
Netherlands
49
47
55
Guyana
50
61
60
Thailand
50
31
34
Italy
50
43
42
Honduras
51
44
45
Mali
51
53
42
Portugal
51
15
16
Korea, Rep.
52
68
65
Senegal
52
43
33
Greece
52
26
21
44
Social Security and Welfare
Name
Transfers to Orgs and HH
Name
Rank1
Rank2
Health
Rank3
Name
Education
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Zimbabwe
53
37
37
Colombia
53
52
53
Trinidad and
Tobago
Name
53
63
72
Guatemala
54
42
46
Niger
54
32
38
Denmark
54
34
45
Venezuela RB
55
73
73
Cote d'Ivoire
55
36
36
Haiti
55
65
64
South Africa
56
52
50
Spain
56
63
52
Romania
56
67
51
Dominican
Republic
57
66
61
Cameroon
57
49
49
Austria
57
31
27
Congo, Rep.
58
57
58
Uruguay
58
59
74
Sweden
58
45
41
53
Senegal
59
49
36
Morocco
59
47
46
Indonesia
59
57
Haiti
60
34
32
Paraguay
60
40
28
Uruguay
60
69
73
Cameroon
61
48
47
Egypt, Arab Rep.
61
69
73
Luxembourg
61
75
75
Indonesia
62
59
59
Mexico
62
58
50
France
62
48
34
Malaysia
63
60
63
Turkey
63
50
48
Argentina
63
50
37
Madagascar
64
45
42
Syrian Arab
Republic
64
71
56
Australia
64
74
74
Cote d'Ivoire
65
40
38
Congo, Dem.
Rep.
65
38
41
South Africa
65
32
40
Burkina Faso
66
43
56
Denmark
66
72
69
Spain
66
54
54
Gambia, The
67
32
44
Argentina
67
64
55
Norway
67
68
61
Jamaica
68
62
67
India
68
18
13
Japan
68
9
8
Thailand
69
47
35
Indonesia
69
60
67
Brazil
69
42
36
Turkey
70
63
55
Luxembourg
70
76
76
Canada
70
73
68
Pakistan
71
56
52
South Africa
71
46
32
Switzerland
71
52
52
Singapore
72
74
74
Sweden
72
74
70
United Kingdom
72
70
70
Niger
73
24
24
Belgium
73
75
75
India
73
5
6
Congo, Dem.
Rep.
74
27
25
Pakistan
74
73
66
Pakistan
74
76
76
Japan
75
44
43
United States
75
37
49
Korea, Rep
76
67
54
Germany
76
72
62
45
Social Security and Welfare
Name
Rank1
Rank2
Transfers to Orgs and HH
Rank3
Correlation Matrix
Name
Rank1
Rank2
Health
Rank3
Correlation Matrix
Rank1
Rank2
Rank1
1
Rank2
0.6419
1
Rank3
0.5765
0.9337
Rank test (p-value)
Rank3
1
Name
Rank1
Education
Rank2
Rank3
Correlation Matrix
Rank1
Rank2
Rank1
1
Rank2
0.7434
1
Rank3
0.6743
0.9313
Rank3
1
Rank test (p-value)
Name
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Correlation Matrix
Rank1
Rank2
Rank1
1
Rank2
0.9341
1
Rank3
0.8614
0.9143
Rank test (p-value)
Rank3
1
Rank1
1
Rank2
0.7163
1
Rank3
0.6848
0.9712
Rank test (p-value)
rank1=rank2
0.8151
rank1=rank2
0.5515
rank1=rank2
1
rank1=rank2
1
rank2=rank3
0.7035
rank2=rank3
0.7493
rank2=rank3
0.6254
rank2=rank3
0.457
Orgs and HH = Organizations and households.
46
1
Table 2d: Spending On Social Safety Nets (SS/GDP), by continent
Social Security and Welfare
Name
Rank1
Rank2
Transfers to Orgs and HH
Rank3
Name
Health
Rank1
Rank2
Rank3
Name
Rank1
Education
Rank2
Rank3
Name
Rank1
Total Government Expenditure
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Latin America
and the
Caribbean
Uruguay
9
3
2
Uruguay
9
5
2
Costa Rica
7
9
12
Guyana
4
3
1
Guyana
1
2
1
Chile
19
19
14
Brazil
19
7
8
Panama
8
6
4
Suriname
5
13
7
Suriname
11
16
11
18
Brazil
24
9
12
Costa Rica
23
13
21
Guyana
14
13
11
Costa Rica
12
14
12
Jamaica
20
12
Argentina
29
49
48
Argentina
24
27
33
Nicaragua
16
12
16
Panama
15
20
18
Nicaragua
26
22
15
Panama
34
33
22
Panama
25
15
6
Jamaica
18
15
14
Chile
25
26
29
Trinidad and
Tobago
35
44
44
Costa Rica
35
36
43
Trinidad and
Tobago
28
40
38
Chile
24
16
28
Trinidad and
Tobago
27
31
41
Panama
38
31
17
Mexico
37
58
65
Nicaragua
30
17
17
Suriname
26
28
20
Venezuela RB
32
30
24
Chile
43
43
40
Guyana
38
44
41
Peru
32
35
39
Trinidad and
Tobago
28
38
46
Nicaragua
35
36
40
Uruguay
44
45
37
Suriname
39
56
58
Paraguay
34
33
34
Venezuela RB
32
36
39
Honduras
38
40
48
Brazil
45
29
29
Nicaragua
40
28
38
Jamaica
46
38
37
Honduras
33
21
17
Peru
41
42
56
Costa Rica
48
47
46
Trinidad and
Tobago
41
69
71
Colombia
47
42
46
Brazil
34
33
40
Mexico
44
45
53
Venezuela RB
60
72
69
Colombia
44
57
57
Dominican
Republic
37
35
43
Colombia
45
43
33
Mexico
63
65
70
Paraguay
45
45
53
Uruguay
45
56
51
Uruguay
54
62
52
Haiti
65
62
59
Venezuela RB
52
72
70
Haiti
49
43
35
Dominican
Republic
58
64
61
Peru
70
67
77
Jamaica
55
61
67
Peru
54
57
59
Guatemala
62
68
72
Honduras
71
66
67
Honduras
59
48
46
Guatemala
57
58
57
Paraguay
64
69
70
Dominican
Republic
72
71
74
Dominican
Republic
63
66
62
Colombia
64
60
67
Haiti
67
72
69
Argentina
73
75
75
Haiti
65
41
33
Mexico
68
67
64
Brazil
69
65
63
Colombia
76
76
73
Guatemala
66
54
51
Paraguay
69
65
62
Argentina
70
74
68
Guatemala
78
77
76
Argentina
72
72
74
Paraguay
79
78
79
Sub-Saharan
Africa
Togo
43
22
29
South Africa
33
29
23
Guinea-Bissau
12
7
9
Cote d'Ivoire
2
5
3
Guinea-Bissau
3
3
3
Guinea-Bissau
46
18
16
Cameroon
36
30
22
Congo, Rep.
30
20
31
Togo
11
12
23
Congo, Rep.
17
7
6
Congo, Rep.
48
46
37
Mali
37
21
20
Gambia, The
31
22
15
Zimbabwe
14
16
19
Togo
24
11
12
South Africa
49
59
54
Gambia, The
38
23
15
Zimbabwe
35
23
29
Guinea-Bissau
17
18
20
Gabon
25
9
9
Zimbabwe
51
39
52
Cote d'Ivoire
39
19
32
Togo
36
17
22
Congo, Rep.
19
22
21
South Africa
39
27
27
Ethiopia
53
11
8
Guinea-Bissau
42
20
24
Cote d'Ivoire
46
39
45
Senegal
21
38
28
Cote d'Ivoire
41
28
34
Senegal
56
42
25
Guinea
44
25
28
Senegal
47
40
30
Gambia, The
42
57
60
Zimbabwe
42
35
38
Cote d'Ivoire
58
32
35
Gabon
45
41
42
Madagascar
51
37
27
Niger
46
53
37
Gambia, The
47
49
53
Mali
60
27
34
Togo
48
28
29
Ethiopia
58
25
47
Mali
48
48
57
Ethiopia
50
38
31
Cameroon
62
47
42
Cameroon
59
50
53
Cameroon
51
55
46
Guinea
53
42
51
47
Social Security and Welfare
Name
Transfers to Orgs and HH
Rank1
Rank2
Rank3
Gambia, The
64
29
32
Madagascar
67
30
28
Name
Rank1
Rank2
Health
Rank3
Name
Education
Rank1
Rank2
Rank3
Burkina Faso
60
41
56
Niger
61
48
36
Name
Total Government Expenditure
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Ethiopia
52
50
54
Mali
54
50
Rank3
60
Burkina Faso
55
60
58
Senegal
55
56
50
48
Burkina Faso
71
34
44
Mali
62
45
26
Madagascar
56
58
55
Cameroon
62
58
Niger
73
24
13
South Africa
65
69
49
South Africa
59
66
59
Madagascar
67
64
62
Congo, Dem.
Rep.
74
20
20
Congo, Dem. Rep.
75
52
58
Congo, Dem.
Rep.
66
63
64
Niger
68
53
47
Congo, Dem. Rep.
75
59
56
Burkina Faso
77
74
64
North America
Canada
21
65
63
United States
15
31
35
United States
17
31
24
Canada
71
70
71
Canada
51
69
71
United States
26
64
55
Canada
16
39
41
Canada
42
66
69
United States
73
51
51
United States
52
61
54
7
Western
Europe
Belgium
1
1
1
Belgium
1
3
5
France
1
2
2
Belgium
1
1
2
Netherlands
4
4
Luxembourg
2
12
19
Netherlands
2
1
1
Netherlands
2
3
3
Netherlands
6
2
6
Belgium
5
6
8
Sweden
3
7
7
France
3
2
3
Germany
3
4
5
Ireland
9
8
4
Italy
9
14
23
Netherlands
4
2
5
Sweden
4
4
4
Ireland
4
1
1
Finland
20
19
15
Ireland
10
15
21
France
5
4
3
Luxembourg
5
32
27
Iceland
5
5
7
Denmark
24
24
26
France
13
18
13
Austria
6
6
4
Austria
6
8
11
United Kingdom
6
8
6
Italy
26
21
32
Luxembourg
14
46
58
Germany
7
10
9
Spain
10
11
9
Austria
9
14
13
Iceland
29
34
38
Sweden
15
39
39
Denmark
8
13
10
Norway
11
34
30
Italy
10
10
8
Sweden
30
25
25
United Kingdom
18
20
22
Italy
11
16
26
United Kingdom
12
14
19
Switzerland
15
46
44
Austria
31
33
30
Austria
21
37
36
Spain
12
8
18
Switzerland
13
43
45
Finland
21
44
50
Cyprus
33
41
34
Denmark
22
36
33
Norway
13
51
39
Cyprus
17
24
25
Norway
25
51
42
Luxembourg
34
28
36
Norway
23
41
42
Ireland
14
14
15
Denmark
20
46
44
Cyprus
29
55
63
France
37
29
22
Germany
28
40
45
Finland
15
60
60
Iceland
22
47
48
Spain
38
62
65
Norway
53
54
39
Cyprus
31
34
30
United
Kingdom
17
23
23
Denmark
53
71
66
Spain
61
61
66
Finland
32
57
49
Switzerland
18
68
64
Luxembourg
55
75
73
United
Kingdom
68
67
73
Iceland
36
54
55
Cyprus
28
50
49
Belgium
56
70
75
Switzerland
72
75
74
Spain
40
52
61
Iceland
30
71
72
Sweden
63
74
72
Germany
76
71
75
Switzerland
56
79
78
Sri Lanka
33
15
21
Pakistan
72
31
30
South Asia
Sri Lanka
29
9
7
Sri Lanka
40
29
25
Sri Lanka
49
49
62
Sri Lanka
37
24
35
India
73
19
18
Pakistan
74
76
76
Pakistan
59
51
52
Pakistan
74
53
48
India
75
6
5
India
74
10
14
Middle East
and North
Africa
Malta
10
5
6
Israel
8
16
12
Malta
11
11
10
Tunisia
7
7
11
Israel
2
1
2
Israel
16
26
47
Portugal
14
10
14
Israel
13
27
19
Israel
8
10
13
Kuwait
6
33
32
Portugal
23
21
24
Tunisia
21
12
16
Kuwait
20
24
21
Morocco
13
15
14
Egypt, Arab Rep.
7
8
5
48
Social Security and Welfare
Name
Transfers to Orgs and HH
Rank1
Rank2
Rank3
Greece
27
43
40
Kuwait
31
74
74
Egypt, Arab
Rep.
32
25
Tunisia
36
38
Iran, Islamic
Rep.
42
Morocco
Syrian Arab
Republic
Name
Health
Rank1
Rank2
Rank3
Egypt, Arab
Rep.
26
18
13
Greece
27
36
18
Name
Education
Rank1
Rank2
Rank3
Portugal
22
26
37
Greece
23
30
17
Tunisia
27
31
Iran, Islamic Rep.
39
52
45
Egypt, Arab Rep.
47
37
36
50
55
50
Name
Total Government Expenditure
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Kuwait
16
9
10
Malta
12
13
20
41
Egypt, Arab
Rep.
18
17
9
Portugal
19
17
19
18
23
Malta
22
27
35
Greece
27
21
10
34
32
Iran, Islamic
Rep.
23
23
17
Syrian Arab
Republic
29
23
28
48
42
38
Portugal
39
39
44
Tunisia
30
26
24
Morocco
50
49
52
Syrian Arab
Republic
47
46
45
Iran, Islamic Rep.
33
30
25
Syrian Arab
Republic
67
63
54
Greece
50
47
31
Morocco
34
25
26
Eastern
Europe and
Central Asia
Romania
22
17
11
Turkey
69
63
59
Poland
7
6
10
Romania
43
68
70
Turkey
40
37
47
Poland
8
5
4
Turkey
66
64
68
Romania
57
73
65
Romania
16
19
16
Turkey
58
55
57
41
East Asia and
the Pacific
Japan
20
35
56
Australia
18
48
47
Australia
19
32
34
Malaysia
10
11
16
Malaysia
46
32
Australia
25
70
69
Korea, Rep.
