Water and sanitation service delivery, pricing, and the poor: An empirical estimate of subsidy incidence in Nairobi, Kenya

Abstract

The increasing block tariff (IBT) is among the most widely used tariffs by water utilities, particularly in developing countries. This is due in part to the perception that the IBT can effectively target subsidies to low-income households. Combining data on households' socioeconomic status and metered water use, this paper examines the distributional incidence of subsidies delivered through the IBT in Nairobi, Kenya. Contrary to conventional wisdom, we find that high-income residential and nonresidential customers receive a disproportionate share of subsidies and that subsidy targeting is poor even among households with a private metered connection. We also find that stated expenditure on water, a commonly used means of estimating water use, is a poor proxy for metered use and that previous studies on subsidy incidence underestimate the magnitude of the subsidy delivered through water tariffs. These findings have implications for both the design and evaluation of water tariffs in developing countries.

Key Points

• The increasing block tariff does not target subsidies to low-income households effectively
• Stated expenditure on water is a poor proxy for metered water use
• Previous studies underestimate the subsidy delivered via water tariffs

1 Introduction

The increasing block tariff (IBT) is among the most widely used tariffs by water utilities, particularly in developing countries. According to a recent survey of water utilities across the globe, 53% of utilities in the sample implement an IBT, with 74% of utilities in developing countries doing so [GWI, 2013]. In a traditional IBT, the marginal price for water use increases from one usage block to the next and customers are charged the marginal price for water use in each block accordingly. The popularity of the IBT reflects two widely held perceptions about its potential merits. First, policy makers believe a low marginal price in the lowest usage block of an IBT, often referred to as a “lifeline block,” will ensure that low-income households have access to a certain quantity of water at a price deemed affordable. Second, they believe that higher prices in the upper block(s) of the IBT can both prevent wasteful or extravagant water use and provide an opportunity to improve cost recovery from households who use more water. The intuitive appeal of the IBT rests on the implicit assumptions that all households have a private piped connection to the water network and that low-income households use less water than high-income households.

Scholars have long questioned whether these assumptions are valid in low-income and middle-income countries [Whittington, 1992; Boland and Whittington, 2000; Komives et al., 2005]. This has led to a body of empirical work that has challenged common intuition about the poor access to water and sanitation services and the relationship between household income and water use [e.g., Komives et al., 2006, 2007; Banerjee et al., 2010; Banerjee and Morella, 2011; Barde and Lehmann, 2014]. In this paper, we examine the distributional incidence of subsidies delivered through the increasing block water tariff in Nairobi, Kenya. We combine socioeconomic data from a household survey with household data on metered water use to estimate the distribution of subsidies among residential customers with a private metered connection in Nairobi. We then use a complete set of customer billing records from Nairobi City Water and Sewer Company (NCWSC) to estimate the distribution of subsidies among all residential customers, including those with shared connections. Finally, we expand the scope of our analysis and examine the distribution of subsidies among residential and nonresidential customers in Nairobi.

Our analysis departs from existing studies in the subsidy incidence literature in three ways. First, studies in the literature typically use stated expenditure on water from household interviews to estimate water use. To our knowledge, this study is the first to combine household-level socioeconomic data with data on metered water use to estimate subsidy incidence in the water sector. Second, unlike the majority of studies in the literature, we use empirical, city-specific estimates of the cost of providing water and wastewater services to estimate subsidy incidence. Finally, all previous studies in the literature focus on the distribution of subsidies among residential customers. Our study extends the literature by examining the distribution of subsidies among all customer classes.

We find that the IBT implemented in Nairobi is not targeting subsidies to low-income households effectively. Among households with a private metered connection, households in the lowest wealth quintile receive less than 20% of the subsidies delivered to these customers. Subsidy targeting improves slightly when we examine subsidy incidence among all residential customers, but higher-income customers still receive a disproportionate share of subsidies. Our analysis of subsidy incidence among all customer classes indicates that nonresidential (e.g., commercial, industrial, and bulk water) customers, who constitute 5% of customer accounts, receive over a third of the subsidies delivered through the tariff. We also find that stated expenditure is a poor proxy for metered water and that the magnitude of the subsidy delivered through the water tariff is substantially larger than previous studies would suggest.

The remainder of the paper is organized as follows. Section 2 of the paper discusses the issue of subsidy incidence and provides a review of the subsidy incidence literature in the water sector. Sections 3 and 4 describe our empirical strategy and the data used in our analysis, respectively. Section 5 presents our results. Section 6 provides a discussion of our results and some concluding remarks.

2 Background and Literature Review

Despite the intuitive appeal of IBTs, there are a number of reasons why the IBT may not effectively target subsidies to low-income households in many low-income and middle-income country contexts. For example, in order for a household to receive a subsidy that is delivered through the water tariff, it must have a piped connection. However, poor households often lack a piped water connection and are thus largely excluded from subsidies provided through low-priced water delivered through a piped connection. Similarly, low-income households are also often more likely than wealthier households to have a shared connection to the piped water network (e.g., a yard tap) and to live in multiunit dwellings that are served by a single meter. Households that share a connection or live in a multiunit dwelling served by a single meter pay a higher volumetric price for water than if they had an individual meter because the collective water use of those who share a connection falls in the upper, more expensive, blocks of the IBT. Finally, the extent to which household income and water use are highly correlated is an empirical question, even among households with a private piped connection. Indeed, the limited empirical evidence in the literature suggests that the correlation between household income and water use is much less than commonly assumed [Whittington et al., 2015].

Concerns about the extent to which the IBT, and utility tariffs more broadly, can be used to effectively target subsidies to low-income households has led to a body of empirical research on subsidy incidence. (See Appendix A for a summary of studies that have been published on subsidy incidence since 2000.) To calculate the distributional incidence of subsidies delivered through the water tariff, the analyst needs information on the magnitude of the subsidy received by each household and the relative income or wealth of each household. The subsidy received by each household is the difference between what it costs to provide the particular household with a particular level of service (e.g., water or water and wastewater service) and what the household actually pays for this service.

The cost of serving each household is a function of households' water use, whether the household has only water or water and wastewater service (i.e., their “level of service”), and the unit cost of providing water and wastewater services. The amount households pay for water and sanitation service is a function of households' water use, their level of service, and the tariff the utility uses to calculate their monthly bill for water and sanitation services. Thus, in total the analyst must have five pieces of information to estimate subsidy incidence: households' water use, households' level of service, the unit cost of providing water and wastewater services, the tariff, and some measure of households' wealth or socioeconomic status. Assembling this information can be quite difficult in practice. (See Gómez-Lobo et al. [2000] for an overview of information and modeling challenges associated with designing water and sanitation tariffs.)

