Unpacking piped water consumption subsidies: Who benefits? New evidence from 10 countries

Universal access to water and sanitation are stated international sustainable development goals (SDGs). The United Nations adopted these SDGs in 2015 with the aim to ‘achieve universal and equitable access to safe and affordable drinking water’ and to ‘achieve access to adequate and equitable sanitation and hygiene for all’ by 2030. However, many countries around the world are still struggling to achieve this goal. In 2015, 29% of the world population lacked access to safely managed water and 61% lacked safely managed sanitation (WHO/UNICEF 2017). The cost of meeting these gaps is estimated to be around $100 billion a year (Hutton & Varughese 2016). Well-designed subsidies stand out as a key instrument to achieve these SDGs, alongside other policies such as investment, technological innovations or better governance and planning. In the context of high levels of poverty, where markets alone do not result in the desired levels of service provision and consumption, targeted subsidies can help address affordability and equity issues. Additionally, water and sanitation subsidies are generally advocated because access to and consumption of water and sanitation services are associated with rising productivity and living standards, positive externalities related to public health, and may free up time spent collecting water (WWAP 2016; Hutton & Chase 2017). For example, Hutton & Chase (2017) show that health and non-health associated costs due to poor water and sanitation are estimated to be over 5% of GDP in 6 of the 30 countries included in their study.

A subsidy represents the difference between the supplier's cost of providing the service and the actual price or tariff paid by a user. Consumption subsidies reduce the cost of consuming water and, as such, are only available to existing customers. These can be distinguished from connection subsidies, which are a one-time reduction in connection charges and hence only available to new, previously unconnected, customers. Water consumption subsidies can take different shapes and sizes: untargeted subsidies (general underpricing of water supply that benefits all consumers); implicit subsidies (generated by flat fees for unmetered services, low meter coverage or low revenue collection); or explicit subsidies (e.g., quantity targeting using increasing block tariffs (IBTs) or subsidies using administrative selection, such as means-tested or geographic targeting). The funding for these subsidies comes from one or a combination of the following two sources: government transfers and rate funds (that arise from charging other users more than the cost of the service, also called cross-subsidies). Finally, the transfer mechanism could be implemented in different ways. In a demand-side subsidy, usually the most transparent type of subsidy, the government provides a monetary transfer directly to the user, who uses this transfer towards their payments to the service provider. In a supply-side subsidy, the government transfers the money to the provider. Andres et al. (2019) and Komives et al. (2005) discuss the different types of WASH subsidies in more detail.

The design and performance of piped water subsidies hinges on the industrial structure and technology of piped water production and delivery and is a contentious area, heavily influenced by political economy considerations. An important challenge for policymakers is to design a subsidy scheme that allows for the recovery of capital and expenditure costs of providing the service (WWAP 2019). Piped water is usually considered a local natural monopoly. The supplier faces large upfront (fixed) costs (resulting in increasing returns to scale) on assets that are long-lived. This feature of water supply, as with other similar public utilities, makes pricing difficult. Allocative efficiency requires prices to be set equal to marginal cost, i.e., the good is consumed up to the point where the benefit consumers obtain from the last unit consumed equals the additional cost incurred by the supplier to produce it. However, pricing at marginal cost in a context of high fixed costs would not allow for full cost recovery (as the marginal cost is lower than the average costs). This feature generates heated debates around how to implement low water tariffs (i.e., close to the marginal cost) while ensuring cost recovery and avoiding financial unviability of utilities. (In the context of lower-income countries, it is recommended that tariffs cover at least the operating expenditures incurred by suppliers, but even this is often not achieved (see, for instance, Baietti & Curiel 2005; Andres et al. 2019).) This, in part, is because it is really difficult to estimate the cost of providing each customer with the service, as costs vary across consumers, depending on geographic location, topography, distance to the source, etc. This situation translates into a high degree of discretion on how to allocate provision costs.

In practice, this often means subsidization of the service across the board. Most countries, whether rich or poor, end up putting in place subsidies schemes funded by the government or designing tariffs so that richer consumers cross-subsidize poorer ones (GWI 2004; Komives et al. 2005; Andres et al. 2019). New estimates by Andres et al. (2019) put total subsidies for networked water and sewerage services at around 0.5% of GDP worldwide, and over 1.5% of GDP for non-advanced economies. In developing countries, this subsidization across the board does not fully address the objectives of universal access to clean water and the affordability of the service, and additionally may distort consumption patterns. Furthermore, high subsidization levels have often undermined the financial sustainability of the system, inducing managers to face soft budget constraints and lowering their financial performance. This, in turn, could reduce the ability of service providers to access commercial finance, which could enable utilities' ability to maintain and expand the service in an affordable manner to new, likely poorer customers (see, for instance, Goksu et al. 2017).

In this paper, we provide new evidence on the distributional incidence and pro-poor targeting of subsidies for piped water consumption for 10 low- and middle-income countries around the world: Ethiopia, Mali, Niger, Nigeria, Uganda, El Salvador, Jamaica, Panama, Bangladesh and Vietnam. We also document the gap between estimates of actual paid tariffs and tariffs that would be needed to recover the full cost of providing the service (also called cost-recovery or cost-reflective tariffs). All these countries capture relevant heterogeneity since they reflect different levels of gross domestic product per capita, connection rates, tariff structures and cost-recovery tariffs. The cases considered include quantity-based tariffs and flat rates. Importantly, in these countries, there is the availability of household surveys with measures of water expenditure and overall consumption expenditures, as well as administrative data on tariffs from the utility and new estimates of country-specific cost-reflective tariffs. These estimates are calculated using an improved methodology and new data presented in Andres et al. (2019) and discussed in more detail in the next section.