31
44
43
Malaysia
41
47
33
Singapore
28
32
27
Australia
49
73
68
Korea, Rep.
54
67
68
Malaysia
35
37
31
Singapore
44
59
61
Thailand
36
35
49
Singapore
57
70
65
Malaysia
57
62
66
Indonesia
40
22
36
Thailand
52
61
55
Korea, Rep.
43
44
43
Indonesia
61
48
43
Indonesia
61
40
27
Japan
41
45
40
Indonesia
70
54
60
Australia
60
56
50
Japan
64
63
66
Thailand
43
26
26
Thailand
68
53
61
Singapore
70
73
73
Japan
71
76
76
Indonesia
63
52
42
Thailand
66
60
63
Korea, Rep.
76
73
71
Japan
65
59
67
Korea, Rep.
69
68
72
Orgs and HH = Organizations and households.
49
Table 2e: Spending On Social Safety Nets (SS/Total Expenditure), by continent
Social Security and Welfare
Name
Transfers to Orgs and HH
Rank1
Rank2
Rank3
•
•
•
•
Uruguay
1
1
1
Argentina
10
3
4
Chile
19
6
Brazil
22
10
Paraguay
29
12
Mexico
31
Costa Rica
33
Colombia
Health
Name
Education
Rank1
Rank2
Rank3
Uruguay
1
2
1
Argentina
5
1
2
6
Panama
19
5
15
Brazil
20
10
14
Costa Rica
22
6
9
38
51
Trinidad and Tobago
25
38
27
Venezuela RB
17
14
22
23
29
Paraguay
31
17
17
Chile
19
18
32
35
43
45
Peru
32
26
22
Guatemala
21
24
26
Panama
36
34
32
Nicaragua
33
19
23
Jamaica
22
23
Nicaragua
38
19
30
Colombia
44
45
48
Trinidad and Tobago
26
34
Trinidad and Tobago
41
68
68
Jamaica
47
43
39
Brazil
31
Suriname
44
60
65
Haiti
Venezuela RB
45
71
71
Suriname
Guatemala
47
53
42
Honduras
49
35
Dominican Republic
52
61
Guyana
57
56
Latin America and the
Caribbean
Name
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Costa Rica
2
2
3
Costa Rica
2
1
4
Panama
4
3
2
Colombia
3
3
1
3
Honduras
13
7
5
Honduras
5
10
10
14
Nicaragua
15
6
18
Jamaica
7
16
14
Dominican Republic
16
10
17
Peru
10
11
12
Mexico
13
22
17
Venezuela RB
15
4
3
Panama
18
31
37
16
Guatemala
20
35
33
49
Suriname
23
32
25
36
36
Nicaragua
28
39
50
36
29
27
Chile
29
25
34
41
37
33
Paraguay
31
53
49
Peru
42
46
34
Trinidad and Tobago
32
28
53
33
Guyana
48
38
19
Dominican Republic
34
44
44
54
Colombia
49
45
54
Guyana
41
50
30
50
Uruguay
52
55
62
Uruguay
59
64
64
Haiti
66
41
40
Paraguay
56
57
42
Argentina
62
69
60
Jamaica
68
64
67
Mexico
61
64
47
Haiti
64
73
74
Argentina
67
68
74
Brazil
69
72
72
2
Sub-Saharan Africa
Togo
46
25
21
Gambia, The
34
23
13
Gambia, The
25
17
8
Cote d'Ivoire
1
2
South Africa
51
66
64
Cameroon
35
29
28
Guinea-Bissau
27
21
30
Senegal
9
18
7
Ethiopia
53
9
7
South Africa
38
34
32
Burkina Faso
32
22
46
Zimbabwe
11
21
20
Zimbabwe
54
42
48
Mali
40
27
16
Zimbabwe
33
27
29
Burkina Faso
16
17
29
Senegal
55
40
31
Cote d'Ivoire
41
25
34
Madagascar
37
25
21
Niger
17
37
23
Cameroon
56
48
53
Guinea
43
30
31
Senegal
46
26
25
Togo
22
38
51
Mali
58
27
19
Gabon
45
44
44
Togo
47
35
38
Mali
25
30
38
Burkina Faso
60
26
44
Guinea-Bissau
46
28
35
Cote d'Ivoire
51
48
41
Cameroon
26
45
48
Cote d'Ivoire
61
46
41
Togo
48
33
25
Congo, Rep.
53
49
57
Madagascar
35
47
45
Congo, Rep.
62
57
58
Niger
54
47
39
Gambia, The
37
56
55
Guinea-Bissau
63
21
23
Cameroon
55
50
58
Ethiopia
42
62
71
Madagascar
65
32
26
Mali
58
42
20
Congo, Rep.
45
61
58
Gambia, The
69
29
25
Ethiopia
59
33
65
Congo, Dem. Rep.
49
48
52
Niger
73
30
20
South Africa
68
72
59
Guinea-Bissau
56
70
69
Congo, Dem. Rep.
74
31
34
Congo, Dem. Rep.
72
58
66
South Africa
66
74
70
50
Social Security and Welfare
Name
Transfers to Orgs and HH
Name
Rank1
Health
Rank2
Rank3
Name
Rank1
Education
Rank1
Rank2
Rank3
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Canada
14
22
22
United States
9
16
18
United States
9
12
14
Canada
70
66
65
United States
20
55
37
Canada
13
21
29
Canada
34
54
56
United States
73
41
43
31
North America
Western Europe
Germany
2
7
8
France
2
3
10
Iceland
1
1
1
Belgium
27
12
Luxembourg
3
14
18
Luxembourg
3
36
26
Germany
3
4
4
Finland
30
9
6
Switzerland
4
15
12
Sweden
4
8
8
France
5
5
6
Iceland
33
26
41
28
Spain
5
2
2
Netherlands
6
12
6
Switzerland
6
11
12
Netherlands
36
23
Sweden
6
5
9
Belgium
7
14
15
Ireland
7
9
7
Ireland
38
42
26
Austria
7
4
3
Austria
8
4
7
United Kingdom
8
8
10
Denmark
40
34
42
47
France
8
13
13
Spain
10
7
4
Austria
10
20
15
Cyprus
43
55
Belgium
9
17
11
Switzerland
11
24
30
Netherlands
11
13
11
Italy
46
33
32
Denmark
11
16
10
Norway
12
32
37
Italy
12
16
9
Austria
52
46
36
22
Netherlands
13
28
39
United Kingdom
15
22
20
Finland
23
30
45
Sweden
53
29
Norway
15
47
43
Cyprus
21
20
36
Norway
30
52
40
Luxembourg
55
19
16
Italy
16
36
49
Iceland
23
46
47
Cyprus
35
62
64
France
60
40
27
Denmark
27
47
45
Finland
18
51
57
Spain
50
63
55
Norway
63
49
40
United Kingdom
23
45
38
Denmark
62
70
69
Spain
65
57
61
Ireland
24
49
55
Luxembourg
65
75
67
Switzerland
71
67
63
Cyprus
28
52
62
Sweden
69
74
71
United Kingdom
72
71
75
Iceland
30
72
72
Belgium
71
73
76
Germany
76
68
67
Sri Lanka
34
11
16
68
Pakistan
72
37
36
South Asia
Sri Lanka
29
9
5
Sri Lanka
45
41
28
Sri Lanka
51
65
India
70
32
24
India
74
7
8
Pakistan
75
60
52
Pakistan
75
75
76
Middle East and North
Africa
Malta
17
8
5
Israel
17
42
40
Malta
18
19
13
Tunisia
8
13
21
Portugal
25
24
27
Portugal
18
11
12
Portugal
20
43
50
Morocco
19
24
24
19
Greece
26
39
63
Tunisia
24
13
11
Greece
24
31
53
Iran, Islamic Rep.
24
27
Israel
32
69
70
Egypt, Arab Rep.
26
18
24
Tunisia
29
28
43
Kuwait
39
8
9
Tunisia
37
44
24
Greece
28
37
38
Kuwait
38
44
37
Egypt, Arab Rep.
44
43
39
62
Egypt, Arab Rep.
39
33
35
Israel
43
65
44
Portugal
47
59
Kuwait
40
73
74
Iran, Islamic Rep.
44
39
35
Malta
48
51
56
Iran, Islamic Rep.
42
54
56
Morocco
60
61
63
Israel
50
54
57
Morocco
48
50
47
Egypt, Arab Rep.
63
56
60
Greece
54
58
54
Syrian Arab Republic
64
62
46
Syrian Arab Republic
73
67
61
Syrian Arab Republic
57
63
59
Eastern Europe and
Central Asia
51
Social Security and Welfare
Name
Transfers to Orgs and HH
Name
Health
Rank1
Rank2
Rank3
Romania
27
18
17
Turkey
71
67
60
Japan
12
20
28
Australia
16
41
Australia
21
59
61
Korea, Rep.
30
40
Korea, Rep.
43
70
69
Malaysia
36
Indonesia
50
58
52
Indonesia
Poland
Rank1
Rank2
Rank3
14
15
21
Name
Education
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Romania
57
69
73
Turkey
21
20
Rank3
Turkey
64
66
70
Romania
68
76
73
42
Australia
14
15
23
Malaysia
4
14
13
41
Singapore
28
40
51
Thailand
6
6
15
39
33
Malaysia
39
53
31
Singapore
12
5
5
37
35
43
Thailand
40
51
48
Korea, Rep.
14
15
11
46
18
East Asia and the
Pacific
Malaysia
59
65
66
Japan
39
48
46
Indonesia
66
59
68
Indonesia
58
52
Thailand
67
63
59
Thailand
42
31
19
Japan
74
76
75
Australia
61
36
35
Korea, Rep.
76
71
72
Japan
67
60
66
Singapore
70
74
73
Orgs and HH = Organizations and households.
52
Table 2f: Spending On Social Safety Nets (SS/Tax Revenues), by continent
Social Security and Welfare
Name
Transfers to Orgs and HH
Rank3
Name
Rank1
Health
Rank1
Rank2
Rank2
Rank3
Uruguay
3
3
3
Uruguay
3
2
1
Argentina
11
7
9
Argentina
4
1
2
Brazil
18
5
4
Brazil
13
3
5
Name
Education
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Panama
3
3
4
Costa Rica
5
4
3
Colombia
5
20
11
Costa Rica
6
4
Nicaragua
10
15
29
Panama
4
7
11
23
Latin America and the
Caribbean
Chile
20
21
23
Panama
16
5
4
Honduras
15
9
10
Honduras
9
24
31
Panama
34
22
26
Costa Rica
24
12
15
Dominican Republic
19
26
30
Jamaica
10
13
20
México
35
33
53
Trinidad and Tobago
28
37
28
Jamaica
23
20
17
Suriname
11
60
67
Paraguay
36
13
12
Peru
30
25
23
Chile
24
41
60
Peru
12
12
10
24
Costa Rica
38
18
15
Nicaragua
31
21
26
Haiti
25
22
18
Mexico
20
19
Colombia
39
55
54
Paraguay
32
23
20
Brazil
27
14
11
Nicaragua
28
61
60
Nicaragua
42
38
41
Colombia
46
46
48
Guatemala
29
17
27
Guatemala
29
25
32
Suriname
44
67
69
Jamaica
47
42
40
Venezuela RB
30
57
65
Guyana
33
59
47
Trinidad and Tobago
46
72
70
Guyana
34
51
35
Venezuela RB
35
55
59
Guyana
50
61
60
Suriname
35
68
71
Chile
36
58
65
Honduras
51
44
45
Peru
42
37
31
Paraguay
45
35
33
Guatemala
54
42
46
Trinidad and Tobago
48
61
72
Dominican Republic
46
62
56
Venezuela RB
55
73
73
Colombia
53
52
53
Trinidad and Tobago
53
63
72
Dominican Republic
57
66
61
Uruguay
58
59
74
Haiti
55
65
64
Haití
60
34
32
Paraguay
60
40
28
Uruguay
60
69
73
Jamaica
68
62
67
Mexico
62
58
50
Argentina
63
50
37
Argentina
67
64
55
Brazil
69
42
36
Sub-Saharan Africa
Guinea-Bissau
32
6
7
Guinea-Bissau
34
18
24
Guinea-Bissau
2
2
2
Guinea-Bissau
2
2
3
Etiopía
45
14
11
Cameroon
35
28
29
Gambia, The
21
25
20
Cote d'Ivoire
4
6
5
Togo
48
30
27
Mali
36
27
13
Madagascar
26
29
25
Zimbabwe
13
21
18
Mali
49
31
30
South Africa
38
35
30
Zimbabwe
33
27
21
Mali
14
38
46
Zimbabwe
53
37
37
Gambia, The
40
29
19
Burkina Faso
38
42
68
Niger
16
23
22
South Africa
56
52
50
Cote d'Ivoire
42
31
39
Congo, Rep.
45
45
51
Senegal
19
36
29
Congo, Rep.
58
57
58
Guinea
44
30
33
Togo
46
39
45
Burkina Faso
21
46
57
44
Senegal
59
49
36
Gabon
45
44
44
Ethiopia
49
48
58
Togo
25
40
Cameroon
61
48
47
Togo
48
36
34
Mali
51
53
42
Madagascar
26
33
38
Madagascar
64
45
42
Senegal
52
43
33
Ethiopia
27
64
63
Cote d'Ivoire
65
40
38
Niger
54
32
38
Cameroon
32
41
39
Burkina Faso
66
43
56
Cote d'Ivoire
55
36
36
Gambia, The
34
66
71
Gambia, The
67
32
44
Cameroon
57
49
49
Congo, Dem. Rep.
37
14
13
Niger
73
24
24
Congo, Dem. Rep.
65
38
41
Congo, Rep.
40
51
43
Congo, Dem. Rep.