For example, data on households' socioeconomic status and demographics are typically available in secondary household survey data, such as national income and expenditure surveys, World Bank Living Standards Measurement Study (LSMS) data, and some national censuses. However, these surveys typically do not contain information on household water use. Similarly, utility billing records contain information on household water use, provided customers are metered, the meters are working, and the utility regularly reads customers' meters. Due to confidentiality requirements, however, it is typically not possible to match household-level socioeconomic data in nationally representative household income and expenditure surveys and customer data in utility billing records.

Because it can be difficult or not possible to obtain good measures of both socioeconomic status and water use for the same household, studies in the literature typically use a single data source to obtain information on both households' socioeconomic status and water use. In particular, most studies use households' stated expenditure on water to estimate households' water use. They collect this information either through primary household surveys [e.g., Foster, 2004; Bardasi and Wodon, 2008; Angel-Urdinola and Wodon, 2012] or from nationally representative household budget and expenditure surveys (e.g., World Bank LSMS data). (See Appendix B for a discussion of why stated expenditure may not be a good proxy for metered water use.)

Studies in the subsidy incidence literature (Appendix A) address the issue of cost in three general ways. (Appendix C provides a summary of cost estimates used in the literature.) First, studies may use generic cost estimates, or international benchmarks, to calculate subsidy incidence [e.g., Komives et al., 2005, 2006; Foster and Yepes, 2006]. Common sources for generic cost estimates include GWI [2004] and W. Kingdom et al. (Full cost recovery in the urban water supply sector: Implications for affordability and subsidy design, unpublished working paper, 2004). Other studies use empirical, site-specific cost estimates [e.g., Groom et al., 2008; Banerjee and Morella, 2011; Walker et al., 2000]. However, these studies typically do not explicitly state what the cost estimates include or precisely how they were derived. Finally, studies may make ad hoc assumptions about the cost of providing water and wastewater services. For example, Barde and Lehmann [2014] assume that the average tariff currently implemented in Lima, Peru (approximately 0.64 USD/m³) represents full cost recovery.

There is broad consensus in the literature that de facto subsidies delivered through the water tariff are poorly targeted and largely regressive (see Appendix A). Indeed, many studies find that subsidies delivered through the water tariff perform worse than if the subsidies were equally distributed among the population. This is principally due to the fact that low-income households are less likely to have a private connection to the piped water network and, thus, do not receive subsidies delivered through the water tariff.

Studies that examine subsidy incidence only among households with a piped connection also find that subsidies are poorly targeted. This is primarily because income and water use are often not highly correlated and the tariff implemented by many utilities is not sufficient to cover the cost of providing service. These empirical results are supported by simulations conducted by Whittington et al. [2015] that suggest little can be done to improve subsidy targeting when tariffs are not sufficient to cover costs.

There are three main gaps in the water literature on subsidy incidence. First, studies in the literature either focus only on subsidies associated with the delivery of piped water service or do not explicitly state whether they include subsidies associated with wastewater service. Piped wastewater services are usually more expensive to provide than piped water services. To the extent that wastewater services are sold below cost and to the extent that higher-income households are more likely to have connections to the piped wastewater network, estimates in the literature may overestimate the performance of subsidies delivered through the tariff.

Second, nearly all of the studies in the literature use stated expenditure to estimate water use, which may be a poor proxy for metered water use. Thus, it is unclear whether the broad consensus in the literature is attributable to the fact that studies use the same, potentially flawed, measure of water use.

Finally, all of the studies in Appendix A focus on subsidy incidence only among residential customers. This is not surprising given that these studies use data from household surveys. As a result, however, the literature ignores the distributional issues between residential and nonresidential (e.g., commercial, industrial, and bulk.) customers. Depending on the tariff applied to nonresidential customers, failing to include nonresidential customers may overstate or understate the magnitude of total subsidies delivered through the water tariff.

3 Empirical Strategy

This study was designed to fill these gaps in the subsidy incidence literature. Our empirical strategy proceeds in three analytical steps. In the first step of our analysis, we combine socioeconomic and demographic data from a survey of 656 households with data on metered water use from NCWSC billing records. We use these data to: (1) estimate the distribution of subsidies among households with a private metered connection, and (2) examine the extent to which stated expenditure is an accurate proxy of metered water use. We focus this first step of the analysis on households with a private metered connection to capture the relationship between household income and water use. Households who shared a connection with another household or family were excluded from our survey sample.

According to the most recent census, less than a quarter of households in Nairobi reported using a private connection to the piped water network as their primary drinking source [Kenya National Bureau of Statistics, 2009]. Approximately half of households used piped water that is not delivered into their dwelling (e.g., a shared tap) as their primary drinking water source. Thus, in the second step of our analysis, we examine the distribution of subsidies among all NCWSC's residential customers, which includes residential customers with shared connections. In the third, final, step, we expand the scope of our analysis to examine the distribution of subsidies among residential and nonresidential customers in Nairobi.

3.1 Subsidy Incidence

We obtained information on customer water use from 21 months of NCWSC's billing records. Like many utilities, NCWSC does not read each meter every month. (NCWSC reads approximately 75% of meters each month.) To address this, we calculate monthly water use directly from actual meter readings in the NCWSC billing data as described in equation 1.

(1)

where

WUSE_i,t

water use for household i in month t;

READING_i,t

meter reading for household i in month t;

READING_i,t-1

previous actual meter reading for household i;

RDATE_i,t

date on which NCWSC read the meter for household i in month t;

RDATE_i,t-1

date of the previous actual meter reading for household i.

We then use the estimates of households' monthly water use obtained in equation 1 to calculate households' average monthly water use over the period covered by the billing records.

We define the subsidy received by each customer as the difference between the cost to serve each household and what a household pays for service. Our analysis of NCWSC's billing records confirms that the utility implements the official tariff (Table 1) to calculate customers' water and sewer bills. Thus, we calculate how much a customer pays by applying NCWSC's official tariff to our estimates of their average monthly water use. NCWSC implements an IBT with 4 usage blocks. In addition to the fixed charge for meter rent, NCWSC applies a minimum charge for 10 m³/month—i.e., households that use less than 10 m³/month are charged for 10 m³/month. A minimum charge irrespective of the volume a household uses is one possible type of positive fixed charge; it is not a necessary or standard feature of an IBT. NCWSC's use of a minimum charge is not unique. According to GWI [2013], approximately 10% of utilities in their tariff database implement a minimum charge. NCWSC charges customers with a connection to the sewer network an additional 75% of the volumetric portion of their water bill for wastewater service.