Our results suggest that, in all the countries analyzed, mean unit prices charged are, on average, lower than the overall cost of producing and distributing piped water, resulting in substantial water consumption subsidies. In addition, subsidies tend to be regressive, with the amount of resources allocated to water consumption subsidies increasing over the expenditure distribution, and richer households in the top deciles usually capturing the lion's share. On average, across the 10 low- and middle-income countries examined, 56% of subsidies end up in the pockets of the richest 20%, but only 6% of subsidies find their way to the poorest 20%.

Our findings also highlight the importance of access to the service in explaining the regressive nature of these subsidies. In particular, (i) poor households live in areas that are not covered by piped water networks (poor households are defined as belonging to the first four deciles of the countrywide national expenditure (or income) per capita distribution in each country), (ii) poor households live in areas with coverage but are not connected to the network and (iii) poor households that are connected to the network appear to be consuming smaller quantities of water than the general population. This implies that there are high errors of inclusion (i.e., households not among the 40% poorest receiving subsidies) and even higher errors of exclusion (households among the 40% poorest not receiving subsidies). This issue is particularly pronounced in the considered African countries, where errors of exclusion fall between 90 and 100%.

Our study contributes to a vast and growing literature on the distributional incidence of fiscal policy in low- and middle-income countries, including a large sub-literature on the distributional incidence of water subsidies. Our findings are consistent with a number of previous studies that use a similar methodology. Key examples of this literature – such as Komives et al. (2006) and Angel-Urdinola & Wodon (2007b) – have shown that quantity-based, targeted subsidies in Nicaragua, Cape Verde, the city of Kathmandu (Nepal), Bangalore (India) and Sri Lanka are usually regressive, with a smaller share of benefits accruing to the poor than the general population. These studies also indicate that poor targeting is mostly associated with low rates of access to water networks in poor neighborhoods, as well as low connection rates for poor households in neighborhoods with access. More recently, the World Bank (2017) found that Tunisian households in the bottom quintile of the income distribution receive 11% of water subsidies, while the top quintile receives 27%. Our study provides systematic cross-country evidence, using a novel and more robust estimate of cost-recovery tariffs based on actual country-specific water providers’ data.

Lack of access to the service by the poor has been identified as an important driver of regressive subsidies in other sectors. There are numerous studies documenting similar patterns in the electricity sector as documented by Komives et al. (2006). Recent studies conducted as part of the Commitment to Equity (CEQ) initiative looking at the distributional incidence of fiscal policy in a range of low- and middle-income countries also confirm these findings using a different methodology (see, for instance, Hounsa et al. (2019) for energy subsidies in Mali). CEQ studies, discussed in Inchauste & Lustig (2017), find that lack of access by the poor to tertiary education is one of the reasons why public education expenditure is regressive in Ethiopia, Ghana, Guatemala and Indonesia. Access also plays a role in the distributional incidence of health subsidies. For example, Chen et al. (2015) find that Chinese government healthcare subsidies are pro-rich and regressive and that lack of access plays an important role. O'Donnell et al. (2008) show the distribution of public healthcare subsidies in Vietnam is pro-rich, although the impact varies by the type of healthcare facility.

Finally, quantity-based consumption subsidies have been found to be regressive, even conditional on having access and being connected to the network. For example, Cardenas & Wittington (2019) use data for connected households in Addis Ababa, Ethiopia and find that water and electricity quantity-based subsidies are large and regressive. They use improved measures of quantities consumed by accessing administrative utilities billing data and matching it to household survey data. In contrast, when subsidies are targeted using administrative-based mechanisms (i.e., geographic targeting or means testing), the evidence suggests that these are likely to be more progressive and pro-poor (Komives et al. 2006).

The rest of the paper is structured as follows: Section ‘Data and methodology’ presents the methodology, the data used and related statistics; Section ‘Findings’ presents the findings and the last section summarizes the findings and discusses the policy implications.

In order to estimate consumption subsidies for piped water at the household level, we need a measure of the cost of delivering a unit (cubic meter, m³) of piped water, a measure of the unit price actually paid by each household and the quantity consumed. To do this, we combine several data sources spanning different periods within and across countries. We use a methodology that closely follows Komives et al. (2005) and Angel-Urdinola & Wodon (2007a), to estimate the distributional and targeting performance of subsidies in each country.

We first describe the multiple data sources used, and then explain the methodology and its implementation. We subsequently provide some descriptive statistics before discussing the limitations of the data and methodology used in this study.

Household level data comes from the latest available socio-economic (or income and expenditure) household surveys collected in each country. Most surveys are from 2015 or 2016, but some surveys are from earlier periods, the earliest being 2012 for Jamaica. Information about the household surveys used for each country considered in the analysis is provided in Supplementary Appendix Table A1.