74
27
25
South Africa
71
46
32
South Africa
65
32
40
53
Social Security and Welfare
Name
Transfers to Orgs and HH
Rank1
Rank2
Rank3
Canada
15
28
29
United States
23
54
49
Germany
2
17
Spain
4
4
Sweden
5
Switzerland
7
Austria
8
Name
Health
Name
Education
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
United States
13
19
24
Canada
40
62
57
Name
Rank1
Rank2
Rank3
Canada
8
14
17
United States
9
19
18
Canada
70
73
68
United States
75
37
49
16
Sweden
1
9
6
Iceland
4
5
5
Cyprus
38
16
14
6
France
2
7
14
Germany
6
8
8
Belgium
39
44
48
19
20
Belgium
5
16
16
France
8
11
9
Ireland
42
27
19
23
17
Netherlands
6
15
11
Ireland
9
7
6
Finland
44
29
30
12
10
Austria
7
6
10
Switzerland
11
16
16
Iceland
47
30
25
North America
Western Europe
Luxembourg
10
64
66
Switzerland
10
20
21
United Kingdom
12
12
19
Netherlands
49
47
55
Belgium
12
35
22
Spain
11
11
8
Italy
14
24
15
Italy
50
43
42
France
13
26
21
Luxembourg
12
39
38
Austria
16
21
14
Denmark
54
34
45
Denmark
14
25
19
Norway
14
32
37
Netherlands
18
33
39
Austria
57
31
27
41
Italy
17
36
51
United Kingdom
18
24
22
Cyprus
31
23
40
Sweden
58
45
Netherlands
19
50
57
Cyprus
21
17
36
Finland
32
54
61
Luxembourg
61
75
75
Finland
21
58
64
Iceland
27
45
45
Norway
47
70
62
France
62
48
34
Norway
22
65
62
Denmark
29
47
43
Spain
56
63
52
Spain
66
54
54
Ireland
24
29
33
Denmark
66
72
69
Norway
67
68
61
United Kingdom
26
51
48
Luxembourg
70
76
76
Switzerland
71
52
52
Cyprus
29
16
43
Sweden
72
74
70
United Kingdom
72
70
70
Iceland
37
71
71
Belgium
73
75
75
Germany
76
72
62
66
South Asia
Sri Lanka
33
10
13
Pakistan
71
56
52
Sri Lanka
26
4
3
Sri Lanka
37
35
37
Sri Lanka
43
56
India
68
18
13
India
73
5
6
Pakistan
74
73
66
Pakistan
74
76
76
1
Middle East and North Africa
Kuwait
1
1
1
Israel
17
34
32
Kuwait
1
1
1
Kuwait
1
1
Malta
9
8
5
Portugal
19
8
9
Iran, Islamic Rep.
7
6
7
Iran, Islamic Rep.
3
3
2
Portugal
25
9
8
Egypt, Arab Rep.
22
10
12
Malta
17
28
22
Morocco
15
17
15
Greece
27
20
34
Greece
23
22
27
Greece
22
13
26
Tunisia
17
22
26
Iran, Islamic Rep.
30
15
18
Tunisia
25
13
7
Portugal
28
10
12
Syrian Arab Republic
24
39
35
50
Israel
31
69
68
Tunisia
36
34
44
Egypt, Arab Rep.
31
49
Egypt, Arab Rep.
40
39
39
Israel
39
65
47
Malta
41
71
Tunisia
41
46
Syrian Arab Republic
43
53
Morocco
47
41
28
Morocco
59
47
46
31
Egypt, Arab Rep.
61
69
40
Syrian Arab Republic
64
71
Eastern Europe and Central
Asia
54
69
Israel
48
53
58
73
Portugal
51
15
16
56
Greece
52
26
21
Social Security and Welfare
Name
Transfers to Orgs and HH
Name
Health
Rank1
Rank2
Rank3
Romania
6
2
2
Turkey
70
63
55
Japan
16
11
14
Australia
20
41
Australia
28
70
72
Korea, Rep.
33
43
Poland
Rank1
Rank2
Rank3
15
26
31
Name
Education
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Romania
44
66
59
Turkey
22
10
12
Turkey
63
50
48
Romania
56
67
51
41
Australia
20
55
64
Malaysia
8
8
9
42
Singapore
41
56
63
Thailand
18
7
7
East Asia and the Pacific
Korea, Rep.
52
68
65
Malaysia
37
38
35
Malaysia
43
30
23
Singapore
23
28
28
Indonesia
62
59
59
Indonesia
39
40
46
Thailand
50
31
34
Korea, Rep.
30
18
17
Malaysia
63
60
63
Japan
41
48
47
Indonesia
69
60
67
Indonesia
59
57
53
Thailand
69
47
35
Thailand
43
33
25
Japan
75
44
43
Australia
64
74
74
Singapore
72
74
74
Korea, Rep.
76
67
54
Japan
68
9
8
Orgs and HH = Organizations and households.
55
At the bottom of table 2a, we report the correlation matrices between R1, R2, and R3 for
each of the types of safety nets. It is interesting to note that, first, the unconditional rank is
not highly correlated with R2 and R3 for social security and transfers. This indicates that
only looking at the unconditional shares spent by governments on safety nets can be
misleading—countries may be spending more or less than we may expect them to, and R2
and R3 allow us to judge the extent to which this is the case. On the whole, R2 and R3 are
highly correlated, although it is typically in exactly those cases when they are not that the
most can be inferred for policy prescriptions. The fact that, in our sample, R2 and R3 are so
highly correlated comes in part from the fact that our measures of institutional quality have
had to be averaged over the entire sample period.
Below the correlation matrices, we report a test of the equality of the ranks using the test
described in Snedecor and Cochrane (1989). The null hypothesis of the test is that the median
of the differences in rank across countries is zero. The test makes no further assumptions. We
report the p-value of the test at the foot of tables 2a to 2c. We generally find no significant
differences across the ranks. On the whole, given the relatively small sample and the lack of
movement of countries, particularly at the highest ranks, this is not altogether surprising. It is
of more interest to look at the movements of individual countries, as opposed to the ranking
as a whole.
So far, we have been expressing safety net expenditures as a share of GDP. In tables 2b
and 2c, we consider the robustness of our results using two alternative normalizations: (i)
expenditures as a share of total government expenditures, and (ii) expenditures as a share of
total government tax revenues. We use these to further capture the notion of a government
budget constraint. On the whole we find our rankings to be largely unchanged when we use
any of these normalizations, although certain oil-dependent economies such as Kuwait
perform better when we take expenditures as a share of total tax revenues. The correlation
between R1 and the other rankings tends to rise when we use total expenditures as the
numeraire, and tend to fall if we use tax revenues.
Tables 2d to 2e repeat the results from tables 2a to 2c except that rankings are given
within each regional grouping. We have done this in order to facilitate analysis of close
neighbors, in the spirit of benchmark competition. Concentrating on table 2d, where we use
safety net expenditures as a share of GDP, we see that Argentina, relative to its neighbors,
performs well with a rank of R1 when we only look at unconditional social security
expenditures. Taking account of structural features, we find that it performs far worse
relative to what Latin American countries are able to do on average. However, once its
institutional features are also taken into account, it again performs relatively well. Other
interesting cases are Indonesia, which appears to perform better than most of its neighbors
when institutional quality is controlled for, and Kuwait, which moves in the opposite
direction relative to its neighbors as we move from R1 to R3. In sub-Saharan Africa, it is
Senegal that appears to have an improved relative performance once institutional quality is
controlled for, and Mali which generally moves down as we move to R3.
Graphical Representation
The information contained in ranks one to three can also be summarized in a simple graph, as
given in figure 1. The graphs show the movements across the three rankings for countries
with respect to the different components of social spending, where the spending measures are
56
expressed as fractions of GDP. The graphs in the leftmost panels plot the R1 ranking against
the R2 ranking, while the right panels plot R2 against R3. The countries far away from the
leading diagonal (diagonal lines not shown) have the greatest shifts moving from one rank to
another. From figure 1 we see clearly that there is more dispersion moving from rank one to
rank two than from rank two to rank three on all spending measures. If a country lies a long
way below the leading diagonal, then it fallen in ranking after controlling for structural (in
the case of R2) or quality of government (in the case of R3). Conversely, if a country lies
above the diagonal, it has improved its ranking.
The topmost panels in figure 1 plot rankings for social security and welfare
expenditures. There is quite wide dispersion evident in the left top panel, indicating large
changes in ranking moving from R1 to R2. For example, Norway, Canada, Cyprus, Finland,
Iceland and Kuwait, among others, all appear far below the diagonal line. This suggests that
after controlling for structural characteristics, these countries perform far worse than we
would expect relative to the international norm—i.e., rankings fall dramatically from R1 to
R2.
Countries that lie above the leading diagonal have the opposite interpretation—they
spend more on social security and welfare than we would expect, controlling for structural
characteristics. Countries in this category for social security and welfare spending include Sri
Lanka, Senegal, and Niger.
The right topmost panel plots rank two against rank three on social security and welfare
spending. As most countries lie close to the leading diagonal here, we see that the correlation
between the two ranks is high. The data on quality of government is simply not able to
introduce that much variation. Nevertheless, in the case of social security and welfare (the
top rightmost panel) it would suggest that countries such as Italy, Israel, Zimbabwe, and Mali
all do worse than we would expect given their level of institutional quality. Panama,
Indonesia, and Niger appear to do better than the international norm, controlling for these
institutional factors in addition to structural characteristics.
The next two panels in figure give the same ranking information for expenditures on
transfers, again expressed as a fraction of GDP. Consistent with the correlation matrix shown
in table 2a, we see that most of the movement is from rank one to rank two. The panels
highlight that countries such as Switzerland and Australia do worse than expected given
structural characteristics, and countries such as Egypt and Costa Rica do better than
expected. Similarly, the last two sets of panels show the rankings for health and education
expenditures as a share of GDP.
The third set of panels in figure 1 presents rankings for education expenditures. As most
countries lie close to the leading diagonals, we see that there are not many strong movements
across the ranks. However, there are some notable exceptions—Malta, Bahrain, and Bolivia
all devote a greater share of GDP to education than we would expect given the structural
features of their economy. India, Indonesia, and the United States devote fewer resources
than we would expect. Controlling for institutional quality in the bottom left panel suggests
that countries such as Italy, Thailand, and Peru perform worse than expected controlling for
institutions. Greece, Egypt, and Norway are among the countries that perform better once
institutions are controlled for.
57
Figure 1: Cross-Country Movements Across Rankings
Social Security and Welfare (share of GDP)
Transfers to Organizations and Households (share of GDP)
Education (share of GDP)
Health (share of GDP)
58
The final set of panels in figure 1 give the corresponding information for health
expenditures. Again, most of the movement is from rank one to rank two. This is confirmed
by the correlation coefficients for the health rankings given in table 2a.
Regression Results
Table 3 reports the actual regression results from (1) and (2), which are used to construct R2
and R3. These results are instructive in and of themselves. The R2 regression (equation (1))
can be thought of as estimating a demand function for safety net expenditure. We see that, for
each type of safety net expenditure as a share of GDP, income is positive and significant.
This suggests that as countries grow richer they spend a greater share of GDP on types of
safety nets. The shock-to-GDP variable is also significant and negative, suggesting that
cyclical factors play a role in determining safety net spending. In particular, safety net
spending appears to be anti-cyclical, as we would expect. The other variables capture societal
needs. The fraction of the population of working age tends to be negative and significant only
for social security. The level of urbanization significantly increases both social security and
transfers spending. This may suggest that the urban poor are better able to get the
government to respond to their interests than are the rural poor.
In table 3, we also report regression (2), which is used to construct R3. This regresses
the time-averaged residual from (1) on various quality-of-government measures. We see that
the increased threat of expropriation and corruption significantly increase government safety
net expenditures, while worsening rule of law and increased government bureaucracy
decrease expenditures. The fact that expenditures may increase as quality of government
decreases may suggest that there is some rent-seeking behavior occurring, or some other nonproductive use of funds.
Table 3: Cross-Country Safety Net Regressions (standard errors in parentheses)
Dependent Var SS/GDP
R2 Regressions: sijt = βXit + γZi +vijt
Independent Variables
Log(GDP per capita)
Shock to GDP per capita
Frac. of popn aged 15-64
(millions)
Frac. of popn residing in
urban regions
Adjusted R-squared
Observations
Social
Spending
Transfers to
Orgs & HH
Health
Education
Total
Expenditure
2.70
2.83
0.641
0.041
2.47
(0.288)
(0.560)
(0.122)
(0.134)
(0.77)
–13.8
–16.7
0.048
–0.644
–25.4
(1.98)
(3.78)
(0.849)
(0.931)
(5.23)
–0.018
–0.010
–0.003
–0.012
–0.045
(0.010)
(0.019)
(0.002)
(0.002)
(0.013)
0.032
0.050
–0.014
–0.004
0.056
(0.018)
(0.033)
(0.008)
(0.008)
(0.048)
0.6166
0.5497
0.1715
0.1058
0.1937
256
101
266
266
303
59
R3 Regressions: v∧ij. = γ + δQi. + ωij.
(dependent variable is time-averaged residual from R2 regression)
Independent Variables
Repudiation of
Government Contract
Expropriation Threat
Corruption
Rule of Law
Government
Bureaucracy
Constant
Adjusted R-squared
Observations
Social
Spending
Transfers to
Orgs & HH
Health
Education
Total
Expenditure
0.048
0.782
0.507
0.291
–0.626
(0.394)
(0.740)
(0.164)
(0.159)
(0.981)
0.328
–0.905
–0.264
–0.388
–0.129
(0.401)
(0.816)
(0.166)
(0.161)
(1.04)
0.336
1.69
0.030
–0.233
0.769
(0.360)
(0.766)
(0.196)
(0.146)
(0.960)
–0.873
0.194
–0.025
–0.072
–0.807
(0.406)
(0.882)
(0.161)
(0.156)
(1.03)
–0.368
–0.955
–0.294
0.136
0.315
(0.401)
(0.709)
(0.166)
(0.161)
(1.04)
–6.15
–4.65
–2.29
0.739
–1.02
(3.03)
(6.85)
(0.125)
(1.208)
(7.94)
0.0295
0.0233
0.0243
0.0527
0.0000
224
90
233
233
265
Dependent Var SS/Total Government Expenditure
R2 Regressions: sijt = βXit + γZi +vijt
Independent Variables
Log(GDP per capita)
Shock to GDP per capita
Frac. of popn aged 15–64
(millions)
Frac. of popn residing in
urban regions
Adjusted R–squared
Observations
Social Spending
Transfers to Orgs
& HH
Health
Education
0.068
0.054
0.015
–0.009
(0.008)
(0.013)
(0.003)
(0.004)
–0.210
–0.334
0.056
0.108
(0.054)
(0.088)
(0.024)
(0.026)
0.0001
0.0001
–0.0001
–0.0004
(0.0003)
(0.0004)
(0.0001)
(0.0001)
0.0008
0.002
–0.001
–0.0005
(0.0005)
(0.001)
(0.0002)
(0.0002)
0.5845
0.5366
0.1259
0.2982
256
100
266
266
60
R3 Regressions: v∧ij. = γ + δQi. + ωij.