Table 1. Summary of the Tariff Implemented by Nairobi City Water and Sewerage Company

Tariff Componenta^a Conversion rate = 90 KSH/USD.
Residential, Commercial, and Industrial
0–10b^b Customers charged for a minimum of 10 m^3/month. m^3/month	0.22 USD/m^3
11–30 m^3/month	0.45 USD/m^3
31–60 m^3/month	0.50 USD/m^3
>60 m^3/month	0.63 USD/m^3
Water Kiosk
All units	0.18 USD/m^3
Bulk Supply
All units	0.31 USD/m^3
Other Charges
Seweragec^c Applied to the volumetric component of the water bill.	75%
Meter Rent	0.59 USD/month
Connection Charges	29 USD

• ^a Conversion rate = 90 KSH/USD.
• ^b Customers charged for a minimum of 10 m³/month.
• ^c Applied to the volumetric component of the water bill.

We define the cost of serving a particular customer as in equation 2.

(2)

where

COST_i

average monthly cost of serving household i (USD/month);

WUSE_i

average water use of household i from equation (1) (m³/month);

WCOST

average volumetric cost of providing water service (USD/m³);

WWCOST

average volumetric cost of providing wastewater service (USD/m³);

I_ww,i

is an indicator variable that takes the value 1 if a household has wastewater service and 0 otherwise.

We develop empirical estimates of the average cost of providing water and wastewater services. Our cost estimates include both operations and maintenance as well as capital costs. They do not include the opportunity cost of the raw water supply.

We estimate subsidy shares, the share of subsidies received by different groups of customers, to assess subsidy incidence. (We do not examine affordability because this requires making ad hoc assumptions about what is and is not “affordable.”) Equation 3 defines the share of subsidies going to a particular group of customers.

(3)

where

S_j

share of subsidies received by customer group j (j = 1…J);

SUB_i

share of subsidies received by household i.

In the first step of our analysis, j indexes the five wealth quintiles of our survey sample. In the second step, j indexes accounts located in low-income areas and accounts in nonlow-income areas. In the final step of our analysis, j indexes residential, nonresidential, kiosk, and bulk customer classes.

3.2 Stated Expenditure as a Proxy for Water Use

We also examine whether stated expenditure is an accurate proxy for metered water use by estimating household water use from households' stated expenditure on water and comparing this to their metered water use. To do this, we ask households if they can recall the amount of their last bill from NCWSC and the number of months of service the bill covered. Equation 4 shows how we impute water use for customers with only piped water service. We use an analogous approach to impute water use for customers with both water and sewer service.

(4)

where

IMPUSE_i

imputed water use for household i (m³/month);

EXPS_i

stated expenditure for household i (Ksh/month);

RENT

monthly meter rent charged in the NCWSC tariff (Table 1);

p_X

volumetric price for water in the Xth block in the NCWSC tariff (Table 1);

b_X

volumetric upper bound for the Xth block in the NCWSC tariff (Table 1);

b_Xmax_w

amount a water customer would be charged for consuming the maximum amount in the Xth block of the NCWSC tariff.

4 Data

The first step of our analysis examines subsidy incidence among households with a private connection to the piped water network. For this analysis, we use data from a sample of 656 households that were randomly drawn from two of Nairobi's six service regions, which were purposefully selected to ensure income heterogeneity in our sample. (See Appendix D for a detailed description of the survey and our sampling strategy.) The survey was conducted between November 2013 and January 2014 and collected a range of socioeconomic and demographic information from households, including data on monthly income, household expenditure, and asset ownership. Following Filmer and Pritchett [2001] and Filmer and Scott [2008], we use principal component analysis to construct an asset index to serve as a proxy for wealth (see Appendix E). We use the asset index as our primary proxy for wealth because approximately 15% of respondents in our sample refused to provide information about their monthly household income. (Assets included in the index include: liquid propane gas (LPG) as a main cooking fuel, biomass or kerosene as a main cooking fuel, separate kitchen, security guard, connection to the electricity grid, mobile phone, internet connection, TV, radio, computer, private car, washing machine, refrigerator, bore well, and additional land in/out of Nairobi.)

We obtain information on customer water use from 21 months of NCWSC's billing records. The billing data cover the period from August 2012 to May 2014. The principal challenge in our empirical strategy was to identify the households in our survey sample in NCWSC's billing records. Like many cities in developing countries, Nairobi does not have a formal system of addresses. Thus, it was not possible to first construct our sample from the billing records and then locate households to conduct the household survey. To address this, we used households' account numbers to identify households in the billing records. Because households do not typically know their NCWSC account number, however, we obtained households' account numbers by matching the serial number on households' water meters with the account numbers on the NCWSC marketing assistants' itineraries. When possible we verified the account number with a physical copy of a household's recent water bill. (Fifty-six percent of households in our sample were able to show enumerators a copy of their water bill. All account numbers matched the accounts associated with the meter serial numbers.)

We use data from 5 years of audited financial statements (FY 2007 to 2012) to estimate NCWSC's average operations and maintenance costs. We derive capital cost estimates from data in NCWSC's water master plan [Ministry of Water and Irrigation and Athi Water Services Board, 2012] and interviews with senior water and sanitation engineers at NCWSC, Athi Water Services Board, and local engineering firms. Table 2 presents the cost estimates we use in our analysis. Assuming 35% nonrevenue water, we estimate the full cost (O&M plus capital costs) of water service to be 1.40 and 1.46 USD/m³ for wastewater service. (Nonrevenue water refers to water the utility produces but for which it does not receive revenue.) These estimates are higher than the cost estimates used in many studies, but of similar magnitude to the cost estimates in GWI [2004] and Kingdom et al. (unpublished working paper, 2004) once nonrevenue is accounted for (see Tables C1 and C2 in Appendix C).

Table 2. Summary of Cost Estimates Used to Calculate Subsidy Incidence

Cost Component	USD/m^3a
Water serviceb^b Cost estimates assume 35% nonrevenue water.	1.40
O&M	0.30
Capital costsc^c Ten percent real discount rate; 30 year average useful life of capital.	1.10
Wastewater serviceb^b Cost estimates assume 35% nonrevenue water.	1.46
Operations and maintenance	0.30
Capital costsc^c Ten percent real discount rate; 30 year average useful life of capital.	1.16

• ^a Conversion rate = 90 KSH/USD.
• ^b Cost estimates assume 35% nonrevenue water.
• ^c Ten percent real discount rate; 30 year average useful life of capital.

For the analysis of subsidy incidence among all residential customers and among all customer classes, we obtain information on the water use of NCWSC's approximately 180,000 residential customers from 21 months of NCWSC's billing records. NCWSC does not have socioeconomic or demographic information about its customers. In the absence of household-level data on income or socioeconomic status, one could potentially use household budget and expenditure survey data or recent census data to obtain aggregate information on household characteristics. This is not possible in Nairobi for two reasons. First, the most recent Kenya Integrated Household Budget and Expenditure Survey (2005–2006) contains only 685 observations from Nairobi. Second, data from the most recent census are not publicly available.