These are general socio-economic surveys that cover a range of dimensions of a household's characteristics and income and spending patterns, and hence are not focused specifically on water and sanitation. Nonetheless, they provide useful self-reported information on water access, connection and water expenditure in the last month or last year, depending on the survey. Details about the exact variables used in each survey are also provided in the Supplementary Appendix. Quantities of consumed water and prices or tariffs paid by each household are usually not reported. Consequently, we combine household survey data on water expenditure with administrative data from providers and government programs detailing the tariff structure for water and sewerage services to impute water quantities and tariffs, as below.

This approach has been used in other studies, such as in Komives et al. (2006) and Angel-Urdinola & Wodon (2007b). However, it is not without limitations. In particular, water expenditure in the last month (some surveys asked about expenditure over the last 12 months) can provide an imperfect proxy for the expenditure on piped water if (i) households use multiple water sources, including technologies for treating non-piped water (in which case it could overestimate piped water expenditure); (ii) meters are not read, shared (Supplementary Appendix Table A2 shows the proportion of households that use the neighbor's tap as their main source of drinking water for the 4 countries for which the data is available: Mali, Niger, El Salvador and Panama. The proportion is low and varies between 1% for all households in Panama and rural households in Niger, to 10% for urban households in Niger and 7% for urban households in Mali) or do not work; or (iii) the water bill includes pro-rated connection charges and arrears (Whittington et al. 2015). Importantly, these surveys also enable the construction of total expenditure (and/or income) variables for each household.

Information about tariff structures comes from the International Benchmarking Network for Water and Sanitation Utilities (IBNET) database. Some countries list most of their water providers in IBNET, while others list only a few. For example, IBNET has data on the tariff structures of all water providers in Jamaica, Mali and Niger, since these have only one piped water provider. Meanwhile, for countries such as Ethiopia, Nigeria, Uganda and Bangladesh – where the provision of services is decentralized and there is a multitude of piped water utilities – IBNET data covers a subgroup of these providers. An additional limitation is that the tariff structure data available to the authors for this study may cover a period that is different than the one covered in the household survey for each country, even though tariffs may not be regularly updated. More information about dates covered for each utility in each country is provided in the Supplementary Appendix.

Cost-reflective tariffs cover the capital and recurrent costs of providing service, including not only the efficient economic cost but also costs arising from inefficiencies of the service provider. We use estimates of cost-reflective tariffs (defined as CRT) that cover both operating expenditure (OPEX) and capital expenditure (CAPEX), obtained from Andres et al. (2019).

Their approach entails answering the following question ‘What is the long-run incremental cost of providing water and sanitation services for a given company?’. It is based on a simple, utility-wide, bottom-up model used by regulators in many utility sectors. In the water and sanitation sector, the model firm approach is used in Chile and a few other Latin American countries. The authors choose efficient firm estimates from Chile as a benchmark to determine the capital stock per customer. The Chilean tariff law seeks to induce efficiency through the use of incremental cost of development pricing. Decree – D.F.L. No 70/1988 – defines this tariff as the ‘value equivalent to a constant per unit price which, when being applied to the incremental forecasted demand, generate revenues to cover incremental operation efficient costs and the required investment from an optimized project of expansion of the firm, such that it should be consistent with a net present value (NPV) of the project equal to zero.’ As pointed out by Bitrán & Arellano (2005): ‘In Chile, to avoid transferring the cost of inefficiencies to users, the rate setting process emulates competitive conditions by using a fictitious company that would theoretically meet demand over the next five years in the most efficient way.' Using the information on operation and maintenance costs for the period 2010–2015 from the IBNET dataset, they estimate efficient cost-reflective tariffs. In addition, this approach accounts for inefficiencies arising from employees and losses, to obtain a total cost-reflective tariff. Despite its stringent assumptions and simplifications, this comprehensive approach improves on the existing estimates that have been used in the literature. For example, the Price Gap Approach uses a single uniform CRT value for all countries of 1,27 USD/m³. There are a range of papers by IMF staff that have used this approach to calculate implicit subsidies to public utilities. Recently Kochhar et al. (2015) have applied it to networked water, while Ebeke & Ngouana (2015) used it to estimate energy subsidies. Another methodology, known as the Hidden Cost Calculator (Ebinger 2004), uses book values of gross assets instead.

In the following subsections, we describe how we classify households by economic status (i.e., poor or not), identify whether they benefit from water consumption subsidies, and estimate the magnitude of the subsidies they receive. All household level statistics are calculated using sampling weights to correct for the bias inherent in representative household surveys (Deaton 2019). Further information on how each variable is constructed for each country is provided in the Supplementary Appendix.

Figure 1 summarizes how we classify households and Table 1 defines the main variables.

Poor households are defined as belonging to the first four deciles of the countrywide national total expenditure (or income) per capita distribution in each country and are denoted by the subscript p. All households in each country are denoted by the subscript h.

Total expenditure includes expenditure on all goods and services, including the water bill, which is a common methodology to rank households according to their resources in developing countries (see, for example, Abramovsky & Phillips (2015)). Annual expenditure figures are converted to equivalent monthly figures. It is important to note that although we conduct our analysis at the household level, we construct deciles of expenditure (or income) on the basis of a household's per capita expenditure, not overall household expenditure. Because we conducted our analysis of expenditure at the household level – and average household size may vary across deciles, particularly in Africa – average total expenditure at the household level may not increase as expected when moving across the household per capita expenditure distribution.