(dependent variable is time-averaged residual from R2 regression)
Independent Variables
Social Spending
Transfers to Orgs
& HH
Health
Education
–0.005
–0.007
0.021
0.018
(0.011)
(0.017)
(0.005)
(0.005)
0.030
0.012
–0.012
–0.018
(0.011)
(0.019)
(0.005)
(0.005)
0.010
0.040
–0.003
–0.009
(0.010)
(0.017)
(0.004)
(0.004)
–0.019
0.008
0.001
0.003
(0.011)
(0.020)
(0.005)
(0.005)
–0.025
–0.044
–0.006
0.005
(0.011)
(0.016)
(0.005)
(0.005)
–0.242
–0.224
–0.053
–0.012
(0.081)
(0.156)
(0.036)
(0.037)
0.0465
0.0629
0.0871
0.0776
224
89
233
233
Repudiation of
Government Contract
Expropriation Threat
Corruption
Rule of Law
Government
Bureaucracy
Constant
Adjusted R-squared
Observations
Dependent Var SS/Total Tax Revenues
R2 Regressions: sijt = βXit + γZi +vijt
Independent
Variables
Log(GDP per
capita)
Shock to GDP per
capita
Frac. of popn aged
15–64 (millions)
Frac. of popn
residing in urban
regions
Adjusted R-squared
Observations
Social Spending
Transfers to Orgs
& HH
Health
Education
0.066
0.061
0.010
–0.033
(0.010)
(0.016)
(0.006)
(0.009)
–0.268
–0.312
0.007
0.003
(0.070)
(0.110)
(0.043)
(0.061)
0.0003
0.0004
–0.0002
–0.0001
(0.0004)
(0.001)
(0.0001)
(0.0001)
0.002
0.002
–0.0003
0.0002
(0.0006)
(0.001)
(0.0004)
(0.001)
0.5270
0.4847
0.0176
0.1941
256
101
266
266
61
R3 Regressions: v∧ij. = γ + δQi. + ωij.
(dependent variable is time-averaged residual from R2 regression)
Independent
Variables
Repudiation of
Government
Contract
Expropriation
Threat
Corruption
Rule of Law
Government
Bureaucracy
Constant
Adjusted R-squared
Observations
Social Spending
Transfers to Orgs
& HH
Health
Education
–0.011
–0.001
0.024
0.011
(0.014)
(0.021)
(0.009)
(0.012)
0.041
0.014
–0.011
–0.012
(0.014)
(0.023)
(0.009)
(0.012)
0.011
0.050
–0.010
–0.014
(0.012)
(0.022)
(0.008)
(0.011)
–0.011
0.009
0.004
–0.001
(0.014)
(0.025)
(0.008)
(0.012)
–0.040
–0.064
–0.006
0.011
(0.014)
(0.020)
(0.009)
(0.012)
–0.273
–0.336
–0.040
0.060
(0.105)
(0.193)
(0.065)
(0.089)
0.0576
0.0792
0.0176
0.0000
224
90
266
233
Note: All first-stage regressions also include a set of continental dummies, the omitted group being Western
Europe.
The remainder of table 3 gives the same set of results with safety nets expressed as
shares of total government expenditures and total tax revenues, respectively. The results are
broadly in line with the previous ones. Income continues to have positive and significant
effects on safety net spending, as does the level of urbanization; when expenditures are
normalized by tax revenues, they are also found to be anti-cyclical. Both social security and
transfers increase with the size of the working-age population.
In terms of the quality-of-institutions regression (2), social security increases as the
threat of expropriation increases and decreases significantly as the rule of law worsens or
government bureaucracy increases. Transfers decrease as government bureaucracy or
corruption worsens. Overall, it is clear that institutional quality can explain some of the
variation in safety net expenditures that cannot be explained when we control only for
structural features of the economy. Even using such poor data with no variation over time,
we find these effects to be present.
Similarly, it is also reassuring to find that, on the whole, the results are robust to the
exact normalization used to measure social safety net expenditures, be it as a share of GDP,
total expenditures, or total tax revenues.
62
To summarize, although at first glance it is somewhat surprising that we find weak
income effects on the whole, this may not be altogether unexpected given that we have
dependent variables expressed as shares. This is also true for the lack of cyclicality that we
find in most safety net expenditures, and generally we find that safety nets do respond to the
level of need in a country.
On the institutional side, the fact that we find some expenditures increasing with
corruption may well be indicative of rent-seeking behavior in that, if government officials are
corrupt, they may make payments to themselves by inflating expenditure levels. The fact that
we control for this in (2) ensures that this possibility is also accounted for in R3. The single
best measure of institutional quality appears to be government bureaucracy, whereas the
repudiation of government contracts has no significant effect on any type of safety net. Given
that our measure of government bureaucracy may be more closely related to governments'
ability to effectively target expenditures on the poor, while repudiation of government
contracts has more influence on the activities of the private sector, our regression results
from (2) are in line with expectations.
Welfare Outcome Ranks
Tables 4a and 4b report the cross-country welfare outcome ranks for life expectancy at birth
and infant mortality rates, respectively. Again, the interpretation is exactly as that for safety
net expenditures. On life expectancy, we see Iceland has the highest unconditional life
expectancy (the time-averaged life expectancy over our sample period), while Ethiopia has
the lowest. As expected, Western Europe and North America are at the top of the ranking,
with sub-Saharan Africa and South Asian countries forming the lowest unconditional
rankings.
Table 4a: Cross-Country Welfare Outcome Regressions: Life Expectancy at Birth
Name
Iceland
Rank1
Rank2
Rank3
1
51
58
Latin America and the
Caribbean
Name
Rank1
Rank2
Rank3
Sweden
2
58
57
Costa Rica
28
3
2
Norway
3
35
27
Uruguay
29
34
47
Netherlands
4
62
81
Jamaica
32
2
7
73
Switzerland
5
59
38
Argentina
35
71
Japan
6
64
82
Panama
37
8
9
Canada
7
42
46
Trinidad and Tobago
38
32
29
France
8
60
39
Venezuela RB
41
49
48
Spain
9
26
54
Chile
42
47
50
Australia
10
66
67
Paraguay
43
6
6
Denmark
11
74
49
Mexico
46
61
77
Italy
12
36
63
Suriname
47
10
14
Greece
13
13
19
Colombia
50
28
26
Israel
14
46
65
Dominican Republic
52
22
15
Cyprus
15
5
16
Ecuador
54
21
23
Belgium
16
81
70
Brazil
55
83
90
United Kingdom
17
72
64
Guyana
57
9
10
New Zealand
18
63
66
El Salvador
62
39
45
63
Name
Rank1
Rank2
Rank3
Hong Kong, China
19
40
44
United States
20
91
75
Austria
21
65
Germany
22
Ireland
23
Malta
24
Name
Rank1
Rank2
Rank3
Peru
65
87
97
Honduras
66
24
21
56
Nicaragua
67
23
32
79
83
Guatemala
71
69
68
19
25
Haiti
81
68
60
18
12
Finland
25
43
61
Sub-Saharan Africa
Luxembourg
26
89
94
South Africa
70
93
85
Singapore
27
86
79
Botswana
76
31
22
Costa Rica
28
3
2
Ghana
77
53
62
Uruguay
29
34
47
Zimbabwe
78
30
34
Poland
30
11
5
Kenya
79
16
18
Portugal
31
12
13
Madagascar
82
38
33
Jamaica
32
2
7
Cameroon
84
78
72
Kuwait
33
82
86
Congo, Dem. Rep.
85
80
80
Hungary
34
14
8
Gabon
86
103
103
Argentina
35
71
73
Congo, Rep.
87
94
95
Romania
36
4
3
Uganda
88
27
36
Panama
37
8
9
Sudan
89
41
55
Trinidad and Tobago
38
32
29
Togo
90
73
71
Sri Lanka
39
1
1
Cote d'Ivoire
91
95
96
Jordan
40
15
20
Nigeria
92
84
69
Venezuela RB
41
49
48
Senegal
93
97
87
Chile
42
47
50
Mozambique
94
67
74
Paraguay
43
6
6
Mali
95
90
88
United Arab Emirates
44
99
98
Malawi
96
70
84
Qatar
45
100
99
Burkina Faso
97
77
92
México
46
61
77
Gambia, The
98
92
93
Suriname
47
10
14
Ethiopia
99
44
35
Malaysia
48
17
17
Niger
100
96
89
Korea, Rep.
49
48
43
Guinea
101
98
100
Colombia
50
28
26
Guinea-Bissau
102
88
91
China
51
29
42
Sierra Leone
103
102
101
Dominican Republic
52
22
15
Thailand
53
7
4
North America
Ecuador
54
21
23
Canada
7
42
46
Brazil
55
83
90
United States
20
91
75
Philippines
56
25
30
Guyana
57
9
10
Western Europe
Tunisia
58
55
28
Iceland
1
51
58
Turkey
59
54
31
Sweden
2
58
57
Syrian Arab Republic
60
20
11
Norway
3
35
27
Iran, Islamic Rep.
61
37
37
Netherlands
4
62
81
El Salvador
62
39
45
Switzerland
5
59
38
Saudi Arabia
63
101
102
France
8
60
39
Algeria
64
50
59
Spain
9
26
54
Peru
65
87
97
Denmark
11
74
49
Honduras
66
24
21
Italy
12
36
63
Nicaragua
67
23
32
Cyprus
15
5
16
Oman
68
85
78
Belgium
16
81
70
64
Name
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Morocco
69
52
52
United Kingdom
17
72
64
South Africa
70
93
85
Austria
21
65
56
Guatemala
71
69
68
Germany
22
79
83
Egypt, Arab Rep.
72
57
40
Ireland
23
19
25
Indonesia
73
56
51
Finland
25
43
61
India
74
75
76
Luxembourg
26
89
94
24
Pakistan
75
33
Botswana
76
31
22
South Asia
Ghana
77
53
62
Sri Lanka
39
1
1
76
Zimbabwe
78
30
34
India
74
75
Kenya
79
16
18
Pakistan
75
33
24
Papua New Guinea
80
76
53
Bangladesh
83
45
41
Haiti
81
68
60
Madagascar
82
38
33
Middle East and North Africa
Bangladesh
83
45
41
Greece
13
13
19
Cameroon
84
78
72
Israel
14
46
65
Congo, Dem. Rep.
85
80
80
Malta
24
18
12
Gabon
86
103
103
Portugal
31
12
13
Congo, Rep.
87
94
95
Kuwait
33
82
86
20
Uganda
88
27
36
Jordan
40
15
Sudan
89
41
55
United Arab Emirates
44
99
98
Togo
90
73
71
Qatar
45
100
99
Cote d'Ivoire
91
95
96
Tunisia
58
55
28
Nigeria
92
84
69
Syrian Arab Republic
60
20
11
Senegal
93
97
87
Iran, Islamic Rep.
61
37
37
Mozambique
94
67
74
Saudi Arabia
63
101
102
Mali
95
90
88
Algeria
64
50
59
Malawi
96
70
84
Oman
68
85
78
Burkina Faso
97
77
92
Morocco
69
52
52
Egypt, Arab Rep.
72
57
40
5
Gambia, The
98
92
93
Ethiopia
99
44
35
Niger
100
96
89
Eastern Europe and Central
Asia
Guinea
101
98
100
Poland
30
11
Guinea-Bissau
102
88
91
Hungary
34
14
8
Sierra Leone
103
102
101
Romania
36
4
3
Turkey
59
54
31
6
64
82
Correlation matrix
rank1
rank1
rank2
rank3
1
rank2
0.3011
1
rank3
0.2581
0.9327
Rank test (p-value)
rank1=rank2
0.0756
rank2=rank3
1
East Asia and the Pacific
Japan
1
Australia
10
66
67
New Zealand
18
63
66
Hong Kong, China
19
40
44
Singapore
27
86
79
Malaysia
48
17
17
Korea, Rep.
49
48
43
42
China
51
29
Thailand
53
7
4
Philippines
56
25
30
Indonesia
73
56
51
Papua New Guinea
80
76
53
65
Table 4b: Cross-Country Welfare Outcome Regressions: Infant Mortality Rate
Name
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Sweden
1
56
58
Latin America and the
Caribbean
Iceland
2
52
60
Costa Rica
33
8
7
Finland
3
29
35
Uruguay
35
40
67
Netherlands
4
66
81
Trinidad and Tobago
37
30
29
Norway
5
39
33
Panama
38
16
14
Japan
6
65
77
Jamaica
39
6
16
Switzerland
7
63
40
Argentina
41
75
83
Denmark
8
67
44
Venezuela RB
42
44
48
18
Australia
9
62
66
Suriname
44
7
France
10
60
42
Paraguay
45
9
9
United Kingdom
11
64
50
Chile
46
69
68
New Zealand
12
50
56
Colombia
47
28
25
Canada
13
45
47
Mexico
50
73
85
Singapore
14
53
45
Guyana
56
17
13
Luxembourg
15
80
86
Brazil
57
90
93
54
Ireland
16
21
23
Ecuador
58
43
Hong Kong, China
17
36
34
Dominican Republic
59
58
39
Belgium
18
77
61
El Salvador
64
49
62
United States
19
85
69
Honduras
65
32
28
Germany
20
76
74
Peru
66
93
97
Austria
21
54
46
Nicaragua
67
31
36
Israel
22
46
65
Guatemala
69
57
51
Haiti
90
83
80
49
Malta
23
19
11
Spain
24
34
64
Cyprus
25
13
26
Sub-Saharan Africa
Italy
26
48
72
South Africa
54
70
Greece
27
24
37
Botswana
60
22
15
Hungary
28
18
10
Zimbabwe
62
10
12
Poland
29
11
4
Kenya
70
5
8
Korea, Rep.
30
23
21
Ghana
74
33
53
Malaysia
31
4
5
Congo, Rep.