To address this, we use the geographic location of customer accounts as a proxy for relative wealth. In particular, we use the GIS location of customer accounts to identify which accounts are located in low-income areas. Information on the GIS location of each account was collected by NCWSC as part of a pilot program to conduct meter reading with smartphones. As of May 2015, approximately 85% of the 180,000 residential customer accounts were geocoded.

We obtain information on the location and extent of low-income areas in Nairobi from the MajiData project of Kenya's Ministry of Water and Irrigation (MWI) and Water Services Trust Fund (WSTF). The MajiData project is a multiyear effort to create a database of information related to water and sanitation service provision in all of Kenya's urban low-income areas. As part of this process, the MajiData team identified and mapped low-income areas in the service area of each Water Service Provider in Kenya using publicly available data, stakeholder consultations, and transect walks of each service area. MajiData's classification of low-income areas includes informal settlements, planned areas with planned low-income housing, and informal housing in planned residential areas. To our knowledge, this information is the most comprehensive, up to date mapping of low-income areas in Nairobi.

For our analysis of subsidy incidence among all customer classes, we obtain water use data from the same set of NCWSC billing records described above. NCWSC billing data include 13 different customer classes. We group these customer classes into four general types: residential, nonresidential, bulk, and kiosk. Our nonresidential customer type includes accounts classified as government, community, and industrial.

5 Results

5.1 Household Survey—Subsidy Incidence

Table 3 presents information on selected characteristics of households in our survey sample. The average household in our sample has four members, which is consistent with the average household size in Nairobi from the latest census. Approximately half of the households in our sample rent their home. Over 90% of households in our sample have a sewer connection. Seventy-eight percent of households in our survey report using their piped water connection as their primary drinking water source. The remaining 22% report using bottled water for their primary drinking water source. Over a quarter of households in our sample report purchasing water from a vendor in the previous year, which reflects the fact that NCWSC does not provide customers 24 × 7 water service.

Table 3. Basic Characteristics of Households Surveyed

Household Characteristic	Value
Household size (s.d.)a^a Standard deviation.	4 (1.78)
Home owner	51%
Primary drinking water source
Piped water connection	78%
Bottled water	22%
Water vendor (previous year)	26%
Household water treatment	68%
Sewer connection	93%

• ^a Standard deviation.

Mean and median water use in our survey sample are 19 and 13 m³/month, respectively (Figure 1). Average water use among all residential customers in the NCWSC billing data is 31 m³/month. However, the mean water use of households on meter-reader itineraries with 100% individual meters is 20 m³/month, similar to what we find in our sample of households. Nearly 40% of households in the sample fall in the lifeline block (0–10 m³/month). Over 80% of the households in the sample have water use that falls in the first two usage blocks (below 30 m³/month). Only 4% of the sample falls in the upper-most block of NCWSC's tariff (>60 m³/month).

Figure 1

Distribution of water use among survey sample with NCWSC tariff blocks.

We find considerable heterogeneity in water use, both within and across wealth quintiles. (This heterogeneity persists if we examine only three wealth groups.) Figure 2 plots household water use versus their wealth index score. The correlation between a household's wealth index score and water use in our sample is 0.15. Mean water use is 16 m³/hh/month for households in the first (lowest) wealth quintile and 30 m³/hh/month for households in the fifth (highest) wealth quintile (Table 4).

Figure 2

Scatterplot of monthly household water use versus wealth. Nine observations with water use above 100 m³/month not shown on the graph for scale purposes.

Table 4. Water Use, Representative Bill, and Average Price Among Wealth Quintiles

	Unit	Wealth Quintile					Overall
		1	2	3	4	5
Mean water use (s.d.)a^a Standard deviation.	m^3/hh/month	16 (30)	14 (15)	14 (17)	24 (25)	30 (32)	19 (26)
Representative water bill	USD/hh/month	10.35	8.39	8.19	14.18	16.76	11.58
Average price	USD/m^3	0.79	0.90	0.83	0.62	0.56	0.74

• ^a Standard deviation.

Table 4 also shows the average monthly bill for households in the five wealth quintiles. The mean bill for households in the lowest quintile is 931 Ksh/hh/month (approximately 10 USD/hh/month). The mean bill for households in the highest wealth quintile is 1509 Ksh/hh/month (approximately 17 USD/hh/month). As a point of comparison, the mean water and sewer bill for households in the lowest quintile is only 60% of what these households report paying for electricity. For the wealthiest households in the sample, the mean bill is less than a quarter of what they report spending on electricity.

Table 4 presents the average price paid by households in each wealth quintile. For the full sample, the mean average price ranges from 79 Ksh/m³ (0.90 USD/m³) to 50 Ksh/m³ (0.56 USD/m³) across wealth quintiles. The mean average price for households in the lowest wealth quintile is 70 Ksh/m³ (0.79 USD/m³) and 50 Ksh/m³ (0.56 USD/m³) for households in the highest wealth quintile. Households in the lowest quintile face a higher average price than households in the highest quintile because the tariff includes both a positive fixed charge and minimum charge (e.g., customers are charged for 10 m³/month regardless of their water use). These average price estimates reflect the fact that over 90% of households in our sample have a sewer connection.

Figure 3 shows the distribution of subsidies across wealth quintiles. If the subsidy were evenly, or randomly, distributed among the population, each wealth quintile would receive 20% of the total subsidy. A well-targeted subsidy would deliver a substantial share of the total subsidies to low-income households. In our sample, households in the lowest quintile receive only 16% of the total subsidy. Households in the top three wealth quintiles receive nearly 70% of the total subsidy, with households in the highest wealth quintile receiving almost 30% of the total subsidy.

Figure 3

Share of subsidies received by each wealth quintile.

5.2 Household Survey—Stated Expenditure as a Proxy for Metered Water Use

During the survey, we asked households if they could recall the amount of their last bill from NCWSC. Nearly 85% of households in our sample indicated that they could, considerably higher than the 30% reported in Foster [2004]. Figure 4 presents a scatterplot of metered versus imputed water use for households who could recall the amount of their previous water bill. The 45° line in Figure 4 traces a line of equality for which imputed and metered water use would be the same for each household. The scatterplot in Figure 4 displays a high degree of dispersion, indicating that stated expenditure does not provide an accurate proxy for metered water use in our sample.

Figure 4

Imputed versus metered water use.