Following Angel-Urdinola & Wodon (2007a), households are considered to have potential access to a network if they are located in a service area (SA_h). (The variable service area (SA) is equivalent to the variable access (A) in Komives et al. (2005) and Angel-Urdinola & Wodon (2007a).) In this paper, we prefer to use service area because it better reflects the variable we are measuring. A household's service area takes the value of 1 if at least one household in their neighborhood self-reported having a water connection in the household survey. We assume that households in an SA have the option of connecting to the piped water network present in their neighborhood. The neighborhood is proxied by the enumeration area where the household is located, according to the household surveys. However, this assumption may not always be correct. If a neighborhood covers a large geographical area, or if the presence of an adjacent water main varies by household, the survey figures may overestimate the actual number of households with potential access to the piped water network.

We set this variable to 1 if a household is connected to the water network, and to zero otherwise. This is determined by the self-reported main source of drinking water in the household surveys.

We construct Q_h|C, the monthly water quantity consumed (measured in m³) by imputing water consumption volumes from self-reported household monthly water expenditure, E_h|C, conditional on being connected to the network. The conversion is undertaken using the corresponding tariff structure from the IBNET data. (Information on how this match is done for each country is provided in the Supplementary Appendix.)

In the first round of calculations, we impute quantities for households for which (i) the total monthly bill for water consumption depends on quantities consumed, (ii) the value of self-reported expenditure on water is greater than zero and (iii) we can map the household to a specific provider for which we have tariff structure information. For households that pay a fixed rate per cubic meter, the monthly quantity of water consumed equals the total expenditure on water divided by the rate. For households facing unit prices that vary by quantity consumed (such as increasing block or volume-differentiated tariffs), each block within the tariff structure is assigned a maximum expenditure level and associated quantity of water consumed. Then, quantities are assigned to the household sample by matching self-reported household water expenditure with the corresponding level of the tariff structure data.

For the remainder of the households for which (i) the bill is a flat rate unrelated to water quantities consumed or (ii) self-reported expenditure on water equals zero, we assume consumption quantities at the median of those households in the same country for which we could impute quantities. In many countries, and particularly Ethiopia and Nigeria, many households appear to be connected to the network and use piped water as their main source of drinking water but report paying zero for their water. Anecdotal evidence supports the fact that many households are connected illegally to the network or that utilities do not invoice and collect revenues as they should.

For those households that report being connected to the piped water network but are located in areas for which we cannot map them to a specific provider, we use the median water quantity consumed for those we could impute the quantity (disaggregated by rural/urban and income deciles if enough observations are available).

Supplementary Appendix Table A1 shows the breakdown of the sample according to whether connected households are matched to corresponding volumetric tariffs and have positive water expenditures, allowing for the estimation of quantity consumed, or whether water quantities and the unit price paid have to be assumed using the information from other households in the data. The proportion of observations relying upon requiring the latter approach varies across countries significantly, with Nigeria and Bangladesh exhibiting the highest proportions.

We define the unit price for a household h (UP_h) as the ratio of self-reported water expenditure from household surveys and quantity consumed. For a household with a water expenditure of zero, the unit price would also be zero.

We define variable B_h as equal to 1 if a household is connected and pays a unit price (UP_h) for water that is lower than the cost-reflective tariff. That is, a household is a subsidy beneficiary.

The monthly value of the subsidy (S_h) is constructed using the variables described above; for each household, this is equal to S_h = Q_h|B × (CRT–UP_h), conditional on UP_h < CRT (i.e., B_h = 1).

The rate of subsidization for a household h, conditional on receiving a subsidy, is R_h|B = 1–E_h|B/(Q_h|B × CRT) = 1–UP_h/CRT.

Having identified beneficiary households (B_h) and the size of subsidies (S_h) they receive, we estimate the share of total subsidies accrued to each expenditure (or income) decile and consider whether they are progressive or not. This methodology is arithmetic, nonbehavioral, and expresses a partial equilibrium, which is well suited to investigate the incidence of subsidies at a point in time or with small marginal changes to prices or subsidies. This approach has limitations related to bigger changes in prices or subsidies. See, for instance, Abramovsky & Phillips (2015) for a description of various microsimulation models for tax and benefits using household survey data.

• (1)

  Progressive subsidies: Subsidies are considered progressive when the share of subsidies accrued to each decile tends to decrease over the expenditure (or income) per capita distribution. This means that poor households capture a higher portion of the subsidy pie.

• (2)

  Regressive subsidies: Conversely, subsidies are considered regressive when the share of subsidies accrued to each decile tends to increase over the expenditure (or income) distribution.

This relationship can sometimes be non-monotonic. In those cases, it is useful to compare the share of subsidies that the top decile receives relative to, for example, the bottom 40% of the distribution, which is known as the Palma ratio and widely used in developing countries (Cobham et al. 2016). Or the 20/20 ratio, the share of subsidies the top two deciles receive relative to the bottom 20% of the distribution. The higher the ratio, the more regressive the subsidies are.