76
55
55
Portugal
32
20
22
Sudan
78
15
17
52
Costa Rica
33
8
7
Cameroon
79
47
Kuwait
34
79
82
Uganda
81
26
27
Uruguay
35
40
67
Togo
82
59
57
Romania
36
2
2
Senegal
83
51
30
Trinidad and Tobago
37
30
29
Nigeria
85
61
43
Panama
38
16
14
Congo, Dem. Rep.
86
74
70
Jamaica
39
6
16
Cote d'Ivoire
87
86
84
Sri Lanka
40
1
1
Gabon
91
102
102
Argentina
41
75
83
Burkina Faso
93
38
71
Venezuela RB
42
44
48
Madagascar
94
68
59
Jordan
43
12
20
Ethiopia
95
42
32
Suriname
44
7
18
Gambia, The
96
91
92
66
Name
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Paraguay
45
9
9
Niger
97
94
89
Chile
46
69
68
Mozambique
98
71
76
Colombia
47
28
25
Guinea
99
98
100
Qatar
48
95
95
Guinea-Bissau
100
88
87
Philippines
49
14
19
Malawi
101
92
96
Mexico
50
73
85
Mali
102
99
98
China
51
27
41
Sierra Leone
103
103
101
Thailand
52
3
3
United Arab Emirates
53
100
99
North America
South Africa
54
70
49
Canada
13
45
47
Syrian Arab Republic
55
25
6
United States
19
85
69
Guyana
56
17
13
Brazil
57
90
93
Western Europe
Ecuador
58
43
54
Sweden
1
56
58
Dominican Republic
59
58
39
Iceland
2
52
60
Botswana
60
22
15
Finland
3
29
35
Saudi Arabia
61
101
103
Netherlands
4
66
81
Zimbabwe
62
10
12
Norway
5
39
33
Oman
63
84
78
Switzerland
7
63
40
El Salvador
64
49
62
Denmark
8
67
44
Honduras
65
32
28
France
10
60
42
Peru
66
93
97
United Kingdom
11
64
50
Nicaragua
67
31
36
Luxembourg
15
80
86
Tunisia
68
82
63
Ireland
16
21
23
Guatemala
69
57
51
Belgium
18
77
61
Kenya
70
5
8
Germany
20
76
74
Indonesia
71
35
31
Austria
21
54
46
Iran, Islamic Rep.
72
72
73
Spain
24
34
64
Papua New Guinea
73
37
24
Cyprus
25
13
26
Ghana
74
33
53
Italy
26
48
72
Algeria
75
87
88
Congo, Rep.
76
55
55
South Asia
Morocco
77
78
79
Sri Lanka
40
1
1
Sudan
78
15
17
India
84
89
91
Cameroon
79
47
52
Bangladesh
89
41
38
Turkey
80
97
90
Pakistan
92
81
75
Uganda
81
26
27
Togo
82
59
57
Senegal
83
51
30
Israel
22
46
65
India
84
89
91
Malta
23
19
11
Middle East and North Africa
Nigeria
85
61
43
Greece
27
24
37
Congo, Dem. Rep.
86
74
70
Portugal
32
20
22
Cote d'Ivoire
87
86
84
Kuwait
34
79
82
Egypt, Arab Rep.
88
96
94
Jordan
43
12
20
Bangladesh
89
41
38
Qatar
48
95
95
Haiti
90
83
80
United Arab Emirates
53
100
99
Gabon
91
102
102
Syrian Arab Republic
55
25
6
Pakistan
92
81
75
Saudi Arabia
61
101
103
Burkina Faso
93
38
71
Oman
63
84
78
Madagascar
94
68
59
Tunisia
68
82
63
67
Name
Rank1
Rank2
Rank3
Name
Rank1
Rank2
Rank3
Ethiopia
95
42
32
Iran, Islamic Rep.
72
72
73
Gambia, The
96
91
92
Algeria
75
87
88
Níger
97
94
89
Morocco
77
78
79
Mozambique
98
71
76
Egypt, Arab Rep.
88
96
94
Guinea
99
98
100
Guinea-Bissau
100
88
87
Eastern Europe and Central
Asia
Malawi
101
92
96
Hungary
28
18
10
Mali
102
99
98
Poland
29
11
4
Sierra Leone
103
103
101
Romania
36
2
2
Turkey
80
97
90
Correlation matrix
rank1
rank2
Rank1
1
Rank2
0.3329
1
Rank3
0.3148
0.9267
Rank test (p-value)
Rank1=rank2
0.1933
Rank2=rank3
0.9179
rank3
1
East Asia and the Pacific
Japan
6
65
77
Australia
9
62
66
New Zealand
12
50
56
Singapore
14
53
45
Hong Kong, China
17
36
34
Korea, Rep.
30
23
21
Malaysia
31
4
5
Philippines
49
14
19
China
51
27
41
Thailand
52
3
3
Indonesia
71
35
31
Papua New Guinea
73
37
24
When we control for structural characteristics, we find that many developing countries,
such as Paraguay, Jamaica, Sri Lanka, and Madagascar, do far better than we would expect
compared to the regional norm, and most developed economies drop significantly in rank
from R1 to R2. This is in part driven by the fact that there is a biological upper bound on life
expectancy, so that the there are probably diminishing marginal benefits from increases in
safety net expenditures on this welfare indicator. However, this does not disguise the fact that
many developing countries actually do have higher life expectancies, conditional on
structural characteristics relative to regional norms, than we would typically believe from just
examining unconditional figures (R1). There tends to be relatively little movement from R2
to R3 (these ranks have a correlation of .95), suggesting that most of the variation that we
observe in life expectancy figures across countries is due to differences in these structural
characteristics rather than being driven by institutional factors.
The pattern of analysis is very similar in table 4b when we look at IMR. Again, most of
the variation in the welfare indicator is due to structural features, not institutions. Examining
the regional results, we see that Malaysia, despite having a high unconditional level of IMR,
actually is one of the best performers in its region once we condition for structural factors.
This is in contrast to Indonesia, which, although it does improve its rank moving to R2, it
falls further behind Malaysia. In Latin America and the Caribbean, Trinidad and Tobago’s
apparently good unconditional record on IMR actually is far worse than we would expect
once structural factors are accounted for, i.e., given its economic characteristics, we would
expect it to perform far better than it does given what other countries manage to achieve on
average.
68
Welfare Outcome Regression Results
Table 5 reports the regressions (5) and (6) from which we have derived R2 and R3 above.
For life expectancy at birth, we see that this is increasing in both health and education
spending, state income per capita, it is pro-cyclical, and increases as the proportion of the
population that is of working age increases, or the level of urbanization increases. In
regression (6) when we see how much of the unexplained variation from (5) can be
accounted for by quality of government factors, none of the factors turn out to be significant.
This is consistent with our earlier observation that most of the movement across ranks occurs
between R1 and R2. A similar set of factors is significant in the IMR regressions, although
here the threat of expropriation surprisingly decreases IMR, while worsening rule of law and
government bureaucracy increase it. Again, in line with our earlier results on safety net
expenditures, it is the level of government bureaucracy which is the single most important
determinant among the quality-of-government variables controlled for.
Table 5: Cross-Country Outcome Regressions
Life Expectancy at Birth
Independent Variables
Log(GDP per capita)
Shock to GDP per capita
Frac. of popn aged 15–64
(millions)
Frac. of popn residing in urban
regions
Adjusted R-squared
Observations
R2 Regression
R3 Regression
Infant Mortality Rate
R2 Regression
4.48
–19.6
(0.304)
(10.54)
–3.30
20.5
(2.01)
(10.2)
0.023
–0.094
(0.004)
(0.018)
0.151
–0.621
(0.018)
(0.092)
0.7517
0.6812
618
618
Repudiation of Government
Contract
Expropriation Threat
Corruption
Rule of Law
Government Bureaucracy
Adjusted R-squared
Observations
69
R3 Regression
0.184
0.074
(0.362)
(1.85)
1.14
–7.39
(0.413)
(2.10)
–0.172
0.897
(0.351)
(1.79)
–0.449
3.29
(0.413)
(2.10)
–1.64
6.89
(0.389)
(1.98)
0.0667
0.0557
525
526
Robustness Checks
In order to see how sensitive our rankings were to our data, we decided to re-run our analysis
using a slightly different set of quality-of-government indices. This was done in two ways: (i)
using different subsets of the existing quality indices; and (ii) using a measure of corruption
taken from a more independent source than the ICRG, namely Transparency International
(TI).
There was no significant difference created in each of the cross-country ranks reported,
either in terms of safety net expenditure or welfare outcome rankings. Using the corruption
measure from TI also gave similar results. Typically, the correlation coefficient between the
rankings using the corruption index from the ICRG data and that from the TI data was over
0.94.
Indian State-Level Analysis
We now turn to the same analysis but at the level of Indian states. Our data set runs over the
period 1960–92, and because of the much more complete series that we have at this level
compared to the cross-country-level analysis, there was no need to average the data into fiveyear time periods. The data series come from Indian government sources, details of which are
in the data appendix. This analysis allows us to control for common macroeconomic shocks
and institutional features across Indian states. The analysis suggests how our methodology
can apply equally to sub-national policy, where the notions of benchmarking and yardstick
competition apply equally as to across neighboring countries.
Table 6: Indian State Safety Net Expenditures (as percentage of state GDP)
States
Food
Calamity
Health
Education
Social
Development1
Andhra Pradesh
0.171
3.941
11.254
28.800
78.707
101.001
Assam
0.269
3.420
10.379
32.780
69.769
95.211
Bihar
0.089
1.491
5.827
18.881
42.460
52.372
Gujarat
1.226
5.217
11.898
33.496
80.301
110.321
Haryana
0.524
2.840
14.429
37.040
77.812
137.947
Jammu & Kashmir
5.503
3.585
24.090
47.263
110.165
203.648
Karnataka
1.028
1.663
11.188
30.727
73.647
99.715
Kerala
0.304
1.662
12.660
45.097
87.682
97.656
Madhya Pradesh
0.201
1.383
10.030
23.239
59.176
80.183
Maharashtra
0.371
2.752
13.087
32.640
76.794
106.728
Orissa
0.347
4.673
10.815
25.597
69.502
91.134
Punjab
0.057
4.978
15.386
43.145
95.059
134.557
Rajasthan
0.252
4.997
11.279
25.408
59.127
79.201
Tamil Nadu
2.641
1.479
13.313
34.087
83.611
110.387
Uttar Pradesh
0.015
1.505
8.133
21.720
49.783
66.795
West Bengal
0.104
3.430
12.594
29.704
74.706
84.795
All States
0.819
3.064
12.273
31.851
74.269
103.228
1. Development expenditures can be greater than 100% of state GDP if net development transfers are
positive.
70
Public Food Distribution ('000 tons)
Voter Turnout
States
Mean
Sd
Min
Max
Mean
Sd
Andhra Pradesh
618.14
595.17
33
2451
68.719
3.515
Assam
402.09
209.79
68
823
62.978
11.530
Bihar
665.58
304.74
304
2092
51.764
5.903
Gujarat
572.39
363.08
142
1402
55.906
5.678
Haryana
121.74
53.94
15
209
67.431
5.108
Jammu & Kashmir
225.72
105.65
71
447
68.965
5.533
Karnataka
532.94
343.11
43
1165
63.372
5.825
Kerala
1075.47
568.18
90
2088
77.572
3.772
Madhya Pradesh
365.44
276.74
4
1102
49.089
6.056
Maharashtra
1515.11
501.15
521
2404
59.347
4.384
Orissa
270.03
164.93
17
625
44.939
7.489
Punjab
227.44
195.73
4
1209
66.139
4.077
Rajasthan
327.86
341.63
4
1263
52.992
6.219
Tamil Nadu
969.58
641.58
89
2269
69.700
4.160
Uttar Pradesh
781.78
329.45
166
1893
52.075
6.034
West Bengal
1620.08
573.54
232
2944
66.573
8.616
All States
643.213
348.026
112.688
1524.125
61.098
5.869
R1, R2, and R3
Table 6 reports the (time-averaged) expenditures on each of these types of safety net, as a
percentage of state GDP. The figures for public food distribution, an in-kind transfer of
grains, is given in thousands of tons. Again, the first point to note is that there are large
variations across states in each of these series, although, on the whole, if we ignore Jammu
and Kashmir, the variation drops considerably and is less than that we observed across
countries.
Table 7a reports the rankings R1 to R3 for each type of safety net with the exception of
public food distribution (PFD), which we consider separately because it is the only in-kind
transfer we have. In table 7a, we take our dependent variable to be safety net expenditures as
a share of state production. We will focus on the food subsidy and calamity expenditures
here, as these are really targeted at the most vulnerable subpopulation of the poor, the
chronically poor.
When we run regression (3) and use this to form R2, we find that the movements across
R1 to R2 are far more stark for food subsidy than for calamity expenditures. For instance,
although unconditionally it appears as if Assam performs well in the level of food subsidies it
provides, once we take account of its ability to pay and the level of needs in that state, it
performs poorly relative to the benchmark comparison of what Indian states are able to do on
average. The same applies to Gujarat and Orissa. When we control for institutional quality
and construct R3, states such as Tamil Nadu and Kerala slip down in rank, suggesting that
they are now spending less than we would have predicted, while other states such as
Rajasthan improve their rankings.
The movements for calamities are much less pronounced moving across the rankings—
the correlation coefficient between R1 and R2 is 0.9393, and between R2 and R3 it is 0.9107.
71
This is as we may well expect, given that calamity expenditures are only responsive to
natural disasters, which are randomly distributed across states, and so accounting for
structural and institutional features should have less effect on relative performance.
A similar pattern emerges when we consider other Indian state-level types of safety net
expenditures. Noticeably, there are large movements across all three rankings for social
security expenditures and development expenditures.
Tables 7b and 7c repeat the analysis except now we use safety net expenditures as
shares of total state government expenditures and state tax revenues, respectively. On the
whole, these lead to higher correlation coefficients across the rankings than using state GDP,
although the pattern of movements across rankings is robust to using any of these three
normalizations.