We find that stated expenditure typically overestimates households' water use, often by a substantial amount. This is reflected in Table 5, which provides summary statistics of metered water use and imputed water use from households in our sample. Average metered water use among households who could recall the amount of their last water bill was approximately 19 m³/month The average water use imputed from stated expenditure among the sample, however, was 27 m³/month (42% higher).

Table 5. Summary Statistics for the Distributions of Metered and Imputed Water Use

Water Use	Unit	Mean	Std. Dev.	Min	Max
Metered	m^3/month	19	24	0.7	292
Imputed	m^3/month	27	34	0.3	436

5.3 Subsidy Incidence Among All Residential Customers

The results presented above examine subsidy incidence among our survey sample, all of whom had a private piped connection. In the second step of our analysis, we expand the scope of our inquiry to examine subsidy incidence among all residential customers. This analysis includes households with a private piped connection as well as households served by a shared connection. NCWSC does not have information on whether accounts are served by shared or individual connections. Thus, the results below examine subsidy incidence among all residential customers and do not directly address water use or subsidy incidence among accounts with shared and individual connections.

Approximately 20% of residential customer accounts in NCWSCs billing records are located in low-income areas identified in the MajiData database. Table 6 provides a summary of water use among residential accounts of different income levels. The mean and median water use in accounts located in low-income areas is 30 and 12 m³/month, respectively. This is only slightly lower than the mean (33 m³/month) and median (14 m³/month) water use of accounts that are not located in low-income areas.

Table 6. Summary of Water Use Among NCWSC Residential Customers

Residential Area Classification	Water Use (m^3/acct./month)
	Mean	Median	Std. Dev.
Low income	33	14	220
Middle/high income	30	12	127
All residential	31	12	194

Figure 5 provides a summary of subsidy incidence among all NCWSC's residential customers. Accounts located in low-income areas constitute 19% of residential accounts and receive 21% of the total subsidies delivered to residential customers. This is approximately the same amount of subsidies that low-income customers would receive if the subsidy was evenly distributed among residential customers.

Figure 5

Share of total residential accounts and subsidies received by accounts in low-income and middle/high-income areas.

5.4 Subsidy Incidence among all Customer Classes

We now turn to the results for subsidy incidence among all customer classes. Residential accounts constitute 94% of NCWSC customers (Table 7). Nonresidential accounts represent 5% of NCWSC customers. The remaining 1% of accounts are official public kiosks and bulk customers. Despite the fact that residential accounts make up the vast majority of NCWSC customers, they account for only 57% of the overall water use and 56% of total billings (Table 7). Nonresidential customers, on the other hand, account for 35% of the overall water use and 41% of total billings.

Table 7. Summary of the Share of Accounts, Water Use, Billings, and Subsidies Among Four NCWSC Customer Classes

Customer Class	% Total Accounts	% Total Water Use	% Total Billings	% Total Subsidy
Residential	94%	57%	56%	63%
Nonresidential	5%	35%	41%	31%
Kiosk	<1%	3%	1%	2%
Bulk	<1%	4%	2%	3%
Total	100%	100%	100%	100%

We find that nonresidential customers receive 31% of the total subsidy. By contrast, residential customers receive 63% of the total subsidy delivered through the water tariff. Among residential customers, accounts in high-income itineraries represent 21% of accounts and receive 19% of the total subsidy. Accounts in low-income itineraries represent 14% of total accounts and receive only 9% of the total subsidies, far less than if subsidies were randomly distributed among customers.

6 Discussion and Conclusions

Our analysis of subsidy incidence among a sample of 656 households in Nairobi with a private metered connection indicates that households in the lowest wealth quintile receive only 15% of the total subsidies delivered to households in our sample. In contrast, households in the highest wealth quintile receive nearly 30% of the subsidies. Thus, among our sample of customers with a private metered connection, the current water tariff performs worse than if the subsidy was randomly distributed among households.

In Nairobi, the poor targeting of the subsidies even among households with a private metered connection is driven by a combination of three factors. First, very few customers' water use falls in the uppermost blocks of NCWSC's IBT (Figure 1). Indeed, over 80% of households in our sample fall in the first two blocks of NCWSC's tariff. Thus, irrespective of the prices in each block there is not a sufficient number of customers in the upper blocks to enable a meaningful level of cross subsidy. Second, at current prices nearly all customers are being subsidized. The average price paid for water and sanitation services among the wealth quintiles in our sample ranges from 0.56 USD/m³ to 0.90 USD/m³. In contrast, we estimate the full cost of providing water and sanitation services in Nairobi to be approximately 2.86 USD/m³. When nearly all customers are subsidized, it is not possible for a subsidy delivered through the tariff to effectively target subsidies to intended beneficiaries. Finally, contrary to common intuition, we find a low correlation between our wealth proxy and water use, which is consistent with the limited data that exist in the literature [Whittington et al., 2015].

We also find that stated expenditure is a poor proxy for metered water use. Despite the significant measurement error associated with using stated expenditure as a proxy for water use, we find that using stated expenditure to estimate subsidy incidence does not change the policy implications of our findings. This is true in our sample because the majority of NCWSC customers have arrears or credits on their accounts, and we find a low correlation between income and whether customers have arrears or credits. This may not be true in other places. Thus, our findings suggest that researchers should exercise caution when using stated expenditure to estimate water use.

When we expand our analysis to the distribution of subsidies among all NCWSC's 180,000 residential customers, we find that subsidy incidence improves very slightly. Among all residential customers, customers located in low-income areas account for approximately 19% of total residential accounts and receive 21% of the total subsidies delivered to residential customers. This seemingly counterintuitive result can be explained by the fact that low-income customers are more likely to have shared connections, which register high levels of water use, and all water use is subsidized at current prices. While subsidy targeting among all residential customers is slightly better than subsidy incidence among only households with a private connection, errors of inclusion remain high and customers in low-income areas are no better off than if subsidies were randomly distributed among residential customers.

Finally, our analysis of subsidy incidence among all customer classes indicates that nonresidential customers receive over one third of the total subsidies delivered through NCWSC's tariff. Residential customers receive only 63% of the total subsidies. This is not surprising given that all customers are subsidized at current prices and nonresidential customers account for nearly 40% of total water use. However, policy makers often implement an IBT with a lifeline block specifically to target subsidies to low-income, residential customers. We find that this is not occurring in Nairobi. Our results highlight the importance of examining subsidy incidence among all customer classes, which has largely been ignored in the literature.

In addition to our findings related to subsidy incidence, our analysis raises important issues about the magnitude of the subsidy delivered through the water tariff. Most studies on subsidy incidence focus on subsidies associated with piped water service among only residential customers. They do not examine subsidies associated with sewer service or subsidies delivered to nonresidential customers. Our analysis suggests that limiting the scope of subsidy incidence in this manner would lead to a substantial underestimate of the magnitude of the subsidy delivered through the water tariff.