We also estimate the share of overall countrywide household-level expenditure on goods and service captured by each decile to contrast with the share of piped water subsidy captured by each decile. This helps understand whether the subsidy is inequality reducing – that is, whether a greater percentage of subsidies is allocated to households in poorer deciles than is the case for overall countrywide expenditure. When the subsidy distribution is more skewed toward richer deciles than the distribution of expenditure or income, the subsidy is regressive and increases inequality. However, if subsidies are less regressive than the distribution of expenditure or income, they still can reduce inequality. These measures are simple statistics that can be easily displayed in bar charts and are conceptually similar to the quasi-Gini coefficients, described, for example, in Komives et al. (2005:140–41).

Following Angel-Urdinola & Wodon (2007a), we define the targeting performance indicator (Ω), which relates to a subsidy's distributional incidence, as the ratio between the amount of subsidies the poor receive (S_P) and the amount of subsidies accrued to all households (S_H), divided by the proportion of poor households in the total population (P/H).

Access factors (SA and C) are fixed in the short term since they are determined by network expansion and households’ decisions, and will affect the distributional incidence of the subsidy, regardless of the consumption subsidy's type or structure.

Two additional measures to assess the targeting performance of subsidies are the errors of inclusion and exclusion.

The error of exclusion (EE) is the share of households in poverty that is not benefiting from the subsidy: EE = 1 − (B_P/P) = (P – B_P)/P.

The error of inclusion (EI) is the share of beneficiary households that are not in poverty: EI = B_NP/B_H = B_NP/(B_P + B_NP). Figure 2 shows graphically how these errors are defined.

Table 2 presents some descriptive statistics regarding household connection to piped water for each country analyzed. Countries are ordered by descending levels of GDP per capita (column 1). The figures reported in columns 2–6 are from household surveys and are, therefore, nationally representative, with the exception of Vietnam. (In Vietnam, the survey includes only five regions: Hanoi City, Da Nang, Dak Nong, Thu Dau Mot City and Bình Du'ơng Province, and Ho Chi Min City.) Overall, richer countries tend to have a lower proportion of households living in rural areas and a higher proportion of households that report using piped water as their main source of drinking water. It is worth noting that there may be slight differences from the Joint Monitoring Program (JMP) data for each country, due to some minor differences in variable definitions in most cases and differences in sources in others. One common difference between our computation and JMP's computation is that we define a household as connected to piped water if it uses piped water as its main source of drinking water either in the dry or wet season. JMP also includes public standpipes as part of its piped water definition in some countries such as Ethiopia and Mali, whereas we include only house connections. Thus, our figures may be easily compared against JMP figures showing house connections only. For example, figures for Ethiopia and Nigeria are slightly higher in Table 2 (19.8% and 13.6%, respectively, in column 3) than those presented by JMP (15.2% and 12.2%, respectively). Despite wide variation in coverage across countries, all countries demonstrate higher connection levels in urban households than both rural and poor ones. As one might expect, there are also wide variations across the administrative geographic areas covered in the household surveys within each country, which are not shown in this paper. Notably, Nigeria's connection rates are quite low given the country's income per capita, but disparities among different population segments are smaller. Latin American countries have substantially higher connection rates, between 70 and 95% for the general population, and around 50% or over for rural and poor households. Vietnam falls in the middle, with a connection rate of 53% for the whole population and 70% for urban households.

Table 3 summarizes the number of providers from the IBNET dataset whose service areas are also covered in the household surveys, the average cost-reflective tariffs and imputed unit prices (conditional on the reporting of positive expenditure on water). More information about IBNET data for each country can be found in the Supplementary Appendix. Most countries have IBT structures, though two African countries – Nigeria and Uganda – have fixed rates. In Bangladesh, some service providers charge a flat water consumption rate, while others use IBT. El Salvador is the only country that has a volume-differentiated tariff (VDT) structure. Some countries or providers also add a value-added tax (VAT) on tariffs. For the analysis carried out in the following section, we use all the households included in every country's household survey, coupled with information from the corresponding utilities. As explained earlier in this section, when a household is located in an area or region with no available tariff structure, we use information from other households in that country to impute quantities and unit prices paid.

Remarkably, all countries show higher cost-reflective tariffs than the average unit price paid by all households that report paying a positive amount for piped water. This already shows that all countries’ providers operate in a way that, on average, subsidizes residential consumers. However, 6 out of the 10 countries show a unit price higher than the estimated CRT_opex, suggesting that they can at least cover operating expenditures.

The study presents limitations that must be taken into account when interpreting the findings, many of which have already been discussed. The study also has important strengths. We summarize them below.

Due to data constraints, the data used spans different periods within each country. For example, household data for Jamaica is from 2012, whereas the tariff data is from 2013. Or within Nigeria, we have data on tariff structures for nine regional providers, spanning from 2012 to 2018, while the household data is from 2015 to 2016.

The data on water expenditure are self-reported and can sometimes include expenditure not only on piped water but also other water sources. If the piped water supply is shared with other households, water expenditure figures may not accurately reflect the amount paid by the reporting household. Since this variable is used to impute the quantity of water consumed and the average unit price paid by households, these variables can suffer from inaccuracies. For example, if the expenditure is overestimated, then quantities will be overestimated. If the measurement issue varies with the level of income of households, this could be biasing our results for those households that are connected to the network. In addition, in many countries, a significant proportion of households report missing or zero expenditure on water even if they report using piped water on premises as their main source of drinking water. Quantities for these households must be imputed from the other households within the sample.