The result for public food distribution (PFD) is in table 7d. Here there is remarkably
little movement between ranks one and two, implying that again there is little that structural
factors have to do with the levels of provision of these in-kind transfers. When we run (4) to
control for institutional features, we do find some large falls in rank (Tamil Nadu, Andrha
Pradesh) and some large rises (Bihar, Orissa). We may use this result to argue that, because
this latter group of states appears to be underperforming relative to the state norm, they
should increase their levels of PFD. Furthermore, the reasons that these states appear to
underperform have more to do with their poorer-quality institutions than a less favorable
economic situation.
72
Table 7a: Indian State Social Safety Net Expenditures (SS/State production)
Food Subsidy
State
Calamity
State
Health
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Jammu & Kashmir
1
1
1
Jammu & Kashmir
1
1
1
Jammu & Kashmir
State
1
1
1
Tamil Nadu
2
3
8
Orissa
2
3
6
Kerala
2
6
8
Kerala
3
6
6
Rajasthan
3
4
5
Assam
3
2
2
Assam
4
10
10
Assam
4
2
2
Orissa
4
10
7
Karnataka
5
9
11
Gujarat
5
7
8
Rajasthan
5
4
6
Gujarat
6
14
12
Punjab
6
5
3
Punjab
6
3
10
Orissa
7
13
9
West Bengal
7
6
7
Karnataka
7
9
11
Rajasthan
8
5
5
Andhra Pradesh
8
8
9
Gujarat
8
13
12
Madhya Pradesh
9
8
3
Karnataka
9
10
11
Madhya Pradesh
9
11
5
West Bengal
10
4
4
Madhya Pradesh
10
12
10
Tamil Nadu
10
8
9
Maharashtra
11
2
2
Maharashtra
11
9
4
Andhra Pradesh
11
12
15
Andhra Pradesh
12
11
13
Kerala
12
14
13
West Bengal
12
7
4
Punjab
13
12
14
Bihar
13
13
14
Maharashtra
13
5
3
Bihar
14
7
7
Tamil Nadu
14
11
12
Bihar
14
14
13
Uttar Pradesh
15
15
15
Uttar Pradesh
15
15
15
Uttar Pradesh
15
15
14
rank1
rank2
rank3
rank1
rank2
rank3
rank1
rank2
rank3
Correlation matrix
Correlation matrix
rank1
1
rank2
0.3929
1
rank3
0.3393
0.8536
Rank test(p-value)
1
Correlation matrix
rank1
1
rank2
0.9393
1
rank3
0.8179
0.9107
Rank test(p-value)
1
rank1
1
rank2
0.6607
1
rank3
0.4750
0.7643
Rank test(p-value)
rank1=rank2
1
rank1=rank2
1
rank1=rank2
1
rank2=rank3
1
rank2=rank3
0.3877
rank2=rank3
1
73
1
Education
State
Social
Rank1
Rank2
Rank3
Jammu & Kashmir
1
1
1
Kerala
2
3
7
Assam
3
2
Orissa
4
State
Development
Rank1
Rank2
Rank3
Jammu & Kashmir
1
1
1
Kerala
2
3
6
2
Orissa
3
6
12
9
Rajasthan
4
State
Rank1
Rank2
Rank3
Jammu & Kashmir
1
1
1
Assam
2
2
2
4
Orissa
3
10
5
4
9
Kerala
4
4
7
Rajasthan
5
5
12
Karnataka
5
7
10
Rajasthan
5
6
8
Punjab
6
4
10
Punjab
6
2
11
Karnataka
6
9
11
Karnataka
7
9
14
Gujarat
7
11
5
Punjab
7
3
12
Gujarat
8
13
5
Tamil Nadu
8
8
7
Gujarat
8
13
10
Tamil Nadu
9
8
8
Andhra Pradesh
9
9
13
Tamil Nadu
9
8
9
Madhya Pradesh
10
11
15
Madhya Pradesh
10
12
8
Madhya Pradesh
10
12
6
Andhra Pradesh
11
10
13
West Bengal
11
5
3
Andhra Pradesh
11
11
15
West Bengal
12
7
6
Bihar
12
13
14
West Bengal
12
7
4
Maharashtra
13
6
3
Maharashtra
13
10
2
Maharashtra
13
5
3
Bihar
14
14
4
Uttar Pradesh
14
14
12
Bihar
14
14
13
Uttar Pradesh
15
15
11
Uttar Pradesh
15
15
14
Rank1
Rank2
Rank3
Correlation matrix
Correlation matrix
Rank1
Rank2
Rank1
1
Rank2
0.6857
1
Rank3
0.1857
0.375
Rank test(p-value)
Correlation matrix
Rank3
1
Rank1
Rank2
Rank1
1
Rank2
0.7890
1
Rank3
0.3187
0.4066
Rank test(p-value)
Rank3
1
Rank1
1
Rank2
0.6536
1
Rank3
0.5107
0.6429
Rank test(p-value)
rank1=rank2
1
rank1=rank2
0.5078
rank1=rank2
1
rank2=rank3
1
rank2=rank3
1
rank2=rank3
1
74
1
Table 7b: Indian State Social Safety Net Expenditures (SS/Total state expenditures)
Food Subsidy
State
Calamity
State
Health
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Jammu & Kashmir
1
1
1
Rajasthan
1
1
1
Rajasthan
State
1
3
2
Tamil Nadu
2
2
2
Gujarat
2
3
3
West Bengal
2
1
1
Gujarat
3
7
7
Orissa
3
4
6
Kerala
3
4
10
Karnataka
4
5
6
West Bengal
4
5
5
Madhya Pradesh
4
7
3
Kerala
5
6
8
Andhra Pradesh
5
6
4
Tamil Nadu
5
5
5
Orissa
6
13
10
Bihar
6
10
14
Andhra Pradesh
6
9
11
Rajasthan
7
4
4
Assam
7
7
7
Orissa
7
11
9
Madhya Pradesh
8
8
5
Punjab
8
2
2
Maharashtra
8
10
8
Maharashtra
9
3
3
Maharashtra
9
12
8
Jammu & Kashmir
9
8
6
Assam
10
11
14
Madhya Pradesh
10
11
9
Gujarat
10
12
12
West Bengal
11
9
11
Uttar Pradesh
11
9
10
Punjab
11
2
4
Andhra Pradesh
12
12
12
Karnataka
12
13
13
Karnataka
12
13
14
Bihar
13
10
9
Tamil Nadu
13
15
12
Uttar Pradesh
13
14
13
Punjab
14
15
15
Jammu & Kashmir
14
8
15
Assam
14
6
7
Uttar Pradesh
15
14
13
Kerala
15
14
11
Bihar
15
15
15
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Correlation Matrix
Correlation Matrix
Rank1
1
Rank2
0.7714
1
Rank3
0.7571
0.9321
Rank Test(P-Value)
1
Correlation Matrix
Rank1
1
Rank2
0.8000
1
Rank3
0.7607
0.8000
Rank Test(P-Value)
1
Rank1
1
Rank2
0.6500
1
Rank3
0.6500
0.8643
Rank Test(P-Value)
Rank1=Rank2
1
Rank1=Rank2
0.2668
Rank1=Rank2
0.2668
Rank2=Rank3
1
Rank2=Rank3
1
Rank2=Rank3
1
75
1
Education
State
Rank1
Rank2
Social
Rank3
State
Rank1
Rank2
Development
Rank3
State
Rank1
Rank2
Rank3
Kerala
1
1
2
Kerala
1
1
5
Kerala
1
3
9
Assam
2
3
1
West Bengal
2
2
2
Andhra Pradesh
2
5
1
Bihar
3
7
3
Andhra Pradesh
3
4
6
Tamil Nadu
3
2
2
Tamil Nadu
4
5
8
Tamil Nadu
4
3
7
Madhya Pradesh
4
4
7
West Bengal
5
4
7
Gujarat
5
5
8
Gujarat
5
7
6
Punjab
6
2
9
Orissa
6
7
9
Assam
6
1
3
Rajasthan
7
6
4
Bihar
7
8
4
Karnataka
7
8
10
Madhya Pradesh
8
8
5
Rajasthan
8
9
3
Jammu & Kashmir
8
10
11
Karnataka
9
9
10
Madhya Pradesh
9
10
1
Orissa
9
6
4
Gujarat
10
11
12
Punjab
10
6
10
Rajasthan
10
12
13
Andhra Pradesh
11
10
11
Karnataka
11
11
11
Punjab
11
15
12
Uttar Pradesh
12
12
6
Maharashtra
12
12
13
West Bengal
12
11
8
Maharashtra
13
13
14
Uttar Pradesh
13
13
12
Bihar
13
9
5
Orissa
14
14
13
Jammu & Kashmir
14
14
14
Uttar Pradesh
14
13
14
Jammu & Kashmir
15
15
15
Maharashtra
15
14
15
Rank1
Rank2
Rank3
Correlation Matrix
Correlation Matrix
Rank1
Rank2
Rank3
Rank1
1
Rank1
1
Rank2
0.9321
1
Rank3
0.8357
0.7464
Rank Test(P-Value)
1
Correlation Matrix
Rank1
Rank2
Rank2
0.9516
1
Rank3
0.6659
0.5604
Rank Test(P-Value)
Rank3
1
Rank1
1
Rank2
0.8286
1
Rank3
0.6107
0.8000
Rank Test(P-Value)
Rank1=Rank2
1
Rank1=Rank2
0.4531
Rank1=Rank2
1
Rank2=Rank3
0.7905
Rank2=Rank3
0.5488
Rank2=Rank3
0.3953
76
1
Table 7c: Indian State Social Safety Net Expenditures (SS/Total state tax revenues)
Food Subsidy
State
Calamity
State
Health
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Jammu & Kashmir
1
1
1
Rajasthan
1
1
1
Jammu & Kashmir
State
1
1
Rank3
1
Tamil Nadu
2
2
3
Orissa
2
2
4
Rajasthan
2
2
3
Gujarat
3
11
10
Gujarat
3
7
5
Orissa
3
4
5
Karnataka
4
7
9
Assam
4
3
3
Assam
4
3
2
Kerala
5
5
6
West Bengal
5
6
7
Kerala
5
7
8
Orissa
6
13
11
Andhra Pradesh
6
8
6
Madhya Pradesh
6
8
6
4
Rajasthan
7
4
5
Bihar
7
11
15
West Bengal
7
5
Assam
8
10
12
Punjab
8
5
2
Tamil Nadu
8
9
9
Madhya Pradesh
9
8
4
Jammu & Kashmir
9
4
9
Andhra Pradesh
9
10
13
Maharashtra
10
3
2
Madhya Pradesh
10
10
11
Karnataka
10
12
12
West Bengal
11
6
7
Maharashtra
11
13
8
Uttar Pradesh
11
13
11
Bihar
12
9
8
Karnataka
12
12
14
Maharashtra
12
11
7
Andhra Pradesh
13
12
13
Uttar Pradesh
13
9
13
Gujarat
13
14
14
Punjab
14
14
15
Tamil Nadu
14
14
12
Punjab
14
6
10
Uttar Pradesh
15
15
14
Kerala
15
15
10
Bihar
15
15
15
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
Correlation Matrix
Correlation Matrix
Rank1
1
Rank2
0.6071
1
Rank3
0.5643
0.9321
Rank Test(P-Value)
1
Correlation Matrix
Rank1
1
Rank2
0.8357
1
Rank3
0.7214
0.7536
Rank Test(P-Value)
1
Rank1
1
Rank2
0.8393
1
Rank3
0.8393
0.9036
Rank Test (P-Value)
Rank1=Rank2
0.7539
Rank1=Rank2
1
Rank1=Rank2
0.3877
Rank2=Rank3
0.7905
Rank2=Rank3
1
Rank2=Rank3
1
77
1
Education
State
Social
State
Development
Rank1
Rank2
Rank3
Rank1
Rank2
Rank3
State
Rank1
Rank2
Rank3
Jammu & Kashmir
1
1
2
Kerala
2
3
3
Jammu & Kashmir
1
1
1
Jammu & Kashmir
1
1
1
Orissa
2
2
2
Assam
2
2
2
Assam
3
2
1
Rajasthan
3
4
4
Orissa
3
3
3
Rajasthan
4
4
6
Kerala
4
3
3
Rajasthan
4
4
6
Orissa
5
6
7
West Bengal
5
5
5
Kerala
5
5
4
Madhya Pradesh
6
8
13
Madhya Pradesh
6
6
6
Madhya Pradesh
6
6
13
Karnataka
7
10
15
Andhra Pradesh
7
7
7
Andhra Pradesh
7
7
12
Bihar
8
13
4
Bihar
8
13
13
Karnataka
8
8
11
Tamil Nadu
9
9
9
Gujarat
9
9
11
Tamil Nadu
9
9
8
West Bengal
10
7
8
Tamil Nadu
10
10
10
Gujarat
10
13
10
Punjab
11
5
14
Punjab
11
8
14
Punjab
11
10
14
Andhra Pradesh
12
11
11
Uttar Pradesh
12
11
8
Bihar
12
14
15
Gujarat
13
14
12
Karnataka
13
12
12
Uttar Pradesh
13
12
9
Uttar Pradesh
14
12
5
Maharashtra
14
14
9
West Bengal
14
11
7
Maharashtra
15
15
10
Maharashtra
15
15
5
Rank1
Rank2
Rank3
Correlation Matrix
Correlation Matrix
Rank1
Rank2
Rank1
1
Rank2
0.8357
1
Rank3
0.5286
0.4679
Rank Test(P-Value)
Correlation Matrix
Rank3
1
Rank1
Rank2
Rank1
1
Rank2
0.9165
1
Rank3
0.8198
0.8374
Rank Test(P-Value)
Rank3
1
Rank1
1
Rank2
0.9571
1
Rank3
0.5143
0.5714
Rank Test (P-Value)
Rank1=Rank2
1
Rank1=Rank2
0.6875
Rank1=Rank2
1
Rank2=Rank3
0.7744
Rank2=Rank3
1
Rank2=Rank3
1
78
1
Table 7d: Indian State Social Safety Net Expenditures: Public Food Distribution
State
Rank1
Rank2
Rank3
Kerala
1
1
5
Jammu & Kashmir
2
2
1
West Bengal
3
3
2
Maharashtra
4
4
3
Assam
5
5
4
Tamil Nadu
6
6
11
Gujarat
7
7
6
Punjab
8
9
15
Karnataka
9
8
12
Andhra Pradesh
10
10
14
Bihar
11
13
7
Orissa
12
11
8
Rajasthan
13
12
10
Uttar Pradesh
14
14
9
Madhya Pradesh
15
15
13
Rank1
Rank2
Rank3
Rank1
1
Rank2
0.9857
1
Rank3
0.6607
0.6571
1
Rank test(p-value)
Rank1=Rank2
1
Rank2=Rank3
0.3018
The estimates for the regressions (3) and (4), which lie behind tables 7a through 7d are
omitted for brevity. There are generally negative income effects for all types of safety net.
