WPS3941 1 Water Markets, Demand and Cost Recovery for Piped Water Supply Services: Evidence from Southwest Sri Lanka1 Céline Nauges and Caroline van den Berg2 In many countries water supply is a service that is seriously underpriced, especially for residential consumers. This has led to a call for setting cost recovery policies to ensure that the tariffs charged for water supply cover the full cost of providing the service. Yet, the question arises how consumers will react to such price increases. We illustrate the impact of price increases on consumption of piped water through a study of the demand for water of piped and non-piped households using cross-sectional data from 1,800 households in Southwest Sri Lanka. The (marginal) price elasticity is estimated at -0.74 for households exclusively relying on piped water, and at –0.69 for households using piped water but supplementing their supply with other water sources, with no significant differences between income groups. Those households that depend on non-piped water sources have a time cost elasticity (as a proxy for price elasticity) of only -0.06. We discuss the implications of these results in terms of pricing policy. World Bank Policy Research Working Paper 3941, June 2006 The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. 1 Financial support from the Bank-Netherlands Water Partnership for Water Supply and Sanitation (BNWP- WSS), a facility that enhances World Bank operations to increase delivery of water supply and sanitation services to the poor, is gratefully acknowledged. 2 Céline Nauges is a Senior Research Fellow at the French National Institute for Research in Agriculture at the University of Toulouse. Caroline van den Berg is a Senior Economist at the Energy and Water Department of the World Bank. The opinions reflected in this paper are the opinions of the authors and not opinions of the World Bank, the French National Institute for Research in Agriculture, or the BNWP-WSS. 2 1. Introduction An extensive empirical literature exists on residential water consumption in developed countries (see Hanemann 1998, Arbués et al. 2003, or Dalhuisen et al. 2003 for comprehensive surveys). Yet, few such studies exist for residential water consumption in developing countries. Most of the studies on residential water demand are mainly in the form of contingent valuation studies to derive willingness-to-pay for getting a house connection to a piped water network (North and Griffin 1997; Whittington et al. 2002; Pattanayak et al. 2006). In most developing countries the quality of residential water consumption datasets often pose a problem, especially as metering is not a very common phenomenon. Yet, the market in which utilities operate in many of these countries is also startling different. In contrast to developed countries, where almost all households obtain water from the utility through a piped network, the market for residential water demand in many developing countries shows much more variation. Households may have a connection to the piped network and use exclusively water from their private tap, but they may also combine piped water with water collected from wells, public taps, or purchase water from vendors; or they may have no connection and rely exclusively on non-piped water. Little is known about households’ behavior in developing countries regarding the factors driving their choices and in particular the substitution/complementarity relationship between piped and non-piped water for piped households or the combination of non-piped water from different sources for non-piped households. As a result, policy decisions are often not well informed and it is usually assumed that residential water demand in developing countries mimics that of developed countries. A more detailed knowledge of the structure of water demand of piped and non-piped households in developing countries can help to better understand consumer behavior. For planning purposes, it is essential to be able to predict the change in residential water demand for utility services that will result from any policy that would involve some change in tariffs and/or income for the household. As underpricing of piped water supply occurs often and makes tariff increases necessary to ensure the long-term sustainability of the service provision, understanding how customers might react to such price increases is of importance. Secondly, many households cannot expect to be connected to the piped network in the near 3 future. For these households one may want to make improvements in the non-piped water distribution system to improve access to safe water. Few studies have estimated residential water demand in developing countries. Using household survey data from 17 cities in Central America and Venezuela, Strand and Walker (2005) derive price elasticities for piped (non-piped) households equal to -0.3 and -0.1 (similar to that of many developing countries). Nauges and Strand (2005), using the same dataset, estimated water demand of non-piped households in four cities in El Salvador and Honduras, where the vast majority of the surveyed households relied on one water source only (private tap, public tap, public well, or truck). They found non-tap water demand elasticities with respect to total water cost (defined as the sum of water price and collection time costs) of between –0.4 and –0.7. Basani et al. (2004), using cross-sectional household-level data from seven provincial Cambodian towns, estimated the price elasticity of water demand of connected households to lie in a range between -0.4 and -0.5. Rietveld et al. (2000), using data from Indonesia, found a price elasticity for connected households of -1.2. The present paper contributes to this literature by providing an empirical analysis of the water demand function of piped and non-piped households from Southwest Sri Lanka. Data come from a survey of 1,800 households conducted in August-October 2003. Section 2 describes the background and data. In section 3, we discuss the specification of the water demand models and estimation strategy. Estimation results are described in section 4, while policy implications and conclusions are found in section 5. 2. Background and data The population of surveyed households covered three districts in Southwest Sri Lanka: Gampaha, Kalutara and Galle. The survey was undertaken to support the design of two private sector transactions in this part of Sri Lanka which the then Government of Sri Lanka (GoSL) was proposing: one for the town of Negombo, north of Colombo, and one stretching along a coastal strip south of Colombo, from the town of Kalutara to the town of Galle. The population in these two service areas in 2001 was slightly more than 1.6 million3. 3 The total population considered in the Greater Negombo service area was about 367,000, while the service area covered by the coastal strip from Kalutara to Galle had a total population of 1,254,000 in 2001. 4 The survey data are rather unique. Because of widespread metering of households with a piped water connection, the consumption data have a high degree of accuracy that is not often found. In addition, the dataset is complemented by a large set of socio-economic and health variables. 2.1. Piped households Among the surveyed households, 38 percent had a private connection to the piped network (for further purposes of the study, we removed from the sample the 84 households who did not report any monthly water use). Of the households with a private connection 23 percent had an in-house private connection, 19 percent had (only) a yard connection, and 58 percent had both. Piped households consume on average about 135 liters of water per capita per day from the piped network. Piped households had to pay SLK 8,415 (equivalent to US$87 in 2003) in order to get a private connection to the piped network (including road cutting, pipe laying, meter installation). This represents about half of the monthly wage for a piped household. Water from the piped network is charged through an increasing five-block tariff. The same tariff applies to all piped households in our sample. Marginal price varies from SLK 1.25 per cubic meter in the lowest block (for any unit below 10 cubic meters per month) to SLK 45 per cubic meter for any unit above 25 cubic meters per month. Households are almost equally distributed across the five blocks.4 The water bill, which includes a fixed fee of SLK 50, is sent every month to each household connected to the piped network. The typical or median monthly water bill is SLK 89, while a typical household spends SLK 10,300 on household expenses – suggesting that the costs of piped water supply makes up less than 1 percent of household expenditure. The typical water bill for the poor (defined as a piped household with an income falling in the first quartile of the income distribution) is SLK 72 (which represents about 1 percent of household expenditure). Piped households have been asked to give their opinion about the quality of the piped water service. Overall, 25 percent of households with piped water declared themselves to be 4 Block 1: [1-10 m3], price is 1.25 SLK/m3; block 2: [11-15 m3], 2.50 SLK/m3; block 3: [16-20 m3], 6.50 SLK/m3; block 4: [21-25 m3], 20.00 SLK/m3; block 5: [>25 m3], 45.00 SLK/m3 . 5 satisfied with the service. The most frequent complaint is about piped water being available less than 24 hours a day (41 percent of households), followed by complaints about frequent breakdowns (9 percent of households), too high a monthly bill (5 percent of households), and poor water quality (3 percent of households). Piped water availability varies across households. In the rainy (dry) season, 31 percent (22 percent) of piped households have a 24 hour service of piped water; 36 percent (42 percent) have access to piped water for 12 hours or less; 10 percent (13 percent) for 6 hours or less. Non-continuous piped water service may be one of the reasons why some piped households get water from other (i.e., non-piped) sources in the neighborhood. Among the piped households, almost 95 percent have access to other water sources, namely public taps (112 households), public wells (172), private wells (352), neighbors (492), vendors (31), rainwater (93), surface water (76) or bottled water (396). As can be seen from this list, many piped households have access to more than one additional source. Despite the widespread access to other sources, only 40 percent of piped households use water from other sources, mostly from private wells. Piped households (who also get water from other sources) consume on average 10 cubic meters per month from these sources (Table 1). Consuming non-piped water imposes different types of “costs” on a household, when compared to using the water directly from their private tap. First, the household may spend time to go to the source and wait at the source to obtain the water. Secondly, water from most non-piped water sources in general involves collection costs (the household may need to buy equipment to abstract the water such as a hand pump or an electric pump). Thirdly, the household may need to pay a fee to get access to the water, in particular if bought from vendors or community sources. Finally, there is the inconvenience of not having access to piped water as such, including a possible lower quality of the non-piped water. In our sample, walking time for piped households who collect water from a private well or from community sources is on average less than 5 minutes, whatever the source (Table 1). The shortest walking time is observed, as expected, for those households who get water from a private well. Only households collecting water from public taps have to wait at the source (7 to 8 minutes on average). Public wells are all of the “dug well” type; private wells are too, although in a small number of cases (12 percent) they are of the “tubewell” variety. Households who collect water from wells have to buy equipment to collect water. The most 6 common equipment is a bucket and rope (as expected from the prevalence of dug wells) followed by hand and electric pumps. Households relying on public (private) well spend on average respectively SLK 2,600 and SLK 13,600 to buy the necessary equipment. Operating costs for households collecting water from public (private) wells represent on average respectively SLK 10 and SLK 34 per month. Households in our sample do not pay any fee for buying non-piped water, whatever the source, but they do pay for installing equipment to obtain access to the source of water. Households were also surveyed regarding water treatment and hygiene practices. Overall, 45 percent of the piped households declared to treat or filter water before drinking it (see Nauges and van den Berg 2006, for a detailed analysis of risk perception and hygiene practices). 2.2. Non-piped households About 62 percent of the households in the sample do not have any piped connection. Among them, 98 get water from public taps, 102 from public wells, 313 from their neighbors, 967 from private wells, 11 from vendors, 29 from surface water, 8 collect rainwater, and 8 buy bottled water. Some households combine water from different sources. The most frequent combination of water sources among the surveyed households is neighbors with private well, public tap with private well, and public well with private well. Households relying on private wells consume on average 759 litres per day, more than twice the amount of water collected on average from community sources: public wells (367 litres), neighbors (243 litres), and public taps (119 litres). The average one-way walking time to go to the source varies between 1 (for accessing a private well) and 6 minutes (public wells). Waiting time at the source varies from none (private well) to 24 minutes (public taps). The cost of installing (operating) equipment to collect water from wells is SLK 6,600 (or SLK 36 per month) on average for households using public wells and SLK 15,400 (or SLK 67 per month) for households using private wells (Table 2). Most public wells (91 percent) are of the dug well type. A vast majority of households relying on public wells (86 percent) collect water using a bucket and rope, 7 percent uses a hand pump, and 4 percent an electric pump. The picture is different for households relying on private wells. A smaller percentage – albeit still the large majority – of private wells (76 7 percent) is of the “dug well” variety. Most households with a private well use pumps: 47 percent an electric pump, 10 percent a handpump and the remainder buckets and ropes to abstract the water. Overall, non-piped households are satisfied with the non-piped water. More than 80 percent of households collecting water from public taps, neighbors, and private wells judge the taste of water as excellent or good (in the rainy season). The percentage of households satisfied with the taste of water is slightly lower among households relying on public wells (52 percent). As far as safety of the water is concerned, 90 percent of all households relying on public taps, neighbors, and private wells think that there is no risk or little risk in drinking the water. This percentage is again lower for households collecting water from public wells (60 percent). Households’ confidence about non-piped water safety is confirmed by the fact that only 40 percent of non-piped households treat or filter their water before drinking it (this percentage is higher among the group of piped households). There is no significant difference across sources. Descriptive statistics on household demographics and socioeconomics, and water treatment are presented in Table 3, for both non-piped and piped households. Mean comparison tests show that piped households in general are characterized by having more household members, higher income, and higher education than non-piped households. 3. Specification and estimation procedure We estimate separate water demand models, one for piped water and the other one for non- piped water, as quality of the water from the piped network may differ from quality of water collected from a private well or from community sources. Consistency of estimation techniques relies on the randomness of the samples considered. Yet, because it is likely that the households’ characteristics for the two groups are different, we have to control for selection bias by first estimating a model that explains the differences between households that have or do not have a connection to the piped network. 8 3.1. Determinants of the connection status The discrete choice model is specified as follows: the discrete variable (di) takes the value of 1 if the household has a private connection to the piped network and 0 otherwise. We assume that this decision is the outcome of a latent model of indirect utility maximisation by the household. Under the assumption of normality of the error term u1, the decision model takes the form of a Probit model. (1) Pr(di = 1) = Pr( x1i β1 p u1 ) = Φ ( x1i β1 ) where x1i is the vector of explanatory variables in the latent model, and β1 is the vector of associated parameters. Getting a private connection may be quite expensive for some households (about LKR 8,500), so we would expect low-income households to less likely have a connection. Also, we would expect that households that have easy access to other sources and in particular households owning a private well are less willing to pay for a private connection to the piped network. We also control for the role of household demographics and socioeconomics (household size, income, education) and opinions about water quality. Note that we use contemporaneous variables to explain a decision which could have been taken years before – we do not know when piped households got the private connection to the water network. We thus need to assume that the explanatory variables that we use have not changed “too much” after getting the connection. This is likely to be the case for variables such as education of the head, and access to other sources. This may not be true for income and opinion about water taste and safety. We will compute Mill’s ratio from the estimated parameters. This ratio will be added to the water demand models to control for selection bias (Heckman, 1979): (2) ( M i = φ x1i β1 ) ( ) ⎡1 − Φ x β ⎤ , ⎣ 1i 1 ⎦ 9 where φ (.) is the standard normal probability density function and Φ (.) is the cumulative of the normal distribution. 3.2. Water demand of piped households The water demand function of the representative household connected to the piped network is traditionally specified as a single equation of the form: (3) W P = f P (P P , I , Z ) describes the relationship between piped water consumption (WP), the price of piped water (PP), household income (I), and a vector of household characteristics (Z) to control for heterogeneity of preferences and outside variables affecting water demand.5 This approach, which provides a satisfactory description of the behavior of piped households collecting water from their private tap only, does not allow to measure substitutability/complementarity relationship between piped and non-piped water for those households who combine water from the piped network with water from other sources such as private or public wells, vendors, or get it from their neighbors. In the latter case, a simultaneous two-equation model is better suited. A two-equation model also allows one to consider piped and non-piped water as two different goods, with different organoleptic (smell, taste, colour) and sanitary properties. For households combining piped water with non-piped water, we thus specify the model as follows: ⎧ ⎪W = f ( P , P , I , Z ) P P P NP (4) ⎨ NP ⎩W = f ( P , P , I , Z ) ⎪ NP NP P where WNP and PNP represent non-piped water consumption and non-piped water price, respectively. In our sample, about 40 percent of piped households combine water from the piped network with non-piped water, the latter being essentially a private well. We estimate separate water demand models: a single-equation model (see equation (3)) for piped households using piped 5 The Mill’s ratio will be added to the list of regressors to control for potential selection bias. 10 water only, and a two-equation model (see model (4)) for piped households combining water from the piped network with non-piped water. Explanatory variables in water demand models commonly include water price, income, and household demographic and socioeconomic characteristics. Some discussion is needed here regarding the specification of the price variable for piped and non-piped water. For all households in our sample water from the piped network is sold under the same five- block increasing tariff, and all piped households have to pay a fixed fee of SLK 50, whatever their monthly consumption. Homogeneous pricing in our sample makes it impossible to estimate water demand using the (consistent) two-step approach describing the choice of the block (first step) and the choice of consumption inside the block (second step), see Hewitt and Hanemann (1995). We estimate a linear demand equation in which the price variable is instrumented to control for endogeneity. The specification of the price variable in the case of non-linear block pricing has been extensively debated during the last 30 years (see Espey et al. 1997, Arbués et al. 2003, and Dalhuisen et al. 2003, for related discussions). If theory advocates the use of marginal price (the price of the last cubic meter), average price (computed as total bill divided by total consumption) has however often been preferred. Authors considering average price argue that households are rarely well informed on the price structure and are thus more likely to react to average price than to marginal price. In the present study, one could argue that average price should be chosen because the water tariff structure is quite complex (it is made of five different blocks, and the fixed fee makes up a large part of the total cost especially for lower-volume users) and so households are less likely to know in which block they are and which marginal price will be charged to them. However, it is also very well possible that households know the marginal price because the price in each block varies significantly (from SLK 1.25 per cubic meter in the low block (for any unit below 10 cubic meters per month) to SLK 45 per cubic meter for any unit above 25 cubic meters per month), and widespread occurrence of metering assumes that households have (some) control over their consumption.6 6 Distribution graphs of households inside each of the five blocks show that households in the first four blocks tend to choose the “right-end” of the block, while households in the fifth block are gathering around the “left- 11 We test which price households are sensitive to using Shin’s price perception test (1985). Shin proposed to introduce in the demand model the following variable: k ⎛ AP ⎞ (5) MPi ⎜ i ⎟ ⎝ MPi ⎠ where MP and AP stand for marginal price and average price, respectively, and k is called the perception parameter. If the consumer responds only to marginal price, then k=0 , and if the consumer responds only to the average price, then k=1. If the consumer’s perceived price lies between the average price and the marginal price, then 0