In some countries, we only have data on tariff structures for some service areas, which means that, for a significant proportion of connected households, we must use average prices and quantities consumed from those areas with data. We have explained how we impute quantities in the section ‘Quantity of water consumed among those connected (Q)’, and we have discussed that this limitation is more important in countries like Nigeria and Bangladesh. This could affect the distributional impact of subsidies across deciles, the calculation of omega and the calculation of inclusion and exclusion errors, but it is not clear in which direction. Having said this, the main driver of regressivity is the lack of access or connection. The unavailability of data on expenditure or tariffs would not affect these measures. However, they could affect the magnitude of the regressivity conditional on being connected.

The definition of the service area may be too large and inaccurate, and this may overestimate the actual number of households with potential access to the piped water network. This may affect the calculation of omega and underestimates the magnitude of the impact of access factors on targeting performance. It will not, however, affect the distributional impact analysis of subsidies nor the calculation of the errors of inclusion or exclusion. Additionally, we are comparing countries with varying proportions of rural populations and differing degrees to which piped water is a feasible solution to close the access gap.

One of the main strengths of this study is that it uses representative samples that cover the poorest households; these households are least likely to have access and be connected to piped water. Studies that restrict their sample to only those households for which water quantity data are available to fail to be representative at the country level. If the objective of policymakers is to use limited public funds to help the poorest access safely managed water, then the strengths of our approach outweigh its limitations.

In this section, we show the estimated overall distributional incidence of water consumption subsidies across households by expenditure decile (or income decile for El Salvador and Panama).

As explained in Section ‘Distributional incidence of subsidies’, each figure shows the percentage of subsidy captured by each decile (i.e., percentage of money spent on subsidies accruing to all households in the given decile, classifying all households in the country along the countrywide expenditure distribution). In addition, we present the share of total household income or expenditure on goods and services in the economy accruing to each expenditure decile and compare it with the share of subsidies accruing to each decile. When the subsidy distribution is more skewed toward richer deciles than the distribution of expenditure or income, it is regressive and increases inequality. If the subsidy shares are less skewed toward richer deciles that total expenditure, but still regressive, they can be regressive but with inequality reducing.

Presented in Figure 3, the estimates for the five African countries demonstrate a common trend: richer households appear to enjoy a greater share of subsidies. In Niger and Uganda, households in the top decile are estimated to receive over 60% of all water consumption subsidies. In Mali, Niger and Uganda, the estimates show that the poorest hardly benefit from the subsidy at all. Furthermore, Nigeria seems to be the only country where subsidies reduce rather than increase inequality according to these estimates. Table 4 further shows the summary measures of distributional incidence. It shows that the subsidies are most regressive in Niger, with the top 20% of households receiving almost 85% of subsidies and the bottom poorest households receiving 0%.

Figure 4 shows the estimated distribution of beneficiaries (households that receive the subsidy) across expenditure deciles for the same five countries to complement the picture painted by distributional incidence. This figure helps to visualize the poor performance of these subsidies in targeting the poor by presenting the number of beneficiaries rather than the value of subsidies received across the deciles of total expenditure. Higher deciles appear to show the highest share of total subsidy beneficiaries (dark gray bars) and the highest share of all households within each decile that are beneficiaries (light gray bars). The differences between the top three deciles and the remaining ones are remarkable in all countries, although less pronounced in Nigeria. As noted above, in Niger, the poor (first four deciles) do not seem to benefit from the subsidy at all; Uganda shows a similar pattern, and in Mali, only the fourth decile shows a very small number of beneficiaries. We conclude that in the five African countries analyzed, and using the household data, tariff data and the methodology described, consumption subsidies are regressive (albeit relatively less so in Nigeria) and not well targeted to the poor. This seems to be due to the poor appearing to have less access to piped water and, even when connected, appearing to consume less water generally as shown in Table 5 and discussed in more detail in the section ‘Targeting performance of existing subsidies: are subsidies pro-poor?’.

Presented in Figure 5, the results for Panama, Jamaica and El Salvador have a similar pattern, demonstrating subsidies’ regressivity. However, the figures suggest that subsidy schemes in all three Latin American countries reduce inequality since subsidy incidence is less skewed toward richer deciles than total expenditure/income. The degree of subsidy regressivity in Latin American countries is lower than in African countries, as evidenced by the average Palma and 20:20 ratios shown in Table 4. As explored in Section ‘Targeting performance of existing subsidies: are subsidies pro-poor?’, this is partly because both a greater number of poor households reside in service areas, and that, conditional on this potential access, a greater percentage of poor households are connected to the network.

Figure 6 shows the distribution of beneficiaries (households that receive the subsidy) across expenditure or income deciles for the three countries. As with the African countries, it complements the picture painted by the distributional incidence of the amount of subsidies. Higher deciles show the largest share of total subsidy beneficiaries (dark gray bars) and the largest share of all households within each decile that are beneficiaries (light gray bars), although in Panama the pattern is less pronounced.

Figure 7 shows the results for two Asian countries, Bangladesh and Vietnam. In Bangladesh, water consumption subsidies are strongly regressive and increase inequality. While subsidies in Vietnam are still regressive, albeit less so, they are actually inequality reducing, suggesting slightly better targeting compared to Bangladesh. Figure 8, which depicts the distribution of beneficiary households across deciles, shows strong regressivity in Bangladesh but a more even distribution across expenditure deciles in Vietnam, once again suggesting better targeting.