Higher rural headcount measures decrease spending while urban headcount increases
spending, which may be suggestive of an urban bias in government responses to poverty
alleviation. Both rural and urban income ginis tend to decrease expenditures, implying that,
as income inequality widens, spending as a share of state GDP rises. Consistent with the
results from table 7a, none of these structural factors except urban headcounts are significant
determinants of calamity spending.
The levels of voter turnout increase health, education, social, and development
expenditures in the regression. Political competition, surprisingly, reduces expenditures.
However, across all expenditure types, we consistently find that increased literacy rates, both
among males and females, significantly increase the levels of safety net expenditures.
Finally, as shocks become more variable, they tend to decrease education and social
spending, but the average level of shocks plays no role.
79
Finally, the regression findings for public food distribution (PFD) suggest that richer
states distribute more, and that urban headcounts and income inequality have greater effects
than the same rural factors. Voter turnout has a very strong positive impact on PFD, as well
as both shock variables. Political competition has no effect and, in line with the earlier
results, male and female literacy rates significantly increase food distribution.
Welfare Outcome Results
The final part of our analysis concerns the construction of rankings R1 to R3 for the welfare
outcome of infant mortality rates across Indian states. These are reported in table 8. States
such as Jammu and Kashmir and Punjab which appear to have low IMR actually do much
worse given their structural features relative to other Indian states, implying they ought to
spend more, while Gujarat and Bihar all manage to outperform the Indian benchmark level,
controlling for structural economic factors. Controlling for institutional quality as we move
to R3, there is still a high degree of movement—West Bengal and Mahrashtra would appear
to have poor-quality institutions, while Bihar and Rajasthan move up the rankings once
institutional quality is controlled for.
Table 8: Indian State-Level Outcome Measures
Infant Mortality Rate (per 1000 live births, urban and rural combined)
State
Rank1
Rank2
Rank3
1
1
3
Jammu & Kashmir
2
14
13
Maharashtra
3
2
6
West Bengal
4
7
14
Karnataka
5
3
4
Punjab
6
13
12
Tamil Nadu
7
5
9
Andhra Pradesh
8
6
5
Bihar
9
4
1
Gujarat
10
8
8
Rajasthan
11
9
2
Madhya Pradesh
12
10
7
Orissa
13
11
11
Uttar Pradesh
14
12
10
Rank1
Rank2
Rank3
Kerala
Rank1
1
Rank2
0.4286
1
Rank3
-0.0505
0.6484
Rank test(p-value)
Rank1=Rank2
0.0923
Rank2=Rank3
0.7744
80
1
The regression from which these rankings are derived (not shown), seem to suggest that
income levels will decrease infant mortality rates, and urban and rural poverty both increase
IMR. In terms of the institutional controls, political competition, voter turnout, and literacy
rates all significantly reduce infant mortality rates.
VII. Toward a Constructive Policy Dialogue
The Case for Safety Nets
There is much debate regarding how much government ought to spend on safety nets both in
developing and developed countries. However, there are no easy answers to this question.
What we have attempted to do in this paper is to create a framework in which a more
constructive policy debate can take place. We lay out this structure below.
The first issue to settle concerns what safety nets are, who they are for, and what are the
benefits of providing them. We summarize our earlier discussion in a simple matrix form in
table 9.
Table 9: A Policy Dialogue
Definition of Safety Nets
(i) redistributive role transferring resources toward the poorer members of
society to bring them out of poverty
(ii) provide greater opportunities for individuals to mitigate risks from
unforeseen contingencies
Who Are They For?
(i) chronically poor
(ii) temporarily poor
(iii) movers into poverty during periods of adjustment or crisis
The Benefits of Safety Nets
Equity
- societal aversion to inequality
- redistribution need not be at the expense of growth
Efficiency
- correction of market failures
- insurance mechanisms for the poor in place of shortterm coping strategies
- reduced uncertainty may increase human and physical
capital investments by the poor
- less incentive to engage in marginalized economic
activities
Political
Economy
- reduces probability of political failure
- raises participation of the poor in policymaking
process
- demonstrates government's commitment to tackling
poverty
Social Cohesion
- raises the political acceptability of market-based
reforms that often need to be made in the aftermath of
economic crisis
- societal aversion to inequality of opportunity
81
We hope that the approach to safety nets taken here helps us to move away from a
program-based definition of safety nets and toward a more objective-based or functional
definition of what safety nets are.
Table 9 also helps us to provide government officials with a clear way of thinking
through what benefits may be conferred to the economy through the expansion of safety nets,
in terms of both equity and efficiency objectives. This set of arguments, many of which are
recent in the literature, provide a case for safety nets that is independent of how we think of
how much governments ought to spend, and whether or not we use the benchmarking
exercise advocated here. Presenting the arguments in favor of safety net expenditures in this
way allows policymakers to stop thinking of such programs purely in terms of the costs of
provision.
Benchmarking
The approach here addresses the question of how much governments ought to spend by
establishing what governments are able to do on average and then using this as a point of
comparison for what any given government is actually doing. This is the notion of
benchmarking, whereby a summary statistic is created by which one country's performance
can be compared to that of its neighbors in a straightforward manner.
Comparison across the different rankings provides a powerful and easily presented tool
for officials to help determine how their country is performing relative to other countries, and
perhaps more pertinently, relative to their closest economic neighbors. In discussing the
results, we have emphasized many cases where neighboring countries had radically different
rankings. This result in itself might sometimes motivate safety net expenditures.
We have applied this idea of benchmarking both to safety net expenditures and to
welfare outcomes. We have done this in terms of cross-country comparisons, but we have
also shown how the methodology can be equally well applied at the sub-national level by
analyzing Indian states.
We have made the following arguments as to why benchmarking can be an effective
tool for policymakers. First, benchmarking allows governments to compare their own
performance to that of their neighbors. Benchmarking through the use of rankings very
clearly conveys information on the underlying effort being made by governments on these
policy outcomes. The data requirements to construct these rankings are minimal compared to
alternative methodologies.
The methodology we have described may enhance the policy debate by presenting
expenditure rankings in terms of the underlying structural features of country economies and
the design of institutions. This is important if governments believe that the structural factors
are only influenced by policy in the long run, while the institutional factors can be subject to
short- and medium-run constitutional reform. It is probably often going to be advisable to
think of the two types of reform as complementary, rather than viewing any single potential
policy reform as always of a higher priority than another. We hope to have demonstrated that
the exact policy prescription will depend on both country-specific factors and the type of
safety net we are talking about.
82
Constructing the Rankings
We have benchmarked countries by constructing three ranks—first, a ranking based on
unconditional expenditures on safety nets; second, a ranking conditioning on structural
features of the economy; and third, a ranking also conditioning on the quality of institutions
within the country.
We have argued that the presentation of our analysis in the form of rankings is useful for
several reasons. First, it is a simple summary statistic on which to base arguments to
policymakers for changes in spending. Second, the use of rankings makes our results more
robust to econometric concerns arising from the potential endogeneity of some covariates, as
well as omitted variables due to lack of data. The data requirements for this approach are also
minimal. For instance, we do not need to define any welfare criterion by which to judge
optimal levels of spending. Third, international or regional norms provide a natural point of
comparison for how much effort governments are seen to be putting into safety net spending.
They can also be used to motivate policy discussions.
Formally, our three ranking systems have been the following:
R1
R2
R3
Time-averaged unconditional ranking of safety net spending. A lower
ranking means that the country unconditionally spends less on safety nets.
Ranking of social spending conditional on structural factors. A lower ranking
means that safety net expenditures are lower than we would expect compared
to the countries in the same region, controlling for structural features of the
economy.
Ranking also taking into account quality of government. Again, a lower
ranking means that safety net expenditures are lower than we would expect
compared to the countries in the same region, controlling both for structural
and institutional features of the economy.
Movements across these rankings as well as the ranks themselves provide information
on the recommended policy prescriptions. The comparison between R1 and R2 is instructive
as it demonstrates that the level of safety net spending, what we have previously referred to
as the demand function for safety nets, is dependent partly on structural features of the
economy such as the distribution of income as well as its per capita level, or the level of
urbanization, which are probably factors that can only be changed through government
policy in the very long run.
Comparing rankings R2 and R3 indicates how controlling for institutional factors, which
are subject to influence through policy, can change a country’s relative performance. The
comparison of rankings R2 and R3 thus points to institutional reform as a means of
improving a country’s/state’s performance relative to the benchmark.
An Example of Benchmarking
An illustrative example can be given here of the type of dialogue that may follow from this
analysis. Consider the relative performance of South Asian countries with regard to life
expectancy. Looking at the unconditional rankings (R1) in table 4a, Sri Lanka is ranked 39th
in the world, India is 73rd, Pakistan is 75th, and Bangladesh is 83rd, out of a total of 103
83
countries. At face value, this suggests that Sri Lanka appears to put much more effort into
improving welfare as measured by life expectancy than do the other South Asian countries.
India, Pakistan, and Bangladesh have much more similar performance, unconditional on any
other factors.
Moving to R2, where we control for structural features of the economy, we find that the
relative performance of South Asian countries changes, and to dramatically different extents.
Pakistan rises to 33rd in the world, India falls to 76th and is overtaken by Bangladesh, which
rises to 45th. This suggests that, given India’s ability to finance such welfare improvements,
and the need for them, it performs poorly compared to its neighbors. In this respect, Sri
Lanka’s performance is impressive—controlling for structural factors, Sri Lanka moves to
first place, suggesting its government puts the most effort into welfare improvement as
measured in this dimension.
Moving to rank three, where we also control for institutional factors, we see that Sri
Lanka remains first, and Pakistan and Bangladesh continue to improve their international
rankings, while India’s performance remains around the same level at 76.
We would argue that the presentation of such information makes it clear that
unconditional rankings can give a misleading perception to the relative performance of
countries. In addition, the method is able to say where the strengths and weaknesses of each
country's policies appear to be—whether structural or institutional.
The value of benchmarking is that it relates the social spending levels of a country to
other countries at similar levels of income (R2) or institutional development (R3). Making
this ranking public information and open for debate may tend to strengthen the incentives for
low-performing governments to spend more effectively on safety nets and other social
expenditures.
84
Data Appendix
Cross-Country Data
All of the data are from the IMF Government Finance Statistics Yearbook, and each series is
available over the period 1972–97. The safety net variable definitions are:
Transfers to Organizations and Households
Current transfer payments to private social institutions such as hospitals and schools, learned
societies, associations, and sports clubs that are not operated as enterprises; current payments
in cash (not in kind) to households, adding to their disposable income without any
simultaneous, equivalent counterpart provided in exchange by the beneficiary, and neither
generating nor extinguishing a financial claim; usually intended to cover charges incurred by
households because of the appearance, or existence, of certain risks and needs.
Social Security and Welfare
Transfer payments (including in kind) to compensate for reduction or loss in income or
inadequate earning capacity; sickness, maternity, and temporary disablement benefits;
government employee pension schemes; old age, disability, or survivors’ benefits;
unemployment compensation benefits; family and child allowances; other social assistance to
persons and to residential institutions for children or the elderly.
Education
Pre-primary, primary, secondary (vocational and technical), tertiary, university, and
subsidiary services to education.
Health
General and specialized hospital services; nursing and convalescent home services; clinics;
medical, dental, and paramedical practitioners; public health affairs and services; medication;
prostheses; medical equipment; applied research; and experimental development.
The data source for the quality-of-government measures is the International Country Risk
Guide (ICRG), a monthly publication of Political Risk Services. This data set has been
compiled by Knack and Keefer (1995), where the following definitions are given:
Repudiation of Government Contracts
Indicates the “risk of a modification in a contract taking the form of a repudiation,
postponement, or scaling down,” due to “budget cutbacks, indigenization pressure, a change
in government, or a change in government economic and social priorities.” Scored 0–10, with
higher scores for higher risks.
Expropriation Threat
Assessment of risk of “outright confiscation” or “forced nationalization.” Scored 0–10, with
higher scores for higher risks.
Corruption in Government
Higher scores indicate “high government officials are likely to demand special payments”
and “illegal payments are generally accepted throughout lower levels of government” and in
the form of “bribes connected with import and export licenses, exchange controls, tax
assessment, policy protection, or loans.” Scored 0–6.
85
Rule of Law
This variable “reflects the degree to which the citizens of a country are willing to accept the
established institutions to make and implement laws and adjudicate disputes.” Lower scores
indicate “sound political institutions, a strong court system, and provisions for an orderly
succession of power.” Higher scores indicate “a tradition of depending on physical force or
illegal means to settle claims.” Upon changes in government in countries scoring low on this
measure, new leaders “may be less likely to accept the obligations of the previous regime.”
Scored 0–6.
Government Bureaucracy
Low scores indicate “autonomy from political pressure” and “strength and expertise to
govern without drastic changes in policy or interruptions in government services”; also
existence of an “established mechanism for recruiting and training”. Scored 0–6.
Indian State-Level Analysis
Public Finance Variables
Development expenditure includes expenditure on economic and social services. Economic
services include agriculture and allied activities, rural development, special area programs,
irrigation and flood control, energy, industry and minerals, transport and communications,
science, technology, and environment. Social services include education, medical and public
health, family welfare, water supply and sanitation, housing, urban development, labor and
labor welfare, social security and welfare, nutrition, and relief on account of natural
calamities. Health, education, food subsidies, and calamity relief expenditures are just a
component of social expenditures. The primary source for state-level information on taxes
and expenditures is an annual publication, Public Finance Statistics (Ministry of Finance,
Government of India). This information is also collated in the Reserve Bank of India’s
annual publication Report on Currency and Finance.