In Nairobi, examining subsidies associated with piped water service among residential customers would result in a total subsidy that is approximately 40% less than the subsidy associated with both piped water and sanitation services for residential customers. Similarly, we find that examining subsidies associated with both piped water and sewer services among only residential customers would underestimate the total subsidy delivered through the water tariff by 45%. In total, focusing only on subsidies associated with providing water service to residential customers would underestimate the magnitude of the subsidy delivered through the water tariff by 65%. We estimate that the total subsidy delivered through the tariff is approximately one and half times NCWSC's total billings.

Policy makers in the water sector often express concern about the affordability of water and sanitation services, especially for low-income households. Indeed, concern about affordability is often the primary justification for keeping water prices low and for implementing an IBT that includes a lifeline block. However, the poorest residents in many cities often rely on vended water from public kiosks or private vendors and pay more for water on a volumetric basis than households with a private connection to the piped network.

Our findings add to a growing body of empirical literature that suggests that IBTs implemented by many utilities do not effectively target subsidies to low-income households. In Nairobi, we find this is particularly true when examining subsidy incidence among all customer classes, but also when we restrict our analysis to households with private metered connections. This is striking given that the poorest households often lack access to piped water and sanitation services altogether. This growing body of evidence suggests that the IBT is an ineffective and often expensive means of delivering subsidies to low-income households. Thus, if policy makers want to subsidize water and sanitation services for low-income households, they should explore alternative subsidy delivery mechanisms, including both connection subsidies and means-tested subsidies (i.e., subsidies for which households must meet an income or wealth criteria).

Acknowledgments

Primary funding for this work was provided by the Environment for Development Initiative. Additional financial support was provided by the Graduate School at the University of North Carolina at Chapel Hill. We thank Nairobi City Water and Sewer Company for their close collaboration on this project. We also thank Gunnar Köhlin, Sanford Berg, Noreen McDonald, Daniel Rodriguez, Meenu Tewari, participants in the 2014 CAMP Resources workshop, and participants in the 8th Annual Meeting of the Environment for Development Initiative for helpful comments on previous drafts. Data sets upon which the analyses were based are available upon request from David Fuente ([e-mail]).

Appendix A: Summary of Subsidy Incidence Literature

Table Appendix A provides a summary of 21 studies that have been published on subsidy incidence since 2000.

Table A1. Summary of Studies in the Subsidy Incidence Literature (Adapted from Whittington et al. [2015])

Study	Country	Data Sourcea^a Aggregate refers to data averaged over a geographic area (e.g., service region, metropolitan area, and county).	Data Year	Sample Size	Water Use Measurea^a Aggregate refers to data averaged over a geographic area (e.g., service region, metropolitan area, and county).	Indicator(s)b^b EOE = Errors of exclusion. EOI = Errors of inclusion.		Subsidy Targetingc^c “Poor” = worse than if subsidies were equally or randomly distributed; “Moderate” = slightly better than if subsidies were equally or randomly distributed; “Excellent” = large proportion of subsidies targeted to low-income households.
Whittington et al. [2015]	Hypothetical	Hypothetical	n.a.	n.a.	Hypothetical	Subsidy share		Poor
Barde and Lehmann [2014]	Lima, Peru	Billing data, expenditure survey, tariff	2010	2570	Stated expenditure	Affordability; subsidy share; EOI; EOE; leakage rate		Poor (nonmeans tested); Excellent (means tested)
Angel-Urdinola and Wodon [2012]	Nicaragua	HH survey data and tariffs	2001 & 2005	3641 (2001) 6102 (2005)	Stated expenditure	Concentration coefficient		Poor
Banerjee and Morella [2011]	Multicountry—Africa	HH surveys and tariffs	Varies	Varies	Stated expenditure	Affordability (share of HH total expenditure); concentration coefficient;		Poor
Banerjee et al. [2010]	45 utilities in 23 African Countries	LSMS and tariffs	Varies	Varies	Stated expenditure	Affordability (share of HH total expenditure); concentration coefficient;		Poor
García-Valinas et al. [2010]	Spain	Municipal surveys	2005	301 municipalities	Aggregate	Affordability		n.a.
Diakité et al. [2009]	Cote d'Ivore	HH panel data	1998–2002	780 total in panel (aggregate data)	Aggregate	Welfare gain/loss		n.a.
Ruijs [2009]	Sao Paolo, Brazil	HH data	1997–2002	63 MRSP	Aggregate	Welfare gain/loss		n.a.
Ruijs et al [2008]	Sao Paolo, Brazil	Aggregate panel data for demand est.	1997–2002	Panel of 39 MRSPs (aggregate data)	Aggregate	Affordability		n.a.
Bardasi and Wodon [2008]	Niger	HH survey	1998	533	Stated use	Average price		n.a.
Groom et al. [2008]	Beijing, China	HH income and expenditure survey—Panel 1987 2002	1987–2002	645 HH plus aggregate data on quintiles	Stated expenditure	Welfare gain/loss		Poor
Fankhauser and Tepic [2007]	Transition countries	LSMS	Varies	Varies	Stated expenditure	Affordability (% of HH expenditure)		n.a.
Angel-Urdinola and Wodon [2007]	Cape Verde, Sao Tome, Rwanda	Nationally rep HH surveys	Varies 1999–2002	Varies	Stated expenditure	Concentration coefficient		Poor
Foster and Yepes [2006]	Multicountry—Latin America	LSMS	Not stated	Not stated	Stated expenditure	Affordability (% of HH that would spend more than x% if tariffs were raised)		Poor
Komives et al. [2006]	Multicountry	Secondary literature	Varies	Varies	Stated expenditure		EOE; concentration coefficient	Poor
Komives et al. [2005]	Multicountry	LSMS	Varies	Varies	Stated expenditure		Concentration coefficient; EOI, EOE; “Material impact”	Poor
Foster and Araujo [2004]	Guatemala	LSMS style national survey (ENCOVI 2000)	2000	7,276	Stated expenditure		EOE; EOI	Poor
Foster [2004]	Argentina	Primary HH Survey (2500 HH)	2002	2,500	Previous bill; Stated expenditure; Imputed using regression		Cumulative dist; concentration coefficient; EOI, EOE	Moderate
Gómez-Lobo and Contreras [2003]	Chile and Columbia	National HH surveys (Chile—CASEN 1998; Columbia—1997 NQLS)	1997/1998	Chile 48,107; Columbia 4,094	Stated expenditure		Concentration curves; EOI; EOE	n.a.
Foster et al. [2000]	Panama	LSMS	1997	n.a.	Stated expenditure		EOE, EOI	n.a.
Walker et al. [2000]	Central America	Household survey	Varies 1995–1998	Varies	Previous bill		EOI; EOE; Average subsidy per HH per month; subsidy share	Poor-moderate

• ^a Aggregate refers to data averaged over a geographic area (e.g., service region, metropolitan area, and county).
• ^b EOE = Errors of exclusion. EOI = Errors of inclusion.
• ^c “Poor” = worse than if subsidies were equally or randomly distributed; “Moderate” = slightly better than if subsidies were equally or randomly distributed; “Excellent” = large proportion of subsidies targeted to low-income households.

Appendix B: Discussion of Stated Expenditure as a Proxy for Metered Water Use

There are several reasons why imputed water use may not be a good proxy for metered water use. Households may not be able to accurately recall how much they spend on water and sanitation services. Households incur a variety of expenses each month and throughout the year and survey evidence suggests that water constitutes a very small portion of monthly household expenditure (often less than 3%) for households with piped connections [Komives et al., 2005, Appendix C.4]. Thus, it is possible that households may have difficulty recalling expenditure on water and sanitation services because they do not represent a major share of their total expenditures. Indeed, in a 2500 household survey conducted in Argentina, Foster [2004] reports that only 30% of the households were able to recall the amount of their most recent bill.

Even if households can perfectly recall their monthly expenditure on water and sanitation services, there are additional reasons why expenditure on these services might be a poor proxy for metered water use. For example, income and expenditure surveys often do not contain information on whether a household connection is metered. If households do have metered connections, the meters may not be working or the utility may not read them on a regular basis. Households may also have a shared connection. In these instances, households' water bills will not reflect their metered water use.

Additionally, household budget and expenditure surveys ask households how much they spent on water last month. They typically do not ask households specifically how much they spent on piped water services, nor do they ask households how much they spent on sanitation services. For example, the most recent Kenya Integrated Household Budget and Expenditure survey asks households “What was the total cost of water for your household last month?” [Kenya National Bureau of Statistics, 2006]. Thus, household recall of expenditure on water in these surveys may include the amount they spent on water from vendors and sewer services.

Water bills may also include fees that are unrelated to water consumption in the most recent billing period. This could include fees for other services (e.g., solid waste collection), pro-rated connection charges, arrears, or penalties for nonpayment. Additionally, countries in Latin America and elsewhere are experimenting with including payment for environmental services in water bills to promote watershed protection [see Whittington and Pagiola, 2012].

Appendix C: Cost Estimates Used in the Literature

This appendix provides a summary of cost estimates used in the subsidy incidence literature. Table C1 provides an overview of cost estimates used in nine studies. Table C2 summarizes the cost estimates from GWI [2004] used commonly in the literature.

Table C1. Summary of Cost Estimates Used in the Literature

Study	Location	Cost Estimates (USD/m^3)	Service	Includes	Source
Foster and Araujo [2004]	Guatemala	0.30–0.40	Water	Indicates “full cost”	Cites “international benchmarks”
Komives et al. [2005]	Multicountry	See Table 2C	Water	Varies	Not stated
Komives et al. [2006]	Multicountry	See Table 2C	Water	Varies	Not stated
Foster and Yepes [2006]	Multicountry	0.30	Water	O&M	Kingdom et al. (unpublished working paper, 2004)
Foster and Yepes [2006]	Multicountry	0.90	Water	O&M plus capital costs	Kingdom et al. (unpublished working paper, 2004)
Groom et al. [2008]	China	0.85	Water	“Full financial” cost recovery	Not stated
Walker et al. [2000]	Multicountry	0.09–0.27	Water	O&M	Not stated
Walker et al. [2000]	Multicountry	0.17–0.47	Water	Capital costs including “financing charges plus depreciation”	Not stated
Barde and Lehmann [2014]	Peru	0.64	Water	Not stated	Average tariff

Table C2. Cost Estimates From GWI [2004]a^a Source: GWI [2004] in Komives et al. [2005].

	Developing Country	Industrialized Countries
<0.20 USD/m^3	Tariff insufficient to cover basic operations and maintenance costs	Tariff insufficient to cover basic operations and maintenance costs
0.20–0.40 USD/m^3	Tariff sufficient to cover operation and some maintenance costs	Tariff insufficient to cover basic operations and maintenance costs
0.40–1.00 USD/m^3	Tariff sufficient to cover operations and maintenance costs and most investment needs	Tariff sufficient to cover basic operations and maintenance costs
>1.00 USD/m^3	Tariff sufficient to cover operations and maintenance costs and most investment needs in the face of extreme supply shortage	Tariff sufficient to cover full cost of modern water systems in most high-income cities

• ^a Source: GWI [2004] in Komives et al. [2005].

Appendix D: Survey Description and Sampling Strategy

This paper uses data collected through a survey of approximately 741 NCWSC customers conducted between November 2013 and January 2014. (The study received UNC IRB approval under Study No. 13-1932 as well as approval from Kenya's National Council for Science and Technology (Ref No. NACOSTU/P/13/8073/406.)

The survey was comprised of eight modules. The first module recorded basic logistical information about the survey, including informed consent, enumerator code, date, and start time. Importantly this module recorded each household's account number, which the enumerators recorded from the NCWSC meter readers' log books. The account number allows us to identify households in the NCWSC billing records.

The second module contained screening questions to ensure the households met the criteria for participation in the study and collected information related to the household's piped water connection. This module also contained questions related to household recall about expenditure on water and sanitation services. The third module of the survey collected information about the household composition and demographics. The fourth module contained questions related to their socioeconomic status, including questions about household expenditure, income, and asset ownership. The fifth module asked households about additional water sources they might use, including private boreholes, water vendors, and bottled water. The sixth module contained a number of questions about the extent to which and under what circumstances households treated their drinking water. The seventh module collected information related to household sanitation facilities. Finally, the eighth module contained wrap up information about the survey, including the enumerators' perceptions of the quality of the survey and the GPS coordinates of the households' location.

In general, wealthier households are more likely to have a connection to the piped water and sewer network than lower income households. Because the objective of the household survey was to investigate the relationship between income and water use among NCWSC customers, the primary challenge for the survey was to ensure the sample had adequate income heterogeneity.

To address limitations of the NCWSC administrative data, our sampling strategy combined purposive and stratified random sampling. NSWCS service area consists of six regions. Each region consists of zones, which are further subdivided into itineraries. There are 26 zones in total and approximately 2000 itineraries. Each itinerary contains between 100 and 200 accounts. NCWSC marketing associates (meter readers) use these itineraries on their daily meter reading routes.

The project team and a committee from NCWSC first assigned an income category to each of the 25 zones based on the subjective, local knowledge of the project team and NCWSC staff members. Due to the fact that the primary sampling challenge was to ensure that the sample had adequate income heterogeneity, the project team then identified the two regions with the highest representation of low-income zones (eastern and northeastern) and randomly selected one of these regions (northeastern). Similarly, we identified the two regions with the highest representation of high income zones (southern and western) and randomly selected one of these regions (western).

Once the two regions were selected, the following sampling strategy was employed. Each day the project team, in collaboration with head of billing and metering from the appropriate regional office randomly assigned enumerators to marketing associates, with two enumerators paired with a single marketing associate. Each pair of enumerators would then shadow a marketing associate on their meter reading route for that day. Starting at the beginning of the itinerary the marketing associate was reading that day, the enumerators were instructed to select the tenth customer account as the first household. The marketing associate would then introduce one of the enumerators to the household and continue on their meter reading route. The second enumerator would select the twentieth account on the list and do the same.

Once the enumerators completed an interview, they would call the marketing associate and meet them where he/she was in their current meter reading route. The enumerator would then use the next account as a sample household. If nobody was at the household, enumerators were instructed to note the address and attempt two call backs. If someone from the household was home, but did not have time to complete the survey or the head of household or their spouse was not home, enumerators were instructed to take the contact information of the head of household and attempt to schedule a call back two times before replacing the household in the sample.

The survey included a number of quality control measures. First, each day the survey supervisor would collect and review completed questionnaires from the enumerators. If the supervisor identified problems with the questionnaire, enumerators were instructed to revisit the households to verify or correct the information. Second, the project team conducted random spot checks on the enumerators to ensure they followed the prescribed sampling protocol and to ensure that they administered the survey instrument appropriately. This included spontaneously meeting enumerators in the field to observe interviews, conversations with the marketing associates about how the enumerators selected households, and visits or calls to households who had been surveyed to discuss the types of questions the enumerator asked them and how long they spent with them. We also employed double, independent data entry.

As described in the main text, our final sample consists of 656 households. Of the initial 741 households surveyed 83 households were dropped from the sample. One household was dropped due to a duplicate account number; 14 accounts could not be located in the NCWSC billing data; 30 accounts did not have an actual meter reading during the billing period we observed; and 38 accounts had no nonzero meter readings during the billing period we observed.

Appendix E: Wealth Index

We constructed a wealth index using principle component analysis following Filmer and Pritchett [2001] and Filmer and Scott [2008]. Table E1 presents the 28 variables we include in the wealth index. All variables were converted to indicator variables (0–1) or continuous variables as appropriate. The first column of Table E1 presents the first component (factor score) of the principle component analysis. The first component has an eignenvalue of 5.93 and explains 21.2% of variation in the 28 variables in our index. For binary variables, the factor score can be interpreted as the marginal change in a household's wealth index score by going from not owning the asset to owning the asset. For example, cooking with biomass decreases a household's wealth index score by 0.243. Similarly, having a security guard on premise increases a household's wealth index score by 0.201.

Table E1. Factor Scores and Mean Values of Components of the Wealth Index by Predicted Wealth Quintilea^a Notes: n = 656. Each quintile has 131 or 132 households.

	Factor Score	Lowest	Second	Third	Fourth	Highest	All
Cooks with LPG	0.234	0.39	0.87	0.99	0.98	0.98	0.84
Cooks with biomass	−0.243	0.60	0.11	0.00	0.01	0.00	0.14
Level of school completed	0.144	11.89	13.69	14.61	16.16	16.88	14.64
Own/rent	0.039	0.52	0.34	0.48	0.49	0.74	0.51
Separate kitchen in house	0.179	0.83	0.99	1.00	1.00	1.00	0.96
Security guard	0.201	0.21	0.56	0.80	0.98	0.98	0.71
Electric security fence	0.161	0.02	0.04	0.14	0.36	0.82	0.27
Electricity connection	0.147	0.90	0.99	1.00	1.00	1.00	0.98
Mobile phone	0.032	0.99	0.99	1.00	1.00	1.00	1.00
Phone with data plan	0.126	0.57	0.76	0.82	0.89	0.94	0.79
Internet connection	0.168	0.01	0.05	0.15	0.40	0.71	0.26
TV	0.126	0.89	0.98	0.99	1.00	1.00	0.97
Radio	0.104	0.70	0.82	0.89	0.84	0.83	0.81
Computer/laptop	0.264	0.17	0.42	0.68	0.85	0.98	0.62
Bicycle	0.214	0.12	0.24	0.20	0.37	0.50	0.29
Motorcycle	0.224	0.02	0.02	0.02	0.06	0.08	0.04
Car	0.277	0.08	0.31	0.56	0.85	0.98	0.56
Washing machine	0.244	0.02	0.07	0.11	0.13	0.72	0.21
Water heater	0.258	0.11	0.31	0.61	0.61	0.86	0.50
Refrigerator	0.258	0.39	0.88	0.97	0.99	0.99	0.84
Gas cooker	0.264	0.30	0.81	0.95	0.96	1.00	0.80
Meko (local stove)	0.086	0.48	0.52	0.36	0.31	0.19	0.37
Additional house in Nairobi	0.209	0.07	0.12	0.15	0.16	0.20	0.14
Additional house outside Nairobi	0.139	0.33	0.30	0.37	0.16	0.23	0.28
Land in Nairobi	0.203	0.09	0.09	0.12	0.11	0.04	0.09
Land outside Nairobi	0.091	0.48	0.51	0.56	0.34	0.18	0.42
Borehole on property	0.070	0.00	0.01	0.00	0.05	0.17	0.05
Toilet inside home	0.207	0.77	0.99	1.00	1.00	1.00	0.95

• ^a Notes: n = 656. Each quintile has 131 or 132 households.

The distribution of the wealth index scores is relatively smooth and skewed to the right (Figure E1). The verticle lines in Figure E1 indicate the cutpoints for the wealth index that divide our sample into five quintiles. Each quintile in our sample has 131 or 132 househholds.

Figure E1

Distribution of factor scores used to construct the wealth index with quintile cutpoints.

The second through sixth columns of Table E1 show the mean of each variable included in the index for households in each predicted wealth quintile. The final column in Table E1 presents the mean for each variable for the entire sample. The mean values for each variable across the wealth quintiles largely agree with our prior assumptions about the relationship between asset ownership and wealth. For example, 1% of households in the lowest wealth quintile report having an internet connection. In contrast, 71% of households in the highest wealth quintile report having an internet connection. Similarly, less than 10% of households in the lowest wealth quintile report owning a car in contrast to 98% of households in the highest wealth quintile. Exceptions to this include the extent to which households own or rent their home and own land/property in or outside of Nairobi.