In the next section, we look at a more synthetic measure of targeting performance that complements this analysis and consider both access and subsidy design factors driving this performance.

Figure 9 and Table 5 show the targeting performance of water consumption subsidies in the 10 countries under analysis. Figure 9 shows how access factors (SA*C) and subsidy design factors (B*R) drive omega (Ω) across countries. Table 5 presents the values of each factor for each country. Factor values below 1 indicate that the factor contributes to reduced targeting performance, whereas factor values above 1 indicate that the factor contributes to increased performance.

As shown in Figure 9, all African, Latin American and Asian countries present Ω values below 1, demonstrating that water consumption subsidies are poorly targeted to the most vulnerable populations. Access factors (dark gray dots) are the driving force explaining low Ω values in most countries. Subsidy design factors (light gray dots) are, in general, around or above 1. These indicators suggest that government improvement strategies should mainly focus on increasing access to services (for each country, we explore whether strategies should primarily focus on expanding service areas, increasing the proportion of connected households within existing service areas, or both in detail for each country), although enhancing subsidy design could also be beneficial.

Different factors influence the severity of poor subsidy targeting performance across countries. In Table 5, we explore the different access and subsidy design factors in more detail.

Four African countries – Ethiopia, Mali, Niger and Uganda – show an estimated value of Ω lower than 0.1. In all four, the main factors that reduce subsidy efficiency in targeting poor households relative to all households include (i) poor households’ lower probability of being located in a service area and (ii) their lower probability of being connected to a water network conditional on being located in a service area. Additionally, in Ethiopia and Niger, the relatively low water quantities consumed by the poor significantly decrease the targeting performance.

Nigeria shows an Ω above 0.5, revealing a significant problem, although less acute than in Ethiopia, Mali, Niger and Uganda. SA, C and Q are closer to 1, and B = 1 and R = 1.03. Consistent with the evidence in the previous section, Nigeria's distributional incidence of subsidies by decile is regressive, although less pronounced than in the other four African countries.

In order to improve the targeting performance of water consumption subsidies, these governments should expand the service area (i.e., SA) of their utilities and connect more households within this service area (i.e., C) to the water network. One way to do this is by implementing connection subsidies in areas where there is already network infrastructure present (Andres et al. 2019 discuss different types of connection subsidies and beneficiaries and the relevant considerations when thinking of how to design them depending on the context). Supplementary Appendix Table A2 shows the number of households that are not connected to the network that uses a public tap, which could be connected to the existing infrastructure. This varies significantly across countries and across households classified as rural, urban and poor. Such considerations are important when planning connection subsidies for a specific context. Additionally, these countries should improve the design of their water consumption subsidies to better target poor households (increasing B to values over 1).

In El Salvador, Jamaica and Panama, the problem is less acute, with an Ω of 0.84, 0.69 and 0.86, respectively. In these countries, the factors lowering Ω to below 1 are different. In all three, the probability of living in a service area is high for both types of households, poor and non-poor (SA is close to 1). However, connection rates conditional on location in a service area and water quantities consumed are both significantly lower for poor households, driving Ω to below 1. These countries would benefit from increasing service connection rates, as well as improving subsidy design to better target the poor (resulting in an increase of variables B and R further above 1).

Bangladesh falls in between African and Latin American countries, with an Ω of 0.35. Both the variables SA and C are low in general, and lower for poor households. Water quantities consumed by the poor are similar to those of other households, but this is potentially driven by the lack of information on households’ expenditure and the flat rates being charged, which complicates imputing Bangladesh's household-specific water consumption (therefore, a median consumption quantity is assumed for most households). A parameter Q close to 1 is potentially reasonable given that any connected household regardless of consumption pays a flat rate. Nonetheless, potentially all factors (SA, C, B and R), in addition to the tariff structure, could be improved. In Vietnam, the value for Ω is much higher, though still below 1, and a focus on subsidy design (B and R) could increase targeting performance.

Table 6 shows the errors of inclusion (the proportion of subsidy beneficiary households that are in the top six deciles, or rich) and the error of exclusion (the proportion of poor households that do not receive a water consumption subsidy). These errors are not conditional on location in a service area or connection, so they differ from those in Table 5. The estimates show that for most countries covered here the errors of exclusion range from 78% to nearly 100%, except for Jamaica, Panama and Vietnam (column 2). This is consistent with the very low levels of Ω presented in Table 5. The countries on the lower end of this spectrum exhibit errors of exclusion around 50%, suggesting that their level of outreach to the poor could still be improved.

Inclusion errors (column 1 of Table 6) are extremely high in all countries, suggesting a wide margin for the improvement of targeting, consistent with the analysis presented so far. The inclusion errors are particularly high in African countries, with figures close to 100% in all countries besides Nigeria. Bangladesh also stands out with figures close to 90%.

This paper provides new evidence on the recent performance of piped water consumption subsidies in terms of pro-poor targeting for 10 low- and middle-income countries around the world.

We use detailed household survey data with information on whether the households’ main source of drinking water is from piped water to define access, and the households’ monthly expenditure on water services combined with available information on tariff structures in each country to impute water quantities and unit prices paid by each household. We supplement this information with new estimates of the cost incurred by the supplier of providing piped water to households to calculate the average subsidy per unit of water consumed at the household level.

Our accounting methodology presents some limitations, most importantly due to the lack of information on water quantity consumed at the household level and the tariff structure faced by each household in each country. This means that water quantities have to be imputed. Furthermore, the tariff structures for some households have to be assumed based upon suppliers operating in other regions within the same country. Finally, we are comparing countries with varying levels of urbanization and rural contexts. As a result, piped water is more feasible to close the access gap in some countries than in others.

With these caveats in mind, our analysis of 10 low- and middle-income countries suggests that piped water is substantially subsidized, as existing tariffs do not seem to fully recover the costs of service provision. Moreover, these consumption subsidies do not appear to efficiently address service gaps among the poor due to ineffective targeting.

Overall, we find that water consumption subsidies in all 10 countries are regressive and, therefore, do not adequately target the poor (the estimated Ω has a value under 1 in all cases, with an average value of 0.45 across all 10 countries). This is despite their high diversity of economic development, national income levels, proportion of rural populations, piped water coverage and the costs of producing and distributing piped water. We also determine that the severity of regressivity varies significantly (from 0.006 in Niger to 0.87 in Panama). On average, across the 10 low and middle-income countries examined, 56% of subsidies end up in the pockets of the richest 20%, but only 6% of subsidies find their way to the poorest 20%.

Only a very small number of poor households benefit from subsidies, particularly within the African countries in our sample. This is because poor households are less likely to be located in areas serviced by utility providers (the variable SA in the analysis), and even where they are, they are less likely to be connected and consuming piped water (variable C) than the general population.

This suggests that access factors primarily drive this poor targeting performance (SA takes a value of 0.7 and C takes a value of 0.6, on average, across all 10 countries). This problem is particularly pressing in the African countries (with an average SA of 0.47 and average C of 0.39 across the sample used) and Bangladesh (which has an SA of 0.75 but a C of 0.46). This is expected since these countries have large rural populations, aggravating the feasibility of extending the network to new service areas. However, connection rates are far from 100% even in urban areas in these countries (Table 2), implying that there is also scope to improve connection rates. Furthermore, in Ethiopia, Mali and Niger, available data show that there is a significant proportion of households that are not connected to the network and that use a public tap as the main source of drinking water, both in urban and rural areas (Supplementary Appendix Table A2). This could indicate the potential feasibility of increasing connection rates for these households.

Aside from improving access for the poor, subsidy design could also be improved to better target the poor. The share of poor households that receive a subsidy conditional on being connected to the network relative to the equivalent share for the whole population (B) averages 0.99 and the rate of subsidization conditional on receiving a subsidy for the poor relative to the equivalent rate for the whole population (R) averages at 1.02, across all 10 countries. This signals a need for improvement in subsidy design even given subpar network coverage and connection rates. This trend is observed in all countries.

It is worth noting that there is not an optimal value of omega; it is a policy choice that will vary across countries according to their particular fairness preferences and political equilibriums. If the objective is to use available and limited public funds to ensure poor households can access and consume piped water in an affordable and sustainable way, then an omega below 1 demonstrates that this objective is not being achieved.

To summarize, our results show that in most developing countries, improving the targeting of current water consumption subsidy schemes to the poorest households will primarily require improving their access to the service. This is because, by definition, unconnected households are excluded from the pool of subsidy beneficiaries. Therefore, network expansion into poorer neighborhoods, if technologically feasible and policies that facilitate household connections should be pursued. Angel-Urdinola & Wodon (2012) provide evidence from Nicaragua that increased access rates among the poor over time improves the targeting performance of water consumption subsidies.

It is also important to note that the targeting performance of connection subsidies would also depend on the context, design of the subsidy and the behavior of both utilities and households. For example, universal connection subsidies were simulated by Komives et al. (2005), assuming that all unconnected households were offered and accepted a subsidized connection. In countries where the proportion of unconnected households is higher among the poor than among the overall population, subsidies are likely to be pro-poor. Yet, in practice, poor households may be located in areas more difficult to reach or face additional financial or technical barriers to connection, such as the inability to afford the fixtures required to connect to the network, preventing them from benefiting from connection subsidies (Komives et al. 2006).

In countries where water access is no longer a significant issue and connection rates are relatively high, such as El Salvador, Jamaica and Panama, simply modifying a subsidy's design could greatly improve its targeting of the poor. In some countries, quantity-based subsidies are used in combination with subsidies targeted using administrative selection. But in other countries, only administrative selection, like geographic targeting or mean testing, is used. In general, administrative selection performs better than quantity-based subsidies in targeting the poor (Komives et al. 2006). Having said this, the error of exclusion can be quite high since in an effort to target the poor accurately, a significant proportion of poor households are excluded.

Further detailed analysis of each context would be needed to design the most cost-effective set of policy instruments for improving the pro-poor performance of consumption subsidies, including those that improve access to safely managed water among the poor. The most desirable approach will vary across, and even within, countries. It will depend on state capacity to implement more refined and explicit mechanisms to target the poor, such as administrative selection, as well as whether the poor live in areas with an existing water network, whether they are able to undertake the necessary upgrades within the home for connection, and whether network expansion, if required, is feasible.

The Supplementary Material for this paper is available online at https://dx.doi.org/10.2166/washdev.2020.073.