Public Food Distribution
Issues/public distribution of food grains (both from central and state governments) divided by
(interpolated) state population measured in tons per person. The source is the Bulletin on
Food Statistics, Ministry of Food and Agriculture, Directorate of Economics and Statistics.
State Income Per Capita
The primary source for data on state income is an annual government publication Estimates
of State Domestic Product (Department of Statistics, Ministry of Planning). The primary
sources for the Consumer Price Index for Agricultural Laborers (CPIAL) and Consumer
Price Index for Industrial Workers (CPIIW), which are used to deflate the agricultural and
non-agricultural components of state domestic product, respectively, are a number of
Government of India publications including the Indian Labor Handbook, the Indian Labor
Journal, the Indian Labor Gazette and the Reserve Bank of India's Report on Currency and
Finance. Ozler, Datt, and Ravallion (1996) have further corrected CPIAL and CPIIW to take
account of interstate cost-of-living differentials and have also adjusted CPIAL to take
account of rising firewood prices. Using their data allows us to put together a consistent and
complete series on real total, agricultural, and non-agricultural state income for the period
1960-92.
86
Poverty and Inequality
We use the headcount and gini measures for the rural and urban areas of India’s 16 major
states, spanning 1957–58 to 1991–92 put together by Ozler, Datt, and Ravallion (1996).
These measures are based on 22 rounds of the National Sample Survey (NSS), which span
this period.
Voter Turnout
The percentage of eligible voters in the state that actually voted in the last elections for the
state legislature. This data is from Butler, Lahiri, and Roy (1991).
Variance of Social Spending
The variance of social spending that cannot be explained by the variance of state income and
natural calamities.
Deviation from State Mean of Level of Social Spending
The deviation from state means of social spending, controlling for mean income levels and
the occurrence of natural calamities.
Construction of the Shock Variables
In the cross-country analysis, we constructed a “shock to GDP variable,” which is essentially
a residual from a standard growth regression. It thus proxies for how far a country is from its
long-run sustainable growth level. The form of the growth regression is standard, being taken
from the growth literature. We simply regress log GDP on initial GDP, population growth,
and capital per worker. A similarly constructed variable is used in the Indian state-level
analysis.
A second set of shock variables is constructed only for the Indian analysis. The variables
are intended to capture the deviations of social expenditures from their long-run stable paths.
The methodology is detailed in Wei (1997). The two variables are constructed by regressing
each type of expenditure on state income growth and state natural disasters, and forming the
mean and variance of the estimated residuals.
87
References
The word processed describes informally reproduced works that may not be commonly
available through libraries.
Alderman, H., J. R. Behrman, S. Khan, D. R. Ross, and R. Sabot. 1995. “Public Schooling
Expenditures in Rural Pakistan: Efficiently Targeting Girls and a Lagging Region.” In
D. van de Walle and K. Nead, eds., Public Spending and the Poor: Theory and
Evidence. Baltimore: The Johns Hopkins University Press.
Andreoni, J. 1990. “Impure Altruism and Donations to Public Goods: A Theory of WarmGlow Giving.” Economic Journal 100: 464–77.
Atkinson, A. B. 1995. “On Targeting Social Security: Theory and Western Experience with
Family Benefits.” In D. van de Walle and K. Nead, eds., Public Spending and the
Poor: Theory and Evidence. Baltimore: The Johns Hopkins University Press.
Bahl, R. C., K. Kim, and C. L. Park. 1986. Public Finances During the Korean
Modernization Process. Cambridge, Mass.: Harvard University Press.
Barr, N. 1994. “On the Design of Social Safety Nets.” London School of Economics.
Processed.
Barro, R. 1974. “Are Government Bonds Net Wealth?” Journal of Political Economy 82:
1095-1119.
Benabou, R. 1996. “Inequality and Growth.” NBER Macroeconomics Annual (Cambridge,
Mass.: MIT Press).
Besley, Timothy, and R. S. L. Burgess. 2000a. “Land Reform, Poverty Reduction and
Growth: Evidence from India.” Quarterly Journal of Economics 105(2): 389-430.
_____. 2000b. “Does Media Make Government More Responsive? Theory and Evidence
from Indian Famine Relief Policy.” London School of Economics. Processed.
Besley, Timothy, and A. Case. 1994. “Unnatural Experiments? Estimating the Incidence of
Endogenous Policies.” NBER Working Paper No. 4956. National Bureau of
Economic Research.
Browning, M. 1992. “Children and Household Behavior.” Journal of Economic Literature
30: 1434-75.
Butler, D., A. Lahiri, and P. Roy. 1991. India Decides: Elections 1952-1991. New Delhi:
Aroom Purie for Living Media India.
Card, J.D. and A. Krueger. 1994. "Minimum Wages and Employment: A Case Study of the
Fast-Food Industry In New Jersey and Pennslyvania." American Economic Review 84
(September): 772-93.
88
Cox, D. and E. Jimenez. 1995. "Private Transfers and the Effectiveness of Public Income
Redistribution In the Philippines." In D. van de Walle and K. Nead, eds., Public
Spending and the Poor: Theory and Evidence. Baltimore: The Johns Hopkins
University Press.
Deaton, A., and J. Muellbauer. 1986. “On Measuring Child Costs: With Application to Poor
Countries.” Journal of Political Economy 94: 720-44.
Deaton, A., and C. H. Paxson. 1996. “Economies of Scale, Household Size, and the Demand
for Food.” Princeton University, Research Program in Development Studies,
Princeton, N.J. Processed.
Deaton, A., K. Parikh, and S. Subramanian. 1994. “Food Demand Patterns and Pricing Policy
in Maharashtra: An Analysis Using Household-Level Survey Data.” Sarvekshana 17:
11-34.
Deolalikar, A. B. 1995. “Government Health Spending in Indonesia: Impacts on Children in
Different Economic Groups.” In D. van de Walle and K. Nead, eds., Public Spending
and the Poor: Theory and Evidence. Baltimore: The Johns Hopkins University Press.
Dreze, J. and A. Sen. 1990. The Political Economy of Hunger Vol II: Famine Prevention.
Oxford: Clarendon Press.
Gertler, P., and P. Glewwe. 1989. “The Willingness to Pay for Education in Developing
Countries.” Living Standards Measurement Survey Working Paper No. 54. World
Bank, Washington D.C.
Gertler, P., and J. van der Gaag. 1990. The Willingness to Pay for Medical Care: Evidence
From Two Developing Countries. Baltimore: The Johns Hopkins University Press.
Gertler, P., L. Locay, and W. Sanderson. 1987. “Are User Fees Regressive? The Welfare
Implications of Health Care Financing Proposals in Peru.” Journal of Econometrics
33: 67-88.
Gruske, K. D. 1985. “Redistributive Effects of the Integrated Financial and Social Budgets in
West Germany.” In Public Finance and Social Policy: Proceedings of the 39th
Conference of the International Institute of Public Finance. Detroit: Wayne State
University Press.
Haddad, L., and R. Kanbur. 1990. “How Serious is the Neglect of Intra-Household
Inequality?” Economic Journal 100: 866-81.
Haddad, L., and R. Kanbur. 1993. "The Value of Intrahousehold Survey Data for Age-Based
Nutritional Targeting." Annales d'Economie et de Statistique 29: 65-81.
Hammer, J. S. 1997. “Economic Analysis for Health Projects.” World Bank Research
Observer 12: 47-71.
89
Hanmer, L. C., L. Bijlmakers, M. T. Basset, D. Sanders, and G. Chapman. 1998. “Human
Capital, Targeting and Social Safety Nets: An Analysis of Household Data from
Zimbabwe.” Oxford Development Studies 26(2): 245-265.
Hertel, T. W. 1989. “Applied General Equilibrium Analysis of Agricultural Policies.” Purdue
University. Processed.
Holzmann, R., and S. Jorgensen. 1999. “Social Protection as Social Risk Management:
Conceptual Underpinnings for the Social Protection Sector Strategy Paper.” Social
Protection Discussion Paper No. 9904. World Bank.
IMF (International Monetary Fund). IMF Government Finance Statistics Yearbook. Various
years, 1973-1998.
India, Ministry of Labour. Indian Labor Gazette. New Delhi.
India, Ministry of Labour. Indian Labor Handbook. New Delhi.
India, Ministry of Labour. Indian Labor Journal. New Delhi.
India, Ministry of Finance. Public Finance Statistics. New Delhi.
India, Ministry of Food and Agriculture, Directorate of Economics and Statistics. Bulletin on
Food Statistics. New Delhi.
India, Ministry of Planning, Department of Statistics. Estimates of State Domestic Product.
New Delhi.
Jarvis, S. J., and J. Micklewright. 1995. “The Targeting of Family Allowances in Hungary.”
In D. van de Walle and K. Nead, eds., Public Spending and the Poor: Theory and
Evidence. Baltimore: The Johns Hopkins University Press.
Jimenez, E., and D. Cox. 1992. “Private Transfers and Public Transfers in Developing
Countries: A Case of Peru.” The World Bank Economic Review 6: 155-69.
Knack, S., and P. Keefer. 1995. “Institutions and Economic Performance: Cross Country
Tests Using Alternative Institutional Measures.” Economics and Politics 7: 207-27.
LeGrand, J. 1982. “The Distribution of Public Expenditure on Education.” Economica 49:
63-8.
Lindert, P. H. 1994. “The Rise of Social Spending, 1880-1930.” Explorations in Economic
History 31: 1-36.
Lindert, P. H. 1996. “What Limits Social Spending?” Explorations in Economic History 33:
1-34.
Lipton, M. and M. Ravallion. 1995. "Poverty and Policy" In J. Behrman and T.N. Srinivasan,
eds. Handbook of Development Economics Volume 3. Amsterdam: North-Holland.
90
McKenzie, G. W. 1983. Measuring Economic Welfare: New Methods. Cambridge:
Cambridge University Press.
Meerman, J. 1979. Public Expenditure in Malaysia: Who Benefits and Why? New York:
Oxford University Press.
Ozler, B., G. Datt, and M. Ravallion. 1996. “A Data Base on Poverty and Growth in India.”
World Bank Research Working Paper Washington, D.C. Processed.
Parikh, K., and T. N. Srinivasan. 1989. "Poverty Alleviation Policies in India: Food
Consumption Subsidy, Food Production Subsidy and Employment Generation.” Yale
University, New Haven, Conn. Processed.
Piggott, J., and J. Whalley. 1987. “Interpreting Net Fiscal Incidence Calculations.” Review of
Economics and Statistics 69: 685-94.
Pitt, M., M. R. Rosenzweig, and D. M. Gibbons. 1995. “The Determinants and Consequences
of the Placement of Government Programs in Indonesia.” In D. van de Walle and K.
Nead, eds., Public Spending and the Poor: Theory and Evidence. Baltimore: The
Johns Hopkins University Press.
Political Risk Services. International Country Risk Guide. 1995.
Prescott, N. 1997. Choices in Financing Health Care and Old Age Security. Proceedings of a
Conference Sponsored by the Institute of Policy Studies, Singapore, and the World
Bank. World Bank Discussion Paper No. 392. Washington, D.C.
Ravallion, M. 1998. "Appraising Workfare Programs." World Bank Research Working Paper
No. 1955.
Ravallion, M., D. van de Walle, and M. Gautman. 1995. “Testing a Social Safety Net.”
Journal of Public Economics 57: 175-99.
Reserve Bank of India. Report on Currency and Finance. New Delhi.
Riboud, M. 1990. “Costa Rica: Public Sector Social Spending.” World Bank, Latin American
Region, Washington, D.C. Processed.
Reynolds, M., and E. Smolensky. 1977. Public Expenditures, Taxes, and the Distribution of
Income: The United States, 1950, 1961, and 1970. NewYork: Academic Press.
Rodrik, D. 1998. “Globalization, Social Conflict and Economic Growth.” Harvard
University, Cambridge, Mass. Processed.
Sala-i-Martin, X. 1997.“Transfers, Social Safety Nets, and Growth.” IMF Staff Papers 44(1):
81-102.
91
Selden, T. M., and M. J. Wasylenko. 1992. “Benefit Incidence Analysis in Developing
Countries.” World Bank Research Working Paper No. 1015 Country Economic
Department, Washington, D.C.
---------. 1995. “Measuring the Distributional Effects of Public Education in Peru.” In D. van
de Walle and K. Nead, eds., Public Spending and the Poor: Theory and Evidence.
Baltimore: The Johns Hopkins University Press.
Selowsky, M. 1979. Who Benefits From Government Expenditure? NewYork: Oxford
University Press.
Sen. 1985. "Goals, Commitment and Identity." Journal of Law, Economics and
Organization. Vol. 1 (2): 341-55.
Snedecor, G. W., and W. G. Cochrane. 1989. Statistical Methods, 8th ed. Ames: Iowa State
University Press.
Subbarao, K. A. Bonnerjee, J. Braithwaite, S. Carvalho, K. Ezemenari, C. Graham, and A.
Thompson. Safety Net Programs and Poverty Reduction: Lessons From CrossCountry Experience. Directions In Development, World Bank: Washington D.C.
Todaro, M. P. 1969. “A Model of Labor, Migration and Urban Unemployment in Less
Developed Countries.” American Economic Review 59: 138-48.
van de Walle, D. 1998.“Assessing the Welfare Impacts of Public Spending.” World
Development 26(3): 365-79.
van de Walle, D. and K. Nead, eds. 1995. Public Spending and the Poor: Theory and
Evidence. Baltimore: The Johns Hopkins University Press.
van de Walle, D., M. Ravallion, and M. Gautman. 1994. “How Well Does the Social Safety
Net Work? The Incidence of Cash Benefits in Hungary 1987- 89. ” Living Standards
Measurement Study Working Paper No. 102. World Bank, Washington, D.C.
Wei, S. 1997. "Why Is Corruption So Much More Taxing than Tax ? Arbitrariness Kills."
NBER Working Papers No. 6255. National Bureau of Economic Research.
Williamson, J. G. 1988. “Migration and Urbanization.” In H. Chenery and T. N. Srinivasan,
eds., The Handbook of Development Economics, vol.1. Amsterdam: North Holland.
World Bank. World Development Indicators database.
http://www.worldbank.org/data/wdi/home.html.
92
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement